This document presents a machine learning technique using supervised deep learning embeddings to predict cervical cancer diagnosis. The technique allows for joint optimization of dimensionality reduction and classification models in a fully supervised manner. It achieves accurate prediction results (top AUC of 0.6875) outperforming previous methods like denoising autoencoders. Visualization of the embedding spaces revealed clinical findings that were validated in medical literature, making the results useful for physicians. The technique is instantiated using deep variational autoencoders and feed-forward neural networks on a dataset of 858 medical records related to cervical cancer screening.
Breast cancer detection from histopathological images is done using deep learning and transfer learning techniques. Image processing is done for better accuracy. CNN and DenseNet-121 algorithms are used. 90.9 % accuracy is achieved using CNN and 88% accuracy is achieved using Transfer learning.
Multiclass skin lesion classification with CNN and Transfer LearningYashIyengar
A multiclass classifier has been developed using transfer learning approach to detect seven different types of skin lesions. A modified VGG19 architecture is used for this purpose. The proposed model has achieved 78 % validation accuracy and 74 % test accuracy.
FRACTAL PARAMETERS OF TUMOUR MICROSCOPIC IMAGES AS PROGNOSTIC INDICATORS OF C...csandit
Research in the field of breast cancer outcome prognosis has been focused on molecular biomarkers, while neglecting the discovery of novel tumour histology structural clues. We thus
aimed to improve breast cancer prognosis by fractal analysis of tumour histomorphology. This study included 92 breast cancer patients without systemic treatment. Fractal parametersfractal dimension and lacunarity of the breast tumour microscopic histology possess prognostic value comparable to the major clinicopathological prognostic parameters. Fractal analysis was performed for the first time on routinely produced archived pan-tissue stained primary breast tumour sections, indicating its potential for clinical use as a simple and cost-effective prognostic indicator of distant metastasis risk to complement the molecular approaches for
cancer risk prognosis.
A novel approach to jointly address localization and classification of breast...IJECEIAES
Localization of the cancerous region as well as classification of the type of the cancer is highly inter-linked with each other. However, investigation towards existing approaches depicts that these problems are always iindividually solved where there is still a big research gap for a generalized solution towards addressing both the problems. Therefore, the proposed manuscript presents a simple, novel, and less-iterative computational model that jointly address the localization-classification problems taking the case study of early diagnosis of breast cancer. The proposed study harnesses the potential of simple bio-inspired optimization technique in order to obtained better local and global best outcome to confirm the accuracy of the outcome. The study outcome of the proposed system exhibits that proposed system offers higher accuracy and lower response time in contrast with other existing classifiers that are freqently witnessed in existing approaches of classification in medical image process.
BRAIN TUMOR MRIIMAGE CLASSIFICATION WITH FEATURE SELECTION AND EXTRACTION USI...ijistjournal
Feature extraction is a method of capturing visual content of an image. The feature extraction is the process to represent raw image in its reduced form to facilitate decision making such as pattern classification. We have tried to address the problem of classification MRI brain images by creating a robust and more accurate classifier which can act as an expert assistant to medical practitioners. The objective of this paper is to present a novel method of feature selection and extraction. This approach combines the Intensity, Texture, shape based features and classifies the tumor as white matter, Gray matter, CSF, abnormal and normal area. The experiment is performed on 140 tumor contained brain MR images from the Internet Brain Segmentation Repository. The proposed technique has been carried out over a larger database as compare to any previous work and is more robust and effective. PCA and Linear Discriminant Analysis (LDA) were applied on the training sets. The Support Vector Machine (SVM) classifier served as a comparison of nonlinear techniques Vs linear ones. PCA and LDA methods are used to reduce the number of features used. The feature selection using the proposed technique is more beneficial as it analyses the data according to grouping class variable and gives reduced feature set with high classification accuracy.
Breast cancer detection from histopathological images is done using deep learning and transfer learning techniques. Image processing is done for better accuracy. CNN and DenseNet-121 algorithms are used. 90.9 % accuracy is achieved using CNN and 88% accuracy is achieved using Transfer learning.
Multiclass skin lesion classification with CNN and Transfer LearningYashIyengar
A multiclass classifier has been developed using transfer learning approach to detect seven different types of skin lesions. A modified VGG19 architecture is used for this purpose. The proposed model has achieved 78 % validation accuracy and 74 % test accuracy.
FRACTAL PARAMETERS OF TUMOUR MICROSCOPIC IMAGES AS PROGNOSTIC INDICATORS OF C...csandit
Research in the field of breast cancer outcome prognosis has been focused on molecular biomarkers, while neglecting the discovery of novel tumour histology structural clues. We thus
aimed to improve breast cancer prognosis by fractal analysis of tumour histomorphology. This study included 92 breast cancer patients without systemic treatment. Fractal parametersfractal dimension and lacunarity of the breast tumour microscopic histology possess prognostic value comparable to the major clinicopathological prognostic parameters. Fractal analysis was performed for the first time on routinely produced archived pan-tissue stained primary breast tumour sections, indicating its potential for clinical use as a simple and cost-effective prognostic indicator of distant metastasis risk to complement the molecular approaches for
cancer risk prognosis.
A novel approach to jointly address localization and classification of breast...IJECEIAES
Localization of the cancerous region as well as classification of the type of the cancer is highly inter-linked with each other. However, investigation towards existing approaches depicts that these problems are always iindividually solved where there is still a big research gap for a generalized solution towards addressing both the problems. Therefore, the proposed manuscript presents a simple, novel, and less-iterative computational model that jointly address the localization-classification problems taking the case study of early diagnosis of breast cancer. The proposed study harnesses the potential of simple bio-inspired optimization technique in order to obtained better local and global best outcome to confirm the accuracy of the outcome. The study outcome of the proposed system exhibits that proposed system offers higher accuracy and lower response time in contrast with other existing classifiers that are freqently witnessed in existing approaches of classification in medical image process.
BRAIN TUMOR MRIIMAGE CLASSIFICATION WITH FEATURE SELECTION AND EXTRACTION USI...ijistjournal
Feature extraction is a method of capturing visual content of an image. The feature extraction is the process to represent raw image in its reduced form to facilitate decision making such as pattern classification. We have tried to address the problem of classification MRI brain images by creating a robust and more accurate classifier which can act as an expert assistant to medical practitioners. The objective of this paper is to present a novel method of feature selection and extraction. This approach combines the Intensity, Texture, shape based features and classifies the tumor as white matter, Gray matter, CSF, abnormal and normal area. The experiment is performed on 140 tumor contained brain MR images from the Internet Brain Segmentation Repository. The proposed technique has been carried out over a larger database as compare to any previous work and is more robust and effective. PCA and Linear Discriminant Analysis (LDA) were applied on the training sets. The Support Vector Machine (SVM) classifier served as a comparison of nonlinear techniques Vs linear ones. PCA and LDA methods are used to reduce the number of features used. The feature selection using the proposed technique is more beneficial as it analyses the data according to grouping class variable and gives reduced feature set with high classification accuracy.
SVM &GA-CLUSTERING BASED FEATURE SELECTION APPROACH FOR BREAST CANCER DETECTIONijscai
Mortality leading among women in developed countries is breast cancer. Breast cancer is women's second most prominent cause of cancer mortality worldwide. In recent decades, women's high prevalence of breast cancer has risen dramatically. This paper discussed several data analysis methods used to detect breast cancer early. Breast cancer diagnosis distinguishes benign and malignant breast lumps. Using data processing tools, we tackled this disease analysis. Data mining is an important step of library discovery where intelligent methods are used to detect patterns. Several clinical breast cancer studies were conducted using soft computing and machine learning techniques. Sometimes their algorithms are easier, easier, or more comprehensive than others. This research is focused on genetic programming and machine learning algorithms to reliably identify benign and malignant breast cancer. This study aimed to optimise the testing algorithm. We used genetic programming methods to choose classification machines' best features and parameter values. Data mining is an important step of library discovery where intelligent methods are used to detect patterns. We are analysing data accessible from the U.C.I. deep-learning data set in Wisconsin. In this experiment, we equate four Weka clustering strategies with genetic clustering. A comparison of results reveals that sequential minimal optimization (S.M.O.) is better than I.B.K. and B.F. Tree processes, i.e. 97.71%.
Breast cancer diagnosis via data mining performance analysis of seven differe...cseij
According to World Health Organization (WHO), breast cancer is the top cancer in women both in the
developed and the developing world. Increased life expectancy, urbanization and adoption of western
lifestyles trigger the occurrence of breast cancer in the developing world. Most cancer events are
diagnosed in the late phases of the illness and so, early detection in order to improve breast cancer
outcome and survival is very crucial.
In this study, it is intended to contribute to the early diagnosis of breast cancer. An analysis on breast
cancer diagnoses for the patients is given. For the purpose, first of all, data about the patients whose
cancers’ have already been diagnosed is gathered and they are arranged, and then whether the other
patients are in trouble with breast cancer is tried to be predicted under cover of those data. Predictions of
the other patients are realized through seven different algorithms and the accuracies of those have been
given. The data about the patients have been taken from UCI Machine Learning Repository thanks to Dr.
William H. Wolberg from the University of Wisconsin Hospitals, Madison. During the prediction process,
RapidMiner 5.0 data mining tool is used to apply data mining with the desired algorithms.
3D Segmentation of Brain Tumor ImagingIJAEMSJORNAL
A brain tumor is a collection of anomalous cells that grow in or around the brain. Brain tumors affect the humans badly, it can disrupt proper brain function and be life-threatening. In this project, we have proposed a system to detect, segment, and classify the tumors present in the brain. Once the brain tumor is identified at the very beginning, proper treatments can be done and it may be cured.
MEDICAL IMAGING MUTIFRACTAL ANALYSIS IN PREDICTION OF EFFICIENCY OF CANCER TH...csandit
Based on pressing need for predictive performance improvement, we explored the value of pretherapy
tumour histology image analysis to predict chemotherapy response. It was shown that
multifractal analysis of breast tumour tissue prior to chemotherapy indeed has the capacity to
distinguish between histological images of the different chemotherapy responder groups with
accuracies of 91.4% for pPR, 82.9% for pCR and 82.1% for PD/SD.
Pneumonia Classification using Transfer LearningTushar Dalvi
Pneumonia can be life-threatening for people with weak immune systems, in which the alveoli filled with fluid that makes it hard to pass oxygen throughout the bloodstream. Detecting pneumonia is from a chest X-ray is not only expansive but also time-consuming for normal people. Throughout this research introduced a machine learning technique to classify pneumonia from Chest X-ray Images. Most of the medical datasets having class imbalance issues in the dataset. The Data augmentation technique used to reduce the class imbalance from the dataset, Horizontal Flip, width shift and height shift techniques used to complete the augmentation technique. Used VGG19 as a base architecture and ImageNet weights added for the transfer learning approach, also Removing initial layers and adding
some more dense layers helped to discover new possibilities. After testing the proposed model on testing data, we are able to achieve 98% recall and 82% of precision. As compare with state of the art technique, the proposed method able to achieve high
recall but that compromises with Precision.
Developed Project with 3 more colleagues for Pneumonia Detection from Chest X-ray images using Convolutional Neural Network. Used confusion matrix, Recall, Precision for check the model performance on testing Data
Twin support vector machine using kernel function for colorectal cancer detec...journalBEEI
Nowadays, machine learning technology is needed in the medical field. therefore, this research is useful for solving problems in the medical field by using machine learning. Many cases of colorectal cancer are diagnosed late. When colorectal cancer is detected, the cancer is usually well developed. Machine learning is an approach that is part of artificial intelligence and can detect colorectal cancer early. This study discusses colorectal cancer detection using twin support vector machine (SVM) method and kernel function i.e. linear kernels, polynomial kernels, RBF kernels, and gaussian kernels. By comparing the accuracy and running time, then we will know which method is better in classifying the colorectal cancer dataset that we get from Al-Islam Hospital, Bandung, Indonesia. The results showed that polynomial kernels has better accuracy and running time. It can be seen with a maximum accuracy of twin SVM using polynomial kernels 86% and 0.502 seconds running time.
SVM &GA-CLUSTERING BASED FEATURE SELECTION APPROACH FOR BREAST CANCER DETECTIONijscai
Mortality leading among women in developed countries is breast cancer. Breast cancer is women's second most prominent cause of cancer mortality worldwide. In recent decades, women's high prevalence of breast cancer has risen dramatically. This paper discussed several data analysis methods used to detect breast cancer early. Breast cancer diagnosis distinguishes benign and malignant breast lumps. Using data processing tools, we tackled this disease analysis. Data mining is an important step of library discovery where intelligent methods are used to detect patterns. Several clinical breast cancer studies were conducted using soft computing and machine learning techniques. Sometimes their algorithms are easier, easier, or more comprehensive than others. This research is focused on genetic programming and machine
learning algorithms to reliably identify benign and malignant breast cancer. This study aimed to optimise the testing algorithm. We used genetic programming methods to choose classification machines' best features and parameter values. Data mining is an important step of library discovery where intelligent methods are used to detect patterns. We are analysing data accessible from the U.C.I. deep-learning data
set in Wisconsin. In this experiment, we equate four Weka clustering strategies with genetic clustering. A comparison of results reveals that sequential minimal optimization (S.M.O.) is better than I.B.K. and B.F. Tree processes, i.e. 97.71%.
USING DATA MINING TECHNIQUES FOR DIAGNOSIS AND PROGNOSIS OF CANCER DISEASEIJCSEIT Journal
Breast cancer is one of the leading cancers for women in developed countries including India. It is the
second most common cause of cancer death in women. The high incidence of breast cancer in women has
increased significantly in the last years. In this paper we have discussed various data mining approaches
that have been utilized for breast cancer diagnosis and prognosis. Breast Cancer Diagnosis is
distinguishing of benign from malignant breast lumps and Breast Cancer Prognosis predicts when Breast
Cancer is to recur in patients that have had their cancers excised. This study paper summarizes various
review and technical articles on breast cancer diagnosis and prognosis also we focus on current research
being carried out using the data mining techniques to enhance the breast cancer diagnosis and prognosis.
GRAPHICAL MODEL AND CLUSTERINGREGRESSION BASED METHODS FOR CAUSAL INTERACTION...ijaia
The early detection of Breast Cancer, the deadly disease that mostly affects women is extremely complex because it requires various features of the cell type. Therefore, the efficient approach to diagnosing Breast Cancer at the early stage was to apply artificial intelligence where machines are simulated with intelligence and programmed to think and act like a human. This allows machines to passively learn and find a pattern, which can be used later to detect any new changes that may occur. In general, machine learning is quite useful particularly in the medical field, which depends on complex genomic measurements such as microarray technique and would increase the accuracy and precision of results. With this technology, doctors can easily diagnose patients with cancer quickly and apply the proper treatment in a timely manner. Therefore, the goal of this paper is to address and propose a robust Breast Cancer diagnostic system using complex genomic analysis via microarray technology. The system will combine two machine learning methods, K-means cluster, and linear regression.
An approach of cervical cancer diagnosis using class weighting and oversampli...TELKOMNIKA JOURNAL
Globally, cervical cancer caused 604,127 new cases and 341,831 deaths in 2020, according to the global cancer observatory. In addition, the number of cervical cancer patients who have no symptoms has grown recently. Therefore, giving patients early notice of the possibility of cervical cancer is a useful task since it would enable them to have a clear understanding of their health state. The use of artificial intelligence (AI), particularly in machine learning, in this work is continually uncovering cervical cancer. With the help of a logit model and a new deep learning technique, we hope to identify cervical cancer using patient-provided data. For better outcomes, we employ Keras deep learning and its technique, which includes class weighting and oversampling. In comparison to the actual diagnostic result, the experimental result with model accuracy is 94.18%, and it also demonstrates a successful logit model cervical cancer prediction.
Early detection of breast cancer using mammography images and software engine...TELKOMNIKA JOURNAL
The breast cancer has affected a wide region of women as a particular case. Therefore, different researchers have focused on the early detection of this disease to overcome it in efficient way. In this paper, an early breast cancer detection system has been proposed based on mammography images. The proposed system adopts deep-learning technique to increase the accuracy of detection. The convolutional neural network (CNN) model is considered for preparing the datasets of training and test. It is important to note that the software engineering process model has been adopted in constructing the proposed algorithm. This is to increase the reliably, flexibility and extendibility of the system. The user interfaces of the system are designed as a website used at country side general purpose (GP) health centers for early detection to the disease under lacking in specialist medical staff. The obtained results show the efficiency of the proposed system in terms of accuracy up to more than 90% and decrease the efforts of medical staff as well as helping the patients. As a conclusion, the proposed system can help patients by early detecting the breast cancer at far places from hospital and referring them to nearest specialist center.
Breast cancer is the leading cause of death for women worldwide. Cancer can be discovered early, lowering the rate of death. Machine learning techniques are a hot field of research, and they have been shown to be helpful in cancer prediction and early detection. The primary purpose of this research is to identify which machine learning algorithms are the most successful in predicting and diagnosing breast cancer, according to five criteria: specificity, sensitivity, precision, accuracy, and F1 score. The project is finished in the Anaconda environment, which uses Python's NumPy and SciPy numerical and scientific libraries as well as matplotlib and Pandas. In this study, the Wisconsin diagnostic breast cancer dataset was used to evaluate eleven machine learning classifiers: decision tree, quadratic discriminant analysis, AdaBoost, Bagging meta estimator, Extra randomized trees, Gaussian process classifier, Ridge, Gaussian nave Bayes, k-Nearest neighbors, multilayer perceptron, and support vector classifier. During performance analysis, extremely randomized trees outperformed all other classifiers with an F1-score of 96.77% after data collection and data analysis.
Breast Cancer Detection from Mammography Images Using Machine Learning Algorithms (U-Net Segmentation and Dense Net Classifier implementation are in progress)
SVM &GA-CLUSTERING BASED FEATURE SELECTION APPROACH FOR BREAST CANCER DETECTIONijscai
Mortality leading among women in developed countries is breast cancer. Breast cancer is women's second most prominent cause of cancer mortality worldwide. In recent decades, women's high prevalence of breast cancer has risen dramatically. This paper discussed several data analysis methods used to detect breast cancer early. Breast cancer diagnosis distinguishes benign and malignant breast lumps. Using data processing tools, we tackled this disease analysis. Data mining is an important step of library discovery where intelligent methods are used to detect patterns. Several clinical breast cancer studies were conducted using soft computing and machine learning techniques. Sometimes their algorithms are easier, easier, or more comprehensive than others. This research is focused on genetic programming and machine learning algorithms to reliably identify benign and malignant breast cancer. This study aimed to optimise the testing algorithm. We used genetic programming methods to choose classification machines' best features and parameter values. Data mining is an important step of library discovery where intelligent methods are used to detect patterns. We are analysing data accessible from the U.C.I. deep-learning data set in Wisconsin. In this experiment, we equate four Weka clustering strategies with genetic clustering. A comparison of results reveals that sequential minimal optimization (S.M.O.) is better than I.B.K. and B.F. Tree processes, i.e. 97.71%.
Breast cancer diagnosis via data mining performance analysis of seven differe...cseij
According to World Health Organization (WHO), breast cancer is the top cancer in women both in the
developed and the developing world. Increased life expectancy, urbanization and adoption of western
lifestyles trigger the occurrence of breast cancer in the developing world. Most cancer events are
diagnosed in the late phases of the illness and so, early detection in order to improve breast cancer
outcome and survival is very crucial.
In this study, it is intended to contribute to the early diagnosis of breast cancer. An analysis on breast
cancer diagnoses for the patients is given. For the purpose, first of all, data about the patients whose
cancers’ have already been diagnosed is gathered and they are arranged, and then whether the other
patients are in trouble with breast cancer is tried to be predicted under cover of those data. Predictions of
the other patients are realized through seven different algorithms and the accuracies of those have been
given. The data about the patients have been taken from UCI Machine Learning Repository thanks to Dr.
William H. Wolberg from the University of Wisconsin Hospitals, Madison. During the prediction process,
RapidMiner 5.0 data mining tool is used to apply data mining with the desired algorithms.
3D Segmentation of Brain Tumor ImagingIJAEMSJORNAL
A brain tumor is a collection of anomalous cells that grow in or around the brain. Brain tumors affect the humans badly, it can disrupt proper brain function and be life-threatening. In this project, we have proposed a system to detect, segment, and classify the tumors present in the brain. Once the brain tumor is identified at the very beginning, proper treatments can be done and it may be cured.
MEDICAL IMAGING MUTIFRACTAL ANALYSIS IN PREDICTION OF EFFICIENCY OF CANCER TH...csandit
Based on pressing need for predictive performance improvement, we explored the value of pretherapy
tumour histology image analysis to predict chemotherapy response. It was shown that
multifractal analysis of breast tumour tissue prior to chemotherapy indeed has the capacity to
distinguish between histological images of the different chemotherapy responder groups with
accuracies of 91.4% for pPR, 82.9% for pCR and 82.1% for PD/SD.
Pneumonia Classification using Transfer LearningTushar Dalvi
Pneumonia can be life-threatening for people with weak immune systems, in which the alveoli filled with fluid that makes it hard to pass oxygen throughout the bloodstream. Detecting pneumonia is from a chest X-ray is not only expansive but also time-consuming for normal people. Throughout this research introduced a machine learning technique to classify pneumonia from Chest X-ray Images. Most of the medical datasets having class imbalance issues in the dataset. The Data augmentation technique used to reduce the class imbalance from the dataset, Horizontal Flip, width shift and height shift techniques used to complete the augmentation technique. Used VGG19 as a base architecture and ImageNet weights added for the transfer learning approach, also Removing initial layers and adding
some more dense layers helped to discover new possibilities. After testing the proposed model on testing data, we are able to achieve 98% recall and 82% of precision. As compare with state of the art technique, the proposed method able to achieve high
recall but that compromises with Precision.
Developed Project with 3 more colleagues for Pneumonia Detection from Chest X-ray images using Convolutional Neural Network. Used confusion matrix, Recall, Precision for check the model performance on testing Data
Twin support vector machine using kernel function for colorectal cancer detec...journalBEEI
Nowadays, machine learning technology is needed in the medical field. therefore, this research is useful for solving problems in the medical field by using machine learning. Many cases of colorectal cancer are diagnosed late. When colorectal cancer is detected, the cancer is usually well developed. Machine learning is an approach that is part of artificial intelligence and can detect colorectal cancer early. This study discusses colorectal cancer detection using twin support vector machine (SVM) method and kernel function i.e. linear kernels, polynomial kernels, RBF kernels, and gaussian kernels. By comparing the accuracy and running time, then we will know which method is better in classifying the colorectal cancer dataset that we get from Al-Islam Hospital, Bandung, Indonesia. The results showed that polynomial kernels has better accuracy and running time. It can be seen with a maximum accuracy of twin SVM using polynomial kernels 86% and 0.502 seconds running time.
SVM &GA-CLUSTERING BASED FEATURE SELECTION APPROACH FOR BREAST CANCER DETECTIONijscai
Mortality leading among women in developed countries is breast cancer. Breast cancer is women's second most prominent cause of cancer mortality worldwide. In recent decades, women's high prevalence of breast cancer has risen dramatically. This paper discussed several data analysis methods used to detect breast cancer early. Breast cancer diagnosis distinguishes benign and malignant breast lumps. Using data processing tools, we tackled this disease analysis. Data mining is an important step of library discovery where intelligent methods are used to detect patterns. Several clinical breast cancer studies were conducted using soft computing and machine learning techniques. Sometimes their algorithms are easier, easier, or more comprehensive than others. This research is focused on genetic programming and machine
learning algorithms to reliably identify benign and malignant breast cancer. This study aimed to optimise the testing algorithm. We used genetic programming methods to choose classification machines' best features and parameter values. Data mining is an important step of library discovery where intelligent methods are used to detect patterns. We are analysing data accessible from the U.C.I. deep-learning data
set in Wisconsin. In this experiment, we equate four Weka clustering strategies with genetic clustering. A comparison of results reveals that sequential minimal optimization (S.M.O.) is better than I.B.K. and B.F. Tree processes, i.e. 97.71%.
USING DATA MINING TECHNIQUES FOR DIAGNOSIS AND PROGNOSIS OF CANCER DISEASEIJCSEIT Journal
Breast cancer is one of the leading cancers for women in developed countries including India. It is the
second most common cause of cancer death in women. The high incidence of breast cancer in women has
increased significantly in the last years. In this paper we have discussed various data mining approaches
that have been utilized for breast cancer diagnosis and prognosis. Breast Cancer Diagnosis is
distinguishing of benign from malignant breast lumps and Breast Cancer Prognosis predicts when Breast
Cancer is to recur in patients that have had their cancers excised. This study paper summarizes various
review and technical articles on breast cancer diagnosis and prognosis also we focus on current research
being carried out using the data mining techniques to enhance the breast cancer diagnosis and prognosis.
GRAPHICAL MODEL AND CLUSTERINGREGRESSION BASED METHODS FOR CAUSAL INTERACTION...ijaia
The early detection of Breast Cancer, the deadly disease that mostly affects women is extremely complex because it requires various features of the cell type. Therefore, the efficient approach to diagnosing Breast Cancer at the early stage was to apply artificial intelligence where machines are simulated with intelligence and programmed to think and act like a human. This allows machines to passively learn and find a pattern, which can be used later to detect any new changes that may occur. In general, machine learning is quite useful particularly in the medical field, which depends on complex genomic measurements such as microarray technique and would increase the accuracy and precision of results. With this technology, doctors can easily diagnose patients with cancer quickly and apply the proper treatment in a timely manner. Therefore, the goal of this paper is to address and propose a robust Breast Cancer diagnostic system using complex genomic analysis via microarray technology. The system will combine two machine learning methods, K-means cluster, and linear regression.
An approach of cervical cancer diagnosis using class weighting and oversampli...TELKOMNIKA JOURNAL
Globally, cervical cancer caused 604,127 new cases and 341,831 deaths in 2020, according to the global cancer observatory. In addition, the number of cervical cancer patients who have no symptoms has grown recently. Therefore, giving patients early notice of the possibility of cervical cancer is a useful task since it would enable them to have a clear understanding of their health state. The use of artificial intelligence (AI), particularly in machine learning, in this work is continually uncovering cervical cancer. With the help of a logit model and a new deep learning technique, we hope to identify cervical cancer using patient-provided data. For better outcomes, we employ Keras deep learning and its technique, which includes class weighting and oversampling. In comparison to the actual diagnostic result, the experimental result with model accuracy is 94.18%, and it also demonstrates a successful logit model cervical cancer prediction.
Early detection of breast cancer using mammography images and software engine...TELKOMNIKA JOURNAL
The breast cancer has affected a wide region of women as a particular case. Therefore, different researchers have focused on the early detection of this disease to overcome it in efficient way. In this paper, an early breast cancer detection system has been proposed based on mammography images. The proposed system adopts deep-learning technique to increase the accuracy of detection. The convolutional neural network (CNN) model is considered for preparing the datasets of training and test. It is important to note that the software engineering process model has been adopted in constructing the proposed algorithm. This is to increase the reliably, flexibility and extendibility of the system. The user interfaces of the system are designed as a website used at country side general purpose (GP) health centers for early detection to the disease under lacking in specialist medical staff. The obtained results show the efficiency of the proposed system in terms of accuracy up to more than 90% and decrease the efforts of medical staff as well as helping the patients. As a conclusion, the proposed system can help patients by early detecting the breast cancer at far places from hospital and referring them to nearest specialist center.
Breast cancer is the leading cause of death for women worldwide. Cancer can be discovered early, lowering the rate of death. Machine learning techniques are a hot field of research, and they have been shown to be helpful in cancer prediction and early detection. The primary purpose of this research is to identify which machine learning algorithms are the most successful in predicting and diagnosing breast cancer, according to five criteria: specificity, sensitivity, precision, accuracy, and F1 score. The project is finished in the Anaconda environment, which uses Python's NumPy and SciPy numerical and scientific libraries as well as matplotlib and Pandas. In this study, the Wisconsin diagnostic breast cancer dataset was used to evaluate eleven machine learning classifiers: decision tree, quadratic discriminant analysis, AdaBoost, Bagging meta estimator, Extra randomized trees, Gaussian process classifier, Ridge, Gaussian nave Bayes, k-Nearest neighbors, multilayer perceptron, and support vector classifier. During performance analysis, extremely randomized trees outperformed all other classifiers with an F1-score of 96.77% after data collection and data analysis.
Breast Cancer Detection from Mammography Images Using Machine Learning Algorithms (U-Net Segmentation and Dense Net Classifier implementation are in progress)
The current big challenge facing radiologists in healthcare is the automatic detection and classification of masses in breast mammogram images. In the last few years, many researchers have proposed various solutions to this problem. These solutions are effectively dependent and work on annotated breast image data. But these solutions fail when applied to unlabeled and non-annotated breast image data. Therefore, this paper provides the solution to this problem with the help of a neural network that considers any kind of unlabeled data for its procedure. In this solution, the algorithm automatically extracts tumors in images using a segmentation approach, and after that, the features of the tumor are extracted for further processing. This approach used a double thresholding-based segmentation technique to obtain a perfect location of the tumor region, which was not possible in existing techniques in the literature. The experimental results also show that the proposed algorithm provides better accuracy compared to the accuracy of existing algorithms in the literature.
Breast cancer histological images nuclei segmentation and optimized classifi...IJECEIAES
Breast cancer incidences have grown worldwide during the previous few years. The histological images obtained from a biopsy of breast tissues are regarded as being the highest accurate approach to determine whether any cells exhibit symptoms of cancer. The visible position of nuclei inside the image is achieved through the use of instance segmentation, nevertheless, this work involves nucleus segmentation and features classification of the predicted nucleus for the achievement of best accuracy. The extracted features map using the feature pyramid network has been modified using segmenting objects by locations (SOLO) convolution with grasshopper optimization for multiclass classification. A breast cancer multiclassification technique based on a suggested deep learning algorithm was examined to achieve the accuracy of 99.2% using a huge database of ICIAR 2018, demonstrating the method’s efficacy in offering an important weapon for breast cancer multi-classification in a medical setting. The segmentation accuracy achieved is 88.46%.
Breast Tumor Detection Using Efficient Machine Learning and Deep Learning Tec...mlaij
Breast cancer tissues grow when cells in the breast expand and divide uncontrollably, resulting in a lump of tissue commonly called and named tumor. Breast cancer is the second most prevalent cancer among women, following skin cancer. While it is more commonly diagnosed in women aged 50 and above, it can affect individuals of any age. Although it is rare, men can also develop breast cancer, accounting for less than 1% of all cases, with approximately 2,600 cases reported annually in the United States. Early detection of breast tumors is crucial in reducing the risk of developing breast cancer. A publicly available dataset containing features of breast tumors was utilized to identify breast tumors using machine learning and deep learning techniques. Various prediction models were constructed, including logistic regression (LR), decision tree (DT), random forest (RF), support vector machine (SVM), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), Light GBM, and a recurrent neural network (RNN) model. These models were trained to classify and predict breast tumor cases based on the provided features.
BREAST TUMOR DETECTION USING EFFICIENT MACHINE LEARNING AND DEEP LEARNING TEC...mlaij
Breast cancer tissues grow when cells in the breast expand and divide uncontrollably, resulting in a lump
of tissue commonly called and named tumor. Breast cancer is the second most prevalent cancer among
women, following skin cancer. While it is more commonly diagnosed in women aged 50 and above, it can
affect individuals of any age. Although it is rare, men can also develop breast cancer, accounting for less
than 1% of all cases, with approximately 2,600 cases reported annually in the United States. Early
detection of breast tumors is crucial in reducing the risk of developing breast cancer. A publicly available
dataset containing features of breast tumors was utilized to identify breast tumors using machine learning
and deep learning techniques. Various prediction models were constructed, including logistic regression
(LR), decision tree (DT), random forest (RF), support vector machine (SVM), Gradient Boosting (GB),
Extreme Gradient Boosting (XGB), Light GBM, and a recurrent neural network (RNN) model. These
models were trained to classify and predict breast tumor cases based on the provided features.
Breast Tumor Detection Using Efficient Machine Learning and Deep Learning Tec...mlaij
Machine Learning and Applications: An International Journal (MLAIJ) is a quarterly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the machine learning. The journal is devoted to the publication of high quality papers on theoretical and practical aspects of machine learning and applications.The goal of this journal is to bring together researchers and practitioners from academia and industry to focus on machine learning advancements, and establishing new collaborations in these areas. Original research papers, state-of-the-art reviews are invited for publication in all areas of machine learning.
Authors are solicited to contribute to the journal by submitting articles that illustrate research results, projects, surveying works and industrial experiences that describe significant advances in the areas of machine learning.
BRAIN TUMOR MRIIMAGE CLASSIFICATION WITH FEATURE SELECTION AND EXTRACTION USI...ijistjournal
Feature extraction is a method of capturing visual content of an image. The feature extraction is the process to represent raw image in its reduced form to facilitate decision making such as pattern classification. We have tried to address the problem of classification MRI brain images by creating a robust and more accurate classifier which can act as an expert assistant to medical practitioners. The objective of this paper is to present a novel method of feature selection and extraction. This approach combines the Intensity, Texture, shape based features and classifies the tumor as white matter, Gray matter, CSF, abnormal and normal area. The experiment is performed on 140 tumor contained brain MR images from the Internet Brain Segmentation Repository. The proposed technique has been carried out over a larger database as compare to any previous work and is more robust and effective. PCA and Linear Discriminant Analysis (LDA) were applied on the training sets. The Support Vector Machine (SVM) classifier served as a comparison of nonlinear techniques Vs linear ones. PCA and LDA methods are used to reduce the number of features used. The feature selection using the proposed technique is more beneficial as it analyses the data according to grouping class variable and gives reduced feature set with high classification accuracy.
deep learning applications in medical image analysis brain tumorVenkat Projects
The tremendous success of machine learning algorithms at image recognition tasks in recent years intersects with a time of dramatically increased use of electronic medical records and diagnostic imaging. This review introduces the machine learning algorithms as applied to medical image analysis, focusing on convolutional neural networks, and emphasizing clinical aspects of the _eld. The advantage of machine learning in an era of medical big data is that signi_cant hierarchal relationships within the data can be discovered algorithmically without laborious hand-crafting of features. We cover key research areas and applications of medical image classi_cation, localization, detection, segmentation, and registration. We conclude by discussing research obstacles, emerging trends, and possible future directions.
A deep learning framework for accurate diagnosis of colorectal cancer using h...IJECEIAES
Colorectal cancer (CRC) is one of the most prevalent malignancies worldwide, with high mortality and incidence rates. Early detection of the disease may increase the probability of survival, making it critical to develop effective procedures for precise treatment. In the past few years, there has been an increased use of deep learning techniques in image classification that aid in the detection of various types of cancer. In this study, convolutional neural network (CNN) models were used to classify colorectal cancer into benign and malignant. After applying various data preprocessing techniques to the image dataset, we evaluated our prototypes using three distinct subsets of testing data, representing 20%, 30%, and 40% of the total dataset. Additionally, four pre-trained CNN models (ResNet-18, ResNet-50, GoogLeNet, and MobileNetV2) were trained, and the network architectural techniques were compared by applying the Adam optimizer. Finally, we assessed the performance of algorithms in terms of accuracy, sensitivity, specificity, precision, F1-score, and area under the receiver operating characteristic (ROC) curve (AUC). In this research, deep learning approaches demonstrated high efficacy in accurately diagnosing colorectal cancer. This indicates that these techniques have an important and significant value for advancing medical research.
Welcome to WIPAC Monthly the magazine brought to you by the LinkedIn Group Water Industry Process Automation & Control.
In this month's edition, along with this month's industry news to celebrate the 13 years since the group was created we have articles including
A case study of the used of Advanced Process Control at the Wastewater Treatment works at Lleida in Spain
A look back on an article on smart wastewater networks in order to see how the industry has measured up in the interim around the adoption of Digital Transformation in the Water Industry.
Student information management system project report ii.pdfKamal Acharya
Our project explains about the student management. This project mainly explains the various actions related to student details. This project shows some ease in adding, editing and deleting the student details. It also provides a less time consuming process for viewing, adding, editing and deleting the marks of the students.
NO1 Uk best vashikaran specialist in delhi vashikaran baba near me online vas...Amil Baba Dawood bangali
Contact with Dawood Bhai Just call on +92322-6382012 and we'll help you. We'll solve all your problems within 12 to 24 hours and with 101% guarantee and with astrology systematic. If you want to take any personal or professional advice then also you can call us on +92322-6382012 , ONLINE LOVE PROBLEM & Other all types of Daily Life Problem's.Then CALL or WHATSAPP us on +92322-6382012 and Get all these problems solutions here by Amil Baba DAWOOD BANGALI
#vashikaranspecialist #astrologer #palmistry #amliyaat #taweez #manpasandshadi #horoscope #spiritual #lovelife #lovespell #marriagespell#aamilbabainpakistan #amilbabainkarachi #powerfullblackmagicspell #kalajadumantarspecialist #realamilbaba #AmilbabainPakistan #astrologerincanada #astrologerindubai #lovespellsmaster #kalajaduspecialist #lovespellsthatwork #aamilbabainlahore#blackmagicformarriage #aamilbaba #kalajadu #kalailam #taweez #wazifaexpert #jadumantar #vashikaranspecialist #astrologer #palmistry #amliyaat #taweez #manpasandshadi #horoscope #spiritual #lovelife #lovespell #marriagespell#aamilbabainpakistan #amilbabainkarachi #powerfullblackmagicspell #kalajadumantarspecialist #realamilbaba #AmilbabainPakistan #astrologerincanada #astrologerindubai #lovespellsmaster #kalajaduspecialist #lovespellsthatwork #aamilbabainlahore #blackmagicforlove #blackmagicformarriage #aamilbaba #kalajadu #kalailam #taweez #wazifaexpert #jadumantar #vashikaranspecialist #astrologer #palmistry #amliyaat #taweez #manpasandshadi #horoscope #spiritual #lovelife #lovespell #marriagespell#aamilbabainpakistan #amilbabainkarachi #powerfullblackmagicspell #kalajadumantarspecialist #realamilbaba #AmilbabainPakistan #astrologerincanada #astrologerindubai #lovespellsmaster #kalajaduspecialist #lovespellsthatwork #aamilbabainlahore #Amilbabainuk #amilbabainspain #amilbabaindubai #Amilbabainnorway #amilbabainkrachi #amilbabainlahore #amilbabaingujranwalan #amilbabainislamabad
CFD Simulation of By-pass Flow in a HRSG module by R&R Consult.pptxR&R Consult
CFD analysis is incredibly effective at solving mysteries and improving the performance of complex systems!
Here's a great example: At a large natural gas-fired power plant, where they use waste heat to generate steam and energy, they were puzzled that their boiler wasn't producing as much steam as expected.
R&R and Tetra Engineering Group Inc. were asked to solve the issue with reduced steam production.
An inspection had shown that a significant amount of hot flue gas was bypassing the boiler tubes, where the heat was supposed to be transferred.
R&R Consult conducted a CFD analysis, which revealed that 6.3% of the flue gas was bypassing the boiler tubes without transferring heat. The analysis also showed that the flue gas was instead being directed along the sides of the boiler and between the modules that were supposed to capture the heat. This was the cause of the reduced performance.
Based on our results, Tetra Engineering installed covering plates to reduce the bypass flow. This improved the boiler's performance and increased electricity production.
It is always satisfying when we can help solve complex challenges like this. Do your systems also need a check-up or optimization? Give us a call!
Work done in cooperation with James Malloy and David Moelling from Tetra Engineering.
More examples of our work https://www.r-r-consult.dk/en/cases-en/
Water scarcity is the lack of fresh water resources to meet the standard water demand. There are two type of water scarcity. One is physical. The other is economic water scarcity.
Final project report on grocery store management system..pdfKamal Acharya
In today’s fast-changing business environment, it’s extremely important to be able to respond to client needs in the most effective and timely manner. If your customers wish to see your business online and have instant access to your products or services.
Online Grocery Store is an e-commerce website, which retails various grocery products. This project allows viewing various products available enables registered users to purchase desired products instantly using Paytm, UPI payment processor (Instant Pay) and also can place order by using Cash on Delivery (Pay Later) option. This project provides an easy access to Administrators and Managers to view orders placed using Pay Later and Instant Pay options.
In order to develop an e-commerce website, a number of Technologies must be studied and understood. These include multi-tiered architecture, server and client-side scripting techniques, implementation technologies, programming language (such as PHP, HTML, CSS, JavaScript) and MySQL relational databases. This is a project with the objective to develop a basic website where a consumer is provided with a shopping cart website and also to know about the technologies used to develop such a website.
This document will discuss each of the underlying technologies to create and implement an e- commerce website.
Supervised deep learning_embeddings_for_the_predic
1. Supervised deep learning embeddings for
the prediction of cervical cancer diagnosis
Kelwin Fernandes1,2
, Davide Chicco3
, Jaime S. Cardoso1,2
and Jessica
Fernandes4
1
Instituto de Engenharia de Sistemas e Computadores Tecnologia e Ciencia (INESC TEC), Porto,
Portugal
2
Universidade do Porto, Porto, Portugal
3
Princess Margaret Cancer Centre, Toronto, ON, Canada
4
Universidad Central de Venezuela, Caracas, Venezuela
ABSTRACT
Cervical cancer remains a significant cause of mortality all around the world, even if
it can be prevented and cured by removing affected tissues in early stages. Providing
universal and efficient access to cervical screening programs is a challenge that
requires identifying vulnerable individuals in the population, among other steps.
In this work, we present a computationally automated strategy for predicting
the outcome of the patient biopsy, given risk patterns from individual medical
records. We propose a machine learning technique that allows a joint and fully
supervised optimization of dimensionality reduction and classification models.
We also build a model able to highlight relevant properties in the low dimensional
space, to ease the classification of patients. We instantiated the proposed approach
with deep learning architectures, and achieved accurate prediction results (top area
under the curve AUC = 0.6875) which outperform previously developed methods,
such as denoising autoencoders. Additionally, we explored some clinical findings
from the embedding spaces, and we validated them through the medical literature,
making them reliable for physicians and biomedical researchers.
Subjects Bioinformatics, Computational Biology, Artificial Intelligence, Data Mining and
Machine Learning
Keywords Dimensionality reduction, Health-care informatics, Denoising autoencoder,
Autoencoder, Biomedical informatics, Binary classification, Deep learning, Cervical cancer,
Artificial neural networks, Health informatics
INTRODUCTION
Despite the possibility of prevention with regular cytological screening, cervical cancer
remains a significant cause of mortality in low-income countries (Kauffman et al., 2013).
The cervical tumor is the cause of more than 500,000 cases per year, and kills more than
250,000 patients in the same period, worldwide (Fernandes, Cardoso & Fernandes, 2015).
However, cervical cancer can be prevented by means of the human papillomavirus
infection (HPV) vaccine, and regular low-cost screening programs (Centers for Disease
Control and Prevention (CDC), 2013). The two most widespread techniques in screening
programs are conventional or liquid cytology and colposcopy (Fernandes, Cardoso &
Fernandes, 2015; Plissiti & Nikou, 2013; Fernandes, Cardoso & Fernandes, 2017b; Xu et al.,
2016). Furthermore, this cancer can be cured by removing the affected tissues when
How to cite this article Fernandes et al. (2018), Supervised deep learning embeddings for the prediction of cervical cancer diagnosis. PeerJ
Comput. Sci. 4:e154; DOI 10.7717/peerj-cs.154
Submitted 17 February 2018
Accepted 26 April 2018
Published 14 May 2018
Corresponding author
Kelwin Fernandes, kafc@inesctec.pt
Academic editor
Sebastian Ventura
Additional Information and
Declarations can be found on
page 16
DOI 10.7717/peerj-cs.154
Copyright
2018 Fernandes et al.
Distributed under
Creative Commons CC-BY 4.0
2. identified in early stages (Fernandes, Cardoso & Fernandes, 2015; Centers for Disease
Control and Prevention (CDC), 2013), in most cases.
The development of cervical cancer is usually slow and preceded by abnormalities in the
cervix (dysplasia). However, the absence of early stage symptoms might cause carelessness in
prevention. Additionally, in developing countries, there is a lack of resources, and patients
usually have poor adherence to routine screening due to low problem awareness.
While improving the resection of lesions in the first visits has a direct impact on
patients that attend screening programs, the most vulnerable populations have poor or
even non-existent adherence to treatment programs. Scarce awareness of the problem and
patients’ discomfort with the medical procedure might be the main causes of this
problem. Furthermore, in low-income countries, this issue can be due to lack of access
to vulnerable populations with low access to information and medical centers.
Consequently, the computational prediction of individual patient risk has a key role in
this context. Identifying patients with the highest risk of developing cervical cancer can
improve the targeting efficacy of cervical cancer screening programs: our software
performs this operation computationally in a few minutes by producing accurate
prediction scores.
Fernandes, Cardoso & Fernandes (2017b) performed a preliminary attempt to tackle the
problem of predicting the patient’s risk to develop cervical cancer through machine
learning software. In that project, the authors employed transfer learning strategies for the
prediction of the individual patient risk on a dataset of cervical patient medical tests. They
focused on transferring knowledge between linear classifiers on similar tasks, to predict
the patient’s risk (Fernandes, Cardoso & Fernandes, 2017b).
Given the high sparsity of the associated risk factors in the population, dimensionality
reduction techniques can improve the robustness of the machine learning predictive
models. However, many projects that take advantage of dimensionality reduction and
classification use suboptimal approaches, where each component is learned separately
(Li et al., 2012; Bessa et al., 2014; Lacoste-Julien, Sha & Jordan, 2009).
In this work, we propose a joint strategy to learn the low-dimensional space and the
classifier itself in a fully supervised way. Our strategy is able to reduce class overlap by
concentrating observations from the healthy patients class into a single point of the space,
while retaining as much information as possible from the patients with high risk of
developing cervical cancer.
We based our prediction algorithm on artificial neural networks (ANNs), which are
machine learning methods able to discover non-linear patterns by means of aggregation
of functions with non-linear activations. A recent trend in this field is deep learning
(LeCun, Bengio & Hinton, 2015), which involves large neural network architectures
with successive applications of such functions. Deep learning, in fact, has been able
to provide accurate predictions of patient diagnosis in multiple medical domains
(Xu et al., 2016; Chicco, Sadowski & Baldi, 2014; Fernandes, Cardoso & Astrup, 2017a;
Cangelosi et al., 2016; Alipanahi et al., 2015). We applied our learning scheme to deep
variational autoencoders and feed-forward neural networks. Finally, we explored
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 2/20
3. visualization techniques to understand and validate the medical concepts captured by
the embeddings.
We organize the rest of the paper as follows. After this Introduction, we describe the
proposed method and the dataset analyzed in the Methods and Dataset sections.
Afterwards, we describe the computational prediction results in the Results section, the
model outcome interpretation in the Discussion section, and we conclude the manuscript
outlining some conclusion and future development.
METHODS
High dimensional data can lead to several problems: in addition to high computational
costs (in memory and time), it often leads to overfitting (Van Der Maaten, Postma & Van
den Herik, 2009; Chicco, 2017; Moore, 2004). Dimensionality reduction can limit these
problems and, additionally, can improve the visualization and interpretation of the
dataset, because it allows researchers to focus on a reduced number of features. For
these reasons, we decided to map the original dataset features into a reduced
dimensionality before performing the classification task.
Generally, to tackle high-dimensional classification problems, machine learning
traditional approaches attempt to reduce the high-dimensional feature space to a
low-dimensional one, to facilitate the posterior fitting of a predictive model. In many
cases, researchers perform these two steps separately, deriving suboptimal combined
models (Li et al., 2012; Bessa et al., 2014; Lacoste-Julien, Sha & Jordan, 2009). Moreover,
since dimensionality reduction techniques are often learned in an unsupervised fashion,
they are unable to preserve and exploit the separability between observations from
different classes.
In dimensionality reduction, researchers use two categories of objective functions: one
for maximizing the model capability of recovering the original feature space from the
compressed low dimensional one, and another one for maximizing the consistency of
pairwise similarities in both high and low dimensional spaces.
Since defining a similarity metric in a high-dimensional space might be difficult,
we limit the scope of this work to minimizing the reconstruction loss. In this sense,
given a set of labeled input vectors X = {x1, x2, : : : , xn}, where xi ∈ Rd
, ∀i ∈ 1, : : : ,n and
Y is a vector with the labels associated to each observation, we want to obtain
two functions C: Rd
/ Rm
and D: Rm
/ Rd
such that m < d and that minimizes the
following loss:
LrðC; D; XÞ ¼
1
jXj
X
x2X
ðD CÞðxÞ x
ð Þ2
(1)
Namely, the composition (○) of the compressing (C), and decompressing (D) functions
approximate the identity function.
In the following sections, we describe the proposed dimensionality reduction technique
and its instantiation to deep learning architectures.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 3/20
4. Joint dimensionality reduction and classification
Since our final goal is to classify the data instances (observations), we need to achieve a
good low-dimensional mapping and build the classifier independently. Thereby, we
propose a joint loss function that minimizes the trade-off between data reconstruction
and classification performance:
LðM; C; D; X; YÞ ¼ LcððM CÞðXÞ; YÞ þ LrðC; D; XÞ (2)
where M is a classifier that receives as input the vectors in the low dimensional space
(C(X)), Lc is a classification loss function such as categorical cross-entropy, and 0. In
this case, we focus on the classification performance using Eq. (1) as a regularization
factor of the models of interest. Hereafter, we will denote this method as semi-supervised
dimensionality reduction.
Fully supervised embeddings
The previously proposed loss function consists of two components: a supervised
component given by the classification task, and an unsupervised component given by
the low-dimensional mapping. However, the scientific community aims at understanding
the properties captured in the embeddings, especially on visual and text embeddings
(Kiros, Salakhutdinov Zemel, 2014; Levy, Goldberg Ramat-Gan, 2014). Moreover,
inducing properties in the low-dimensional space can improve the class separability.
To apply this enhancement, we introduce partial supervision in the Lr loss.
We can explore these properties by learning the dimensionality reduction process
in a supervised way. Namely, learning a bottleneck supervised mapping function
((D ○ C)(x) ≈ M(x, y)) instead of the traditional identity function ((D ○ C)(x) ≈ x) used
in reconstruction-based dimensionality reduction techniques. The reconstruction loss
Lr(C, D, X) becomes:
LM ðC; D; X; YÞ ¼
1
jXj
X
hx;yi2X;Y
ðD CÞðxÞ Mðx; yÞ
ð Þ2
(3)
where M(x) is the desired supervised mapping.
To facilitate the classification task, removing the overlap between both classes should be
captured in low-dimensional spaces. Without loss of generality, we assume that the feature
space is non-negative. Thereby we favor models with high linear separability between
observations by using the mapping function Eq. (4) in Eq. (3).
Symðx; yÞ ¼
x; if y
x; if:y
(4)
In our application, if all the features are non-negative, the optimal patient’s behavior
associates to the zero vector with total lack of risk patterns. On the other hand, a patient
with high feature values is prone to have cancer. Within the context of cervical cancer
screening, we propose the mapping given by Eq. (5), where the decoded version of the
healthy patients is the zero vector. This idea resembles the fact that their risk conduct
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 4/20
5. has not contributed to the disease occurrence. On the other hand, we mapped ill patients
to their original feature space, for promoting the low-dimensional vectors to explain
the original risk patterns that originated the disease.
Zeroðx; yÞ ¼ 1 ðyÞ x (5)
While the definition of the properties of interest to be captured by the low-dimensional
space is application-dependent, the strategy to promote such behavior can be adapted
to other contexts.
Deep supervised autoencoders
Autoencoders are special cases of deep neural networks for dimensionality reduction
(Chicco, Sadowski Baldi, 2014; Vincent et al., 2008). They can be seen as general feed-
forward neural networks with two main sub-components: the first part of the neural
network is known as the encoder, and its main purpose is to compress the feature space.
The neural network achieves this step by using hidden layers with fewer units than the
input features, or by enforcing sparsity in the hidden representation. The second part of
the neural network, also known as the decoder, behaves in the opposite way, and tries to
approximate the inverse encoding function. While these two components correspond
to the C and D functions in Eq. (1), respectively, they can be broadly seen as a single
ANN that learns the identity function through a bottleneck, a low number of units, or
through sparse activations. Autoencoders are usually learned in an unsupervised fashion
by minimizing the quadratic error Eq. (1).
Denoising autoencoders (DA) represent a special case of deep autoencoders that
attempt to reconstruct the input vector when given a corrupted version (Vincent et al.,
2008). DA can learn valuable representations even in the presence of noise. Scientists can
experiment this task by adding an artificial source of noise in the input vectors. In the
neural network architecture (Fig. 1), we also included a dropout layer after the input
layer that randomly turns off at maximum one feature per patient (Srivastava et al., 2014).
Thereby, we aim to build stable classifiers that produce similar outcomes for patients with
small differences in their historical records. Furthermore, we aim at producing stable
decisions when patients lie on a subset of the answers to the doctors’ questions during
the medical visit, by indicating absence of a given risk behavior (for example, high number
of sexual partners, drug consumption, and others). We use a Parametric Rectifier Linear
Unit (PReLU) (He et al., 2015) as activation function in the hidden layers of our
architectures (Fig. 1). PReLU is a generalization of standard rectifier activation units,
which can improve model fitting with low additional computational cost (He et al., 2015).
The loss functions (Eqs. 2 and 3) can learn a joint classification and encoding–decoding
network in a multitask fashion (Fig. 2). Additionally, to allow the neural network to
use either the learned or the original representation, we include a bypass layer that
concatenates the hidden representation with the corrupted input. In the past, researchers
have used this technique in biomedical image segmentation with U-net architectures
(Ronneberger, Fischer Brox, 2015) to recover possible losses in the compression process,
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 5/20
6. and to reduce the problem of vanishing gradients. We use this bypass layer with
cross-validation.
In a nutshell, our contribution can be summarized as follows: (i) we formalized a
loss function to handle dimensionality reduction and classification in a joint fashion,
leading to a global optimal pipeline; (ii) in order to induce desired properties on the
compressed space, we proposed a loss that measures the model’s capability to recreate a
mapping with the desired property instead of the identity function usually applied in
dimensionality reduction; (iii) we showed that multitask autoencoders based on neural
networks can be used as a specific instance to solve this problem, and we instantiated
this idea to model an individual patient’s risk of having cervical cancer.
DATASET
The dataset we analyze contains medical records of 858 patients, and covers a random
sampling of patients between 2012 and 2013 who attended the gynecology service at
Figure 1 Deep denoising autoencoder. The blocks in blue and red represent the encoding (C) and
decoding (D) components of the network, respectively. Full-size DOI: 10.7717/peerj-cs.154/fig-1
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 6/20
7. Hospital Universitario de Caracas in Caracas, Venezuela. Most of the patients belong to
the lowest socioeconomic status (Graffar classification: IV–V (Graffar, 1956)) with low
income and educational level, being the population with the highest risk. The age of the
patients spans between 13 and 84 years old (27 years old on average). All patients are
sexually active and most of them (98%) have been pregnant at least once. The screening
process covers traditional cytology, the colposcopic assessment with acetic acid and the
Schiller test (Lugol’s iodine solution) (Fernandes, Cardoso Fernandes, 2017b). The
medical records include the age of the patient, sexual activity (number of sexual partners
and age of first sexual intercourse), number of pregnancies, smoking behavior, use of
contraceptives (hormonal and intrauterine devices), and historical records of sexually
transmitted diseases (STDs) (Table 1). Hence, we encoded the features denoted by
bool T, T ∈ {bool, int} as two independent values: whether or not the patient answered
the question and, if she did, the answered value. In some cases, the patients decided not to
answer some questions for privacy concerns. This behavior is often associated with risk
behaviors being a relevant feature to explore when modeling risk patterns. Therefore,
we added a flag feature that allows the model to identify if the question was answered
or not after missing value imputation. We encoded the categorical features using the
one-of-K scheme. The hospital anonymized all the records before releasing the dataset.
The dataset is now publically available on the Machine Learning Repository website of
the University of California Irvine (UCI ML) (University of California Irvine, 1987),
Figure 2 Supervised deep embedding architecture. The blocks in blue, red, and green represent the
encoding (C), decoding (D), and classification (M) components of the network, respectively.
Full-size DOI: 10.7717/peerj-cs.154/fig-2
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 7/20
8. which also contains a description of the features (University of California Irvine Machine
Learning Repository, 2017).
To avoid problems of the algorithm behavior related to different value ranges of
each feature, we scaled all the features in our experiments using [0,1] normalization,
and we input missing data using the average value (Chicco, 2017). While more
complex pre-processing schemes could be introduced, such as inferring the missing
value with a k-nearest neighbor model (Santos et al., 2015), we decided to use this
methodology to avoid additional complexity that would make it difficult to fairly
compare the explored techniques. In most cases, the features positively correlate to the
cancer variable, with 0 representing the lack of that risk pattern and 1 representing
the maximum risk.
RESULTS
We measured the performance of the proposed methods with the area under the
Precision–Recall (PR) curves (Davis Goadrich, 2006; Chicco, 2017) and the logistic loss
(also known as cross-entropy loss) function.
As baseline, we use a deep feed-forward neural network with a softmax activation in the
output layer. The remaining parameters (such as the initial dropout layer, depth and
optimization algorithm) conform to the ones used in the proposed methodologies
(Table 2). The main hyper-parameters related to the network topology are the depth
and width, which define the number of layers in the architecture and the size of the
low-dimensional representation.
Table 1 Feature names and data type acquired in the risk factors dataset (Fernandes, Cardoso Fernandes, 2017b).
Feature Type Feature Type
Age int IUD (years) int
Number of sexual partners bool int Sexually transmitted diseases (STDs) (yes/no) bool bool
Age of first sexual intercourse bool int Number of STDs int
Number of pregnancies bool int Diagnosed STDs Categorical
Smokes (yes/no) bool bool STDs (years since first diagnosis) int
Smokes (years and packs) int int STDs (years last diagnosis) int
Hormonal contraceptives (yes/no) bool Previous cervical diagnosis (yes/no) bool
Hormonal contraceptives (years) int Previous cervical diagnosis (years) int
Intrauterine device (IUD) (yes/no) bool Previous cervical diagnosis Categorical
Note:
int, integer; bool, boolean.
Table 2 Set of possible options for fine-tuning each parameter.
Parameter Values
Depth {1, : : : , 6}
Width {10, 20}
Regularization {0.01, 0.1}
Bypass usage {false, true}
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 8/20
9. We used a stratified 10-fold cross-validation in the assessment of the proposed
methods. We optimized the neural networks by using the RMSProp optimization strategy
(Tieleman Hinton, 2012) for a maximum number of 500 epochs, with early stopping
after 100 iterations without improvement and a batch size of 32. We validated these
parameters empirically, and it was enough to ensure model convergence in all cases. We
also validated the performance of other optimization strategies such as Adam and
stochastic gradient descent. However, we did not observe any gain in terms of predictive
performance or convergence. We use sparse autoencoders by adding an L1 penalization
term, to ensure that each unit combines a small subset of risk factors, as would be done
by a human expert.
We fine-tuned all the hyper-parameters using a grid search strategy with nested
stratified threefold cross-validation. In this sense, we validated the performance of each
network configuration on three training-validation partitions and choose the one that
maximizes the area under the PR curve. Then, for the best configuration, we re-trained the
model using the entire training set. We chose the size of the low-dimensional space as part
of this nested cross-validation procedure, and chose empirically the parameters related
to the optimization algorithm (that are strategy, number of epochs, early stopping).
To recreate the decisions made by the physician at different configurations of the
screening process, we consider the observability of all possible subsets of screening
outcomes when predicting the biopsy results. Thereby, we cover scenarios where only
behavioral and demographic information is observable (first line of each table with empty
subset) up to settings where cytology and colposcopy (Hinselmann and Schiller)
results are available.
Diagnosis prediction results
Our proposed architectures with embedding regularization achieved the best diagnosis
prediction results in most cases (Tables 3 and 4) when compared with other neural
Table 3 Performance of the proposed architectures in terms of area under the Precision–Recall
curve.
Subset Baseline Semi Sym Zero SVM k-NN DecTree
0.1334 0.1424 0.1534 0.1744 0.0877 0.0345 0.1941
C 0.1998 0.1853 0.2115 0.2174 0.1550 0.3033 0.2560
H 0.4536 0.4459 0.4407 0.4625 0.4192 0.3885 0.3616
S 0.6416 0.6411 0.6335 0.6668 0.5905 0.5681 0.6242
CH 0.4752 0.4684 0.4754 0.4609 0.4423 0.4095 0.4023
CS 0.6265 0.6424 0.6388 0.5985 0.6205 0.5379 0.6089
HS 0.6200 0.6356 0.6277 0.5864 0.6199 0.6335 0.5956
CHS 0.6665 0.6351 0.6875 0.6404 0.6374 0.6653 0.5542
Best 0 2 2 2 0 1 1
Notes:
The subset of observable screening strategies include: Cytology (C), Hinselmann (H), and Schiller (S).
Baseline, deep feed-forward neural network; Semi, semi-supervised dimensionality reduction (Eq. 2); Sym, symmetry
mapping dimensionality reduction (Eq. 4); Zero, zero mapping dimensionality reduction (Eq. 5); SVM, support vector
machine; k-NN, k-nearest neighbors; DecTree, decision tree.
We highlight the best performing models in bold.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 9/20
10. network approaches. Furthermore, the fully supervised embeddings improved the
performance of the semi-supervised approach (Eq. 2), through both the strategies
(symmetric and zero mapping). The relative gains in terms of area under the PR
curve depend on the subset of observable modalities, ranging from 30.7% when only
medical records are observed to 3.3% when the outcome of all the screening procedures
is known.
Using a paired difference Student’s t-test (Menke Martinez, 2004) with a 95%
confidence level, zero-mapping methodology achieved better results than the baseline and
semi-supervised learning schemes. We found no statistical differences between the
symmetry and zero mappings.
We validated the performance of traditional machine learning models such as
support vector machines (SVM) with radial basis function kernel (Scholkopf et al., 1997),
k-nearest neighbors (Peterson, 2009), and decision trees (Quinlan, 1986). In general, the
proposed models surpassed the performance of the classical methodologies in terms of
area under the PR curve. The SVM model achieved better logarithmic loss given the
post-processing of its scores using the Logistic Regression model that directly optimize
this metric. Further improvements could be observed by post-processing the outcome
of the other strategies.
The gains achieved by the mapping-based supervised embeddings happen because the
proposed fully-supervised strategies aim to reduce the overlap between observations from
both classes. In the past, researchers showed that class overlap has higher correlation with
the model performance than the imbalance ratio in highly unbalanced datasets (Cruz
et al., 2016). The visualization of the embeddings through the t-distributed stochastic
neighbor embedding (t-SNE) (Van Der Maaten Hinton, 2008) confirms this aspect,
because in t-SNE fully supervised embeddings achieve better separability and fewer
overlapping clusters (Figs. 3–6).
Table 4 Performance of the proposed architectures in terms of logarithmic loss.
Subset Baseline Semi Sym Zero SVM k-NN DecTree
0.3004 0.2708 0.2657 0.2716 0.2421 4.3670 4.1889
C 0.2829 0.2757 0.2868 0.2609 0.2614 2.6884 3.5001
H 0.2169 0.2274 0.2422 0.2031 0.1984 0.7178 3.2175
S 0.1710 0.1475 0.1489 0.1359 0.1273 0.9366 1.6893
CH 0.2210 0.2054 0.2286 0.2123 0.2196 1.0477 2.8509
CS 0.1594 0.1469 0.1240 0.1464 0.1248 0.4036 1.7687
HS 0.1632 0.1786 0.1615 0.1622 0.1225 0.3238 1.8098
CHS 0.1563 0.1577 0.1494 0.1514 0.1099 0.4037 1.8906
Best 0 1 1 1 4 0 0
Notes:
The subset of observable screening strategies include: Cytology (C), Hinselmann (H), and Schiller (S). Area under the
Precision–Recall curve.
Baseline, deep feed-forward neural network; Semi, semi-supervised dimensionality reduction (Eq. 2); Sym, symmetry
mapping dimensionality reduction (Eq. 4); Zero, zero mapping dimensionality reduction (Eq. 5); SVM, support vector
machine; k-NN, k-nearest neighbors; DecTree, decision tree.
We highlight the best performing models in bold.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 10/20
11. Figure 3 Two-dimensional projection of the unsupervised embedding using t-distributed stochastic
neighbor embedding (t-SNE) (Van Der Maaten Hinton, 2008).
Full-size DOI: 10.7717/peerj-cs.154/fig-3
Figure 4 Two-dimensional projection of the semi-supervised embedding using t-distributed
stochastic neighbor embedding (t-SNE) (Van Der Maaten Hinton, 2008).
Full-size DOI: 10.7717/peerj-cs.154/fig-4
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 11/20
12. Figure 5 Two-dimensional projection of the semi-supervised embedding with symmetry mapping
using t-distributed stochastic neighbor embedding (t-SNE) (Van Der Maaten Hinton, 2008).
Full-size DOI: 10.7717/peerj-cs.154/fig-5
Figure 6 Two-dimensional projection of the supervised embedding with zero mapping using
t-distributed stochastic neighbor embedding (t-SNE) (Van Der Maaten Hinton, 2008).
Full-size DOI: 10.7717/peerj-cs.154/fig-6
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 12/20
13. For visualization purposes, we are using t-SNE based upon neighborhood similarities,
since learning a valuable representation in a two-dimensional space raises difficulties.
Moreover, because of the high dimensionality of our embeddings, their reduction
capabilities rely on their sparsity.
Results in other applications
To observe the impact of our method, we validated the performance of the
aforementioned model architectures on several biomedical datasets available on the
UC Irvine Machine Learning Repository. Thus, we assessed the model’s performance on
nine datasets. The machine learning models we proposed achieved high prediction results,
being the zero-mapping approach the best model in most cases (Tables 5 and 6).
Table 5 Performance of the proposed architectures on other datasets downloaded from UC Irvine
Machine Learning Repository (University of California Irvine, 1987), measured through the area
under the Precision–Recall curve.
Dataset Baseline Semi Sym Zero
Breast cancer Mangasarian, Street Wolberg (1995) 0.9795 0.9864 0.9835 0.9856
Mammography Elter, Schulz-Wendtland Wittenberg (2007) 0.8551 0.8539 0.8533 0.8489
Parkinson Little et al. (2007) 0.9517 0.9526 0.9573 0.9604
Pima diabetes Smith et al. (1988) 0.7328 0.7262 0.7095 0.7331
Lung cancer Hong Yang (1991) 0.7083 0.6042 0.6927 0.8021
Cardiotocography Ayres-de Campos et al. (2000) 0.9948 0.9948 0.9925 0.9958
SPECTF heart Kurgan et al. (2001) 0.9470 0.9492 0.9462 0.9463
Arcene Guyon et al. (2005) 0.8108 0.8433 0.8900 0.8455
Colposcopy QA Fernandes, Cardoso Fernandes (2017b) 0.7760 0.8122 0.7961 0.8470
Best 1 2 1 5
Note:
We highlight the best performing models in bold.
Table 6 Performance of the proposed architectures on other datasets downloaded from UC Irvine
Machine Learning Repository (University of California Irvine, 1987), measured through logarithmic
loss.
Dataset Baseline Semi Sym Zero
Breast cancer Mangasarian, Street Wolberg (1995) 0.0984 0.0888 0.0966 0.0930
Mammographic Elter, Schulz-Wendtland Wittenberg (2007) 0.5122 0.5051 0.4973 0.4822
Parkinson Little et al. (2007) 0.3945 0.4042 0.3883 0.4323
Pima diabetes Smith et al. (1988) 0.5269 0.5229 0.5250 0.5472
Lung cancer Hong Yang (1991) 1.1083 0.8017 0.6050 0.8328
Cardiotocography Ayres-de Campos et al. (2000) 0.0113 0.0118 0.0116 0.0110
SPECTF heart Kurgan et al. (2001) 0.4107 0.4205 0.4121 0.4196
Arcene Guyon et al. (2005) 1.3516 0.8855 1.0230 1.1518
Colposcopy QA Fernandes, Cardoso Fernandes (2017b) 0.5429 0.5406 0.5195 0.4850
Best 1 3 2 3
Note:
We highlight the best performing models in bold.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 13/20
14. This outcome suggests that mapping the majority class to a unique point in the space
might improve the learning effectiveness in unbalanced settings. This idea draws a link
between binary and one-class classification, and we plan to explore it more in the future.
DISCUSSION
As shown in the Results section, our deep learning algorithm can predict cervical cancer
diagnosis with high accuracy. To further understand the clinical interpretability of our
prediction model, we investigated which dataset risk features have the highest impact
in the cervical cancer diagnosis for the patients.
Figure 7 Agglomerative clustering of features by impact on the embedding space. Full-size DOI: 10.7717/peerj-cs.154/fig-7
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 14/20
15. In fact, pre-invasive intra-epithelial lesions of the cervix and cervical cancer relate to
HPV infection of oncological serotypes that progress to oncological lesions, and multiple
factors contribute to this progress without a definite cause-dependent relation. The
patterns that have highest acceptance in the literature regard presence of human
immunodeficiency virus and smoking, followed by sexual risk behaviors such as early
sexual initiation, promiscuity, multiple pregnancies, and a history of sexually transmitted
infections. Another factor involved is the use of oral contraceptives.
From a technical point of view, while black-box machine learning techniques have
achieved state-of-the-art results in several applications, the lack of interpretability of the
induced models can limit their general acceptance by the medical community. Thus,
we tried to understand the relationships by using our prediction model to corroborate if
they are supported by the medical literature.
In this context, we studied the impact of the original features on the embedding space
to find correlations in the decision process. To determine this impact, we perturbed each
feature using all the other values from the feature’s domain, and then we computed
the maximum impact of the features in the embedded space. Finally, we applied an
agglomerative clustering technique to aggregate features with similar impact in the
embedding features. From a medical point of view, we validated several properties of
interest (Fig. 7).
For instance, risky sexual patterns such as an early sexual initiation and the presence
(and lifespan) of STDs (with a special focus on HPV) have the most similar impact in the
predictive outcome of the model. Also, smoking habits are associated by the model as
having a similar effect as these sexual patterns. These relationships were already studied in
the medical literature (Louie et al., 2009; Deacon et al., 2000).
The similarity between the use of hormonal contraceptives with condylomatosis and
the use of intrauterine devices with STDs shows another interesting pattern that has not
been quantified yet to the best of our knowledge. These patterns might be evidence
of sexual patterns with high risk.
CONCLUSION
Cervical cancer is still a widespread disease nowadays, and its diagnosis often requires
frequent and very time-consuming clinical exams. In this context, machine learning can
provide effective tools to speed up the diagnosis process, by processing high-scale patients’
datasets in a few minutes.
In this manuscript, we presented a computational system for the prediction of cervical
patient diagnosis, and for the interpretation of its results. Our system consists of a loss
function that allows joint optimization of dimensionality reduction, and classification
techniques able to promote relevant properties in the embedded spaces. Our deep learning
methods predicted the diagnosis of the patients with high accuracy, and their application
to other datasets showed that their robustness and effectiveness is not bounded to cervical
cancer. Our methods can be used to analyze profiles of patients where the biopsy and
potentially other screening results are missing, and are able to predict confidently if
they have cervical cancer.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 15/20
16. In the future, we plan to employ alternative approaches for data missing imputation,
such as oversampling through k-nearest neighbors (Santos et al., 2015) or latent semantic
indexing similarity (Chicco Masseroli, 2015). We also plan to try alternative prediction
models, like probabilistic latent semantic analysis (Pinoli, Chicco Masseroli, 2015).
Finally, we plan to extend our computational system by adding a feature selection step,
able to state the most relevant features among the dataset.
ACKNOWLEDGEMENTS
The authors thank the Gynecology Service of the Hospital Universitario de Caracas, and
Francis Nguyen (Princess Margaret Cancer Centre) for the English proof-reading of
this manuscript.
ADDITIONAL INFORMATION AND DECLARATIONS
Funding
This work was funded by the Project “NanoSTIMA: Macro-to-Nano Human Sensing:
Towards Integrated Multimodal Health Monitoring and Analytics/NORTE-01-0145-
FEDER-000016” financed by the North Portugal Regional Operational Programme
(NORTE 2020), under the PORTUGAL 2020 Partnership Agreement, and through the
European Regional Development Fund (ERDF), and also by Fundacao para a Ciencia e a
Tecnologia (FCT) within the PhD grant number SFRH/BD/93012/2013. The funders had
no role in study design, data collection and analysis, decision to publish, or preparation
of the manuscript.
Grant Disclosures
The following grant information was disclosed by the authors:
NanoSTIMA: Macro-to-Nano Human Sensing: Towards Integrated Multimodal Health
Monitoring and Analytics: NORTE-01-0145-FEDER-000016.
North Portugal Regional Operational Programme: NORTE 2020.
PORTUGAL 2020 Partnership Agreement.
European Regional Development Fund (ERDF).
Fundacao para a Ciencia e a Tecnologia (FCT): SFRH/BD/93012/2013.
Competing Interests
The authors declare that they have no competing interests.
Author Contributions
Kelwin Fernandes conceived and designed the experiments, performed the experiments,
analyzed the data, contributed reagents/materials/analysis tools, prepared figures and/or
tables, performed the computation work, authored or reviewed drafts of the paper,
approved the final draft.
Davide Chicco conceived and designed the experiments, analyzed the data, contributed
reagents/materials/analysis tools, prepared figures and/or tables, authored or reviewed
drafts of the paper, approved the final draft.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 16/20
17. Jaime S. Cardoso conceived and designed the experiments, contributed reagents/
materials/analysis tools, authored or reviewed drafts of the paper, approved the final
draft, proposed parts of the general strategy of the project.
Jessica Fernandes analyzed the data, contributed reagents/materials/analysis tools,
authored or reviewed drafts of the paper, approved the final draft, construction of the
dataset and domain expertise about the application.
Data Availability
The following information was supplied regarding data availability:
The dataset is publicly available at the University of California, Irvine Machine
Learning Repository: https://archive.ics.uci.edu/ml/datasets/Cervical+cancer+%28Risk
+Factors%29
The dataset is also available at Github: https://github.com/kelwinfc/cervical-cancer-
screening/tree/master/risk-factors/data
The software code of the methods used in the project is available at Github: https://
github.com/kelwinfc/cervical-cancer-screening/
We implemented the software in Python 2.7 using the Keras (Chollet, 2015) and
TensorFlow (Abadi et al., 2016) frameworks, and tested it on a computer running the
Linux Ubuntu 16.04 operating system.
REFERENCES
Abadi M, Barham P, Chen J, Chen Z, Davis A, Dean J, Devin M, Ghemawat S, Irving G, Isard M,
Kudlur M, Levenberg J, Monga R, Moore S, Murray DG, Steiner B, Tucker P, Vasudevan V,
Warden P, Wicke M, Yu Y, Zheng X, Google Brain. 2016. TensorFlow: a system for large-scale
machine learning. In: Proceedings of the 12th USENIX Symposium on Operating Systems Design
and Implementation (OSDI 2016), Savannah, GA, USA. Vol. 16, 265–283.
Alipanahi B, Delong A, Weirauch MT, Frey BJ. 2015. Predicting the sequence specificities of
DNA- and RNA-binding proteins by deep learning. Nature Biotechnology 33(8):831–838
DOI 10.1038/nbt.3300.
Ayres-de Campos D, Bernardes J, Garrido A, Marques-de Sa J, Pereira-Leite L. 2000. Sisporto
2.0: a program for automated analysis of cardiotocograms. Journal of Maternal-Fetal Medicine
9(5):311–318 DOI 10.3109/14767050009053454.
Bessa S, Domingues I, Cardosos JS, Passarinho P, Cardoso P, Rodrigues V, Lage F. 2014.
Normal breast identification in screening mammography: a study on 18,000 images. In: 2014
IEEE International Conference on Bioinformatics and Biomedicine (BIBM). Belfast: IEEE,
325–330.
Cangelosi D, Pelassa S, Morini M, Conte M, Bosco MC, Eva A, Sementa AR, Varesio L. 2016.
Artificial neural network classifier predicts neuroblastoma patients’ outcome. BMC
Bioinformatics 17(12):83.
Centers for Disease Control and Prevention (CDC). 2013. Cervical cancer screening among
women aged 18–30 years—United States, 2000–2010. Morbidity and Mortality Weekly Report
61(51–52):1038.
Chicco D. 2017. Ten quick tips for machine learning in computational biology. BioData Mining
10(1):35 DOI 10.1186/s13040-017-0155-3.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 17/20
18. Chicco D, Masseroli M. 2015. Software suite for gene and protein annotation prediction and
similarity search. IEEE/ACM Transactions on Computational Biology and Bioinformatics
12(4):837–843 DOI 10.1109/tcbb.2014.2382127.
Chicco D, Sadowski P, Baldi P. 2014. Deep autoencoder neural networks for Gene Ontology
annotation predictions. In: Proceedings of ACM BCB 2014. Newport Beach: ACM, 533–540.
Chollet F. 2015. Keras. Available at https://github.com/keras-team/keras.
Cruz R, Fernandes K, Cardoso JS, Costa JFP. 2016. Tackling class imbalance with ranking. In: The
2016 International Joint Conference on Neural Networks. Vancouver: IEEE, 2182–2187.
Davis J, Goadrich M. 2006. The relationship between precision-recall and ROC curves. In:
Proceedings of the 23rd International Conference on Machine Learning. Pittsburgh: ACM,
233–240.
Deacon JM, Evans CD, Yule R, Desai M, Binns W, Taylor C, Peto J. 2000. Sexual behaviour and
smoking as determinants of cervical HPV infection and of CIN3 among those infected: a case–
control study nested within the Manchester cohort. British Journal of Cancer 83(11):1565–1572
DOI 10.1054/bjoc.2000.1523.
Elter M, Schulz-Wendtland R, Wittenberg T. 2007. The prediction of breast cancer biopsy
outcomes using two CAD approaches that both emphasize an intelligible decision process.
Medical Physics 34(11):4164–4172 DOI 10.1118/1.2786864.
Fernandes K, Cardoso JS, Astrup BS. 2017a. Automated detection and categorization of genital
injuries using digital colposcopy. In: Alexandre L, Salvador Sánchez J, Rodrigues J, eds. Iberian
Conference on Pattern Recognition and Image Analysis. Faro: Springer, 251–258.
Fernandes K, Cardoso JS, Fernandes J. 2017b. Transfer learning with partial observability applied
to cervical cancer screening. In: Alexandre L, Salvador Sánchez J, Rodrigues J, eds. Iberian
Conference on Pattern Recognition and Image Analysis. Faro: Springer, 243–250.
Fernandes K, Cardoso JS, Fernandes J. 2015. Temporal segmentation of digital colposcopies. In:
Paredes R, Cardoso J, Pardo X, eds. Iberian Conference on Pattern Recognition and Image
Analysis. Santiago de Compostela: Springer, 262–271.
Graffar M. 1956. Une méthode de classification sociale d’échantillons de population. Courrier
6(8):455–459.
Guyon I, Gunn S, Ben-Hur A, Dror G. 2005. Result analysis of the nips 2003 feature selection
challenge. In: Weiss Y, Schölkopf B, Platt JC, eds. Advances in Neural Information Processing
Systems. Vancouver: Neural Information Processing Systems Foundation, 545–552.
He K, Zhang X, Ren S, Sun J. 2015. Delving deep into rectifiers: Surpassing human-level
performance on imagenet classification. In: Proceedings of IEEE ICCV 2015, Santiago, Chile,
1026–1034.
Hong Z-Q, Yang J-Y. 1991. Optimal discriminant plane for a small number of samples and
design method of classifier on the plane. Pattern Recognition 24(4):317–324
DOI 10.1016/0031-3203(91)90074-f.
Kauffman RP, Griffin SJ, Lund JD, Tullar PE. 2013. Current recommendations for cervical cancer
screening: do they render the annual pelvic examination obsolete? Medical Principles and
Practice 22(4):313–322 DOI 10.1159/000346137.
Kiros R, Salakhutdinov R, Zemel RS. 2014. Unifying visual-semantic embeddings with
multimodal neural language models. Available at http://arxiv.org/abs/1411.2539.
Kurgan LA, Cios KJ, Tadeusiewicz R, Ogiela M, Goodenday LS. 2001. Knowledge discovery
approach to automated cardiac SPECT diagnosis. Artificial Intelligence in Medicine
23(2):149–169 DOI 10.1016/s0933-3657(01)00082-3.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 18/20
19. Lacoste-Julien S, Sha F, Jordan MI. 2009. Disclda: discriminative learning for dimensionality
reduction and classification. In: Bengio Y, Schuurmans D, Lafferty JD, Williams CKI, Culotta A,
eds. Advances in Neural Information Processing Systems. Vancouver: Neural Information
Processing Systems Foundation, 897–904.
LeCun Y, Bengio Y, Hinton G. 2015. Deep learning. Nature 521(7553):436–444
DOI 10.1038/nature14539.
Levy O, Goldberg Y, Ramat-Gan I. 2014. Linguistic regularities in sparse and explicit word
representations. In: Morante R, Yih SW, eds. CoNLL. Baltimore: Association for Computational
Linguistics, 171–180.
Li W, Prasad S, Fowler JE, Bruce LM. 2012. Locality-preserving dimensionality reduction and
classification for hyperspectral image analysis. IEEE Transactions on Geoscience and Remote
Sensing 50(4):1185–1198 DOI 10.1109/tgrs.2011.2165957.
Little MA, McSharry PE, Roberts SJ, Costello DA, Moroz IM. 2007. Exploiting nonlinear
recurrence and fractal scaling properties for voice disorder detection. Biomedical Engineering
Online 6(1):23 DOI 10.1186/1475-925x-6-23.
Louie KS, De Sanjose S, Diaz M, Castellsague X, Herrero R, Meijer CJ, Shah K, Franceschi S,
Munoz N, Bosch FX. 2009. Early age at first sexual intercourse and early pregnancy are risk
factors for cervical cancer in developing countries. British Journal of Cancer 100(7):1191–1197
DOI 10.1038/sj.bjc.6604974.
Mangasarian OL, Street WN, Wolberg WH. 1995. Breast cancer diagnosis and prognosis via
linear programming. Operations Research 43(4):570–577 DOI 10.1287/opre.43.4.570.
Menke J, Martinez TR. 2004. Using permutations instead of student’s t distribution for p-values in
paired-difference algorithm comparisons. In: 2004 IEEE International Joint Conference on
Neural Networks, Proceedings. Vol. 2. Budapest: IEEE, 1331–1335.
Moore JH. 2004. Computational analysis of gene-gene interactions using multifactor
dimensionality reduction. Expert Review of Molecular Diagnostics 4(6):795–803
DOI 10.1586/14737159.4.6.795.
Peterson LE. 2009. K-nearest neighbor. Scholarpedia 4(2):1883 DOI 10.4249/scholarpedia.1883.
Pinoli P, Chicco D, Masseroli M. 2015. Computational algorithms to predict gene ontology
annotations. BMC Bioinformatics 16(Suppl. 6):S4 DOI 10.1186/1471-2105-16-s6-s4.
Plissiti ME, Nikou C. 2013. A review of automated techniques for cervical cell image analysis and
classification. In: Andreaus U, Iacoviello D, eds. Biomedical Imaging and Computational
Modeling in Biomechanics. Dordrecht: Springer, 1–18.
Quinlan JR. 1986. Induction of decision trees. Machine Learning 1(1):81–106
DOI 10.1007/bf00116251.
Ronneberger O, Fischer P, Brox T. 2015. U-net: convolutional networks for biomedical image
segmentation. In: International Conference on Medical Image Computing and Computer-Assisted
Intervention. Munich: Springer, 234–241.
Santos MS, Abreu PH, Garca-Laencina PJ, Simão A, Carvalho A. 2015. A new cluster-based
oversampling method for improving survival prediction of hepatocellular carcinoma patients.
Journal of Biomedical Informatics 58:49–59 DOI 10.1016/j.jbi.2015.09.012.
Scholkopf B, Sung K-K, Burges CJ, Girosi F, Niyogi P, Poggio T, Vapnik V. 1997. Comparing
support vector machines with Gaussian kernels to radial basis function classifiers. IEEE
Transactions on Signal Processing 45(11):2758–2765 DOI 10.1109/78.650102.
Smith JW, Everhart J, Dickson W, Knowler W, Johannes R. 1988. Using the ADAP learning
algorithm to forecast the onset of diabetes mellitus. In: Proceedings of the Annual Symposium on
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 19/20
20. Computer Application in Medical Care. New York: American Medical Informatics Association,
261.
Srivastava N, Hinton GE, Krizhevsky A, Sutskever I, Salakhutdinov R. 2014. Dropout: a simple
way to prevent neural networks from overfitting. Journal of Machine Learning Research
15(1):1929–1958.
Tieleman T, Hinton G. 2012. Lecture 6.5—rmsprop: divide the gradient by a running average of
its recent magnitude. Coursera: Neural Networks for Machine Learning 4(2):26–31.
University of California Irvine. 1987. Machine Learning Repository. Available at http://archive.ics.
uci.edu/ml/ (accessed 10 August 2017).
University of California Irvine Machine Learning Repository. 2017. Cervical Cancer (Risk
Factors) Data Set. Available at https://archive.ics.uci.edu/ml/datasets/Cervical+cancer+%28Risk
+Factors%29 (accessed 1 February 2018).
Van Der Maaten L, Hinton G. 2008. Visualizing data using t-SNE. Journal of Machine Learning
Research 9:2579–2605.
Van Der Maaten L, Postma E, Van den Herik J. 2009. Dimensionality reduction: a comparative.
Journal of Machine Learning Research 10:66–71.
Vincent P, Larochelle H, Bengio Y, Manzagol P-A. 2008. Extracting and composing robust
features with denoising autoencoders. In: Proceedings of ICML 2008. Helsinki: ACM,
1096–1103.
Xu T, Zhang H, Huang X, Zhang S, Metaxas DN. 2016. Multimodal deep learning for cervical
dysplasia diagnosis. In: International Conference on Medical Image Computing and Computer-
Assisted Intervention. Athens: Springer, 115–123.
Fernandes et al. (2018), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.154 20/20