This document describes a new algorithm for fully automatic brain tumor segmentation using 3D convolutional neural networks. The algorithm uses 3D convolutional filters to preserve spatial information, and a high-bias CNN architecture to increase effective data size and reduce model variance. On a dataset of 274 brain MR images, the algorithm achieved a median Dice score of 89% for whole tumor segmentation, significantly outperforming past methods. This demonstrates the effectiveness of generalizing low-bias high-variance methods like CNNs to learn from medium-sized datasets.
Performance analysis of automated brain tumor detection from MR imaging and CT scan using basic image processing techniques based on various hard and soft computing has been performed in our work. Moreover, we applied six traditional classifiers to detect brain tumor in the images. Then we applied CNN for brain tumor detection to include deep learning method in our work. We compared the result of the traditional one having the best accuracy (SVM) with the result of CNN. Furthermore, our work presents a generic method of tumor detection and extraction of its various features.
Brain Tumor Detection Using Artificial Neural Network Fuzzy Inference System ...Editor IJCATR
Manual classification of brain tumor is time devastating and bestows ambiguous results. Automatic image classification is
emergent thriving research area in medical field. In the proposed methodology, features are extracted from raw images which are then
fed to ANFIS (Artificial neural fuzzy inference system).ANFIS being neuro-fuzzy system harness power of both hence it proves to be
a sophisticated framework for multiobject classification. A comprehensive feature set and fuzzy rules are selected to classify an
abnormal image to the corresponding tumor type. This proposed technique is fast in execution, efficient in classification and easy in
implementation.
Brain tumor classification using artificial neural network on mri imageseSAT Journals
Abstract
In this paper, an attempt has been made to summarize the multi-resolution transformation and the different classifiers useful to
analyze the brain tumor using MRI. X-ray, MRI, Ultrasound etc. are different techniques used to scan brain tumor images.
Radiologist prefers MRI to get detail information about tumor to help him diagnoses. In this paper we have used MRI of brain
tumor for analysis. We have used Digital image processing tool for detection of the tumor. The identification, detection and
classification of brain tumor have been done by extracting features from MRI with the help of wavelet transformation. The MRI of
brain tumor is classified into two categories normal and abnormal brain. In this work Digital image processing has been used as
a tool for getting clear and exact details about tumor in earlier stages. This helps the physicians and practitioners for diagnoses.
Key word – Brain tumor, Wavelet transform, segmentation.
Brain tumour segmentation based on local independent projection based classif...eSAT Journals
Abstract
Brain tumour detection and segmentation is most important and challenging task in early tumour diagnosis. There are various
segmentation methods available but they are still challenging methods because of its complex characteristics such as ambiguous
boundaries and high diversity. To overcome this problem we are going to implement automatic brain tumour detection and
segmentation method by using local independent projection based classification. In this method we are going to consider tumour
segmentation as a classification problem. In this paper locality is important in calculations of projections. Also local anchor
embedding is used to solve linear projection weights. The softmax regression model is used to improve classification performance.
In this study we used MRI images as training and testing data. Finally the brain tumour is classified into tumour and edema
region. The area of tumour region is calculated in pixels.
Key Words: Brain tumour detection & segmentation, local independent projection based classification, local anchor
embedding and softmax regression.
Performance analysis of automated brain tumor detection from MR imaging and CT scan using basic image processing techniques based on various hard and soft computing has been performed in our work. Moreover, we applied six traditional classifiers to detect brain tumor in the images. Then we applied CNN for brain tumor detection to include deep learning method in our work. We compared the result of the traditional one having the best accuracy (SVM) with the result of CNN. Furthermore, our work presents a generic method of tumor detection and extraction of its various features.
Brain Tumor Detection Using Artificial Neural Network Fuzzy Inference System ...Editor IJCATR
Manual classification of brain tumor is time devastating and bestows ambiguous results. Automatic image classification is
emergent thriving research area in medical field. In the proposed methodology, features are extracted from raw images which are then
fed to ANFIS (Artificial neural fuzzy inference system).ANFIS being neuro-fuzzy system harness power of both hence it proves to be
a sophisticated framework for multiobject classification. A comprehensive feature set and fuzzy rules are selected to classify an
abnormal image to the corresponding tumor type. This proposed technique is fast in execution, efficient in classification and easy in
implementation.
Brain tumor classification using artificial neural network on mri imageseSAT Journals
Abstract
In this paper, an attempt has been made to summarize the multi-resolution transformation and the different classifiers useful to
analyze the brain tumor using MRI. X-ray, MRI, Ultrasound etc. are different techniques used to scan brain tumor images.
Radiologist prefers MRI to get detail information about tumor to help him diagnoses. In this paper we have used MRI of brain
tumor for analysis. We have used Digital image processing tool for detection of the tumor. The identification, detection and
classification of brain tumor have been done by extracting features from MRI with the help of wavelet transformation. The MRI of
brain tumor is classified into two categories normal and abnormal brain. In this work Digital image processing has been used as
a tool for getting clear and exact details about tumor in earlier stages. This helps the physicians and practitioners for diagnoses.
Key word – Brain tumor, Wavelet transform, segmentation.
Brain tumour segmentation based on local independent projection based classif...eSAT Journals
Abstract
Brain tumour detection and segmentation is most important and challenging task in early tumour diagnosis. There are various
segmentation methods available but they are still challenging methods because of its complex characteristics such as ambiguous
boundaries and high diversity. To overcome this problem we are going to implement automatic brain tumour detection and
segmentation method by using local independent projection based classification. In this method we are going to consider tumour
segmentation as a classification problem. In this paper locality is important in calculations of projections. Also local anchor
embedding is used to solve linear projection weights. The softmax regression model is used to improve classification performance.
In this study we used MRI images as training and testing data. Finally the brain tumour is classified into tumour and edema
region. The area of tumour region is calculated in pixels.
Key Words: Brain tumour detection & segmentation, local independent projection based classification, local anchor
embedding and softmax regression.
A Survey on Segmentation Techniques Used For Brain Tumor DetectionEditor IJMTER
In recent years Brain tumor is one of the most commonly found causes for death among
children and adults. Early detection of tumor is a must in order to reduce the death rate. For tumor
detection various image techniques can be used. In this paper we mainly concentrate on the images
obtained from MRI scans. In MRI images, the tumor may appear clearly, but for further treatment
the physician need to be a qualified and well experienced person. In order to help the radiologist in
detection computer-aided diagnosis was developed. The generation of a CAD system consists of
several processes and among them segmentation is considered to the most important process. Image
Segmentation is a process of partitioning an image into multiple segments. The main objective of
segmentation is to represent the image into a simplified form so as to increase the efficiency and
accuracy of the system. Therefore the segmentation of brain tumor can be considered as an important
role in the medical image process. Hence in this paper we concentrate on the recently used
segmentation techniques for the detection of tumor using MRI images.
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
Brain tumor detection and localization in magnetic resonance imagingijitcs
A tumor also known as neoplasm is a growth in the abnormal tissue which can be differentiated from the
surrounding tissue by its structure. A tumor may lead to cancer, which is a major leading cause of death and
responsible for around 13% of all deaths world-wide. Cancer incidence rate is growing at an alarming rate
in the world. Great knowledge and experience on radiology are required for accurate tumor detection in
medical imaging. Automation of tumor detection is required because there might be a shortage of skilled
radiologists at a time of great need. We propose an automatic brain tumor detectionand localization
framework that can detect and localize brain tumor in magnetic resonance imaging. The proposed brain
tumor detection and localization framework comprises five steps: image acquisition, pre-processing, edge
detection, modified histogram clustering and morphological operations. After morphological operations,
tumors appear as pure white color on pure black backgrounds. We used 50 neuroimages to optimize our
system and 100 out-of-sample neuroimages to test our system. The proposed tumor detection and localization
system was found to be able to accurately detect and localize brain tumor in magnetic resonance imaging.
The preliminary results demonstrate how a simple machine learning classifier with a set of simple
image-based features can result in high classification accuracy. The preliminary results also demonstrate the
efficacy and efficiency of our five-step brain tumor detection and localization approach and motivate us to
extend this framework to detect and localize a variety of other types of tumors in other types of medical
imagery.
DETECTING BRAIN TUMOUR FROM MRI IMAGE USING MATLAB GUI PROGRAMMEIJCSES Journal
Engineers have been actively developing tools to detect tumors and to process medical images. Medical image segmentation is a powerful tool that is often used to detect tumors. Many scientists and researchers are working to develop and add more features to this tool. This project is about detecting Brain tumors from MRI images using an interface of GUI in Matlab. Using the GUI, this program can use various combinations of segmentation, filters, and other image processing algorithms to achieve the best results.
We start with filtering the image using Prewitt horizontal edge-emphasizing filter. The next step for detecting tumor is "watershed pixels." The most important part of this project is that all the Matlab programs work with GUI “Matlab guide”. This allows us to use various combinations of filters, and other
image processing techniques to arrive at the best result that can help us detect brain tumors in their early stages.
Brain tumor detection and segmentation using watershed segmentation and morph...eSAT Publishing House
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
details about brain tumor
literature survey on many reference papers related to brain tumor detection using various techniques
our proposed novel methodology for brain tumor detection
BRAIN TUMOR MRI IMAGE SEGMENTATION AND DETECTION IN IMAGE PROCESSINGDharshika Shreeganesh
Image processing is an active research area in which medical image processing is a highly challenging field. Medical imaging
techniques are used to image the inner portions of the human body for medical diagnosis. Brain tumor is a serious life altering
disease condition. Image segmentation plays a significant role in image processing as it helps in the extraction of suspicious regions
from the medical images. In this paper we have proposed segmentation of brain MRI image using K-means clustering algorithm
followed by morphological filtering which avoids the misclustered regions that can inevitably be formed after segmentation of the brain MRI image for detection of tumor location.
A Survey on Segmentation Techniques Used For Brain Tumor DetectionEditor IJMTER
In recent years Brain tumor is one of the most commonly found causes for death among
children and adults. Early detection of tumor is a must in order to reduce the death rate. For tumor
detection various image techniques can be used. In this paper we mainly concentrate on the images
obtained from MRI scans. In MRI images, the tumor may appear clearly, but for further treatment
the physician need to be a qualified and well experienced person. In order to help the radiologist in
detection computer-aided diagnosis was developed. The generation of a CAD system consists of
several processes and among them segmentation is considered to the most important process. Image
Segmentation is a process of partitioning an image into multiple segments. The main objective of
segmentation is to represent the image into a simplified form so as to increase the efficiency and
accuracy of the system. Therefore the segmentation of brain tumor can be considered as an important
role in the medical image process. Hence in this paper we concentrate on the recently used
segmentation techniques for the detection of tumor using MRI images.
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
Brain tumor detection and localization in magnetic resonance imagingijitcs
A tumor also known as neoplasm is a growth in the abnormal tissue which can be differentiated from the
surrounding tissue by its structure. A tumor may lead to cancer, which is a major leading cause of death and
responsible for around 13% of all deaths world-wide. Cancer incidence rate is growing at an alarming rate
in the world. Great knowledge and experience on radiology are required for accurate tumor detection in
medical imaging. Automation of tumor detection is required because there might be a shortage of skilled
radiologists at a time of great need. We propose an automatic brain tumor detectionand localization
framework that can detect and localize brain tumor in magnetic resonance imaging. The proposed brain
tumor detection and localization framework comprises five steps: image acquisition, pre-processing, edge
detection, modified histogram clustering and morphological operations. After morphological operations,
tumors appear as pure white color on pure black backgrounds. We used 50 neuroimages to optimize our
system and 100 out-of-sample neuroimages to test our system. The proposed tumor detection and localization
system was found to be able to accurately detect and localize brain tumor in magnetic resonance imaging.
The preliminary results demonstrate how a simple machine learning classifier with a set of simple
image-based features can result in high classification accuracy. The preliminary results also demonstrate the
efficacy and efficiency of our five-step brain tumor detection and localization approach and motivate us to
extend this framework to detect and localize a variety of other types of tumors in other types of medical
imagery.
DETECTING BRAIN TUMOUR FROM MRI IMAGE USING MATLAB GUI PROGRAMMEIJCSES Journal
Engineers have been actively developing tools to detect tumors and to process medical images. Medical image segmentation is a powerful tool that is often used to detect tumors. Many scientists and researchers are working to develop and add more features to this tool. This project is about detecting Brain tumors from MRI images using an interface of GUI in Matlab. Using the GUI, this program can use various combinations of segmentation, filters, and other image processing algorithms to achieve the best results.
We start with filtering the image using Prewitt horizontal edge-emphasizing filter. The next step for detecting tumor is "watershed pixels." The most important part of this project is that all the Matlab programs work with GUI “Matlab guide”. This allows us to use various combinations of filters, and other
image processing techniques to arrive at the best result that can help us detect brain tumors in their early stages.
Brain tumor detection and segmentation using watershed segmentation and morph...eSAT Publishing House
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
details about brain tumor
literature survey on many reference papers related to brain tumor detection using various techniques
our proposed novel methodology for brain tumor detection
BRAIN TUMOR MRI IMAGE SEGMENTATION AND DETECTION IN IMAGE PROCESSINGDharshika Shreeganesh
Image processing is an active research area in which medical image processing is a highly challenging field. Medical imaging
techniques are used to image the inner portions of the human body for medical diagnosis. Brain tumor is a serious life altering
disease condition. Image segmentation plays a significant role in image processing as it helps in the extraction of suspicious regions
from the medical images. In this paper we have proposed segmentation of brain MRI image using K-means clustering algorithm
followed by morphological filtering which avoids the misclustered regions that can inevitably be formed after segmentation of the brain MRI image for detection of tumor location.
Lung Cancer Detection using transfer learning.pptx.pdfjagan477830
Lung cancer is one of the deadliest cancers worldwide. However, the early detection of lung cancer significantly improves survival rate. Cancerous (malignant) and noncancerous (benign) pulmonary nodules are the small growths of cells inside the lung. Detection of malignant lung nodules at an early stage is necessary for the crucial prognosis.
AN ANN BASED BRAIN ABNORMALITY DETECTION USING MR IMAGEScscpconf
The Main purpose of this paper is to design, implement and evaluate a strong automatic diagnostic system that increases the accuracy of tumor diagnosis in brain using MR images.This presented work classifies the brain tissues as normal or abnormal automatically, usingcomputer vision. This saves lot of radiologist time to carryout monotonous repeated job. The
acquired MR images are processed using image preprocessing techniques. The preprocessed images are then segmented, and the various features are extracted. The extracted features are
fed to the artificial neural network as input that trains the network using error back propagation algorithm for correct decision making.
International Journal of Research in Engineering and Science is an open access peer-reviewed international forum for scientists involved in research to publish quality and refereed papers. Papers reporting original research or experimentally proved review work are welcome. Papers for publication are selected through peer review to ensure originality, relevance, and readability.
USING SINGULAR VALUE DECOMPOSITION IN A CONVOLUTIONAL NEURAL NETWORK TO IMPRO...ijcsit
A brain tumor consists of cells showing abnormal brain growth. The area of the brain tumor significantly
affects choosing the type of treatment and following the course of the disease during the treatment. At the
same time, pictures of Brain MRIs are accompanied by noise. Eliminating existing noises can significantly
impact the better segmentation and diagnosis of brain tumors. In this work, we have tried using the
analysis of eigenvalues. We have used the MSVD algorithm, reducing the image noise and then using the
deep neural network to segment the tumor in the images. The proposed method's accuracy was increased
by 2.4% compared to using the original images. With Using the MSVD method, convergence speed has
also increased, showing the proposed method's effectiveness.
A brain tumor consists of cells showing abnormal brain growth. The area of the brain tumor significantly
affects choosing the type of treatment and following the course of the disease during the treatment. At the
same time, pictures of Brain MRIs are accompanied by noise. Eliminating existing noises can significantly
impact the better segmentation and diagnosis of brain tumors. In this work, we have tried using the
analysis of eigenvalues. We have used the MSVD algorithm, reducing the image noise and then using the
deep neural network to segment the tumor in the images. The proposed method's accuracy was increased
by 2.4% compared to using the original images. With Using the MSVD method, convergence speed has
also increased, showing the proposed method's effectiveness
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.
Comparison of Image Segmentation Algorithms for Brain Tumor DetectionIJMTST Journal
This paper deals with the implementation of Simple Algorithms for detection of size and shape of tumor in brain using MRI images. Generally, CT scan or MRI that is directed into intracranial cavity produces a complete image of brain. This image is visually examined by the physician for detection & diagnosis of brain tumor. However this method of detection resists the accurate determination of stage & size of tumor. To avoid that, this project uses computer aided method for segmentation (detection) of brain tumor by applying Fuzzy C-Means, K-Means, Gaussian Kernel and Pillar K-means algorithms. This segmentation process includes a new mechanism for clustering the elements of high-resolution images in order to improve precision and reduce computation time. The system applies FCM, Gaussian kernel and K-means clustering to the image later optimized by Pillar Algorithm. It designates the initial centroids’ positions by calculating the Euclidian distance metric between each data point and all previous centroids. Then it selects data points which have the maximum distance as new initial centroids. This algorithm distributes all initial centroids according to the maximum accumulated distance metric. In addition, it also reduces the time for analysis. At the end of the process the tumor is extracted from the MRI image and its exact position and the shape is also determined. This paper evaluates the proposed approach for Brain tumor detection by comparing with K-means, Fuzzy C means, Gaussian Kernel and manually segmented algorithms. The experimental results clarify the effectiveness of proposed approach to improve the segmentation quality in aspects of precision and computational time.
The biomedical profession has gained importance due to the rapid and accurate diagnosis of clinical patients using computer-aided diagnosis (CAD) tools.
The diagnosis and treatment of Alzheimer’s disease (AD) using complementary multimodalities can improve the quality of life and mental state of patients.
In this study, we integrated a lightweight custom convolutional neural network
(CNN) model and nature-inspired optimization techniques to enhance the performance, robustness, and stability of progress detection in AD. A multi-modal
fusion database approach was implemented, including positron emission tomography (PET) and magnetic resonance imaging (MRI) datasets, to create a fused
database. We compared the performance of custom and pre-trained deep learning models with and without optimization and found that employing natureinspired algorithms like the particle swarm optimization algorithm (PSO) algorithm significantly improved system performance. The proposed methodology,
which includes a fused multimodality database and optimization strategy, improved performance metrics such as training, validation, test accuracy, precision, and recall. Furthermore, PSO was found to improve the performance of
pre-trained models by 3-5% and custom models by up to 22%. Combining different medical imaging modalities improved the overall model performance by
2-5%. In conclusion, a customized lightweight CNN model and nature-inspired
optimization techniques can significantly enhance progress detection, leading to
better biomedical research and patient care.
A Dualistic Sub-Image Histogram Equalization Based Enhancement and Segmentati...inventy
Research Inventy : International Journal of Engineering and Science is published by the group of young academic and industrial researchers with 12 Issues per year. It is an online as well as print version open access journal that provides rapid publication (monthly) of articles in all areas of the subject such as: civil, mechanical, chemical, electronic and computer engineering as well as production and information technology. The Journal welcomes the submission of manuscripts that meet the general criteria of significance and scientific excellence. Papers will be published by rapid process within 20 days after acceptance and peer review process takes only 7 days. All articles published in Research Inventy will be peer-reviewed.
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...kevinkariuki227
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Verified Chapters 1 - 19, Complete Newest Version.pdf
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Verified Chapters 1 - 19, Complete Newest Version.pdf
Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...Oleg Kshivets
RESULTS: Overall life span (LS) was 2252.1±1742.5 days and cumulative 5-year survival (5YS) reached 73.2%, 10 years – 64.8%, 20 years – 42.5%. 513 LCP lived more than 5 years (LS=3124.6±1525.6 days), 148 LCP – more than 10 years (LS=5054.4±1504.1 days).199 LCP died because of LC (LS=562.7±374.5 days). 5YS of LCP after bi/lobectomies was significantly superior in comparison with LCP after pneumonectomies (78.1% vs.63.7%, P=0.00001 by log-rank test). AT significantly improved 5YS (66.3% vs. 34.8%) (P=0.00000 by log-rank test) only for LCP with N1-2. Cox modeling displayed that 5YS of LCP significantly depended on: phase transition (PT) early-invasive LC in terms of synergetics, PT N0—N12, cell ratio factors (ratio between cancer cells- CC and blood cells subpopulations), G1-3, histology, glucose, AT, blood cell circuit, prothrombin index, heparin tolerance, recalcification time (P=0.000-0.038). Neural networks, genetic algorithm selection and bootstrap simulation revealed relationships between 5YS and PT early-invasive LC (rank=1), PT N0—N12 (rank=2), thrombocytes/CC (3), erythrocytes/CC (4), eosinophils/CC (5), healthy cells/CC (6), lymphocytes/CC (7), segmented neutrophils/CC (8), stick neutrophils/CC (9), monocytes/CC (10); leucocytes/CC (11). Correct prediction of 5YS was 100% by neural networks computing (area under ROC curve=1.0; error=0.0).
CONCLUSIONS: 5YS of LCP after radical procedures significantly depended on: 1) PT early-invasive cancer; 2) PT N0--N12; 3) cell ratio factors; 4) blood cell circuit; 5) biochemical factors; 6) hemostasis system; 7) AT; 8) LC characteristics; 9) LC cell dynamics; 10) surgery type: lobectomy/pneumonectomy; 11) anthropometric data. Optimal diagnosis and treatment strategies for LC are: 1) screening and early detection of LC; 2) availability of experienced thoracic surgeons because of complexity of radical procedures; 3) aggressive en block surgery and adequate lymph node dissection for completeness; 4) precise prediction; 5) adjuvant chemoimmunoradiotherapy for LCP with unfavorable prognosis.
Title: Sense of Taste
Presenter: Dr. Faiza, Assistant Professor of Physiology
Qualifications:
MBBS (Best Graduate, AIMC Lahore)
FCPS Physiology
ICMT, CHPE, DHPE (STMU)
MPH (GC University, Faisalabad)
MBA (Virtual University of Pakistan)
Learning Objectives:
Describe the structure and function of taste buds.
Describe the relationship between the taste threshold and taste index of common substances.
Explain the chemical basis and signal transduction of taste perception for each type of primary taste sensation.
Recognize different abnormalities of taste perception and their causes.
Key Topics:
Significance of Taste Sensation:
Differentiation between pleasant and harmful food
Influence on behavior
Selection of food based on metabolic needs
Receptors of Taste:
Taste buds on the tongue
Influence of sense of smell, texture of food, and pain stimulation (e.g., by pepper)
Primary and Secondary Taste Sensations:
Primary taste sensations: Sweet, Sour, Salty, Bitter, Umami
Chemical basis and signal transduction mechanisms for each taste
Taste Threshold and Index:
Taste threshold values for Sweet (sucrose), Salty (NaCl), Sour (HCl), and Bitter (Quinine)
Taste index relationship: Inversely proportional to taste threshold
Taste Blindness:
Inability to taste certain substances, particularly thiourea compounds
Example: Phenylthiocarbamide
Structure and Function of Taste Buds:
Composition: Epithelial cells, Sustentacular/Supporting cells, Taste cells, Basal cells
Features: Taste pores, Taste hairs/microvilli, and Taste nerve fibers
Location of Taste Buds:
Found in papillae of the tongue (Fungiform, Circumvallate, Foliate)
Also present on the palate, tonsillar pillars, epiglottis, and proximal esophagus
Mechanism of Taste Stimulation:
Interaction of taste substances with receptors on microvilli
Signal transduction pathways for Umami, Sweet, Bitter, Sour, and Salty tastes
Taste Sensitivity and Adaptation:
Decrease in sensitivity with age
Rapid adaptation of taste sensation
Role of Saliva in Taste:
Dissolution of tastants to reach receptors
Washing away the stimulus
Taste Preferences and Aversions:
Mechanisms behind taste preference and aversion
Influence of receptors and neural pathways
Impact of Sensory Nerve Damage:
Degeneration of taste buds if the sensory nerve fiber is cut
Abnormalities of Taste Detection:
Conditions: Ageusia, Hypogeusia, Dysgeusia (parageusia)
Causes: Nerve damage, neurological disorders, infections, poor oral hygiene, adverse drug effects, deficiencies, aging, tobacco use, altered neurotransmitter levels
Neurotransmitters and Taste Threshold:
Effects of serotonin (5-HT) and norepinephrine (NE) on taste sensitivity
Supertasters:
25% of the population with heightened sensitivity to taste, especially bitterness
Increased number of fungiform papillae
ARTIFICIAL INTELLIGENCE IN HEALTHCARE.pdfAnujkumaranit
Artificial intelligence (AI) refers to the simulation of human intelligence processes by machines, especially computer systems. It encompasses tasks such as learning, reasoning, problem-solving, perception, and language understanding. AI technologies are revolutionizing various fields, from healthcare to finance, by enabling machines to perform tasks that typically require human intelligence.
HOT NEW PRODUCT! BIG SALES FAST SHIPPING NOW FROM CHINA!! EU KU DB BK substit...GL Anaacs
Contact us if you are interested:
Email / Skype : kefaya1771@gmail.com
Threema: PXHY5PDH
New BATCH Ku !!! MUCH IN DEMAND FAST SALE EVERY BATCH HAPPY GOOD EFFECT BIG BATCH !
Contact me on Threema or skype to start big business!!
Hot-sale products:
NEW HOT EUTYLONE WHITE CRYSTAL!!
5cl-adba precursor (semi finished )
5cl-adba raw materials
ADBB precursor (semi finished )
ADBB raw materials
APVP powder
5fadb/4f-adb
Jwh018 / Jwh210
Eutylone crystal
Protonitazene (hydrochloride) CAS: 119276-01-6
Flubrotizolam CAS: 57801-95-3
Metonitazene CAS: 14680-51-4
Payment terms: Western Union,MoneyGram,Bitcoin or USDT.
Deliver Time: Usually 7-15days
Shipping method: FedEx, TNT, DHL,UPS etc.Our deliveries are 100% safe, fast, reliable and discreet.
Samples will be sent for your evaluation!If you are interested in, please contact me, let's talk details.
We specializes in exporting high quality Research chemical, medical intermediate, Pharmaceutical chemicals and so on. Products are exported to USA, Canada, France, Korea, Japan,Russia, Southeast Asia and other countries.
Prix Galien International 2024 Forum ProgramLevi Shapiro
June 20, 2024, Prix Galien International and Jerusalem Ethics Forum in ROME. Detailed agenda including panels:
- ADVANCES IN CARDIOLOGY: A NEW PARADIGM IS COMING
- WOMEN’S HEALTH: FERTILITY PRESERVATION
- WHAT’S NEW IN THE TREATMENT OF INFECTIOUS,
ONCOLOGICAL AND INFLAMMATORY SKIN DISEASES?
- ARTIFICIAL INTELLIGENCE AND ETHICS
- GENE THERAPY
- BEYOND BORDERS: GLOBAL INITIATIVES FOR DEMOCRATIZING LIFE SCIENCE TECHNOLOGIES AND PROMOTING ACCESS TO HEALTHCARE
- ETHICAL CHALLENGES IN LIFE SCIENCES
- Prix Galien International Awards Ceremony
The prostate is an exocrine gland of the male mammalian reproductive system
It is a walnut-sized gland that forms part of the male reproductive system and is located in front of the rectum and just below the urinary bladder
Function is to store and secrete a clear, slightly alkaline fluid that constitutes 10-30% of the volume of the seminal fluid that along with the spermatozoa, constitutes semen
A healthy human prostate measures (4cm-vertical, by 3cm-horizontal, 2cm ant-post ).
It surrounds the urethra just below the urinary bladder. It has anterior, median, posterior and two lateral lobes
It’s work is regulated by androgens which are responsible for male sex characteristics
Generalised disease of the prostate due to hormonal derangement which leads to non malignant enlargement of the gland (increase in the number of epithelial cells and stromal tissue)to cause compression of the urethra leading to symptoms (LUTS
Ozempic: Preoperative Management of Patients on GLP-1 Receptor Agonists Saeid Safari
Preoperative Management of Patients on GLP-1 Receptor Agonists like Ozempic and Semiglutide
ASA GUIDELINE
NYSORA Guideline
2 Case Reports of Gastric Ultrasound
Recomendações da OMS sobre cuidados maternos e neonatais para uma experiência pós-natal positiva.
Em consonância com os ODS – Objetivos do Desenvolvimento Sustentável e a Estratégia Global para a Saúde das Mulheres, Crianças e Adolescentes, e aplicando uma abordagem baseada nos direitos humanos, os esforços de cuidados pós-natais devem expandir-se para além da cobertura e da simples sobrevivência, de modo a incluir cuidados de qualidade.
Estas diretrizes visam melhorar a qualidade dos cuidados pós-natais essenciais e de rotina prestados às mulheres e aos recém-nascidos, com o objetivo final de melhorar a saúde e o bem-estar materno e neonatal.
Uma “experiência pós-natal positiva” é um resultado importante para todas as mulheres que dão à luz e para os seus recém-nascidos, estabelecendo as bases para a melhoria da saúde e do bem-estar a curto e longo prazo. Uma experiência pós-natal positiva é definida como aquela em que as mulheres, pessoas que gestam, os recém-nascidos, os casais, os pais, os cuidadores e as famílias recebem informação consistente, garantia e apoio de profissionais de saúde motivados; e onde um sistema de saúde flexível e com recursos reconheça as necessidades das mulheres e dos bebês e respeite o seu contexto cultural.
Estas diretrizes consolidadas apresentam algumas recomendações novas e já bem fundamentadas sobre cuidados pós-natais de rotina para mulheres e neonatos que recebem cuidados no pós-parto em unidades de saúde ou na comunidade, independentemente dos recursos disponíveis.
É fornecido um conjunto abrangente de recomendações para cuidados durante o período puerperal, com ênfase nos cuidados essenciais que todas as mulheres e recém-nascidos devem receber, e com a devida atenção à qualidade dos cuidados; isto é, a entrega e a experiência do cuidado recebido. Estas diretrizes atualizam e ampliam as recomendações da OMS de 2014 sobre cuidados pós-natais da mãe e do recém-nascido e complementam as atuais diretrizes da OMS sobre a gestão de complicações pós-natais.
O estabelecimento da amamentação e o manejo das principais intercorrências é contemplada.
Recomendamos muito.
Vamos discutir essas recomendações no nosso curso de pós-graduação em Aleitamento no Instituto Ciclos.
Esta publicação só está disponível em inglês até o momento.
Prof. Marcus Renato de Carvalho
www.agostodourado.com
Acute scrotum is a general term referring to an emergency condition affecting the contents or the wall of the scrotum.
There are a number of conditions that present acutely, predominantly with pain and/or swelling
A careful and detailed history and examination, and in some cases, investigations allow differentiation between these diagnoses. A prompt diagnosis is essential as the patient may require urgent surgical intervention
Testicular torsion refers to twisting of the spermatic cord, causing ischaemia of the testicle.
Testicular torsion results from inadequate fixation of the testis to the tunica vaginalis producing ischemia from reduced arterial inflow and venous outflow obstruction.
The prevalence of testicular torsion in adult patients hospitalized with acute scrotal pain is approximately 25 to 50 percent
New Directions in Targeted Therapeutic Approaches for Older Adults With Mantl...i3 Health
i3 Health is pleased to make the speaker slides from this activity available for use as a non-accredited self-study or teaching resource.
This slide deck presented by Dr. Kami Maddocks, Professor-Clinical in the Division of Hematology and
Associate Division Director for Ambulatory Operations
The Ohio State University Comprehensive Cancer Center, will provide insight into new directions in targeted therapeutic approaches for older adults with mantle cell lymphoma.
STATEMENT OF NEED
Mantle cell lymphoma (MCL) is a rare, aggressive B-cell non-Hodgkin lymphoma (NHL) accounting for 5% to 7% of all lymphomas. Its prognosis ranges from indolent disease that does not require treatment for years to very aggressive disease, which is associated with poor survival (Silkenstedt et al, 2021). Typically, MCL is diagnosed at advanced stage and in older patients who cannot tolerate intensive therapy (NCCN, 2022). Although recent advances have slightly increased remission rates, recurrence and relapse remain very common, leading to a median overall survival between 3 and 6 years (LLS, 2021). Though there are several effective options, progress is still needed towards establishing an accepted frontline approach for MCL (Castellino et al, 2022). Treatment selection and management of MCL are complicated by the heterogeneity of prognosis, advanced age and comorbidities of patients, and lack of an established standard approach for treatment, making it vital that clinicians be familiar with the latest research and advances in this area. In this activity chaired by Michael Wang, MD, Professor in the Department of Lymphoma & Myeloma at MD Anderson Cancer Center, expert faculty will discuss prognostic factors informing treatment, the promising results of recent trials in new therapeutic approaches, and the implications of treatment resistance in therapeutic selection for MCL.
Target Audience
Hematology/oncology fellows, attending faculty, and other health care professionals involved in the treatment of patients with mantle cell lymphoma (MCL).
Learning Objectives
1.) Identify clinical and biological prognostic factors that can guide treatment decision making for older adults with MCL
2.) Evaluate emerging data on targeted therapeutic approaches for treatment-naive and relapsed/refractory MCL and their applicability to older adults
3.) Assess mechanisms of resistance to targeted therapies for MCL and their implications for treatment selection
Ethanol (CH3CH2OH), or beverage alcohol, is a two-carbon alcohol
that is rapidly distributed in the body and brain. Ethanol alters many
neurochemical systems and has rewarding and addictive properties. It
is the oldest recreational drug and likely contributes to more morbidity,
mortality, and public health costs than all illicit drugs combined. The
5th edition of the Diagnostic and Statistical Manual of Mental Disorders
(DSM-5) integrates alcohol abuse and alcohol dependence into a single
disorder called alcohol use disorder (AUD), with mild, moderate,
and severe subclassifications (American Psychiatric Association, 2013).
In the DSM-5, all types of substance abuse and dependence have been
combined into a single substance use disorder (SUD) on a continuum
from mild to severe. A diagnosis of AUD requires that at least two of
the 11 DSM-5 behaviors be present within a 12-month period (mild
AUD: 2–3 criteria; moderate AUD: 4–5 criteria; severe AUD: 6–11 criteria).
The four main behavioral effects of AUD are impaired control over
drinking, negative social consequences, risky use, and altered physiological
effects (tolerance, withdrawal). This chapter presents an overview
of the prevalence and harmful consequences of AUD in the U.S.,
the systemic nature of the disease, neurocircuitry and stages of AUD,
comorbidities, fetal alcohol spectrum disorders, genetic risk factors, and
pharmacotherapies for AUD.
A New Algorithm for Fully Automatic Brain Tumor Segmentation with 3-D Convolutional Neural Networks
1. A New Algorithm for Fully Automatic Brain Tumor
Segmentation with 3-D Convolutional Neural Networks
Christopher Elamri
Stanford University
mcelamri@stanford.edu
Teun de Planque
Stanford University
teun@stanford.edu
Abstract
Glioblastoma (GBM) is the most common primary tumor
of the central nervous system in the United States, with 3 in
100,000 people diagnosed every year [1]. In this paper,
we present a new algorithm for fully automatic brain tu-
mor segmentation based on 3-D Convolutional Neural Net-
works. Segmentation of the GBM region of the surround-
ing brain makes it easier to access the image data within
MR scans, and consequently can help us better understand
GBMs. Most methods for volumetric image data use 2-
D convolutions; we present a true generalization of CNNs
to 3-D filters to preserve spatial information and increase
robustness. In addition, we present a relatively high bias
CNN architecture that enables us to both expand the effec-
tive data size and reduce the variance of our model. Our
median Dice score accuracy is around 89 percent in the
whole tumor segmentation. This result represents a signif-
icant improvement over past algorithms and demonstrates
the power of our approach in generalizing low-bias high-
variance methods like CNNs to learn from medium-size data
sets.
1. Introduction
Glioblastoma (GBM) is the most prevalent and most ag-
gressive type of brain cancer [2]. As a result of its resistance
to existing therapeutic methods, the most common length
of survival following diagnosis is 12 to 15 months, and less
than 3-5% of patients survive more than five years [1][3].
Much is still being done to understand GBM and its two
subtypes, high grade glioma (HGG) and low grade glioma
(LGG), but there is still a gold mine of untapped data. Chief
amongst these is imaging data in the form of magnetic res-
onance (MR) scans. Most methods of analyzing and ex-
tracting quantitative information from this imaging data re-
quires some form of segmentation of the GBM, and bet-
ter yet, classification of the tumor into four sub-categories:
necrosis, edema, non-enhancing tumor, and enhancing tu-
mor. To date, the gold standard of segmentation is still hu-
man radiologist segmentation. However, with the pure size
of imaging data being accrued, manual segmentation of all
images is no longer a sustainable system.
Nonetheless, existing automated segmentation algo-
rithms are based on high-bias learning algorithms and have
failed to give substantial improvement in classification ac-
curacy as described by Subbanna et al., Menze et al., and
Bauer et al. [4]. As a result, these high-bias systems are in-
creasingly being replaced with data-driven low-bias models.
The development of low-bias systems such as convolutional
neural networks for medical purposes is, nevertheless, chal-
lenging due to their high-variance, as many medical datasets
are limited in size.
In this paper, we propose a novel approach to automatic
brain tumor segmentation. We take as input a 3-D MR im-
age and output one of five labels (normal or one of 4 tu-
mor subtypes) for each voxel (the 3-D analog of pixel).
Our new approach surpasses current methods by reducing
the variance of CNNs applied to medium datasets. First,
where most methods use 2-D convolutions for 3-D data, we
introduce a CNN that uses 3-D filters, made computation-
ally efficient by the transformation to Fourier space. Con-
sequently, our system is more robust and minimizes loss of
spatial information. Second, we present a novel CNN archi-
tecture which differs from those traditionally used in brain
segmentation and computer vision [4][5]; our CNN archi-
tecture uses a higher-bias system than CNNs using pixel-
wise neural networks that enables us to both expand the
effective data size from patient number to total pixel and
reduce the variance of CNNs. Our algorithm for GBM seg-
mentation and characterization exploits the CNNs frame-
work and the structure of medical imaging data to train a
model that not only is highly robust but also generalizes
data-driven low-bias systems to medium data while mini-
mizing variance.
1
2. 2. Problem Statement
Our problem amounts to efficiently classifying brain
voxels into five categories: non-tumor, necrosis, edema,
non-enhancing, and enhancing. The input to our system
consists of a preprocessed 3-D image of a brain. Each 3-
D image is of size 240 × 240 × 155 × 4, where the depth
4 represents the four modalities of our input images. The
goal is to for each voxel correctly and efficiently output its
category (non-tumor, necrosis, edema, non-enhancing, and
enhancing).
2.1. Dataset
We use the data provided by the MICCAI Multimodal
Brain Tumor Image Segmentation Challenge (BRATS) [4].
The 2015 BRATS dataset contains 274 patients: 220 pa-
tients with high grade glioblastomas and 54 patients with
low grade glioblastomas. For each patient our data contains
four MR modalities: T1 post-contrast, T1 pre-contrast, T2
weighted, and FLAIR. In addition, the data contains an ex-
pert segmentation for each patient that we will treat as the
ground truth. The expert segmentation was made by pool-
ing together the segmentation of eleven different radiolo-
gists and includes pixel-wide labeling as one of five cat-
egories: non-tumor, necrosis, edema, non-enhancing, en-
hancing.
All the data has been preprocessed, involving skull strip-
ping, co-registration of the images, and interpolation of the
images so that all images are of the same size (240 × 240 ×
155 pixels). We were not given any information about the
spatial scale of the pixels; and thus, all units for our current
method will be in the given pixel-space.
3. Technical Approach
3.1. Convolutional Neural Network Motivation
Convolutional neural networks (CNNs) are powerful
low-bias systems in deep learning with remarkable perfor-
mance in complex tasks ranging from facial recognition to
image classification [7][8]. CNNs use 2D convolutions as
filters for feature learning and feed the results of these con-
volution layers into a fully connected neural network. Our
algorithm builds on the idea of a CNN as a fully connected
neural network trained on sequential convolution layers to
improve accuracy and computational efficiency in learning
from 3D imaging data. Specifically, we generalize the idea
of 2D convolutions to 3D filters that retain the full spatial
information of the original stack of images. We then decou-
ple the pixels when providing the training data for a fully
connected neural network, thus increasing the size of our
training set and providing a computational speedup.
(a) Modalities (b) Labels
(c) Modalities (d) 3D-Rendering of Tumor
Figure 1: Data Visualization. (a) All four modalities are
co-registered. The modalities are ul: T1-pre, ur: T1-post,
dl: T2W, dr: FLAIR. (b) Corresponds to (a). Better visual-
ization of the four subregions of the tumor (Y: necrosis, G:
Edema, B: Non-Enhancing, P: Enhancing). (c) Visualiza-
tion of the colored labels on top of the modality images for
a different patient. (ul: T1-pre, ur: T1-post, dl: T2W, dr:
FLAIR) (d) 3-D rendering of the tumor within the brain.
3.2. 3-D Convolutions
We can reduce the bias of our system by considering the
input image as a 3-dimensional space of pixels. Expanding
our convolution from our usual 2-dimensional convolutions
into 3-dimensional convolutions, we get the following equa-
tion:
(I ∗ f)[x, y, z] =
240
τx=1
240
τy=1
155
τz=1
I[τx, τy, τz] ˙f[x − τx, y − τy, z − τz]
where I is the 3-dimensional image and f is the 3-
dimensional filter. Given an image of size n × n × n and
a filter of size m × m × m the time complexity of the con-
volution operation is O(m3
n3
). We can reduce the time
complexity of the convolution operation to O(n3
log(n)) by
implementing our convolution as an element-wise multipli-
cation in Fourier space. We use a 3-dimensional Difference
of Gaussian filter:
DoG(σ) =
1
(2πσ2)3/2
exp
−x2 + y2 + z2
2σ2
−
1
(πσ2)3/2
exp
−x2 + y2 + z2
σ2
Each filter is a difference of two 3-dimensional Gaussians
(normalized) with scales
√
2σ and σ. We create 8 of these
filters with the scales [
√
2, 2, 2
√
2, ..., 16]. Difference of
2
3. Figure 2: Visualization of 3-D Convolution. Similar to the
2-dimensional equivalent, in a 3-dimensional convolution,
we create an output image from the dot-product of our 3-
dimensional filter and a sub-image centered at every pixel.
We represent the filter with the red-box, our image with blue
box, and the convolutional product with our white box
Gaussian is an effective blob detector since it is rotationally
symmetric.
To create our final convolutional products, we will con-
sider all four modalities, and the magnitude of the gradient
of those images. In addition to the eight filter products, we
will include the original pixel intensity and the magnitude
of the gradient value. Thus, for each of the four modalities,
we will get an expansion into 18 feature images, giving us a
total feature space of 72-dimensions for each of our pixels.
We choose a filter of size 33 × 33 × 33. Basically, we get
for each voxel [x, y, z] the following set of features:
• 1 voxel intensity V [x, y, z]
• 1 voxel intensify gradient V [x, y, z]
• 8 DoG convolutions (V ∗ DoG)[x, y, z]
• 8 DoG convolutions in gradient space ( V ∗
DoG)[x, y, z]
3.3. 3-D Convolutional Neural Network Architec-
ture
Our input consists of a 3-dimensional image with four
channels (one for each modality). Our first layer is a non-
trained convolutional layer with 72 240×240×155×4 fil-
ters, 3-dimensionally convolved over all channels with our
input image. In this layer, our filters are the hand-selected
Difference of Gaussian filters for a specified modality with
all other values in the other three channels being zero. Our
training does not go through this layer.
The following convolutional layers use convolutions of
filters sized 1 × 1 × 1 over the number of channels in the
preceding layer (either 72 for the first layer or 100 for all
Figure 3: Main Methodology of Learning Pipeline. We
start with our four modalities of input images. To that,
we will convolve our 72 sparse 3-dimensional Difference
of Gaussian filters to get our first convolution layer. From
that point on, we will be convolving a trivial 1 × 1 × 1 (so a
scalar) filter over all channels to get the following convolu-
tional layer. The convolution of a 1 × 1 × 1 filter essentially
decouples each pixels information from each other after the
initial layers convolution. On the last layer, we will get five
channels, each channel giving us a probability score for ev-
ery pixel in our image for each of the five categories.
other layers). The 1×1×1 filter is a scalar that we will con-
stantly multiply over the whole image. In a sense, by only
wiring the convolution layers onward with these 1 × 1 × 1
filters, we are decoupling all of the pixels in our image. Be-
cause the first convolution layer couples neighborhood in-
formation via the 33×33×33 convolutions, all subsequent
layers pixels will maintain some level of information of its
original neighborhood. Thus, we are able to decrease the
number of weights we would have to train while drastically
increasing our effective training data size.
On the last layer, we will get five channels, each channel
giving us a probability score for every pixel in our image
along the five sub-regions (0 = non-tumor, 1 = necrosis, 2 =
edema, 3 = non-enhancing, and 4 = enhancing). We use the
softmax function as the loss function. To train the convolu-
tional neural network we use stochastic gradient descent by
patient. During test time, we classify using a voting met-
ric that assigns a voxel to one of the five categories based
on its classification scores (see algorithm 1). This method
is equivalent to how the expert segmentations were pooled
from 11 different radiologist’s segmentations.
Essentially, our classification priority in descending or-
der goes: enhancing, necrosis, non-enhancing, and edema.
If our neural nets returns positive classifications for multiple
tumor subtypes, we classify to the positive subtype with the
3
4. highest priority. This hierarchical design is based off of the
hierarchical majority vote used to combine several different
algorithmic results [4].
This seemingly arbitrary methodology makes perfect
sense in the context of our classification problem. Tumor
segmentations are judged generally in terms of three accu-
racies: whole tumor accuracy, tumor core accuracy, and en-
hancing tumor accuracy. Thus, because they have their own
accuracy scores, we must prioritize classification of the core
over the non-core (edema), and then also the enhancing core
over the other core. The enhancing core generally covers a
smaller area of the brain, which lends even more reason to
be more sensitive to its detection.
Results are reported as the standard Dice score calculated
via 10-fold cross validation (see: beginning of next section).
We do not use cross validation to select parameters, decid-
ing to keep our neural net parameters set to default values.
This is both because the additional computation time would
be prohibitive, and also because our Dice scores (which are
also calculated from the cross validation) would become bi-
ased upwards.
Algorithm 1 Hierarchical majority decision Our neural
network outputs a number between 0 and 1 per voxel for
each of the four tumor structures (edema, non-enhancing,
necrosis, enhancing), respectively indicated by pedema,
pnonenh, pnecrosis, penh
1: procedure HIERARCHICALDECISION(Score)
2: label ← NormalTissue
3: if pedema >= 0.5 then
4: label ← Edema
5: end if
6: if pnonenh >= 0.5 then
7: label ← Non − Enhancing
8: end if
9: if pnecrosis >= 0.5 then
10: label ← Necrosis
11: end if
12: if penh >= 0.5 then
13: label ← Enhancing
14: end if
15: end procedure
4. Results
We use Dice Coefficient to compute our results (also
known as the Sørensen-Dice Index) [9].
DiceScore =
2|Pred ∩ Ref|
|Pred| + |Ref|
where Pred is the set of voxels designated as the region of
interest in the prediction and Ref is the set of voxels desig-
Figure 4: Performance Comparison. Comparison of our
3D CNN method to the leading methods of both 2013, 2014,
radiologists, and our own 2D implementation of our algo-
rithm. The main metric used for our accuracy is Dice Score.
Whole = (Edema, Non-enhancing, Necrosis, Enhancing)
Figure 5: Histogram of Dice Score Accuracies
nated as the region of interest in the reference for compari-
son (in our case, the expert segmentation).
The median Dice performance on the whole tumor de-
tection is 89%! To wit, the inter-radiologist repeatability
is only 85%, so our accuracy has saturated with respect to
the ground truth. Figure 4 shows a comparison amongst the
leading algorithms in the literature. The first row shows the
performance of human radiologists. The full distribution of
accuracy of the algorithm among the 274 patient is in figure
5. One particularly successful segmentation can be seen in
figure 6.
5. Discussion
Our median performance on the whole tumor detection
revolves around a dice score of 89, which is right before
losing clinical significance. This is very competitive with
previous methods. One clear issue with our algorithm is
dealing with outliers as can be seen in figure 5. More than
half of our segmentations are wildly successful, but some
segmentations return sub-50% scores, which you would not
typically see with a radiologist.
4
5. Figure 6: Slices-by-Slices Result A representative slice
view of a test case giving 98% Dice score accuracy. Each
of the 12 images has three parts. On the left is the T1 post-
contrast image at the given slice, the middle image is the
expert segmentation labeling, and the right-hand side image
is the labelling at the slice by our algorithm. The label color
scheme follows that which was introduced in figure 1.
figure 4 shows a comparison amongst the top results in
the literature. Some notable ones in 2013 include ones by
Zhao and Subbanna which incorporated Markov Random
Fields (MRF), achieving a Dice accuracy of 82, respectively
[4]. Festa from 2013 used random forests to achieve a Dice
of 62 [4]. In 2014, groups used deep learning and convolu-
tion neural nets (CNNs) to achieve accuracies of 85 (Davy)
and 88 (Urban) [10]. Our method improves upon the ex-
isting vanilla convolutional neural network architecture by
making some high biased assumptions that allow us to dras-
tically reduce our variance.
Basically, one of our assumption is that voxels are inde-
pendent, only coupled by information tied to their feature
vectors. Certainly, this is a high-bias assumption, but it al-
lows us to use n = 4.5 billion training examples rather than
only n = 274 patient samples. Neural nets are ultimately
low-bias systems, but our hope is that the improved vari-
ance caused by the enlarged sample space will overcom-
pensate for our high-bias assumptions. Furthermore, we
contrast our method with standard deep learning algorithms
like CNN, which learn the features using convolution ker-
nels on the inputs. Vanilla CNNs use each patient as a single
training sample. We, on the other hand, select Difference of
Gaussian (DoG) convolution filters to relate voxels to their
neighborhoods. Although we are highly biasing our system
compared to usual deep learning framework, our features
may be more adequate since our subsequent feed-forward
neural net can then learn higher level features from each
pixel’s lower level features.
6. Conclusion
The higher-biased CNN 3-D architecture performs re-
markably well on the glioblastoma segmentation problem.
The segmentation results are competitive with those using
much more complex methods, and we argue our success
is due to our smart choice of features along with a greatly
enlarged sample space and flexible training method (neural
nets). Our algorithm is powerful despite its relatively high
bias, and we hope that it may serve the medical community
in their work.
The natural next step of this project is a thorough anal-
ysis of our models asymptotes. We have claimed that our
large data set has significantly reduced model variance, but
it is unknown whether we can further reduce variance with
more data. Given that our segmentation algorithm is al-
ready on par with our reference expert segmentations, we
suspect but would like to confirm that our model has already
reached its large-data asymptotic performance.
7. Acknowledgments
This project would not have been possible without the
help of our research mentor Darvin Yi and the Gevaert
Biomedical Informatics Lab. In addition, we would like to
thank the CS231N course staff for their ongoing support.
References
[1] World Cancer Report 2014. World Health Organization.
2014. pp. Chapter 5.16.
[2] Bleeker, Fonnet E., Remco J. Molenaar, and Sieger
Leenstra. ”Recent Advances in the Molecular
Understanding of Glioblastoma.” J Neurooncol Journal
of Neuro-Oncology 108.1 (2012): 11-27. Web.
[3] Gallego, O. ”Nonsurgical Treatment of Recurrent
Glioblastoma.” Current Oncology Curr. Oncol. 22.4
(2015): 273. Web.
[4] Menze, B. H. The Multimodal Brain Tumor Image
Segmentation Bench-mark (BRATS). IEEE Transactions
on Medical Imaging 2014 (2013).
[5] Njeh, Ines, Lamia Sallemi, Ismail Ben Ayed, Khalil
Chtourou, Stephane Lehericy, Damien Galanaud, and
Ahmed Ben Hamida. ”3D Multimodal MRI Brain
Glioma Tumor and Edema Segmentation: A Graph
Cut Distribution Matching Approach.” Computerized
Medical Imaging and Graphics 40 (2015): 108-19. Web.
[6] Smirniotopoulos, James G., Frances M. Murphy,
Elizabeth J. Rushing, John H. Rees, and Jason W.
Schroeder. ”Patterns of Contrast Enhancement in the
Brain and Meninges1.” RadioGraphics 27.2 (2007):
525-51. Web.
5
6. [7] Lawrence, S., C.l. Giles, Ah Chung Tsoi, and A.d.
Back. ”Face Recognition: A Convolutional Neural-
network Approach.” IEEE Trans. Neural Netw. IEEE
Transactions on Neural Networks 8.1 (1997): 98-113.
Web.
[8] Krizhevsky, A., Sutskever, I., and Hinton, G.E.
”ImageNet Classification with Deep Convolutional
Neural Networks.” Adv. Neural Inf. Process. Syst.
(2012) 1-9. Web.
[9] Srensen, T. A method of establishing groups of equal
amplitude in plant sociology based on similarity of
species and its application to analyses of the vegetation
on Danish commons. Kongelige Danske Videnskabernes
Selskab. (1948) 5 (4):1-34.
[10] Challenge Manuscripts.” (2014) MICCAI 2014.
Harvard Medical School, Boston, Massachusetts
(2014) .
6