Rapple "Scholarly Communications and the Sustainable Development Goals"
PRESENTATION-214718032-印斯麦.pptx
1. Two Goal Learning to Classify Autistic and Non-autistic Face
by Using Transfer Learning
SHAHZAD INZAMAM 214718032
Supervisor : Liu Jin
Master Degree
December 2023
2. 知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Presentation Outline
• Introduction
• Autism Spectrum Disorder
• Current Diagnosis Methods
• Transfer Learning for Autistic Disease Prediction
• Proposed Methodology
• Primary Result Experiment
• Conclusion
3. 01
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Introduction
Autism Spectrum Disorder (ASD) affects millions of people worldwide,
and its diagnosis remains a challenge for healthcare professionals.
However, recent advancements in machine learning have shown promising
results in predicting ASD with high accuracy. Transfer learning, in
particular, has emerged as a powerful tool for improving the accuracy of
ASD diagnosis.
In this presentation, we will explore the potential of transfer learning in
predicting ASD and its importance in revolutionizing the field of
healthcare. We will discuss current diagnosis methods, their limitations, and
how transfer learning can be used to overcome these limitations and
discover how cutting-edge technology can improve the lives of individuals
with ASD.
Fig 1: ASD Samples Images
Kaggle ASD Dataset
4. 02
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Autism Spectrum Disorder
• Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that affects communication, social interaction,
and behavior. It is a spectrum disorder, meaning that the severity of symptoms can vary widely from person to person.
Some individuals with ASD may have difficulty with verbal and nonverbal communication, while others may have
repetitive behaviors or intense interests in specific topics.
• According to the Centers for Disease Control and Prevention (CDC), approximately 1 in 54 children in the United
States are diagnosed with ASD. While the exact causes of ASD are still unknown, research suggests that a
combination of genetic and environmental factors may play a role. It is important to note that ASD is not caused by
bad parenting or vaccines, as some myths suggest.
• Real-life examples of individuals with ASD include Temple Grandin, The individuals demonstrate that having autism
does not limit one's potential or ability to make a positive impact on the world.
5. 03
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Late detection of ASD in patients can be attributed to several key factors:
1. Diagnosis of ASD primarily relies on interactive sessions, necessitating the involvement of clinical experts who can
accurately diagnose children around the age of two [16].
2. Limited access to specialists poses a challenge for parents, particularly in rural communities or underdeveloped
countries, where the availability of such physicians is scarce [17].
3. Lack of familiarity and awareness about ASD among parents often leads them to overlook developmental issues in
their children and not consider them as symptoms of a potential disorder.
4. Furthermore, children from racial and ethnic minority backgrounds who undergo initial screening are less likely to
undergo subsequent medical examinations due to the high costs associated with the required sophisticated
equipment and skilled personnel [18].
6. 04
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
• Currently, the most common method for diagnosing Autism Spectrum Disorder (ASD) is through behavioral
assessments and interviews with parents or caregivers. These assessments can be time-consuming and rely heavily on
subjective observations, which can lead to misdiagnosis or delayed diagnosis. Additionally, these assessments may not
capture subtle differences in behavior that are indicative of ASD.
• Another method for diagnosing ASD is through brain imaging techniques such as magnetic resonance imaging (MRI)
and electroencephalography (EEG). While these methods can provide valuable insights into brain function, they are
expensive and require specialized equipment and expertise. Furthermore, they may not be accessible to all individuals,
particularly those in low-resource settings.
• Transfer learning has the potential to improve the accuracy [5] of ASD diagnosis by leveraging existing data and
models from related fields. By using pre-trained models [6] and adapting them to ASD prediction, transfer learning
can overcome the limitations of current diagnosis methods and provide a more objective and efficient approach to
diagnosis.
Current Diagnosis Methods
7. 05
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
In the realm of ASD patient diagnosis, several distinct technologies are available for the detection of autism. These
include:
1. Child Autism Rating Scale (CARS)
2. Autism Diagnostic Observation Schedule (ADOS)
3. Autism Detection in Early Childhood (ADEC)
4. Autism Diagnostic Interview – revised (ADI-R)
5. Social Communication Questionnaire (SCQ) Fig 2: Images of dataset containing the both autistic and non-
autistic faces
These technologies serve as invaluable resources in the diagnostic process, contributing to a more comprehensive
understanding of ASD and enabling healthcare professionals to make accurate assessments and tailored interventions for
individuals with autism.
8. 06
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Transfer Learning for ASD Prediction
After conducting a comprehensive examination of the current research landscape both domestically and
internationally, it is evident that there is a pressing requirement for the development of a convolutional neural
network (CNN) base on transfer learning architecture capable of detecting ASD with minimal hyperparameters
and low computation cost due to pre-trained weights. This imperative endeavor aims to facilitate the creation of
an effective Transfer learning-based ASD diagnosis model. Presently, the research conducted in this domain is
still in its nascent stages, and numerous unresolved challenges persist.
9. 07
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Main research ideas, and academic innovations
Our proposal introduces a highly robust face recognition framework
utilizing transfer learning, offering the potential for superior accuracy in
the identification of children with autism.
In this research, we conduct an in-depth analysis of an optimized
ResNet101 and EfficientNetB5 model, surpassing hybrid deep
learning approaches by employing deep learning methods to
accurately detect ASD.
ASD is characterized by specific facial traits in children, such as a
notably wide upper face with widely spaced eyes, and a relatively
short middle face section encompassing the cheeks and nose.
Consequently, our primary contribution lies in developing an expert
system that leverages facial landmarks to identify ASD in children,
thereby enhancing the healthcare system in China and enabling
early detection of ASD.
Academic Innovation
The research methodology includes the following steps:
Dataset preprocessing
Ablation study
Optimal hyperparameter selection
Build Hybrid-Net
Analysis of model performance
Main Research Idea
Fig 3: Flow diagram of Convolution Neural Network Model
10. 08
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Proposed Methodology
Table 1 ASD: Kaggle ASD Dataset (Access 2023)
Category Subcategory Dataset Division
Train Validate Test
Disorder Autistic 1263 100 100
No Autistic 1263 100 100
Table 2 Dataset Division
Kaggle ASD Dataset
Fig 4:Datasets Samples of Autistic Images
Category Subcategory Total Image
Disorder Autistic 1463
No Autistic 1463
11. 09
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Fig 5:Proposed Methodology for Autistic Prediction
Proposed Methodology
12. 10
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
ResNet Basic Architecture
Fig 6: ResNet-Based Architecture
13. 11
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
EfficientNetB5 Basic Architecture
Fig 7: EfficientNetB5 Based Architecture
14. 12
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Performance evaluation metrics
Primary Result Experiment
15. 13
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Experiment Setup
The deep learning ResNet101 and EfficientNetB5 models hybridized were applied to detect ASDs. As a result,
Jupyter Notebook, based on Python, was responsible for all of the calculations that were altered for developing a
classification of ASDs in children. Several methods may be used to detect ASDs. Individual classification models
were trained using the 90% training set, and their efficacy was evaluated using the 10% test set. Both sets were
utilized in conjunction with the training set. For the purpose of determining how well these classifiers functioned,
a number of assessment measures, including visualize by Confusion Metrix accuracy, area under the curve
(AUC), sensitivity, specificity, precision, and F1-score, were computed.
16. 14
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Primary Result
Metrics Evaluations
Accuracy Precision Recall F1-Score
Kaggle Autistic
Dataset
ResNet101 (Base) 86 87 86 86
EfficientNetB5 (Base) 83 84 83 83
Table 3 Model Accuracy, Precision, Recall, and F1-Score (Kaggle Autistic Dataset)
17. 15
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Primary Result
Confusion Matrices
ResNet101
88
85
12
15
Autistic
Autistic
No
Autistic
No Autistic
EfficientNetB5
73
93
27
7
Autistic
Autistic
No
Autistic
No Autistic
Fig 8: ResNet101 & EfficientNetB5 Based Architecture Confusion Matrices
18. 16
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Primary Result
Training and Validation Accuracy
Fig 9: RestNet101 & EfficientNetB5 Training Accuracy
19. 17
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Primary Result
Training and Validation Loss
Fig 9: RestNet101 & EfficientNetB5 Training Loss
20. 18
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Primary Result
Training and Validation Loss
Reference Model Dataset Accuracy %
Mujeeb Rahman et al [9] MobileNet Kaggle Dataset 87
Thabtah et al [32] Xception Kaggle Dataset 90
Alsaade et al [31] Xception Kaggle Dataset 91
Alkahtani et al [33] VGG16 + MobileNet Kaggle Dataset 92
Rabbi et al [34] CNN Model Kaggle Dataset 92.31
Arumugam et al [35] CNN Model Kaggle Dataset 91
Akter et al [36] MobileNet-V1+K-Mean
clustering
Kaggle Dataset 92.10
Table 4 Comparative Analysis with Existing Studies
21. 19
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
Conclusion
The early diagnosis of ASD in children has been shown to have significant positive effects on the diagnosed child’s
long-term health outcomes. All the detection methods used at present rely on the judgement of professionals, despite
the fact that this approach is both subjective and costly.
In this study, we recommended using a deep learning system that integrates several facial landmark features in order to
identify children who have ASD. The utilization of this system could reduce costs and increase the effectiveness of the
detection process. To start, we devised a unique approach for recognizing the properties of facial landmarks.
In this study, we made use of a dataset (Kaggle Autistic Dataset) that is available to the general public and includes
images of the faces of both autistic and typically developing children.
Pre-trained models for binary ASD classification were developed and assessed using ResNet101, and EfficientNetB5
models employing these also constructed. When compared to the other models we examined, we discovered that the
ResNet101 model had the best accuracy (86%). The results suggest 2926 19 of 21 that still images of children’s faces
might be used to rapidly gather diagnostic indicators of ASD, thus enabling an ASD screening approach that is both
rapid and accurate.
22. 20
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
1. Lord, C.; Elsabbagh, M.; Baird, G.; Veenstra-Vanderweele, J. Autism Spectrum Disorder. Lancet 2018, 392, 508–520.
2. Kojovic, N.; Natraj, S.; Mohanty, S.P.; Maillart, T.; Schaer, M. Using 2D Video-Based Pose Estimation for Automated Prediction of Autism Spectrum Disorders in Young
Children. Sci. Rep. 2021, 11, 1–10.
3. Georgoula, C.; Ferrin, M.; Pietraszczyk-Kedziora, B.; Hervas, A.; Marret, S.; Oliveira, G.; Rosier, A.; Crutel, V.; Besse, E.; Severo, C.A.; et al. A Phase III Study of
Bumetanide Oral Liquid Formulation for the Treatment of Children and Adolescents Aged between 7 and 17 Years with Autism Spectrum Disorder (SIGN 1 Trial):
Participant Baseline Characteristics. Child Psychiatry Hum. Dev. 2022, 8, 1–13.
4. Zuckerman, K.E.; Broder-Fingert, S.; Sheldrick, R.C. To Reduce the Average Age of Autism Diagnosis, Screen Preschoolers in Primary Care. Autism 2021, 25, 593–596.
5. Goh, K.L.; Morris, S.; Rosalie, S.; Foster, C.; Falkmer, T.; Tan, T. Typically Developed Adults and Adults with Autism Spectrum Disorder Classification Using Centre of
Pressure Measurements. In Proceedings of the 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Shanghai, China, 20–25 March
2016; pp. 844–848, ISSN 2379-190X.
6. Speaks, A. What is autism. Retrieved Novemb. 2011, 17, 2011. Available online: https://www.autismspeaks.org/what-autism (accessed on 20 January 2023).
7. Guillon, Q.; Hadjikhani, N.; Baduel, S.; Rogé, B. Visual Social Attention in Autism Spectrum Disorder: Insights from Eye Tracking Studies. Neurosci. Biobehav. Rev.
2014, 42, 279–297.
8. Kanhirakadavath, M.R.; Chandran, M.S.M. Investigation of Eye-Tracking Scan Path as a Biomarker for Autism Screening Using Machine Learning Algorithms.
Diagnostics 2022, 12, 518.
9. Mujeeb Rahman, K.K.; Subashini, M.M. Identification of Autism in Children Using Static Facial Features and Deep Neural Networks. Brain Sci. 2022, 12, 94. [CrossRef]
Yang, M.; Cao, M.; Chen, Y.; Chen, Y.; Fan, G.; Li, C.; Wang, J.; Liu, T. Large-scale brain functional network integration for discrimination of autism using a 3-D deep
learning model. Front. Hum. Neurosci. 2021, 15, 277.
10. Ahsan, M.M.; E Alam, T.; Trafalis, T.; Huebner, P. Deep MLP-CNN model using mixed-data to distinguish between COVID-19 and Non-COVID-19 patients. Symmetry
2020, 12, 1526.
Reference
23. 21
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
11. Aldhyani, T.H.H.; Nair, R.; Alzain, E.; Alkahtani, H.; Koundal, D. Deep Learning Model for the Detection of Real Time Breast Cancer Images Using Improved Dilation-Based
Method. Diagnostics 2022, 12, 2505.
12. Schelinski, S.; Borowiak, K.; von Kriegstein, K. Temporal Voice Areas Exist in Autism Spectrum Disorder but Are Dysfunctional for Voice Identity Recognition. Soc. Cogn. Affect.
Neurosci. 2016, 11, 1812–1822.
13. Jiang, X.; Chen, Y.F. Facial Image Processing. In Applied Pattern Recognition; Bunke, H., Kandel, A., Last, M., Eds.; Studies in Computational Intelligence; Springer:
Berlin/Heidelberg, Germany, 2008; pp. 29–48.
14. Garcia, C.; Ostermann, J.; Cootes, T. Facial Image Processing. Eurasip J. Image Video Process. 2008, 2007, 070872.
15. Yolcu, G.; Oztel, I.; Kazan, S.; Oz, C.; Palaniappan, K.; Lever, T.E.; Bunyak, F. Facial Expression Recognition for Monitoring Neurological Disorders Based on Convolutional
Neural Network. Multimed. Tools Appl. 2019, 78, 31581–31603.
16. Haque, M.I.U.; Valles, D. A Facial Expression Recognition Approach Using DCNN for Autistic Children to Identify Emotions. In Proceedings of the 2018 IEEE 9th Annual
Information Technology, Electronics and Mobile Communication Conference (IEMCON), Vancouver, BC, Canada, 1–3 November 2018; pp. 546–551.
17. Oosterling, I.J.; Wensing, M.; Swinkels, S.H.; Van Der Gaag, R.J.; Visser, J.C.; Woudenberg, T.; Minderaa, R.; Steenhuis, M.P.; Buitelaar, J.K. Advancing early detection of autism
spectrum disorder by applying an integrated two-stage screening approach. J. Child Psychol. Psychiatry 2010, 51, 250–258.
18. Rossignol, D.A.; Frye, R.E. Melatonin in autism spectrum disorders: A systematic review and meta-analysis. Dev. Med. Child Neurol. 2011, 53, 783–792.
19. Tait, K.; Fung, F.; Hu, A.; Sweller, N.; Wang, W. Understanding Hong Kong Chinese families’ experiences of an autism/ASD diagnosis. J. Autism Dev. Disord. 2016, 46, 1164–1183.
20. Jaiswal, S., et al. Automatic detection of ADHD and ASD from expressive behaviour in RGBD data. in 2017 12th IEEE International Conference on Automatic Face & Gesture
Recognition (FG 2017). 2017. IEEE.
21. Aldrige, D.K. Is It Autism? Facial Features That Show Disorder. Available online: https://www.cbsnews.com/images/is-itautism-facial-features-that-show-disorder/ (accessed on 3
March 2022).
24. 22
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
22. Ahsan, M.M.; Gupta, K.D.; Islam, M.M.; Sen, S.; Rahman, M.; Shakhawat Hossain, M.; COVID-19 symptoms detection based on nasnetmobile with explainable ai using
various imaging modalities. Mach. Learn. Knowl. Extr. 2020, 2, 490–504.
23. Mishra, P.; Passos, D. Realizing transfer learning for updating deep learning models of spectral data to be used in new scenarios. Chemom. Intell. Lab. Syst. 2021, 212, 104283.
24. Allely, C.S. and P. Wilson, Diagnosing autism spectrum disorders in primary care. The Practitioner, 2011. 255(1745): p. 27-31.
25. LaMantia, A.S. Why does the face predict the brain? Neural crest induction, craniofacial morphogenesis, and neural circuit development. Front. Physiol. 2020, 11, 610970.
26. Akter, T.; Ali, M.H.; Khan, M.; Satu, M.; Uddin, M.; Alyami, S.A.; Ali, S.; Azad, A.; Moni, M.A. Improved transfer-learning-based facial recognition framework to detect
autistic children at an early stage. Brain Sci. 2021, 11, 734.
27. Hosseini, M.P.; Beary, M.; Hadsell, A.; Messersmith, R.; Soltanian-Zadeh, H. Deep Learning for Autism Diagnosis and Facial Analysis in Children. Front. Comput. Neurosci.
2021, 15, 789998.
28. Mujeeb Rahman, K.; Subashini, M.M. Identification of Autism in Children Using Static Facial Features and Deep Neural Networks. Brain Sci. 2022, 12, 94. [CrossRef]
29. Khosla, Y.; Ramachandra, P.; Chaitra, N. Detection of autistic individuals using facial images and deep learning. In Proceedings of the 2021 IEEE International Conference on
Computation System and Information Technology for Sustainable Solutions (CSITSS), Bangalore, India, 16–18 December 2021; pp. 1–5.
30. Alsaade, F.W.; Alzahrani, M.S. Classification and Detection of Autism Spectrum Disorder Based on Deep Learning Algorithms. Comput. Intell. Neurosci. 2022, 2022, 8709145.
31. Alsaade, F.W.; Alzahrani, M.S. Classification and Detection of Autism Spectrum Disorder Based on Deep Learning Algorithms. Comput. Intell. Neurosci. 2022, 2022, 8709145.
25. 23
自 强 不 息 厚 德 载 物
知 行 合 一 、 经 世 致 用 C e n t r a l S o u t h U n i v e r s i t y
32. Thabtah, F.; Kamalov, F.; Rajab, K. A New Computational Intelligence Approach to Detect Autistic Features for Autism Screening. Int. J. Med. Inform. 2018, 117, 112–124.
33. Alkahtani, H.; Aldhyani, T.H.H.; Alzahrani, M.Y. Deep Learning Algorithms to Identify Autism Spectrum Disorder in Children-Based Facial Landmarks. Appl. Sci. 2023, 13,
4855. https://doi.org/10.3390/app13084855
34. Rabbi, M.F., Hasan, S.M., Champa, A.I., Zaman, M.A.: A convolutional neural network model for early-stage detection of autism spectrum disorder. In: 2021 International
Conference on Information and Communication Technology for Sustainable Development (ICICT4SD), pp. 110–114. IEEE (2021)
35. Arumugam, S.R., Karuppasamy, S.G., Gowr, S., Manoj, O., Kalaivani, K.: A deep convolutional neural network based detection system for autism spectrum disorder in facial
images. In: 2021 5th International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I-SMAC), pp. 1255–1259. IEEE (2021)
36. Akter, T., et al.: Improved transfer-learning-based facial recognition framework to detect autistic children at an early stage. Brain Sci. 11(6), 734 (2021)
The research methodology includes the following steps:
Dataset preprocessing: The dataset is organized and the images are resized to prepare them for training.
Ablation study: Hyperparameters are tuned during the training process, and the performance of the models is evaluated and validated after each iteration.
Optimal hyperparameter selection: Once the optimal set of hyperparameters is determined, the model training optimizer sheds light on the reasons behind low accuracy by examining prediction probabilities.
Build Hybrid-Net Model with combination of ResNet101 and EfficientNetB5 for classification of Autistic and Normal images with high precision, and recall.
Analysis of model performance: The performance of the models is analyzed to determine the future directions of the research, particularly in terms of dataset preprocessing and the application of feature maps.
基于以上现状,我的博士研究课题为。。。
基于以上现状,我的博士研究课题为。。。
ResNet101 is used as the benchmark experiment. The multi-scale feature extraction module, depth feature extraction module and DDC module are added in ResNet101, and the octave convolution is used to replace ordinary convolution
EfficientNet was released this June (2019) by Google AI and is the new state-of-the-art on ImageNet. It introduces a systematic way to scale CNN (Convolutional Neural Networks) in a nearly optimal way. For this kernel we will use the B5 version, but feel free to play with the larger models. This kernel provides weights for EfficientNetB0 through B5. Weights for EfficientNetB6 and B7 can be found in Google AI's repository for EfficientNet. I highly recommend you to read the EfficientNet paper as it signifies a fundamental shift in how the Deep Learning community will approach model scaling!