Multi-class Alzheimer’s Disease Classification Using Deep Learning Techniques

Multi-class Alzheimer’s Disease Classification Using Deep Learning Techniques

1. PhD Progress Review Tentative Title Multi-class Alzheimer’s Disease Classification Using Deep Learning Techniques Research Scholar Name: S.Gani Lakshmi ID Number: AV.SC.R4CSE23002, Department of CSE, Amrita Vishwa Vidyapeetham, Amrita School of Computing, Amaravati, Andhra Pradesh Under the Guidance Dr Chandan Kumar Assistant Professor (Sl. Gr.), Department of CSE, Amrita Vishwa Vidyapeetham, Amrita School of Computing, Amaravati, Andhra Pradesh

2. Doctoral Committee Members

3. CONTENTS Course work planned/ completed Introduction Literature Survey Work done in first 6 months Plan of action for the next 6 months Tentative date of thesis submission Conclusion

4. I Course Work(Planned / Completed) Subject Name Subject Code L-T-P-C Progress Research Methodolgy-1 RM801 2-0-0-2 Completed Grade : A Advanced Linear Algebra MA833 4-0-0-4 End-Term Next Month Mid-Term:34 Deep Learning CS829 3-0-1-4 End-Term Over Research Methodology-2 RM802 2-0-0-2 End-Term Next Month IoT/ Advanced Digital Image processing 18CS623/CS824 3-0-1-4 2024-2025(Odd Semester) 4

5. II INTRODUCTION Brain: It is one of the largest organs in the body, and coordinates most body activities. It is the center for all thought, memory, judgment, and emotion. Each part of the brain is responsible for controlling different body functions, such as temperature regulation and breathing. The brain is contained in the skull & weighs 1300 - 1400 g made up of about 1000 billion neurons 7 each neuron is surrounded by about 10 glial cells (neuroglia). Neurons cannot multiply & many neurons are lost everyday in life but glial cells can multiply throughout the life.

6. Brain disease: Brain diseases encompass various conditions affecting the brain's structure and function, including neurodegenerative diseases, mental health disorders, brain tumors , cerebrovascular diseases, traumatic brain injuries, and infections. Alzheimer's disease is the most common cause of dementia, characterized by progressive cognitive decline and memory loss. The World Health Organization reports that over 55 million people worldwide live with dementia, with the number expected to rise to 78 million by 2030 and 139 million by 2050.

7. Dementia Dementia is a condition that affects memory, thinking, and social skills due to the loss of healthy brain cells. Common signs include memory problems, poor judgment, confusion, trouble communicating, and behavior changes. Types of Dementia: Alzheimer's Disease : Most common, involves brain changes with amyloid plaques and tau tangles. Frontotemporal Dementia : Affects younger people, and involves abnormal tau and TDP-43 proteins. Lewy Body Dementia : Involves alpha-synuclein protein deposits. Vascular Dementia : Caused by damaged blood vessels in the brain. Mixed Dementia : Combination of different types of dementia.

8. Alzheimer’s disease: Alzheimer’s Disease (AD) is the most common form of dementia that can lead to a neurological brain disorder that causes progressive memory loss as a result of damaging the brain cells and the ability to perform daily activities. Alzheimer's disease is a progressive brain disorder marked by changes that lead to the accumulation of protein deposits in the brain. This causes brain shrinkage and cell death, making it the most common cause of dementia, characterized by a continuous decline in memory, thinking, behavior , and social skills, impairing a person's ability to function. In the United States, approximately 6.5 million individuals aged 65 and older are living with Alzheimer's, with over 70% being 75 or older. Globally, of the 55 million people with dementia, an estimated 60% to 70% have Alzheimer's disease. The damage most often starts in the region of the brain that controls memory, but the process begins years before the first symptoms. The loss of neurons spreads in a somewhat predictable pattern to other regions of the brains. By the late stage of the disease, the brain has shrunk significantly.

9. Alzheimer's Disease Impact

10. Differences between a healthy brain and a brain affected by Alzheimer's disease using MRI scans

11. Alzheimer's disease

12. SIGNS of AD Ten warning signs of Alzheimer's disease 1) Memory loss 2 ) Difficulty performing familiar tasks 3) Problems with language 4) Disorientation to time and place 5) Poor or decreased judgment 6) Problems with abstract thinking 7) Misplacing things 8) Changes in mood or behavior 9) Changes in personality 10) Loss of initiative

13. Classification of Stages:

14. Amyloid precursor protein(APP) The amyloid precursor protein (APP) is important in Alzheimer's disease. It is found in neurons and breaks down into amyloid-β (Aβ) peptides. Rare mutations in APP can lead to familial Alzheimer's disease by increasing Aβ levels or changing the Aβ42/Aβ40 ratio. Mutations and duplications of the APP gene have similar effects. Some mutations are linked to familial cerebral amyloid angiopathy (CAA). A protective mutation in APP can reduce Aβ production by 20%. Genes like APOE and ABCA7 also affect how APP works. APP interacts with other cell proteins, and its fragments have roles in brain development, like helping neurons stick together and form connections. In adults, APP may help respond to brain damage

15. APP

16. APP (Amyloid Precursor Protein) is initially good as it is a large protein found in the brain, but it can lead to problems if it breaks down incorrectly. When APP breaks down, it produces smaller pieces, including Aβ, which can cause Alzheimer's disease if not properly regulated. Aβ (Amyloid-β) is generally bad because it is a small protein fragment that comes from APP and can clump together to form amyloid plaques. These plaques are harmful to brain cells and disrupt their function, playing a key role in the development of Alzheimer's disease. α (Alpha-secretase) is a good enzyme because it cuts APP in a way that prevents the formation of harmful Aβ fragments. By doing so, alpha-secretase helps maintain normal brain function and prevents the formation of amyloid plaques. β (Beta-secretase) is a bad enzyme as it cuts APP to produce harmful Aβ fragments. This promotes the formation of amyloid plaques, which contribute to the progression of Alzheimer's disease. γ (Gamma-secretase) is also bad because it further processes the fragments produced by β-secretase to finalize the production of Aβ fragments. Gamma-secretase works with β-secretase to produce Aβ, leading to plaque formation and the advancement of Alzheimer's disease

17. Plaque Tau Amyloid Plaques and Tau Tangles are believed to be the two molecules responsible for the brain Plaque (Amyloid Plaques) damage associated with Alzheimer's disease BAD Effect: Amyloid plaques are clumps of amyloid-β (Aβ) protein that accumulate in the brain. They are harmful because they disrupt brain cell function, leading to the death of neurons and contributing to the progression of Alzheimer's disease. Tau (Tau Protein) Good in its normal state, but bad when it becomes abnormal. Effect: Tau protein normally helps stabilize microtubules in neurons, which are essential for cell structure and transport. However, in Alzheimer's disease, tau proteins become abnormally modified and form tangles inside neurons. These tangles disrupt the transport system in cells, leading to cell death and further contributing to the symptoms and progression of Alzheimer's disease.

18. Plaque Tau Tangles

19. Diagnostic tests Brain imaging. * CT scan * MRI * PET * SPECT

20. MRI( Magnetic Resonance Imaging ) MRI is crucial in Alzheimer's disease for detecting early brain changes, monitoring progression, and aiding research. Key parameters include hippocampal volume, cortical thickness, brain activity, and white matter integrity. Techniques like segmentation and surface-based analysis help calculate these metrics. Findings include hippocampal atrophy, cortical thinning, enlarged ventricles, and reduced brain activity. MRI helps classify AD stages from preclinical to severe based on these brain changes. Class - 1: Mild Demented (896 images) Class - 2: Moderate Demented (64 images) Class - 3: Non-Demented (3200 images) Class - 4: Very Mild Demented (2240 images)

21. CLASSIFICATION OF DEMENTIA LEVELS Level Area Affected Stage Impact (IMP) % Affected at This Stage Type of Cognitive Loss Non-Demented No significant brain damage Preclinical Normal cognitive function N/A None Very Mild Demented Early brain regions (e.g., hippocampus) Early Slight memory issues, occasional confusion 10-20% Minor memory lapses, slight decline in problem-solving Mild Demented Brain regions critical for memory (e.g., hippocampus, entorhinal cortex) Mild Noticeable memory problems, difficulty with complex tasks 20-30% Memory loss, difficulty with planning and organizing Moderate Demented More widespread brain damage Moderate Significant memory loss, impaired judgment, personality changes 30-50% Severe memory loss, confusion, difficulty recognizing people Severe Demented Extensive brain damage Severe Loss of ability to respond to environment, need for extensive help with daily activities 50-70% Loss of ability to carry out daily activities, extensive cognitive decline

22. III LITERATURE REVIEW Reference Technique Number of classes Dataset Accuracy Advantages Limitations Author name and year is required VGG16 Three No dementia: 2000 Very mild dementia: 2000 Mild dementia: >2000 (Total >6000 images) Accuracy: 99.68% High accuracy of 99.68% on the testing set. Model fine-tuned with hyperparameters. Effective feature extraction using transfer learning with VGG16. Limited to MRI images for Alzheimer’s disease. Requires significant computational resources for training. May overfit if not properly regularized with dropout and normalization.

23. LITERATURE REVIEW Reference Technique Number of classes Dataset Accuracy Advantages Limitations [2] CNN F our Mild Demented: 717 Moderate Demented: 52 Non-demented: 2560 Very Mild Demented: 1792 Kaggle Training Accuracy: 99.72% Validation Accuracy: 98.71% Precision: 98%Recall: 96%F1-Score: 97%ROC-AUC: 1.00 High accuracy in training and validation datasets. The data imbalance problem is not considered Considered single optimizer and learning rate Potential overfitting if not properly regularized with dropout layers

24. LITERATURE REVIEW Reference Technique Number of classes Dataset Accuracy Advantages Limitations [3] Resnet50 Two Mild Demented: 717 Moderate Demented: 52 Non-demented: 2560 Very Mild Demented: 1792 Kaggle Training Accuracy: 95%Validation Accuracy: 95% Transfer learning with Fastai and ResNet-50 improved performance despite limited data. The development of a web application allows practical deployment, making the model accessible for medical professionals. It is at risk of overfitting without proper regularization and is limited to Alzheimer's disease classification, necessitating further validation on diverse datasets.

25. LITERATURE REVIEW Reference Technique Number of classes Dataset Accuracy Advantages Limitations [4] CNN VGG-19 F our ADNI dataset: 300 patients AD: 5764 images EMCI: 5817 images LMCI: 3460 images NC: 6775 images CNN (2D): 93.61% CNN (3D): 95.17% VGG19: 97% High accuracy using VGG19 with 97% for multi-class AD stage classifications. End-to-end framework reduces computational complexity and memory requirements. Web application for remote AD stage checking. Computationally intensive requiring significant resources. Risk of overfitting without proper regularization. Limited to Alzheimer's disease MRI classification.

26. LITERATURE REVIEW Reference Technique Number of classes Dataset Accuracy Advantages Limitations [5] R esnet18 Alexnet GoogleNet Five Cognitively Normal (CN), Early Mild Cognitive Impairment (EMCI) Mild Cognitive Impairment (MCI) Late Mild Cognitive Impairment (LMCI) Alzheimer’s Disease (AD) Accuracy:97.22% AUC: 96.483% Accuracy:96.22% AUC: 95.483% Accuracy:94.22% AUC: 94.483% optimizing it for classifying various stages of Alzheimer's disease. Efficient transfer learning with ResNet-18. Limited to Alzheimer's disease MRI classification.

27. IV Work done in first 6 months Technique Operations Dataset Exisisting Results Proposed Proposed Method Results Epohcs CNN with Smotek Algorithm Convolution blocks 4 No.of Filters 128 Input image(176*208) Mild Demented: 717 Moderate Demented: 52 Non-demented: 2560 Very Mild Demented: 1792 Loss:0.0649% accuracy: 96.63% AUC: 97.76%, , F1-score :98.61%, Precision:97.63%, recall:97.58%, Added one convolution Block No of filters 256 Added one max pool Testing Loss : 0.037692 Testing Accuracy : 98.779297 % Testing AC : 99.878949 % Testing F1-Score : 98.785162 % Testing Precision : 98.826981 % Testing Recall : 98.730469 50 Vgg19 19 layers Mild Demented: Moderate Demented: Non-demented: Very Mild Demented: Kaggle loss: 0.6345 - auc : 0.9366 - val_loss : 0.5520 - val_auc : 0.9447 Epoch 95: early stopping Changed hyper parameters loss: 0.5345 - auc : 0.9466 - val_loss : 0.4950 - val_auc : 0.9547 Epoch 95: early stopping 100 Densenet 169 169 Mild Demented: Moderate Demented: Non-demented: Very Mild Demented Training Accuracy: 97.94% Validation Accuracy: 92.26% Testing Accuracy: 93.34% Changed hyper parameters Training Accuracy: 99.94% Validation Accuracy: 93.26% Testing Accuracy: 94.34% 100

28. Implemented codes Resnet 152 152 Mild Demented: Moderate Demented: Non-demented: Very Mild Demented Training Accuracy: 77.94% Validation Accuracy: 73.26% Testing Accuracy: 76.42% Changed hyper parameters Training Accuracy: 79.94% Validation Accuracy: 74.26% Testing Accuracy: 78.34% 20 Inception V3 105 Mild Demented: Moderate Demented: Non-demented: Very Mild Demented Training Accuracy: 94.94% Validation Accuracy: 95.26% Testing Accuracy: 96.42% Changed hyper parameters Training Accuracy: 71.94% Validation Accuracy: 70.26% Testing Accuracy: 69.42% 20

29. Proposed Architecture DCNN Changed model added one convolution Block 5 No.of Filters=256 one Average Pool accuracy-97% Precision Recall F1 Score-

30. IMPLEMENTED PAPERS CNN with SMOTEK Algorithm[6] RNN [7] LSTM+Convolutional auto encoder[8] 3D deep convolutional neural network [9] CNN+Vgg19[10] CNN+Vgg16+vgg19+Resnet 50+Densenet 121[11]

31. Implemented ML models Classifier Accuracy Precision Recall F1 Score RandomForestClassifier 96.43% 0.98 0.95 0.96 GradientBoostingClassifier 92.00% 0.9459459459459459 0.8974358974358975 0.9210526315789473 AdaBoostClassifier 91.96% 0.9636363636363636 0.8833333333333333 0.9217391304347826 DecisionTreeClassifier 89.33% 0.9487179487179487 0.8604651162790697 0.9024390243902439 RandomForestClassifier 89.33% 0.972972972972973 0.8372093023255814 0.9 ExtraTreesClassifier 93.33% 0.9473684210526315 0.9230769230769231 0.935064935064935

32. Classification of Alzheimer’s disease using ML codes https://www.kaggle.com/code/rodrigox93/multi-class-model-detect-cdr-acc-83-f1-78/input?select=oasis_longitudinal.csv

33. ADD-Net: An Effective Deep Learning Model for Early Detection of Alzheimer Disease in MRI Scans(IEEE) Metric proposed Model Current Model Loss 0.037692 0.0649 Accuracy 98.779297% 96.63% F1-Score 98.785162% 98.61% Precision 98.826981% 97.63% Recall 98.730469% 97.58% AUC 9 6.233% 97.76%

34. VGG-19

35. Vgg19(Executed)

36. Densenet 169(executed) Training Accuracy: 99.94% Validation Accuracy: 93.26% Testing Accuracy: 94.34%

37. ResNet50

38. ResNet50

39. Dataset Dataset Link : 1)https://www.kaggle.com/datasets/tourist55/alzheimersdataset-4-class-of-images Alzheimer's Disease Neuroimaging Initiative (ADNI) Dataset : ADNI  | Alzheimer's Disease Neuroimaging Initiative (usc.edu) OASIS Brains (wustl.edu) https://www.kaggle.com/code/rodrigox93/multi-class-model-detect-cdr-acc-83-f1-78/input?select=oasis_longitudinal.csv

40. V. PLAN OF ACTION FOR NEXT 6 MONTHS We will try to explore different hospitals to collect the datasets Planning to submit one conference and one journal paper writing conference paper using ML models for early detection of AD We will try to Implement and write a journal paper on “Alzheimer’s disease Early detection and classification using Comparison CNN and 5 pre-trained models” We will try to Implement and write a journal paper on “Alzheimer’s disease classification using GAN and LSTM with autoencoders”. We will find a way to meet some of the radiologists to know more about the diseases.

41. VI. PROPOSED WORK AND TIMELINE Milestone Targeted completion Date Actual Status Course work 9/2024 In progress Comprehensive Exam 12/2024 Qualifying Exam 2/2025 Open Seminar 1 9/2025 Open Seminar 2 3/2026 Synopsis Submission 9/2026 Final Thesis Submission 12/2026

42. VII. Conclusion Brain diseases like Alzheimer's represent an extreme difficulty in a human's life and surrounding relatives and it should be clearly diagnosed to avoid future reduplications or side effects. Nowadays, with the evaluation of artificial intelligence especially DL and different CNN-based architectures in image classifications and predictions, it is easier and more efficient to detect such kind of diseases and classify it in its different stages (i.e., mild demented, moderate demented, non-demented and very mild demented). In this path, this work researches the viability of different DCNN (CNN, VGG-16, VGG-19, ResNet50 and DenseNet121) algorithms to appoint the most precise algorithm for predicting and classifying AD stages depending on sampled MRI images. The algorithms’ performance evaluation has been calculated using accuracy, f1-score that totalizes precision and recall metrics .

43. REFERENCES Liu, Maximus, et al. "A Deep Convolutional Neural Network For Early Diagnosis of Alzheimer’s Disease."  2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES) . IEEE, 2022. Chew, Jordan, and Tan Weng Chun. "A Deep Learning-Based Framework for Alzheimer's Disease (AD) Classification Using Transfer Learning."  Proceedings of the 12th International Conference on Biomedical Engineering and Technology . 2022. Jayanthi, V. S., Blessy C. Simon, and D. Baskar. "Alzheimer’s disease classification using deep learning."  Computational Intelligence and Its Applications in Healthcare . Academic Press, 2020. 157-173. Shanmugam, Jayanthi Venkatraman, et al. "Alzheimer’s disease classification using pre-trained deep networks."  Biomedical Signal Processing and Control  71 (2022): Sadat, Sayed Us, et al. "Alzheimer’s disease detection and classification using transfer learning technique and ensemble on convolutional neural networks."  2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC) . IEEE, 2021. Fareed, Mian Muhammad Sadiq, et al. "ADD-Net: an effective deep learning model for early detection of Alzheimer disease in MRI scans."  IEEE Access  10 (2022 Al Olaimat , Mohammad, et al. "PPAD: A deep learning architecture to predict progression of Alzheimer’s disease."  Bioinformatics  39.Supplement_1 (2023) Chen, L., Saykin , A. J., Yao, B., Zhao, F., & Alzheimer’s Disease Neuroimaging Initiative (ADNI. (2022). Multi-task deep autoencoder to predict Alzheimer’s disease progression using temporal DNA methylation data in peripheral blood.  Computational and Structural Biotechnology Journal ,  20 , 5761-5774. Liu, S., Masurkar , A. V., Rusinek , H., Chen, J., Zhang, B., Zhu, W., ... & Razavian , N. (2022). Generalizable deep learning model for early Alzheimer’s disease detection from structural MRIs.  Scientific reports ,  12 (1), 17106. Helaly , H. A., Badawy , M., & Haikal, A. Y. (2022). Deep learning approach for early detection of Alzheimer’s disease.  Cognitive computation ,  14 (5), 1711-1727. Dawood, A. S. A Comparative Study Using Deep Learning Models And Transfer Learning for Detection And Classification of Alzheimer’s Disease. Purwono , P., Ma'Arif , A., Suwarno , I., Mangkunegara , I. S., Dewi, P., & Setyawati , E. (2023, August). Comparison of the Performance of CNN Transfer Learning in the Classification of Alzheimer's Disease. In  2023 International Conference on Information Technology Research and Innovation (ICITRI)  (pp. 43-48). IEEE.

44. THANK YOU