Abstract

Alzheimer's Disease (AD) is a neurodegenerative disorder requiring early detection. This study compares AI models—Convolutional Neural Networks (CNN), Support Vector Machines (SVM), and Random Forest (RF)—in analyzing neuroimaging data (MRI, PET) to enhance AD prediction and improve early diagnosis using machine learning techniques. Through the application of multi-modal data in the form of genetic, clinical, and neuroimaging data, the study also investigates the effectiveness of combining different data types to enhance the predictability of AI models for AD diagnosis. Feature importance analysis was also performed using methods like SHAP (SHAP (Shapley Additive Explanations) values to determine the most important variables in the model predictions, e.g., certain brain regions or genetic components. The study also investigates the generalizability and real-world applicability of the AI models by training the models on an independent dataset representing diverse clinical settings. The performance of each model was assessed using a variety of statistical measures like accuracy, precision, recall, F1-score, and Area Under the Curve (AUC). The findings showed that CNN performed better compared to that of SVM and RF models in all the performance metrics with the highest accuracy (92%), precision (93%), recall (91%), and AUC (0.95). The findings suggest that CNN effectively detects subtle neuroimaging patterns, making it a strong tool for early Alzheimer's diagnosis. While SVM and RF performed well, CNN showed superior accuracy. Cross-validation confirmed its generalizability, crucial for clinical use. Implementing AI models, especially CNN, may enable earlier detection, timely interventions, and improved patient outcomes in Alzheimer’s care.