Design of An Iterative End-to-End Multi-Modal Deep Learning Framework for Explainable Diagnosis of Alzheimer’s and Parkinson’s Diseases from Brain Imaging Process

The precise and early diagnosis of neurodegenerative diseases such as Alzheimer's Disease (AD) and Parkinson's Disease (PD) continues to pose a critical clinical challenge due to common symptoms as well as the progressive nature of these diseases. Most of the currently existing systems are characteristically mono-modal imaging or handcrafted features; hence their generalizability, robustness, and interpretability get limited. Moreover, the current models do not form any unified chain comprising preprocessing, region segmentation, feature fusion, classification, and explainability; this limits their deployment in real clinical settings. To address these challenges, this work proposes an integrative, end-to-end multi-modal diagnostic framework tailored for AD and PD detection using MRI, CT, and X-ray brain images and samples. Starting with Contrast Limited Adaptive Histogram Equalization (CLAHE) for image enhancement, it improves contrast and noise reduction. UNet++ is used in Region of Interest (ROI) segmentation targeting disease relevant locations, such as the hippocampus and basal ganglia. EfficientNet-B7 provides robust representation learning by extracting high dimensional embeddings from ROI-masked images, utilizing pretrained weights with medically fine-tuned retrievals. A novel use of a multi-head self-attention mechanism is then employed to detect features between modalities for best cross-modal integration. A CNN-Transformer hybrid model that effectively combines local spatiality and a global context awareness is employed to classify the data samples. Grad-CAM++ yields high-fidelity saliency maps for model explainability that closely aligns with radiologist annotations. High-performance metrics (Accuracy ≥ 95%, Sensitivity ≥ 92%, Specificity ≥ 93%) are attained by this system but also provide interpretable outputs and actionable insights in processing. The integration of more advanced deep learning modules within one pipeline marks a significant step towards a reliable, explainable, and clinically proven diagnosis of AD and PD Sets.