Weighted Multi-Modal Contrastive Learning Based Hybrid Network for Alzheimer’s Disease Diagnosis

This project presents a robust approach for diagnosing Alzheimer's disease using a hybrid deep learning network based on Weighted Multi-Modal Contrastive Learning (WMCL). The system categorizes patients into four classes: 'MildDemented,' 'ModerateDemented,' 'NonDemented,' and 'VeryMildDemented.' We applied state-of-the-art deep learning models, including EfficientDet, DenseNet, and ResNet combined with GRU, to classify brain images for early diagnosis of Alzheimer's. The training process involved using MRI images, with preprocessing steps such as normalization and augmentation to enhance model performance. The system leverages Flask for the backend, with a user-friendly interface built using HTML and CSS, allowing users to register, log in, and upload MRI images. Upon uploading, the system processes the images and provides real-time predictions along with a detailed explanation using Grad-CAM++ for visual interpretability. The model's performance was evaluated using key metrics, including accuracy, precision, recall, and F1-score, ensuring high reliability in diagnosis. This system can aid healthcare professionals by offering a tool for early-stage Alzheimer's detection and facilitating interpretability in the decision-making process.

Keywords:

Alzheimer's disease, deep learning, Weighted Multi-Modal Contrastive Learning, EfficientDet, DenseNet, ResNet + GRU, Grad-CAM++, image classification, MRI, Flask, AI explainability, early diagnosis, brain imaging.