Exploration of Explainable AI with Deep Learning Model for Early Detection of Alzheimer's Disease

Detecting of Alzheimer's disease (AD) at early stage is crucial for timely intervention and management. This study explores the application of an Explainable AI with deep learning models for detecting Alzheimer's disease (AD) at early stage using medical imaging data. We developed a Convolutional Neural Network (CNN) to classify brain scans into four distinct stages of Alzheimer's progression and superior spatial hierarchical relationships modeled by Capsule neural network (CapsNet). To enhance interpretability, we employed explainable AI (XAI) techniques, LIME and SHAP, to explain the predictions of a pretrained InceptionV3 model for image classification which allow clinicians to visualize the regions of the brain that model considers most indicative of Alzheimer's. These visualizations provide valuable insights into the decision-making process, fostering trust and facilitating clinical validation. Our results demon- strate that explainable deep learning models not only achieve best performance but also offer critical transparency, making viable tools for aiding in the diagnosis and understanding of Alzheimer's disease. The model achieves best F1 score, precision, recall and accuracy with 96% and 97% for CNN and CapsNet respectively indicating its robustness in detecting subtle differences in brain images.