Enhancing Early Detection and Classification of Alzheimer’s Disease Stages Using Deep Learning and LIME for Interpretability

The issue of transparency in diagnostic methods has been a crucial challenge in Alzheimer’s Disease detection. This paper focuses on analyzing state-of-art deep learning algorithms namely, customized convolutional network, InceptionV3, DenseNet121, AlexNet, VGG16, and ResNet50 for detecting Alzheimer’s disease. The work also focuses on analyzing the transparency and decision-making capabilities of these models by using the Explainable AI tool, Local Interpretable Model-agnostic Explanations, for clinicians and patients. The learnings and decision outputs obtained from these convolutional models are related and explained using LIME, where ResNet and DenseNet lead to efficient solutions with appropriate feature extraction and importance provided in LIME visualization. This explication of the model’s decision-making process affected by LIME can improve diagnostic accuracy and foster trust that can offer a great chance for the medical experts to offer better care for patients, who are suffering from Alzheimer’s Disease.