Explainable Transfer Learning for Precise Alzheimer’s Disease Prediction from MRI Data

Alzheimer’s disease (AD) is one of the most prevalent chronic neurological illnesses among older persons. Early diagnosis of AD enables timely intervention and management, traditionally achieved through clinical assessments, cognitive tests, and neuroimaging techniques like Magnetic Resonance Imaging (MRI). MRI can diagnose AD at various stages, including early mild cognitive impairment (MCI), moderate MCI, and advanced stages of AD. The goal of this study is to categorize AD stages from MRI images using transfer learning. Deep learning has shown remarkable performance in medical image analysis, surpassing conventional machine learning methods. We have used seven transfer learning models: EfficientNetB0, Resnet152, InceptionV3, Resnet101, VGG16, VGG19, and Resnet50 for our imbalanced dataset and got the highest weighted accuracy of 97.42%. Elevated accuracy levels do not necessarily equate to genuine predictive strength, particularly in the presence of uneven data distributions. Therefore, it is crucial to implement different evaluation techniques to more precisely gauge the model’s classifying proficiency. We also have applied the explainable artificial intelligence (XAI) technique, specifically LIME, to identify the most salient regions in the MRI images and to assign feature importance scores to the input variables. We have assessed the quality and usefulness of the explanations using various metrics.