Explainable AI for Medical Image Processing: A Study on MRI in Alzheimer’s Disease
Linda M. Duamwan, Jordan J. Bird
Abstract
Alzheimer’s disease is the most common type of dementia, characterised by memory-related brain changes that impair the patient’s cognitive abilities. Early detection of this disease is critical from a clinical point of view to increase the chances of treating a patient at risk of further cognitive degeneration. The rate of development and expansion in the field of deep learning for medical analysis is rapidly increasing alongside the increased incidences of neurodegenerative diseases. Machine learning algorithms are often applied to automate tasks and alleviate issues, but modern methods such as neural networks often present as black boxes. In the field of medicine, it is crucial to understand why a machine learning algorithm has made a prediction. In this study, we initially utilise a CNN-based approach for computer vision in the detection of Alzheimer’s disease from the ADNI MRI dataset. On unseen magnetic resonance imaging scans, our algorithm achieves a classification accuracy of 94.96%. We then implement the Local Interpretable Model Agnostic Explanations (LIME) algorithm to reveal visual evidence to support the predictions made by the model, and automatically visualise image segments contributing to predictions via Felzenszwalb’s segmentation algorithm. The objectives of explainable AI in this field are to provide medical professionals with specific, easy-to-understand information to support efficient, consistent, and convenient diagnoses.