Towards Explainable Image Analysis for Alzheimer's Disease and Mild Cognitive Impairment Diagnosis

One of the most common neurodegenerative dis-orders is Alzheimer's disease (AD), with almost 25 million people globally diagnosed with tentative AD. The symptom most commonly identified during the earlier stages of AD is a short-term memory loss, which also remains a notable symptom of Mild cognitive impairment (MCI). Due to the disease, memory and mental abilities gradually deteriorate, and thus the ability to cope with daily activities decreases. Early stage diagnosis is critical as it can slow down progression of the symptoms. That is why, in this work, we propose a novel approach to provide early AD diagnosis as well as the quantifiable reasoning behind it using machine learning. We compared three different machine learning algorithms-Support-Vector Machine (SVM), Artificial Neural Network (ANN) and Convolutional Neural Network (CNN)-for the task of classification of Electroencephalogram (EEG) signals. We have also automated the tuning process for ANN and CNN and tested the ability of our explainable artificial intelligence (XAI) to provide computational reasoning of the prediction. Using an open dataset with electroencephalogram (EEG) signals from AD and MCI patients and the Local Interpretable Model-Agnostic (LIME) library, our approach can identify brain regions that are believed to be indicative of the onset or progression of AD/MCI cases.