AUTHOR=Tekkesinoglu Sule , Pudas Sara 

TITLE=Explaining graph convolutional network predictions for clinicians—An explainable AI approach to Alzheimer's disease classification

JOURNAL=Frontiers in Artificial Intelligence

VOLUME=Volume 6 - 2023

YEAR=2024

URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2023.1334613

DOI=10.3389/frai.2023.1334613

ISSN=2624-8212

ABSTRACT=Graph-based representations are becoming more common in the medical domain, where each node defines a patient within a potentially large population, and the edges signify associations between patients, relating individuals with disease and symptoms in a node classification task. In this work, a Graph Convolutional Networks (GCNs) model trained on the Alzheimer's Disease Neuroimaging Initiative (ADNI) database aimed to capture differences in neurocognitive, genetic, and brain atrophy patterns predictive of cognitive status, from Normal Cognition (NC) to Mild Cognitive Impairment (MCI) and Alzheimer's Disease (AD). Explainability is a crucial design feature in medical applications to promote clinical adoption and physician trust. We introduce a decomposition-based explanation method for individual node classification. By measuring the output variations resulting from decomposing input values, we determine the degree of impact of input values on the prediction. We thereby explain how each feature from different modalities at both individual and group-level contributes to a diagnostic result. We also studied relational data by silencing all the edges of a particular class, given that graph data contains critical information in edges. Our functional evaluation analysis of the proposed work showed high stability (e.g., low variations in explanations) concerning minor changes in input values (for edge weights >.80). Our human-grounded evaluation survey of domain experts (N=11) confirmed the validity of the explanations provided by the model as 71% of the responses agreed on the correctness of the explanations. The understandability of the explanations was rated as above six * Data used in preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (http:// adni.loni.usc.edu/). As