GCN-KAN: Graph Kolmogorov-Arnold Networks for Interpretable Alzheimer's Disease Diagnosis from Structural MRI

Liang Dong, Tianqi Ding, Keith Evan Schubert

Confirmation: I have read and agree with the IEEE BHI 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: Alzheimer's disease, explainable artificial intelligence, graph convolutional networks, Kolmogorov-Arnold networks, neuroimaging, interpretable diagnosis.

TL;DR: GCN-KAN combines Graph Convolutional Networks with Kolmogorov-Arnold Networks to enhance both accuracy and interpretability in Alzheimer's disease diagnosis from brain MRI data.

Abstract:

Alzheimer's Disease (AD) is a progressive neurodegenerative disorder that poses significant diagnostic challenges due to its complex etiology. Graph Convolutional Networks (GCNs) have shown promise in modeling brain connectivity for AD diagnosis, yet their reliance on linear transformations limits their ability to capture intricate nonlinear patterns in neuroimaging data. To address this, we propose GCN-KAN, an architecture that integrates Kolmogorov-Arnold Networks (KANs) into GCNs to enhance both diagnostic accuracy and interpretability. Leveraging structural MRI data from 91 subjects, our model employs learnable spline-based transformations to better represent brain region interactions. Evaluated on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, GCN-KAN outperforms traditional GCNs by 5.2\% in classification accuracy (62.6\% vs. 57.4\%) while providing interpretable insights into key brain regions associated with AD.
The model identifies the hippocampus, inferior parietal gyrus, and amygdala as most critical for diagnosis, with normalized importance scores of 0.65, 0.61, and 0.60, respectively.
These identified regions align with established neurological research on AD pathology.
This approach offers a robust and explainable tool for AD diagnosis, potentially facilitating earlier intervention and more personalized treatment planning.

Track: 3. Imaging Informatics

Registration Id: 25NB8JP2JRL

Submission Number: 29