Accurate Brain Age Prediction from MRI: Evaluating Kolmogorov-Arnold and Convolutional Networks

Alessandro Giupponi, Davide De Crescenzo, Marco Pinamonti, Manuela Moretto, Alessandra Bertoldo, Mattia Veronese, Marco Castellaro

Keywords: Brain aging, Convolutional Neural Networks, Kolmogorov-Arnold Networks, Neurological biomarker

TL;DR: This study demonstrates that Kolmogorov-Arnold Networks outperform traditional CNNs in predicting brain age from 3D MRI scans, offering more accurate estimations and promising improvements in neurological health assessment.

Abstract:

Brain age prediction using T1-weighted MRI has become a key biomarker for assessing
neurological health, with application in studying neurodegeneration (Soumya Kumari and
Sundarrajan, 2024; Mishra et al., 2023; Lea et al., 2021) and brain development (Tanveer
et al., 2023). While convolutional neural networks (CNNs) remain a standard approach,
recent advances suggest that Kolmogorov-Arnold Networks (KANs) may offer superior
performance in image-based task (Bodner et al., 2025; Li et al., 2024). In this study, we
present the first use of KANs for brain age prediction from 3D MRI scans, comparing their
performance against traditional CNNs. Experimental results show that KAN-based models
reduce estimation errors, highlighting their potential for improving brain age assessment.

Submission Number: 55