Voxel-Level Importance Maps for Interpretable Brain Age Estimation

Brain aging, and more specifically the difference between chronological biological age of a person, may be promising biomarker for identifying neurodegenerative diseases. For this purpose accurate prediction is important but localisation areas that play significant role in also crucial, order to gain clinicians’ trust reassurance about performance model. Most interpretability methods are focused on classification tasks cannot directly transferred regression tasks. In study, we focus task brain from 3D Magnetic Resonance (MR) images using Convolutional Neural Network, termed We interpret its predictions by extracting importance maps, which discover parts most age. do so, assume voxels not useful resilient noise addition. implement model aims add as much possible input without harming average maps subjects end up with population-based map, displays regions influential task. test our method 13,750 MR UK Biobank, findings consistent existing neuropathology literature, highlighting hippocampus ventricles relevant aging.