ARCHIVES

Brain age estimation using deep learning provides a sensitive holistic biomarker of neurodegeneration, but conventional models generally compress a complex three-dimensional MRI scan into a single global age estimate. This paper proposes an explainable artificial intelligence (XAI) framework that extends brain age estimation to voxel-level Brain Age Gap (vBAG) mapping for differential characterization of Alzheimer’s disease (AD) and Parkinson’s disease (PD). A three-dimensional fully convolutional network (3D-FCN) is trained on cognitively healthy participants to learn normative structural aging patterns from T1-weighted MRI. A localized bias- orrection procedure is then applied to reduce age-dependent regression-to-the-mean effects. The corrected model is evaluated on ADNI and PPMI cohorts, and three-dimensional Grad-CAM is used to identify anatomical regions that contribute to accelerated brain-age predictions. The proposed analysis retains the paper’s reported global Brain Age Gap results and downstream support vector machine (SVM) classification results, while interpreting them in the context of established neuroanatomical evidence. The framework is designed to distinguish disease-specific spatial patterns that may be obscured by a single global brain-age score. The reported maps emphasize medial temporal and hippocampal regions in AD and subcortical/brainstem structures in PD. The approach therefore aims to bridge predictive accuracy and anatomical interpretability in computational neuroimaging.