Explainable AI Reveals Hidden Biomarkers in Parkinson's Brain Imaging

Parkinson”s disease (PD) is the second most common neurodegenerative disorder, characterized by progressive dopaminergic neuron loss and diverse motor and non-motor symptoms. Early and accurate diagnosis remains challenging due to the lack of definitive biomarkers. While DaTSCAN SPECT imaging supports clinical assessment, interpretation often centers on striatal regions, potentially overlooking other areas involved in disease progression. This study presents a robust deep learning approach to classify PD and investigate the role of non-striatal regions using explainable artificial intelligence (XAI). A 3D Convolutional Neural Network (3D CNN) was trained on DaTSCAN volumes from the PPMI database to distinguish controls from PD patients. Model interpretability was enhanced using Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize regions influencing predictions. The CNN model achieved high accuracy (>97<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>) across all stages. Grad-CAM heatmaps consistently revealed attention to cortical and subcortical areas beyond the striatum, suggesting distributed functional changes that may serve as early or complementary biomarkers. These findings highlight the importance of integrating explainable AI in neuroimaging to improve diagnostic precision.