Diffusion Tensor Imaging Biomarkers for Parkinson’s Disease Symptomatology
Résumé
Voxel-based analysis is an invaluable tool for biomarker discovery in population neuroimaging. The traditional approach however is limited to local, linear biomarkers, determining if the linear correlation between the quantitative value of an image is correlated with the disease state at a single voxel. By analysing convolutional neural networks that directly predict clinical scores using a newly proposed voxel-based diktiometry, non-linear and non-local biomarkers can be visualised, leading to an additional tool for biomarker discovery. Our approach using diffusion tensor images to predict UPDRS3 and Hoehn and Yahr scores for Parkinson’s disease patients, shows consistent and explainable results. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.