Parkinson's disease is the second most prevalent neurode-generative disorder after Alzheimer's disease. The brainstem, despite itsearly and crucial involvement in Parkinson's disease, is largely unex-plored in the domain of functional medical imaging. Here we propose adata-driven, connectivity-pattern based framework to extract functionalsub-regions within the brainstem and devise a machine learning basedtool that can discriminate Parkinson's disease from healthy participants.We first propose a novel framework to generate a group model of brain-stem functional sub-regions by optimizing a community quality function,and generate a brainstem regional network. We then extract graph theo-retic features from this brainstem regional network and, after employingan SVM classifier, achieve a sensitivity of disease detection of 94%--comparable to approaches that normally require whole-brain analysis.To the best of our knowledge, this is the first study that employs brain-stem functional sub-regions for Parkinson's disease detection ^[1].

A schematic diagram of the proposed framework of Parkinson's disease detection from brainstem sub-regions.