NigrosomeNet: An Automatic Framework to Quantify Nigrosomal Neurodegeneration in Substantia Nigra Using Deep Learning

Motivation: Early differential diagnosis of parkinsonism-related and unrelated disorders, particularly dementia with Lewy bodies vs Alzheimer's, is crucial for clinical management. Nigrosome 1 (N1) roughly corresponds to the dorsolateral nigral hyperintensity visible on iron-sensitive imaging, aids in identifying early parkinsonian neurodegeneration. Goal(s): Developing an automatic tool for N1 segmentation enhancing diagnostic speed and accuracy for neurodegenerative disorders. Approach: NigrosomeNet employs convolutional neural network, trained and validated on MRI data, to automatically segment N1. Validation included dice similarity comparisons with expert annotations. Results: NigrosomeNet achieved fast and high segmentation accuracy (dice coefficient of 0.96 for parkinsonian patients), showing a significant N1 volume difference between groups. Impact: NigrosomeNet provides a rapid, reliable, and rater-independent solution for N1 analysis, enhancing diagnostic accuracy in clinical settings. This tool could significantly streamline neurodegenerative disease management, support large-scale studies, and reduce the need for specialized training, making early diagnosis more accessible.