Machine learning for Parkinson disease: a prospective analysis

Deep learning using fundus photography is a widely employed machine learning tool for medical image analysis. Recently, these tools have been applied to assess neurologic dysfunction in patients with Parkinson disease (PD).
This study (n=266 individuals with PD, n=349 with non-PD atypical motor abnormalities) demonstrated a high sensitivity of 83.23% and 82.61% for the Hoehn and Yahr and Unified Parkinson’s Disease Rating Scale part III scores assessment, respectively. However, a lower specificity for both assessments (Hoehn and Yahr =66.81% and Unified Parkinson’s Disease Rating Scale part III score =65.75%) were reported.
The authors concluded that deep learning is a promising method for assessing neurologic dysfunction in patients with PD and provides further insight on the association between the retina and brain.