7 Tesla MRI + ML achieves 100% accuracy in Parkinson's motor subtype stratification

A team led by Anne Louise Kristoffersen at the Norwegian University of Science and Technology applied 7 Tesla quantitative MRI combined with machine learning to objectively stratify Parkinson's disease motor subtypes. They included 24 people with PD (PwP) and 21 healthy controls, classifying them into Healthy Controls (HC), Postural Instability and Gait Difficulty (PIGD), and Tremor Dominant (TD). A U-Net deep learning model performed automatic brain segmentation, achieving a mean Dice Similarity Coefficient of 0.86 across all regions of interest. Using two classification approaches—one with all extracted features and one with optimized feature selection via machine learning—the team tested three tasks: HC vs PwP, PIGD vs TD, and multiclass (HC vs PIGD vs TD).

With feature selection (Approach B), performance soared: HC vs PwP reached 82% accuracy (AUC 0.93), PIGD vs TD achieved 100% accuracy (AUC 1.00), and the multiclass task hit 73% accuracy (AUC 0.91)—all dramatically better than using all features (Approach A: 69%, 69%, 62% respectively). The authors emphasize that low-dimensional, interpretable imaging signatures from 7T qMRI can support objective PD diagnosis and phenotype stratification. They note that larger multi-site studies are needed to validate generalizability. The work opens the door for precision medicine in Parkinson's, enabling tailored treatments based on motor subtype imaging biomarkers.