A team from Shanghai Jiao Tong University and Kyoto University releases the first open Niacin Skin-Flushing Response dataset and applies an Efficient-Unet for precise area segmentation, then employs an SVM classifier to distinguish healthy controls from psychiatric patients based on normalized skin-flush metrics.

Key points

• Open NSR dataset: 600 photos from 120 subjects with binary masks and manual scores.
• Segmentation: Efficient-Unet achieved 91.31% Dice and 84.06% IoU without post-processing.
• Classification: SMOTE-balanced SVM with 5-fold CV reached 60–65% sensitivity and 75–88.3% specificity across psychiatric categories.

Why it matters: This device-independent AI approach offers a scalable, objective biomarker platform that could transform psychiatric diagnostics by reducing subjectivity and resource barriers.