Hands as Diagnostics: AI Detects Acromegaly From Hand Images

Japanese researchers trained a modified ResNet-50 AI on more than 11,000 dorsal-hand and clenched-fist images from 716 patients across 15 facilities to diagnose acromegaly, achieving 89% sensitivity, 91% specificity, and an AUC of 0.96, outperforming 10 board-certified endocrinologists; the model employs privacy-by-design by excluding facial data, aiming for deployment in public health settings, though external validation and generalizability beyond Japanese patients remain to be shown.