Seeing beyond BMI: Estimating cardiometabolic risk with smartphone imagery
Google Research 1 month ago 45
Researchers developed PhotoScan, a deep learning system that estimates body composition metrics from smartphone photos to predict insulin resistance risk. The model was trained on 35,323 UK Biobank records and validated on 132 participants, achieving a body fat percentage error of 2.13% and an insulin resistance classification AUROC of 0.760, nearly matching the gold-standard DXA scan at 0.773. This smartphone-based approach offers a scalable alternative to expensive clinical imaging for early metabolic disease screening.