Seeing beyond BMI: Estimating cardiometabolic risk with smartphone imagery

Summary

Google Research demonstrates the feasibility of PhotoScan, a deep learning approach that estimates body composition from smartphone photos to predict insulin resistance with accuracy comparable to DXA scans in a clinical research setting. Insulin resistance, a critical driver of metabolic disease, often precedes type 2 diabetes by years, and a HOMA-IR score above 2.9 indicates insulin resistance. The study highlights that additional body composition biomarkers, such as the Android-to-Gynoid fat ratio and visceral-to-subcutaneous fat ratio, provide deeper clinical insights than total body fat alone. While DXA scans are the gold standard for body composition measurement, they are expensive, require specialized infrastructure, and involve low-dose radiation, making PhotoScan a promising alternative for everyday screening.