Google develops AI tool to estimate body fat via smartphone photos

Google Research has developed an experimental AI framework called PhotoScan that estimates body composition using only smartphone photographs. By analyzing frontal and side-view images, the deep learning model can estimate body fat percentage, the waist-to-hip ratio, and the relationship between visceral and subcutaneous fat.

In research tests, the system demonstrated a strong concordance with medical-grade Dual-energy X-ray Absorptiometry (DXA) scans. The AI's ability to predict insulin resistance showed accuracy levels approaching clinical standards. Notably, the technology outperformed bioelectrical impedance analysis (BIA) sensors commonly found in smartwatches, which often provide less precise estimations.

The model was trained using over 35,000 records from the UK Biobank, utilizing MRI data to create 2D projections for the neural network. While the technology offers a potentially accessible way to monitor metabolic health and fat distribution—factors more critical to health than Body Mass Index (BMI)—it remains a research prototype and is not yet a validated medical diagnostic tool.