Google AI Tool Estimates Body Fat From Smartphone Photos

Google Research has developed an artificial intelligence system called PhotoScan that can estimate body fat percentage and distribution using only standard smartphone photos, potentially offering a more accurate and accessible alternative to wearable sensors. The research, published on the scientific archive Arxiv.org, indicates the tool could provide insights into metabolic health risks like insulin resistance without requiring clinical imaging. The system was trained using data from over 35,000 health records from the UK Biobank and refined with scans from 677 additional adults. Researchers used dual-energy X-ray absorptiometry (DXA) scans, considered the gold standard for body composition analysis, along with MRI data to teach the AI. The model was then further trained on actual smartphone camera images. In tests, PhotoScan's estimates of body fat percentage were within about two percentage points of DXA scan results, outperforming the bioelectrical impedance analysis (BIA) sensors commonly found in smartwatches and scales, which were nearly three points off.