Google Research unveils PhotoScan for body fat assessment
Google Research published PhotoScan on August 17, 2026, using smartphone cameras to assess body fat with accuracy close to DXA scans.
Stock photo for illustration only, not from the actual event
- Google Research published the PhotoScan study on August 17.
- It utilizes 2D smartphone photos to evaluate body fat composition.
- Results show accuracy comparable to clinical DXA scans.
- Integration into wearable products is planned for September 2026.
Google Research published a new research paper detailing PhotoScan on August 17, showcasing an experimental Deep Learning approach designed to evaluate human body composition using standard two-dimensional smartphone camera images. This initiative highlights the tech giant's ongoing push to bring advanced health diagnostic capabilities directly to consumer mobile devices.
According to Google, the PhotoScan system achieved prediction accuracy closely rivaling DXA scans when estimating insulin resistance, while outperforming the BIA sensors typically embedded in modern smartwatches. The evaluation system produces three key metrics upon processing user photographs taken from both front and side angles: body fat percentage, trunk-to-hip fat ratio, and visceral-to-subcutaneous fat ratio.
To develop the system, the research team initially trained a ResNet-50 neural network architecture using data from 35,323 participants in the UK Biobank database. They subsequently fine-tuned the model with the PhotoBIA dataset containing smartphone images from 677 volunteers, before testing it on an independent cohort of 132 individuals from the MetabolicMosaic study based in San Francisco.
Testing against the PhotoBIA group revealed that PhotoScan achieved a mean absolute error of 2.15 for body fat percentage, compared to 2.91 for the BIA sensor model. Furthermore, when evaluating insulin resistance classification, integrating PhotoScan data raised the AUROC score from a baseline demographic model of 0.692 up to 0.760, approaching the 0.773 score achieved by incorporating clinical DXA data.
Stock photo for illustration only, not from the actual event
Google's development of a smartphone-based body composition scanner represents a significant effort to bypass the logistical barriers of clinical DXA equipment, which is costly and inaccessible for daily tracking, while overcoming the precision limits of standard wrist-based BIA sensors that only yield rough total body fat estimates. By applying deep learning models to standard 2D photographs, this technology bridges the gap between sophisticated medical diagnostics and everyday consumer accessibility.
This research directly feeds into upcoming software features such as Insulin Resistance Trends, designed to assess insulin resistance tendencies utilizing heart rate, sleep, and movement metrics without requiring blood draws. This feature is scheduled for rollout starting in September 2026 across devices including Pixel Watch 3, Pixel Watch 4, and Fitbit Air. Meanwhile, competitors like Samsung continue relying on BioActive sensors featuring BIA technology in wearables such as the Galaxy Watch Ultra2 launched on July 22.
Google concluded that while DXA scans remain the clinical gold standard despite being difficult to scale, and wearable BIA sensors offer convenience with limited basic metrics, PhotoScan emerges as a compelling middle-ground alternative. Nonetheless, the research team emphasized that these findings currently represent technical proof-of-concept possibilities and remain strictly in the research prototype phase.
Source: Thairath Lifestyle
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment