Google's PhotoScan estimates body fat from selfies more accurately than wearables

Full text

TL;DR

Wearables can estimate body fat using bioelectrical impedance analysis, but really accurate readings require X-ray scans.

Google Research shares a new technique for delivering high-quality estimates based on selfies.

Beyond just breaking down body composition, the system could also be used to predict insulin resistance.

Most of us have a weight problem. According to the NIH , almost 3 out of 4 American adults (73.1%) are either overweight or obese. But getting the full picture of your body composition is a whole lot more complicated than just stepping on a scale , and some of the most accurate ways to assess how and where we carry fat rely on complicated X-ray analysis. Some Google researchers have been working an easier way to offer that same sort of insight, using little more than the cameras on our phones.

PhotoScan is Google Research’s system for estimating body composition by using AI to analyze a series of 2D images of your body. First, Google trained the system using some of the data from those actual X-ray scans (a technique called dual-energy X-ray absorptiometry, or DXA), measuring both overall body fat levels, as well as ratios between core and waist/leg fat (A/G ratio), and organ and subcutaneous fat (V/S ratio).