Google uses AI to estimate people's body fat based on smartphone photos they take. This method is said to estimate body fat more accurately than wearables. This PhotoScan AI is also said to be able to 'detect' insulin resistance, often a precursor to diabetes.

So their photoscan bodyfat % has a lower RMSE in predicting (DXA) body fat than their BIA bodyfat % ... but what BIA estimation did they use? I've been told that not all BIA is equal. They tell us:

> Fitbit Sense 2 was used to measure BIA on the wrist. It implements a 4-wire sensing mechanism using a 50kHz excitation frequency and utilizes two bottom electrodes, two top electrodes, and TI’s AFE4500 chip.

... huh I wasn't aware that fitbit offered wrist-based BIA? And I've heard that systems that have sensors across a long path in your body are better (e.g. foot and hand contacts)? But most confusing, the specs for Fitbit Sense 2 [1] don't say anything about BIA, just cEDA. I'm not an expert by any means, but google's own AI tells me that EDA sensors cannot be used for BIA. So did they basically use an especially low quality BIA measurement?

[1] https://store.google.com/us/product/fitbit_sense_2_specs?hl=...

Can one try out this photoscanner?