Visceral Fat: What Imaging Shows Scans Can't
A 2023 study in Frontiers in Endocrinology set out to measure how well common scanning methods estimate visceral fat, using full 3D MRI volume as the reference standard. The gap between methods was large enough to matter: some tools tracked the true volume closely, others only loosely. It's a useful explanation for anyone who has seen a body-fat scale and a clinical scan disagree.
Two fats, one reading
Visceral fat sits deep in the abdomen, around the organs. Subcutaneous fat sits just under the skin. Most consumer tools, and even a lot of clinical ones, collapse both into a single body-fat percentage, even though they behave differently and sit in different places in the body.
The researchers behind this study used data from 4,558 UK Biobank participants and compared four ways of estimating visceral fat: whole-body MRI (the reference), DXA, bioelectrical impedance (BIA, the technology in most smart scales), and single-slice CT, the method most clinics actually use day to day.
Where each method lands
DXA correlated very strongly with the full MRI volume, R² = 0.94. BIA, the current running through a smart scale, correlated only moderately, R² = 0.49. That's a meaningful gap between the two technologies, and it means a bioimpedance reading is a rough proxy, not a measurement of visceral fat itself.
DXA's accuracy also wasn't constant. In leaner participants, those with BMI under 20, its correlation with MRI dropped to R² = 0.62. The same tool performs differently depending on who's standing on it, which is worth knowing before treating any single reading as exact.
The problem with a single slice
A single CT slice at the waist, still the default in many studies and clinics, explained only R² = 0.51 to 0.64 of the variance in true whole-abdomen visceral fat volume, even at the best vertebral levels. One slice simply doesn't capture how fat is distributed up and down the torso.
Stacking two slices together, at vertebrae L2 and L3, pushed that correlation up to R² = 0.92, almost matching DXA. The lesson isn't that any one method is useless. It's that a single point-in-time, single-location reading is an approximation of a volume, not the volume itself.
This was a correlational, cross-sectional analysis of existing imaging data, not a trial, and the authors note real limits: BIA equations can be biased across ethnic groups, and hydration status can skew DXA-based fat estimates. None of these tools is wrong so much as incomplete on its own.
What this means for tracking yourself
Trainr builds a 3D scan from two phone photos and tracks waist, chest, arm, and body-fat estimates over time. None of that replaces MRI or DXA, and the body-fat number, like any consumer estimate, should be read as an approximation. What a phone scan can do well is something imaging studies don't optimize for: consistent, frequent tracking of the same body, so the trend line becomes the signal rather than any single number.
If your waist measurement in Trainr moves down over eight weeks while your arm measurement holds steady, that pattern is more informative than one isolated scan, in the same way a stacked CT reading beat a single slice in this study. The first scan is free, and the value compounds the more consistently you use it, because change over time is the thing worth watching.
Scan yourself — the first one is freeA single scan is a snapshot of a volume, not the volume itself — which is exactly why the trend matters more than the number.
Trainr is a wellness tool, not a medical device. Measurements and body-composition figures are estimates for self-tracking; for medical questions, talk to a professional.