Accuracy2

Module 1

Fit3d Accuracy Reference

The one page to keep next to you during a consult. Everything here is expanded in the modules that follow.
‍

SNAP scan · practitioner quick referenceSNAP

The numbers to use

Body composition

Body-fat %

Benchmarked against DXA: within about 3.5 points on average, and more repeatable than DXA itself. Use the value.

Measurements

Waist and waist-to-height ratio

Read straight off the 3D model, as close as a trained tape measurer. Under 0.5 is the common target.

Protect

Lean mass

The number to watch during weight loss: some lean loss is normal, but fat should fall much faster.

Risk signal

Visceral fat

Estimated from body shape, calibrated to DXA. A strong cardiometabolic risk indicator.

Trust a scan when

The patient wore minimal clothing, or fitted gym wear
They wore the same type of clothing as at baseline (minimal, or fitted)
They weighed themselves just before scanning
The change holds across two or more scans

Discount a scan when

Clothing was loose or bulky, or a different type from baseline
Weight wasn't updated
A single scan jumps against the trend

What to tell patients

"Your scan is benchmarked against DXA, the clinical reference for body composition."
"Each scan is useful on its own. Over weeks, it tells us even more."
"Scan in the same type of clothing, at the same time of day, after weighing yourself."

If a result looks off

1 · Check how it was taken. Clothing, the weight entered that day, and scan conditions explain most surprises.

2 · Rescan if needed. Same type of clothing as baseline, weighed just before. A real change shows up again.

3 · Look at the drivers. If it still doesn't add up, Module 3 shows what feeds each number. Very muscular or very lean bodies are the usual reason.

Module 2

How a scan works

A patient turns once in front of their Tablet.

SNAP's body foundation model reads the video and estimates their body composition and 3D body shape directly from the images.
‍

Capture

About 10 seconds of video. The patient turns 360° in an A-pose, phone at waist height.

Foundation model

SNAP's body foundation model reads the frames directly: the body's shape and its surface, skin included.

Estimate

The same model estimates the metrics, from body-fat % to visceral fat and lean mass, trained and validated against DXA.

3D body

It also estimates the body's shape, and the 3D model is built from it. Circumferences are read off it at fixed landmarks.

Report

Values, health bands, and percentiles against population data by age and sex.

The one idea to remember

Almost every number in the report comes from one body-fat estimate. Once body-fat % is known, fat mass, lean mass and muscle follow by arithmetic. Get the scan right and everything downstream is right. Get it wrong (loose clothing, old weight) and every number shifts together.

Body-fat %→ Fat mass · Lean mass→ Limb lean mass→ Skeletal muscle→ Metabolic age · BMR

What the models need from the patient

Body shape and surface, read from the scan. Ideally as much visible skin as possible: the model reads the skin surface, not just the outline. Clothing that hides the body makes it look bigger than it is.
Current weight, entered before every scan. Every mass value is a share of this weight.
Sex, age and height, set once. Health bands and percentiles depend on sex and age.

Module 3

Where the numbers come from

For each metric: what it's built from, what moves it most, and how much weight to give it. The bars show how strongly each input drives the result, relative to the strongest one.

MEASURED · high confidence ESTIMATED · benchmarked against DXA DERIVED · for communication

Waist circumference and waist-to-height ratio

MEASURED

3D body model with waist, stomach and hip measurement rings, and a height line

Read directly off the 3D model, not estimated. No model sits between the scan and the number. It's also the strongest input to body-fat % and visceral fat.

Waist is the high waist, just above the navel.
Stomach (low waist) is the widest point around the navel.
Waist-to-height ratio is waist divided by height. Under 0.5 is the common target.
The landmark is placed by the model the same way on every scan. Tape measurements vary with where the tape sits and how tight it's pulled.

Matching your protocol. If your clinic measures waist at a specific landmark (for example the iliac crest), check which of SNAP's two values sits closest, and use that one consistently.

Body-fat %

ESTIMATED

A model trained against DXA. It combines two views of the body: one driven by trunk shape relative to height, one anchored to BMI.

Trunk girth100

Height45

Limbs (thigh, chest)24

BMI7

Trunk girth (waist, stomach, hips) carries the most weight. When trunk girths rise together, which is what real fat gain looks like, the estimate moves strongly.

Band	Men	Women
Minimum	5 to 10%	13 to 18%
Healthy	10 to 20%	18 to 28%
High	20 to 25%	28 to 32%
Very high	25% and above	32% and above

Bands adapted from ACE and ACSM guidelines.

Visceral fat

ESTIMATED

Fat around the organs, more closely tied to cardiometabolic risk than total fat. Estimated from body-shape ratios and body-fat %, calibrated to DXA. It is not a CT or MRI measurement.

WomenMen

Waist ÷ height100
91

Stomach ÷ height·
100

Hips ÷ height53
·

Body-fat %31
11

Chest ÷ height·
7

Bars are relative to the strongest input for each sex. A dot means the input isn't used.

In women, waist dominates. In men, the model reads the contrast between stomach and waist: a belly that protrudes below the waist line pushes the estimate up.

Fat mass and lean mass

ESTIMATED

A direct split of body weight. No second model.

Fat mass = weight × body-fat %
Lean mass = weight minus fat mass (muscle, water, organs, bone)

Both inherit the accuracy of body-fat %, and both depend on the weight entered that day. For weight-loss programs, lean mass is the number to watch: some lean loss is normal, and the outcome you want is fat falling much faster than lean mass.

Limb lean mass and skeletal muscle

ESTIMATED

Appendicular lean mass (arms and legs) is predicted from body-fat %, BMI and body roundness, trained on DXA. Skeletal muscle mass is then a near-linear function of it.

Body-fat %100

BMI34–60

Body roundness3–7

Body roundness index (BRI) describes how round, rather than tall and narrow, the body is, using waist and height. Two people with the same BMI can have very different BRI.

Reported in kg, and as an index (mass ÷ height²) with a low, healthy or high category and an age and sex percentile. Muscle here is inferred from body shape and fat, not imaged directly.

Metabolic age

DERIVED

Body-fat % translated into an age: the average age of people in a ~20,000-person US population survey (NHANES) with similar body fat.

A communication device, not an independent biomarker. It makes adiposity easier for patients to relate to. The underlying signal is body-fat % and its percentile. The display is capped at 15 years either side of real age.

BMR and daily energy needs

DERIVED

Resting energy expenditure estimated from lean mass, then scaled to a daily total at a sedentary activity level.

Intentionally basic: a sensible starting point, not an indirect calorimetry result. Calorie recommendations never go below 1,200 kcal a day.

Module 4

How accurate it is

Two questions matter clinically. Is the number close to the reference method (accuracy)? And does it give the same answer twice (repeatability)? For tracking a patient over time, repeatability matters most.

Source: SNAP validation data

Waist circumference

vs a multi-camera 3D body scanner · n = 78

Trained tape measurer±1.50%

SNAP vs reference±1.36%

SNAP, scan to scan±0.60%

As close to the reference as a trained tape measurer. On a 100 cm waist, repeat scans agree to within about 0.6 cm.

Body-fat %

vs DXA · n = 271

SNAP vs DXA±3.56 pts

DXA, scan to scan±2.00 pts

SNAP, scan to scan±0.41 pts

On average within about 3.5 points of DXA, close enough to use the value in a consult. And repeat SNAP scans agree far more closely than repeat DXA scans, so it's also very good at tracking change.

Visceral fat

vs DXA · n = 226

SNAP vs DXA±0.23 pts

DXA, scan to scan±0.10 pts

SNAP, scan to scan±0.02 pts

Very stable from scan to scan, so a sustained change in direction is meaningful.

Limb lean mass

vs DXA · n = 185

SNAP vs DXA±1.61 pts

DXA, scan to scan±0.40 pts

SNAP, scan to scan±0.20 pts

Good for direction of travel during weight loss. For a muscle-loss question that needs an exact value, DXA remains the reference.

Group figures, not guarantees

Accuracy is an average across a study population. An individual patient can sit further from DXA than the average.

The same on any phone

Results are consistent across current iPhone and Android devices, so patients don't need special hardware.

Every method has error

DXA, BIA and tape all disagree with each other. Pick one method per patient and track it, rather than comparing numbers across methods.

Module 5

Where it works, where it fails

Every body composition method has blind spots. These are SNAP's, with what you'll see in the results and what to do about it.

Works best for

Adults with overweight or obesity tracking change over a weight-management program. That's the population and use case the models were built and validated on.

Hardest for

Bodies far from the average shape: very muscular builds, very lean people, and people with low muscle at a normal weight. Their measurements (waist, hips, every circumference) stay accurate. It's the body composition estimates built from them that need more caution.

How the scan was taken

👕

Loose or bulky clothing

The scan measures the outer outline. Fabric adds centimetres.

You'll see

Body-fat % too high, lean mass too low.

Ask for a rescan in minimal clothing or fitted gym wear.

↘

Clothing type changes between scans

The most common trap. Patients often start in looser clothing and switch to fitted or minimal clothing later.

You'll see

Body-fat % falling faster than real change, and lean mass that looks better preserved than it is.

Agree the clothing type at baseline (minimal, or fitted). It doesn't need to be the same outfit. If the type changes, treat the next scan as a new baseline.

⚖

Weight not updated

Every mass value is a share of the weight entered.

You'll see

Body composition that doesn't fit the change. If the patient lost weight but the old weight was entered, the scan sees the smaller body but splits the old total, so fat mass or lean mass reads too high, depending on how their body changed.

Weigh on the same scale just before every scan.

≈

One scan against the trend

Day-to-day differences are mostly noise.

You'll see

A single jump or drop that the next scan doesn't confirm.

Act on changes that hold across two or more scans.

The patient's body

💪

Very muscular builds

Large circumferences from muscle look like large circumferences from fat.

You'll see

Body-fat % reads high.

Measurements still hold. Lean on waist and the trend for body-fat %. SNAP's Athlete Mode can be enabled for these patients.

◌

Low muscle at a normal weight

Sometimes called "skinny fat". Shape changes little as muscle is lost.

You'll see

Body-fat % can read low; circumferences barely move.

A normal result doesn't rule out low muscle. Pair it with strength, function and history.

▫

Very lean patients

Below roughly 15% body fat, body shape tells the models less.

You'll see

Larger error relative to the value.

Measurements still hold. Don't use small body-fat differences in this range to make decisions.

∅

Not validated

Pregnancy, patients who can't stand unaided and turn, limb amputation, conditions that change limb shape (lipedema, significant edema).

You'll see

Numbers that look normal but have no validation behind them.

Use waist trends only, or another method.

What each number is, and isn't

Benchmarked against DXA

Close to DXA on average. One patient can sit further from DXA than the average, so don't present it as a DXA result.