Body Composition: InBody vs Smart Scale vs DEXA

A gym analyser says your body fat has fallen 1.3 percentage points since March. The scale at home says it has gone up. A DEXA report puts the figure several points above both. None of the three devices is broken. Each measures one physical quantity, models the rest from a prediction equation you cannot inspect, and carries its own noise.

What follows is what the five common methods measure, how far one reading can sit from a reference method, and the question that decides whether any of it is worth acting on: how large a change has to be before it is signal rather than noise.

Key Takeaways:

• No body-composition device measures fat directly. All estimate it from a surrogate: electrical resistance, air displacement or X-ray absorption, converted via a population equation you cannot inspect.
• Correlation with a reference method tells you about groups. Limits of agreement tell you about individuals. The two can be far apart.
• A reading can shift by 1 to 2 kg on an impedance device with no tissue change at all, from hydration, food, bladder fill or cycle phase.
• The least significant change is the smallest difference between two readings that exceeds measurement noise at 95% confidence. Almost no device printout states this figure, but it is the number that decides whether a change is real.
• The method matters less than the discipline of repeating it under the same conditions each time.

The first thing to know about a body-composition number is how much of it was measured and how much was calculated.

What each method physically measures, and what it infers from a model:

• DXA (DEXA): Physically measures the attenuation of two X-ray beams per pixel, separating bone mineral, lean soft tissue and fat. Infers the soft-tissue split in pixels that also contain bone.
• Multi-frequency segmental impedance (InBody type): Physically measures impedance in arms, legs and trunk at several frequencies, reflecting total body water and its extracellular and intracellular split. Infers fat mass, fat-free mass and regional composition from proprietary equations using water volumes, age, sex, height and weight.
• Consumer smart scale: Physically measures weight plus impedance across a single-frequency foot-to-foot path. Infers fat mass and body-fat percentage from embedded equations that are rarely disclosed.
• Air-displacement plethysmography (Bod Pod): Physically measures whole-body volume from pressure changes in a sealed chamber. Infers density converted to fat percentage by two-compartment equations.
• Skinfolds: Physically measures subcutaneous fat thickness at standard sites with calipers. Infers body density, then fat percentage, from population-specific equations.

A device that infers composition from water content is reporting a hydration estimate dressed as a fat estimate, which is why the impedance family behaves as it does below.

Validation papers usually lead with a correlation coefficient, and a correlation coefficient is close to useless for one person. Correlation is a population-level property and says nothing about the agreement for the individual in front of the device. What matters is bias, meaning how far the average reading sits from the reference, and limits of agreement, meaning how far an individual reading can sit from it.

DXA is the practical reference for the other four and is not itself exact. Against four-compartment models, which combine bone mineral, body volume and total body water, one series reported a mean bias of 0.84 percentage points with limits of agreement from about minus 5.33 to plus 7.01 points. In trained athletes, individual error has reached roughly 9 points and 6.3 kg of fat mass.

Segmental multi-frequency impedance correlates well with DXA, at r-squared often between 0.65 and 0.81, and still misses badly on individuals. In overweight and obese men, one system underestimated fat mass by about 6.6 kg and body-fat percentage by 7 points. The DXA-minus-InBody difference for total fat mass in obese adults ran from about minus 10 to plus 7 kg.

Consumer scales weigh accurately and estimate fat poorly. Median weight error against DXA sits near 0 to 0.3 kg, while a 2021 study found median underestimation of fat mass by 2.2 to 4.4 kg. Across 14 consumer devices tested against a four-compartment model, constant error ranged from minus 3.5 to plus 11.7 percentage points.

The one Indian dataset here deserves its own sentence. In urban south Indians, the CURES-3 validation found no significant mean difference from DXA for leg-to-leg impedance or skinfolds, while hand-held impedance underestimated by 4.45 plus or minus 4.83 points. The means agreed. Individuals still landed about 5 points either side.

This is the number that decides whether a follow-up test means anything, and it is the one almost no report prints.

Precision error describes how much repeated measurements vary on an unchanged body. Least significant change is what that error implies for a pair of readings: in densitometry practice it is 2.77 times the precision error, and it is the smallest difference between two measurements that exceeds noise at 95% confidence.

What each method's least significant change looks like in practice:

• DXA, same system, controlled: About 1.2 percentage points (body fat). Fat mass: 1.0 to 1.5 kg. Lean mass: 0.8 to 1.2 kg. Based on same-day precision studies.
• DXA over weeks to months: 2.5 to 3.5% relative change. Adds biological variation. Lean mass figure not separately published.
• Clinic multi-frequency impedance: 2.1 to 2.7 percentage points. Fat mass: 1.5 to 2.4 kg. Lean mass: 1.6 to 2.3 kg. Based on published change tables.
• Consumer smart scales, controlled: 1.5 to 2.8 percentage points. Fat mass: 1.0 to 1.9 kg. Lean mass: 1.6 to 2.4 kg. Based on device series, not specific brands.
• Air-displacement plethysmography: Around 3 percentage points. Fat mass: at least 2 kg. Based on detectable-change experiments.
• Skinfolds: At least 3 to 4 percentage points. Fat mass threshold not published. Based on limits of agreement.

Clinic impedance devices are more repeatable than they are accurate. Intra-unit coefficients of variation often sit below 1%, and standard errors of measurement against DXA fall between about 0.5 and 0.9 kg for fat mass. For the InBody 770, same-day standard errors of about 0.77 points, 0.54 kg and 0.58 kg give thresholds near 2.1 points, 1.5 kg and 1.6 kg under laboratory conditions. Field data without strict fasting or activity control show standard deviations near 2.4 kg for fat mass and 4.1 kg for fat-free mass.

Consumer devices are technically precise and practically noisy. A day-to-day series covering InBody and Omron units put real-change thresholds at 1.5 to 2.8 percentage points, 1.0 to 1.9 kg of fat mass and 1.6 to 2.4 kg of fat-free mass. Used at home without standardised timing, that is a floor rather than an estimate.

Impedance methods estimate fat-free mass from total body water, so anything that moves water moves the answer.

These are the main sources of within-day reading variation on an unchanged body:

• Fluid shifts from exercise, a sauna or a salty meal can change measured total body water by 0.5 to 1.0 litres and fat-free mass by several hundred grams.
• Glycogen state shifts body weight by 1 to 2 kg and fat-free-mass estimates by a similar amount during loading or depletion.
• Recent exercise raises muscle temperature and blood flow, lowering resistance in a way that reads as lean gain.
• Cycle phase has produced body-fat differences of about 1 to 2 percentage points in women, from fluid changes across the cycle.
• Food, fluid and a full bladder affect readings through stomach and bowel contents, blood volume and thoracic gas. Protocols require 3 to 8 hours of fasting and an emptied bladder. Skipping that can alter estimates by up to about 0.5 to 1.0 kg.

DXA is the least hydration-sensitive of the five acutely, though oedema or dehydration can change soft-tissue attenuation enough to affect lean-mass estimates. Set those magnitudes against the thresholds above and the rule writes itself: a fortnight-to-fortnight comparison on any impedance device is mostly measuring your week.

Cross-device comparison is where readers are misled most often, and the error runs in three directions.

Between impedance and DXA, individual differences commonly span about 5 kg of fat mass. A DXA result of 25% body fat can correspond to an impedance estimate anywhere from roughly 20% to 30% in the same person, so a change measured on one cannot be read against a baseline taken on the other.

Between analysers, the picture is better within a family than across the market. Research comparing InBody 230, 720 and 770 units found coefficients of variation under about 1%. Consumer devices varied widely against a four-compartment model cross-sectionally, while longitudinal validity was better. Tracking one device over time is defensible in a way that comparing two devices is not.

Between DXA machines, manufacturers differ in beam energies, calibration phantoms and soft-tissue algorithms. Absolute values can differ by several hundred grams to a few kilograms across systems, and a software upgrade can shift a baseline as effectively as a new machine. Serial scans belong on the same system at the same centre.

Indian radiation-safety training material puts the typical patient effective dose for a DEXA scan at about 0.001 mSv, roughly one microsievert. Natural background exposure runs near 3.1 mSv a year and the public dose limit is 1 mSv a year, so a body-composition scan sits orders of magnitude below both.

Under the Atomic Energy (Radiation Protection) Rules 2004, the Atomic Energy Regulatory Board sets limits of 20 mSv a year averaged over five years for occupational workers, 30 mSv in any single year, and 1 mSv a year for members of the public. These limits are explicitly not applied to patients undergoing diagnostic X-ray examinations.

AERB requires licensing of DEXA units and emphasises optimisation, but prescribes no maximum number of examinations per patient per year. What limits repeat scanning is clinical justification, which makes it a question for the doctor requesting the scan.

Every method except DXA converts a physical measurement into body composition using an equation derived in some population, and for Indian adults most of those populations were not Indian.

CURES-3 itself used manufacturer software for impedance and generic skinfold equations from the literature rather than Indian-specific ones. Published work on multi-frequency segmental impedance calibrated for Indian adults is limited, and most commercial analysers deploy global or Asia-wide equations that may not account for the higher body-fat percentage at lower BMI documented in Asian Indians.

The ethnicity setting on a device is less informative than it looks. Many analysers offer an "Asian" mode that alters the underlying equations or hydration assumptions, and manufacturers rarely disclose the derivation dataset behind it. Whether that setting reproduces Indian patterns or applies a broad regional average is not established.

Indian DXA reference values for whole-body composition exist only in fragments in the open literature, so there is no complete Indian normal range to read a result against.

The figures below are assembled from provider and manufacturer listings rather than any published tariff, and vary by city, facility and package.

Indicative costs and availability for each method:

• DXA whole body: Available at tertiary hospitals, diagnostic chains and some mobile services. From about ₹2,999, often bundled into health packages.
• Clinic multi-frequency impedance: Available at hospitals, sports centres, gyms and corporate wellness. About ₹500 to ₹2,000, commonly ₹800 to ₹1,200.
• Consumer smart scale: Available via e-commerce and brand channels. About ₹1,500 to ₹6,000 for generic Bluetooth models; multi-frequency home units at ₹22,999 to ₹41,999.
• Air-displacement plethysmography: Available at sports-science and research institutions. No India-wide pricing located.
• Skinfolds: Available at gyms, clinics and field settings. No separate pricing located.

Two caveats belong here. The impedance figures come from manufacturer pages, so they are evidence of what is claimed rather than of what is charged. And no primary-source national fee schedule for any of these tests in India was identified in public regulatory or insurance documents as of August 2026.

Body composition is one of the few measurements where consistency beats accuracy. A smart scale with a four-point bias, used at the same time each morning in the same fasted state, will show a trend more honestly than alternating between a gym analyser and a hospital DXA. A fixed bias cancels within a series. Disagreement between series does not.

For a reading to be meaningfully comparable over time, six things need to be the same each time:

• Same device
• Same time of day
• Same hydration and fasting state
• Same menstrual cycle phase, if applicable
• Same clothing
• Same electrode contact method

Leave enough time between readings for a change to clear the thresholds listed above. When a series does move by more than its threshold, the useful question is what it means alongside your weight trajectory, your strength and your blood results. That conversation belongs with a registered medical practitioner or a registered dietitian.

Want your body-composition reading read alongside your clinical markers rather than in isolation? Book an assessment with a Voy clinician.