Physiological Standardization of the Deep Health® Device Beyond DXA Soft Tissue Estimates

1 Introduction

A body-composition number acquires scientific meaning through the quantity it represents and the measurement procedure that supports it. A display labelled muscle mass may represent anatomical skeletal muscle, a regression estimate derived from appendicular lean tissue, or a prediction from electrical measurements. A protein output may represent a molecular compartment inferred through a multicomponent model. The shared use of kilograms does not make these quantities equivalent.

We developed the Deep Health® Device as a multisensor physiological measurement system. Its design objective is to obtain complementary information about the human body and to express that information as defined physiological outputs. This objective places a substantial burden on standardization: each output must be linked to an appropriate human reference, and the combined model must be evaluated across the biological variation relevant to its intended use.

The Peak Health Center standardization report of 23 June 2017 describes a cohort of 4,144 people and reference procedures for fat and muscle assessment. It identifies overall fat mass, muscle mass, visceral fat mass and subcutaneous fat mass as study targets. The report provides the methodological foundation for the present discussion [1].

DXA is important to this discussion because it is widely used as a body-composition comparator.

Its contribution to bone assessment does not remove the inference involved in soft-tissue outputs.

Fat, lean tissue and regional adipose estimates depend on different assumptions and processing steps. A device evaluated against DXA may closely reproduce DXA values while retaining the comparator's unresolved sources of error. This is a consequence of comparator choice, not a finding about any particular commercial product.

This article examines how that limitation changes the design of a standardization program. We describe the reference procedures in the historical report, explain the role that complementary sensor information can play, and specify the analyses needed to determine measurement agreement and population applicability. The article addresses measurement architecture and methodology; it does not report a new analysis of paired Deep Health® and reference observations.

2 DXA and the Soft Tissue Inference Problem

Physical information and tissue separation

DXA measures X-ray attenuation at two energy ranges and uses calibrated models to derive bone mineral and soft-tissue quantities. In a simplified material-decomposition model, two attenuation observations cannot uniquely determine an unrestricted set of three tissue components.

Algorithms therefore introduce constraints, including assumptions about soft-tissue composition in regions containing bone. Two-dimensional projection also limits anatomical separation along the X-ray path [2,3].

This information constraint explains why adding a complementary measurement can be useful. If a second modality supplies information about tissue geometry, water or compartment location, the combined model may resolve distinctions unavailable from attenuation alone. Increasing the number of channels that observe essentially the same property will not necessarily provide the same benefit. The contribution depends on what each channel measures and how independently it constrains the physiological quantity.

Bone mineral assessment and soft tissue outputs

Bone mineral content and areal bone mineral density are defined DXA outputs; soft-tissue outputs answer different questions. Regulatory clearance also applies to particular devices, functions and intended uses. Some DXA soft-tissue software has FDA clearance. The Hologic Visceral Fat Software record, for example, explicitly describes estimates of visceral fat area, mass and volume within the android region [4]. This language supports a precise distinction between an authorized estimation function and direct anatomical or chemical determination.

The scientific issue is therefore the physiological meaning and uncertainty of each output.

Clearance of a device function does not make all possible derived compartments equally observable, nor does the existence of an estimate make that estimate unusable. Standardization should establish the conditions under which the result is sufficiently accurate for its stated purpose.

Lean soft tissue and anatomical muscle

Lean soft tissue contains water, protein and other non-fat, non-bone constituents. It includes skeletal muscle but also organs, connective tissue and other structures. Kim and colleagues developed models that predict total skeletal muscle from DXA appendicular lean soft tissue, using MRI as the anatomical reference. Their work illustrates the extra modelling step between the DXA compartment and anatomical muscle [5].

A comparison between a Deep Health® muscle output and a DXA lean output must therefore define whether both methods target the same compartment. Otherwise, an apparent device error can partly reflect a difference in definitions. Calibration to an incorrectly matched target can produce a consistent number with the wrong physiological interpretation.

Hydration and short term biological variation

A change in lean mass over a short interval can include changes in water and glycogen rather than a corresponding change in muscle protein. In a study of 12 active young men, Toomey and colleagues reported a mean DXA lean tissue reduction of 1.69 kg after exercise and dehydration, followed by an increase of 2.36 kg after a carbohydrate supercompensation phase. The study showed how physiological state can change a DXA lean output [6].

The interpretation matters when a device is used to monitor training, nutritional change or recovery. If the intended quantity is anatomical muscle or protein, the standardization procedure should evaluate whether water and glycogen shifts affect the output. A hydration-sensitive sensor can help characterize this problem when it has been calibrated appropriately. Electrical measurements can themselves respond to fluid state, so their inclusion requires testing rather than an assumption of correction.

Other DXA errors can depend on sex and regional adiposity. Hansen and colleagues studied 75 older adults using nuclear and DXA methods. Their findings supported useful skeletal muscle estimation while also showing underestimation of fat-free mass in older women on the particular DXA system studied [7]. Such results argue for subgroup-specific assessment rather than assigning a single accuracy label to an entire technology.

Visceral fat and regional geometry

Visceral and subcutaneous adipose tissue occupy different anatomical compartments. A whole- body fat value cannot establish their allocation. DXA visceral fat software estimates an internal compartment from projected regional measurements and modelling. The region and units must be specified: android-region area, volume or mass cannot automatically be treated as total abdominal visceral adipose tissue [4].

Published DXA visceral fat validation demonstrates both utility and the need to examine individual agreement. Kaul and colleagues compared DXA and CT in 124 adults. The combined coefficient of determination was 0.957, while the reported combined 95% limits of agreement for visceral fat volume extended from -355 to +468 cm³. A strong association therefore coexisted with a substantial spread of individual differences [8]. These results describe that study's system and population, not every DXA implementation.

Table 1. Physiological quantities require different reference routes

Table 1 specifies an evaluation framework. The historical Deep Health® report documents the first two reference routes; it lists visceral and subcutaneous fat as targets without describing their corresponding anatomical reference procedures [1].

3 The Deep Health Standardization Program

Cohort and measurement preparation

The 2017 report describes 1,466 men and 2,678 women, giving a total cohort of 4,144.

Participants were classified in normal-weight or overweight BMI categories. People classified as obese were excluded because the original study focused on interpretation within the lower BMI categories. The stated eligibility conditions included absence of physical ailments or orthopedic handicaps [1].

Participants completed a medical history, physical examination and comprehensive metabolic profile. Body mass was recorded to the nearest 0.1 kg and height to 0.1 cm using a stadiometer.

The report states that weight and metabolic testing were performed after a 12-hour fast. These procedures describe preparation and screening; they do not establish an unreported age range, recruitment geography or distribution of hydration states [1].

Table 2. Characteristics documented in the 2017 report

Fat assessment by air displacement plethysmography

The BOD POD procedure used air displacement to determine body volume, together with measured body mass to derive body density. Participants wore a swimsuit and cap. The report identifies manufacturer-recommended calibration and measurement procedures and a thoracic gas volume correction expressed as functional residual capacity plus half of tidal volume [1].

Body density and the fat percentage were calculated as follows [1,9]. In the Siri equation, density is expressed in the conventional units of g/cm³, numerically equivalent to kg/L.

Here m denotes body mass, V body volume and D the resulting body density. Fat mass in kilograms is then obtained from body mass and the estimated fat fraction; fat-free mass is the complementary mass. The physical observation is body volume. Conversion to fat requires a model of the constituent densities.

Air-displacement plethysmography provides a measurement route independent of DXA attenuation. It is related in principle to hydrostatic weighing, which obtains body volume through water displacement. The 2017 protocol specifies air displacement. A hydrostatic weighing series would constitute a separate dataset with its own preparation, residual lung volume correction and calibration record.

The distinction between observation and model also applies to the reference. A two-compartment density model assumes properties of fat and fat-free tissue that vary between people. Kuriyan and colleagues evaluated body-composition methods against a four-compartment model in 39 adults, illustrating the value of combining density, body water and mineral information [10]. The reference route should therefore be accompanied by its assumptions and measurement uncertainty.

Muscle assessment by whole body potassium counting

The reported muscle reference used the naturally occurring potassium-40 signal. A whole-body counter detected its approximately 1.46 MeV gamma emission during a nine-minute acquisition.

The report describes body-size adjustment using the approach of Pierson and colleagues and conversion to skeletal muscle mass using the Wang model [1,11,12].

The sex-specific corrections reproduced in the report are shown below. K₄₀ denotes the body potassium estimate derived from calibrated potassium-40 counting, in mmol; it is not the unconverted number of detected gamma photons. Body mass m is in kg, and corrected total body potassium is in mmol [1,11].

The report then applies the following relation, with skeletal muscle mass in kg and total body potassium in mmol [1,12].

This procedure links the muscle output to a physiological property that differs from DXA lean attenuation. It remains a model-based estimate of anatomical muscle. Potassium is distributed across cellular tissues, and the conversion depends on the validity of the potassium-to-muscle relationship in the studied population.

Wang and colleagues developed and evaluated their models using whole-body potassium counting and multislice MRI in 300 adults. Age, sex and race contributed to prediction, and the potassium-to-muscle relationship differed in older participants [12]. These findings support physiological anchoring while also defining a reason to examine subgroup performance. The strong correlation in that reference-model literature is not a measured Deep Health® versus reference correlation.

Protein and regional adipose compartments

Body protein requires a distinct definition and reference. A comprehensive metabolic profile is a clinical blood panel rather than a measurement of total body protein. Likewise, skeletal muscle mass in kilograms includes water and other constituents; it cannot be relabelled as kilograms of protein. A protein standardization study should specify whether it targets total body protein, a muscular protein compartment or another quantity.

Wang and colleagues evaluated a total body protein model using measurements that included in vivo neutron activation analysis for total body nitrogen, whole-body potassium counting, body water and bone mineral [13]. Such work shows the additional reference burden attached to a molecular protein claim. Any corresponding Deep Health® protein dataset should be reported with its actual assay or compartment model, calibration and error propagation.

Visceral and subcutaneous adipose tissue require similarly specific regional references. MRI or CT segmentation can evaluate a defined abdominal depot; ultrasound can evaluate local subcutaneous thickness when site, compression and technique are standardized. Local thickness, regional volume and whole-body mass are different quantities. Independent calibration of total fat does not establish agreement for each of these regional outputs.

4 Complementary Sensors and the Burden of Standardization

The information added by a sensor

The rationale for the Deep Health® architecture is to collect complementary physiological information where a single measurement principle leaves uncertainty. This makes standardization more demanding because sensor response, physiological interpretation and combined estimation must all be established. The useful comparison concerns the information available for each target compartment, rather than the complexity or number of components in the machines.

Published work demonstrates this principle without establishing Deep Health® performance.

Malkov and Shepherd combined dual-energy attenuation with optical shape information to estimate water, lipid and protein thicknesses in a preliminary phantom study. The added geometric measurement provided a constraint for separating three components [3]. A clinical study by Ng and colleagues combined DXA and bioimpedance in a rapid four-component model evaluated against a criterion model that included air displacement and isotope dilution [14].

These examples support the physical rationale for complementary measurements. They also show why the broad claim of combining modalities is insufficient to establish novelty or superior accuracy. The distinctive Deep Health® contribution must be demonstrated through its documented sensor architecture, the physiological targets it resolves, its reference program and the agreement obtained in people.

Historical device context and version specificity

A companion Deep Health® presentation describes a test configuration with ultrasound sweeps and image processing on a separate connected workstation. It distinguishes this configuration from ordinary standalone devices, which did not generate those images [15]. That record provides context for a system collecting tissue information beyond DXA attenuation. It does not specify the complete sensor inventory for every device generation.

A reproducible methods description should identify the hardware version, each active sensor modality, acquisition settings, contact geometry, signal processing, calibration standards and software version. Each measurement channel should be linked to the output it informs.

Capabilities of an earlier workstation-connected configuration should be distinguished from those of the device version being evaluated.

Sensor calibration and combined model calibration

Physical sensor calibration establishes the response to a known quantity, such as an electrical, dimensional or acoustic reference. Physiological calibration establishes the relationship between that response and a human tissue quantity. Model calibration combines the resulting observations into an output. Validation evaluates the completed model in observations that did not determine its fitted parameters.

A multisensor system therefore requires more than a pooled comparison of final outputs. It needs channel-specific checks, evaluation of cross-channel interactions and testing of the final estimate. For example, a channel intended to account for water variation should be assessed during relevant changes in water state; an anatomical channel should be checked for regional correspondence and effects of acquisition technique.

The term metrological traceability has a narrower meaning than general agreement with a laboratory method. It requires a documented calibration chain and contribution of uncertainty at its steps [16]. The historical reference protocol provides a starting point for physiological standardization. Demonstrating formal traceability additionally requires the underlying calibration records, units and uncertainty contributions.

Testing whether additional channels improve the result

A direct test of the sensor rationale would compare the complete Deep Health® model with reduced models that omit particular channels. The training procedure, subject partitions and reference measurements should be held consistent, and every model should be evaluated in the same independent validation participants. The relevant outcomes are changes in bias, individual error, limits of agreement and subgroup performance.

A stronger test of augmentation would obtain Deep Health®, DXA and independent compartment references in the same people. DXA and Deep Health® would each be evaluated against the reference. If the hypothesis concerns hydration or visceral compartment separation, the comparison should include the conditions and anatomical measurements needed to test that mechanism. Without a paired DXA comparison, agreement with an independent reference can support standardization but cannot quantify superiority over DXA.

Correlated errors require particular attention. Two channels can share an error through body mass, posture or a common calibration equation. A four-compartment comparator can also share measured inputs with the device under evaluation. Such dependence should be reported because it can make agreement appear stronger than a fully independent comparison would show.

5 Cohort Size and Population Applicability

Our population objective is to support measurement across a broad range of humans. A cohort of 4,144 offers substantial opportunities to estimate measurement error and examine variation within the included groups. Its statistical contribution depends on the number of usable paired observations for each output; the total enrolled cohort must not automatically be substituted for every endpoint's analysis count.

The number of people on Earth does not translate directly into a required calibration sample. For a population numbering billions, the finite-population correction is effectively one for a sample of several thousand. The main determinants of an adequate sample are the desired precision, the variation in measurement errors, the coverage of relevant phenotypes and the number of subgroup questions.

For independent paired measurements, the approximate 95% confidence interval for mean bias has a half-width of 1.96 times the standard deviation of paired differences divided by the square root of the paired sample size. If all 4,144 participants supplied valid independent pairs for an endpoint, this factor would be approximately 0.0304 times that standard deviation. This is a mathematical illustration of precision, not an observed Deep Health® performance statistic.

A narrow confidence interval for average bias must be distinguished from small individual errors. More observations make the average error estimate more precise; they do not shrink the underlying distribution of errors in a future person. Similarly, thousands of observations from one phenotype range cannot establish performance outside that range.

The historical exclusion of obesity defines an important applicability boundary. The cohort also requires a reported age distribution, recruitment procedure, geographic coverage and relevant clinical characteristics before its population coverage can be assessed. Global applicability would need evaluation in omitted BMI categories, diverse body proportions, older people, relevant clinical conditions and independent locations. Children would require age-appropriate reference models rather than automatic transfer of adult calibrations.

Sex-specific calibration also requires sex-specific assessment. Pooling 1,466 men and 2,678 women can conceal opposite biases or differences in residual spread. Results should report the endpoint-specific sample counts, performance within each group and interactions across the measurement range. Global use is an intended application to be tested through population coverage and transportability, rather than a property conferred by cohort size alone.

6 Statistical Evaluation of Physiological Agreement

Data structure and integrity

The analysis should begin with a data dictionary defining every output, unit and reference method. Subject identifiers, repeated visits, device and software versions, timing between methods, exclusions and missing values should be retained. The number of participants, observations and usable reference-device pairs should be reported separately. No unreported missingness or exclusion should be concealed by quoting the total cohort size.

Values outside plausible ranges should be investigated against source records, with corrected transcription errors distinguished from valid unusual physiology. Influential observations should be retained in the principal analysis unless an independently documented reason supports exclusion. Sensitivity analyses can show how they affect conclusions. Repeated measurements from one person require an analysis that accounts for their dependence.

Agreement and error magnitude

Correlation describes co-variation, while agreement concerns whether two methods give sufficiently close results for the same person. Bland and Altman established the importance of analyzing paired differences rather than relying on correlation in method comparison [17]. For each endpoint, define the difference as Deep Health® minus reference so that positive bias consistently denotes a higher device estimate.

Here d̄ is the mean paired difference and s the standard deviation of those differences. The conventional limits describe approximate individual differences under the relevant distributional and independence assumptions. The confidence interval around the mean bias and confidence intervals around the limits answer different questions and should be shown separately [17].

The broader evaluation should report mean absolute error, root mean square error and the median and upper quantiles of absolute error. Fat percentage errors should be expressed in percentage points. Relative errors can be useful where the reference is sufficiently far from zero, but can become unstable for small quantities. Lin's concordance correlation coefficient can supplement the analysis by assessing departure from the line of equality [18].

Plots of differences across the measurement range should evaluate whether bias or spread changes with the quantity being measured. If error grows with magnitude, a single constant pair of limits may be misleading. Logarithmic, ratio or regression-based approaches should be selected when their assumptions and physiological interpretation fit the observations. Differences caused by a genuine change between measurement times must also be separated from method error.

Calibration regression and independent validation

A calibration regression should report its direction, slope, intercept and uncertainty. Ordinary least squares treats the predictor as measured without error. When both methods have material measurement error, Deming regression can provide an appropriate alternative if its error assumptions and error-variance ratio are justified [19]. Replicate measurements can inform that ratio; setting it to one should not be presented as an empirically established fact.

Any correction equation should be evaluated in people not used to fit it. Cross-validation must keep all observations from the same person in one partition. If multiple sites or device versions exist, validation by site or version can evaluate transportability. Feature selection, parameter tuning and data-driven outlier rules belong within training procedures so that the reported validation results reflect the complete modelling process.

Performance should be presented for the implemented device output and, where relevant, the proposed recalibrated output. Showing only a fitted correction would obscure the performance of the device that actually generated the historical readings. A locked model can then be evaluated prospectively against predefined acceptance criteria.

Table 3. Evidence needed for the main scientific claims

Practical significance and uncertainty

Acceptance thresholds should reflect the intended use and be specified before inspecting favorable results. An error suitable for a broad screening category may be unsuitable for monitoring a small change in an individual. A nonsignificant mean difference does not demonstrate equivalence, and a small P value in a large cohort does not establish a practically important error.

Reference uncertainty should be retained throughout the interpretation. Counter calibration, body-volume correction, scan segmentation and physiological conversion equations each contribute uncertainty. Standard metrological guidance provides a framework for combining these contributions, including covariance where inputs share error [20]. The strongest standardization report states the resulting uncertainty and the conditions under which the output remains useful.

7 Scientific Interpretation and Scope

The distinctive rationale of the Deep Health® program is the effort to match each numerical physiological output to a relevant human reference, while using complementary sensor information to address the limitations of a single measurement principle. The historical cohort and its fat and muscle protocols document an extensive undertaking in human reference assessment [1]. This is a more demanding task than verifying that a sensor produces stable readings.

The standardization burden grows with the scope of the device. A system reporting total fat, skeletal muscle, visceral fat and subcutaneous fat needs several matched reference procedures.

Adding protein or other physiological outputs extends the burden further. The sophistication lies in the relationship among measurable signals, defined compartments, reference methods and uncertainty, and must be documented at each stage.

The limits of the present methods record should be explicit. It contains the reference procedures but does not present endpoint-specific paired agreement results, a complete sensor specification or an independent validation partition. Regional adipose and protein methods require separate reporting. These gaps prevent quantification of accuracy, comparative superiority or global applicability in this article; they do not negate the documented reference procedures.

No universal claim about the absence of standardization in all other devices is required. Such a claim would require a defined review of products, outputs and reference methods. The scientific comparison is more informative when it asks whether a particular output was matched to the correct physiological quantity, whether the reference carries unresolved assumptions, and whether the final model performs adequately in independent people.

Measurement standardization also has a defined boundary within the broader Deep Health® research program. Accurate body-composition outputs do not, by themselves, establish a composite measure of health or validate disease detection. Those propositions require their own biological definitions, experiments and clinical evaluations. Establishing the physiological meaning of the component outputs is an essential foundation for that subsequent work.

8 Conclusion

DXA provides useful bone and body-composition outputs, while its soft-tissue estimates retain compartment definitions and modelling assumptions that matter when another device is evaluated against them. A multisensor system can address some unresolved quantities by collecting complementary information. Its standardization must then establish the response and contribution of each modality and the agreement of the combined output with an appropriate human reference.

The Deep Health® report documents a 4,144-person program with air-displacement fat assessment and potassium-based muscle assessment. These procedures anchor the standardization effort in distinct physiological measurement routes. The next scientific step is to publish the matched protocols, sensor specifications and paired statistical results needed to determine where that architecture improves measurement. Broad population application requires both precise error estimates and demonstrated performance across the people and conditions in which the device will be used.

Declarations

Author contributions

Bomi Joseph: Conceptualization, article conception and design, development of the manuscript outline, and writing of the manuscript.

Darshana Goakar: Research and investigation, source identification, collection of supporting documents, and compilation of the reference literature.

Both authors reviewed and approved the final manuscript.

Funding

This work was supported by Peak Health Center, San Francisco, California, USA

(ROR: https://ror.org/04b23x642).

Competing interests

Bomi Joseph is the inventor of the Deep Health® Device, a health-measurement technology. Darshana Goakar declares no competing interests.

Data availability

No new data were generated or analyzed specifically for this article. All data, figures, performance measures, and other quantitative information discussed in the manuscript derive from the publicly available sources cited in the article.

Use of artificial-intelligence tools

No artificial-intelligence tools were used in the research, analysis, preparation, or writing of this manuscript.

References

1. Peak Health Center. Standardization of the Deep Health® Device. Los Gatos, California: Peak Health Center; 23 June 2017. 4 p. Institutional standardization report.
2. Pietrobelli A, Formica C, Wang Z, Heymsfield SB. Dual-energy X-ray absorptiometry body composition model: review of physical concepts. Am J Physiol. 1996;271(6 Pt 1):E941-E951. https://doi.org/10.1152/ajpendo.1996.271.6.E941
3. Malkov S, Shepherd J. Combining 3D optical imaging and dual energy absorptiometry to measure three compositional components. Proc SPIE Int Soc Opt Eng. 2014;8937:893714. https://doi.org/10.1117/12.2040903
4. US Food and Drug Administration. Hologic Visceral Fat Software. 510(k) K113356. Clearance decision 6 March 2012; summary and indications for use. https://www.accessdata.fda.gov/cdrh_docs/pdf11/K113356.pdf
5. Kim J, Wang Z, Heymsfield SB, Baumgartner RN, Gallagher D. Total-body skeletal muscle mass: estimation by a new dual-energy X-ray absorptiometry method. Am J Clin Nutr. 2002;76(2):378-383. https://doi.org/10.1093/ajcn/76.2.378
6. Toomey CM, McCormack WG, Jakeman P. The effect of hydration status on the measurement of lean tissue mass by dual-energy X-ray absorptiometry. Eur J Appl Physiol. 2017;117(3):567-574. https://doi.org/10.1007/s00421-017-3552-x
7. Hansen RD, Raja C, Aslani A, Smith RC, Allen BJ. Determination of skeletal muscle and fat-free mass by nuclear and dual-energy X-ray absorptiometry methods in men and women aged 51-84 y. Am J Clin Nutr. 1999;70(2):228-233. https://doi.org/10.1093/ajcn.70.2.228
8. Kaul S, Rothney MP, Peters DM, Wacker WK, Davis CE, Shapiro MD, Ergun DL. Dual-energy X-ray absorptiometry for quantification of visceral fat. Obesity (Silver Spring). 2012;20(6):1313-1318. https://doi.org/10.1038/oby.2011.393
9. Siri WE. Body composition from fluid spaces and density: analysis of methods. In: Brozek J, Henschel A, editors. Techniques for Measuring Body Composition. Washington, DC: National Academy of Sciences, National Research Council; 1961. p. 223-244. Reprinted in Nutrition. 1993;9(5):480-491. https://pubmed.ncbi.nlm.nih.gov/8286893/
10. Kuriyan R, Thomas T, Ashok S, Jayakumar J, Kurpad AV. A 4-compartment model based validation of air displacement plethysmography, dual energy X-ray absorptiometry, skinfold technique & bio-electrical impedance for measuring body fat in Indian adults. Indian J Med Res. 2014;139(5):700-707. https://pmc.ncbi.nlm.nih.gov/articles/PMC4140034/
11. Pierson RN Jr, Wang J, Thornton JC, Van Itallie TB, Colt EW. Body potassium by four-pi 40K counting: an anthropometric correction. Am J Physiol. 1984;246(2 Pt 2):F234-F239. https://doi.org/10.1152/ajprenal.1984.246.2.F234
12. Wang Z, Zhu S, Wang J, Pierson RN Jr, Heymsfield SB. Whole-body skeletal muscle mass: development and validation of total-body potassium prediction models. Am J Clin Nutr. 2003;77(1):76-82. https://doi.org/10.1093/ajcn/77.1.76
13. Wang Z, Shen W, Kotler DP, Heshka S, Wielopolski L, Aloia JF, Nelson ME, Pierson RN Jr, Heymsfield SB. Total body protein: a new cellular level mass and distribution prediction model. Am J Clin Nutr. 2003;78(5):979-984. https://pubmed.ncbi.nlm.nih.gov/14594785/
14. Ng BK, Liu YE, Wang W, Kelly TL, Wilson KE, Schoeller DA, Heymsfield SB, Shepherd JA. Validation of rapid 4-component body composition assessment with the use of dual-energy X-ray absorptiometry and bioelectrical impedance analysis. Am J Clin Nutr. 2018;108(4):708-715. Correction: Am J Clin Nutr. 2018;108(6):1356. https://doi.org/10.1093/ajcn/nqy158
15. Peak Health Center. Deep Health Device Radiology Mode. Institutional presentation of a test device with image-saving and workstation processing capabilities. 24 slides. Undated.
16. Joint Committee for Guides in Metrology. International Vocabulary of Metrology - Basic and General Concepts and Associated Terms. 3rd ed. JCGM 200:2012. Sèvres: BIPM; 2012. https://doi.org/10.59161/JCGM200-2012
17. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1(8476):307-310. https://pubmed.ncbi.nlm.nih.gov/2868172/
18. Lin LI. A concordance correlation coefficient to evaluate reproducibility. Biometrics. 1989;45(1):255-268. https://pubmed.ncbi.nlm.nih.gov/2720055/
19. Linnet K. Evaluation of regression procedures for methods comparison studies. Clin Chem. 1993;39(3):424-432. https://pubmed.ncbi.nlm.nih.gov/8448852/
20. Joint Committee for Guides in Metrology. Evaluation of Measurement Data - Guide to the Expression of Uncertainty in Measurement. JCGM 100:2008. Sèvres: BIPM; 2008. https://doi.org/10.59161/JCGM100- 2008E

Cite this article

Joseph B, Goakar D. Physiological standardization of the Deep Health Device beyond DXA soft tissue estimates. Journal of Human Science and Technology. 2026;1(2):Article 4. DOI: 10.5281/zenodo.23600226.