Monitoring Body Composition Change for Intervention Studies with Advancing 3D Optical Imaging Technology in Comparison to Dual-Energy X-Ray Absorptiometry

INTRODUCTION

Obesity remains an area of concern as global prevalence continues to rise (1-3). According to the Center for Disease Control and Prevention (CDC), more than 42% of adults in the United States were considered obese in 2018, while the prevalence was approximately 30% two decades prior (4). To counter the obesity epidemic, diet and physical activity interventions have been studied extensively to target weight loss (5, 6). However, metanalyses have shown that weight is only loosely associated with metabolic health (7) and initial changes in response to intervention are small and quickly undone long-term (6, 8). On the other hand, a range of changes in body composition (reduced total body, abdominal and visceral fat, and increased muscle mass) can be produced through diet or exercise intervention and are consistently associated with decreased cardiovascular disease risk (9, 10). Further, decrements in skeletal muscle, particularly in the elderly, can lead to losses in strength and endurance, reductions in energy expenditure, and an increased risk of insulin resistance (11). Nevertheless, measures of body composition have been relegated to research and specialized facilities while clinical care continues to rely on weight as a flawed marker of health.

Another pressing reason to limit our reliance on weight is the differing relative weight of the compartments and tissues. Garrow (12) suggested, and Prentice et al. (13) concurred, that weight loss is typically 25% fat-free mass (FFM, i.e. lean soft tissue + bone mineral content) and 75% fat loss. However, recent research suggested this “25/75 rule of thumb” may not accurately describe various weight loss interventions. The amount FFM lost depends on energy intake, diet composition, sex, baseline adiposity, inactivity or type and physical activity level, and potentially metabolic and hormonal responses (14, 15). As such, it may not be appropriate to only monitor weight in interventions as FFM might be lost at a greater proportion. By monitoring body composition, investigators receive a more accurate assessment of their intervention’s efficacy.

Body composition by dual-energy X-ray absorptiometry (DXA) has been used extensively in clinical settings for its accuracy and precision of whole-body and regional measurements (16). Although DXA provides clinically useful measurements (e.g., bone mineral density, VAT, fat, and lean mass), it requires expensive radiological equipment, qualified technicians, and may not be feasible or accessible for routine clinical practice or frequent monitoring. The ideal body composition method should be affordable, accessible, free of ionizing radiation, and not require radiological-qualified technicians.

Three-dimensional optical (3DO) imaging has become increasingly accessible, made large advancements in recent years, and is safe to use repeatedly (17, 18). 3DO scanners provide accurate and precise digital anthropometry in comparison to criterion methods and output a 3D mesh that represents a person’s entire shape (19, 20). 3DO shape has shown to be highly predictive of DXA body composition (21-24). The next pressing issue is whether 3DO could successfully capture changes in body composition as different modalities have not validated well longitudinally to standard methods (25). However, the lack of available longitudinal data has limited the assessment of 3DO in monitoring body composition changes. The hypothesis of the study was that 3DO can monitor change with a similar sensitivity to DXA given previous cross-sectional accuracy and precision. Therefore, the objective of this study was to evaluate 3DO’s accuracy for monitoring body composition changes across a variety of intervention studies in comparison to DXA.

METHODS

RESULTS

One hundred and thirty-three participants (43 female and 85 male) were included in the final analysis (Table 2). One-hundred sixty-four participants were excluded for dropout (n=4), unavailable 3DO or DXA data at one or both timepoints (n=157), movement artifacts (n=2), or mislabeled data (n=1) (Supplemental Figure 1). The time between baseline and follow-up DXA scans across all studies ranged from three to twenty-three weeks (Figure 1). Females and males lost on average 3.5 kg and 5.4 kg of total FM, 1.8 kg and 3.4 kg of total FFM, and 30 g and 100 g of VAT as a result of interventions, respectively, according to DXA. The majority of the body composition changes occurred in the trunk for both sexes. Average body shapes were presented at baseline and follow-up for each study intervention (Figure 2).

• Download figure
• Open in new tab

Figure 1.

Histograms of the time (weeks) between the baseline and follow-up DXA scans for females (top) and males (bottom). The numbers over the bars represent the count. Multiple numbers over a multi-color bar represent the count in the corresponding order.

• Download figure
• Open in new tab

Figure 2.

Average baseline (left) and follow-up (right) body shapes for females (top) and males (bottom) by study.

In females (Figure 3), 3DO and DXA total FM and FFM at baseline (R²s; 0.91 and 0.84; RMSE; 3.3 kg and 3.3 kg, respectively) and follow-up (R²s; 0.89 and 0.84; RMSE; 2.8 kg and 2.8 kg, respectively) were highly correlated, while %fat and ALM were moderately correlated (R²s: 0.61 – 0.79). Female Bland-Altman plots are shown in Supplemental Figure 2. Slight proportional bias was observed in the baseline total FM from a single outlier with high leverage but not seen in the follow-up. In the comparison of 3DO body composition changes to DXA (Supplemental Table 1), 3DO achieved an R² of 0.86, 0.73, 0.23; and 0.70, CCC of 0.90, 0.82, 0.47, and 0.81; RMSE of 1.98 kg, 1.58 kg, 2.2%, 0.37 kg for total FM, total FFM, %fat, and ALM, respectively. Mean differences were observed for VAT, ALM, ALMI, and leg FFM (30 g, 0.19 kg, 0.07 kg/m², 0.16 kg, respectively, p-value < 0.05). After adjustments for possible covariates in the stepwise linear regression models with forward selection (Supplemental Table 2), weight change further explained variance in total FM, FFM, and ALM, while changes in BMI further explained variance in %fat change, which modestly improved the R²s to 0.90, 0.80, 0.51, and 0.79, respectively. The adjustments with demographic covariates alleviated residual bias between 3DO and DXA change. Mean differences (i.e., VAT, ALM, ALMI, and leg FFM) were no longer observed after adjustments (p-value = 0.99). The green glyphs (Figure 3 and 4) symbolizes the participants that exceeded both the DXA and 3DO LSCs, the purple glyphs were those who did not exceed DXA nor 3DO’s LSC, and the orange and blue glyphs exceeded one LSC but not the other. The LSCs were depicted with the orange vertical lines (3DO) or the blue horizontal lines (DXA).

• Download figure
• Open in new tab

Figure 3.

(Female) Scatter plot comparisons between 3DO and DXA body composition at baseline, follow-up, and adjusted change. Blue, horizontal lines and orange, vertical lines are signifying the amount of change needed to pass the least significant change for DXA and 3DO, respectively. Purple (Zone 4) and green glyphs (Zone 1) represent agreement of significant or nonsignificant change by 3DO and DXA. The orange glyphs (Zone 3) represent significant change detected by 3DO but not DXA and vice versa for the blue glyphs (Zone 2). The zones and glyphs are consistent for the proceeding adjusted change plots.

• Download figure
• Open in new tab

Figure 4.

(Male) Scatter plot comparisons between 3DO and DXA body composition at baseline, follow-up, and adjusted change. Blue, horizontal lines and orange, vertical lines are signifying the amount of change needed to pass the least significant change for DXA and 3DO, respectively. Purple (Zone 4) and green glyphs (Zone 1) represent agreement of significant or nonsignificant change by 3DO and DXA. The orange glyphs (Zone 3) represent significant change detected by 3DO but not DXA and vice versa for the blue glyphs (Zone 2). The zones and glyphs are consistent for the proceeding adjusted change plots.

In males (Figure 4), 3DO and DXA total FM and FFM at baseline (R²s; 0.88 and 0.84; RMSE; 3.7 kg and 3.6 kg, respectively) and follow-up (R²s; 0.88 and 0.85; RMSE; 3.3 kg and 3.2 kg, respectively) were highly correlated, while %fat and ALM were moderately correlated (R²s: 0.60 – 0.75). Male Bland-Altman plots are shown in Supplemental Figure 3. Proportional bias was observed in total FM and %fat due to a single outlier with high leverage. Not bias was observed in total FFM and ALM. In the comparison of 3DO body composition changes to DXA (Supplemental Table 1), 3DO achieved an R² of 0.75, 0.75, 0.25, and 0.52; CCC of 0.76, 0.78, 0.39, and 0.64; and RMSE of 2.3 kg, 1.8 kg, 2.4%, and 0.5 kg for total FM, total FFM, %fat, and ALM, respectively. Of the observed variables in Figure 4, mean differences were observed for total FM, total FFM, %fat, and ALM (p-value<0.001). After adjustments for possible covariates in stepwise linear regression with forward selection (Supplemental Table 2), R²s modestly improved for changes in total FM, FFM, %fat, and ALM to 0.80, 0.76, 0.42, and 0.56, respectively. Mean differences in total FM, total FFM, %fat, and ALM were no longer observed after adjustments (p-value = 0.99).

3DO predicted regional body composition changes (i.e., arms, legs, and trunk) with R²s that ranged from 0.39 – 0.91 and RMSEs that ranged from 0.16 – 1.3 kg in females and males (Supplemental Table 1). 3DO and DXA found significant changes in the majority of the sample for total FM (69%), total FFM (78%), %fat (53%), and ALM (78%) (Supplemental Table 3). Cohen’s kappa showed that 3DO had a moderate agreement to DXA for most outputs.

From the total sample, 64% of females and 90% of males lost weight after intervention. Of those that lost weight, % FFM loss relative to weight loss ranged from -1370% to 478% as measured by DXA. Among those who lost FFM, 80% of females and 81% of males lost more than 25% FFM relative to their weight loss (Figure 5). The % FFM loss relative to weight loss according to study was on average 36% for Bariatric surgery, 37% for FB4, 27% for OPS II, 313% for Athletes, 85% for TREAT, and 59% for REALPA. The very high percentages were for small weight changes and/or large compartment changes with little overall weight change (e.g., Δ FFM /absolute(Δweight) * 100 for an Athlete = (−1.96/0.36) * 100 = -544%).

• Download figure
• Open in new tab

Figure 5.

Histograms to show the frequency of percent fat-free mass change relative to total weight change in the female (top) and male (bottom) samples.

The experimental models (Supplemental Tables 4 and 5) for females (models 1-5) and males (models 1-3 and 5) predicted change better than the unadjusted results presented in Supplemental Table 1. Female models 1, 2, and 3 were the same as male models 1, 2, and 3 after the stepwise regression. The prediction of total FM change improved in both females (R² = 0.94, RMSE = 1.29 kg) and males (R² = 0.84, RMSE = 1.79 kg).

DISCUSSION

These present findings support 3DO as a sensitive tool assessing changes in body composition, compared to DXA. The current study evaluated 3DO’s ability to monitor body composition change with body shape via the 3DO mesh. Previously derived shape models and equations were used to estimate body composition values (24), and 3DO change from baseline to follow-up was compared to DXA change. Overall, 3DO change was highly correlated to DXA change, and the two methods agreed on the statistical significance of the change in the majority of the study population. Additional adjustments of demographics explained further variance between 3DO and DXA change. The experimental models showed that specific calibration with changes in shape might improve the body composition change prediction. Additional longitudinal data may further validate the experimental models. Nevertheless, the cross-sectional 3DO models produced comparable body composition estimates compared to DXA, which supports our hypothesis that using 3DO imaging is a feasible method to monitor body composition changes.

With respect to the “rule-of-thumb” by Garrow’s (12) and Prentice et al. (13) that 25% of weight loss comes from FFM, the proportion of FFM lost by the vast majority of our sample was greater than 25% (Figure 5). Thus, this “rule-of-thumb” underestimated the loss of FFM for most participants. The loss of FFM varied by study intervention due to differences in protocol, type of activity, and objectives. This is a critical area of study given that loss of FFM is associated with diminished strength and decrements in physical functioning, exercise endurance and capacity, metabolic rate, and health (11). This further supports the idea that the relative proportion of FM and FFM change is dependent on energy intake, diet composition, sex, baseline adiposity, inactivity or type and level of physical activity level, and potentially metabolic and hormonal responses (14). It should also be noted that the experimental group in the OPS II study received testosterone, which has shown to improve muscle protein kinetics during stress and was a contributing factor to the proportion of FFM loss (32). The LSU athletes and OPS II participants were study examples that monitored body composition for performance. 3DO can also provide an accessible and safe method to monitor body composition changes since muscle build-up can play a pivotal role in strength, endurance, and speed. Some participants had percent changes in FFM that were greater than 100%. This was likely due to an undesirable recomposition where the person increased FM and lost FFM, which resulted in a minimal weight change (small denominator). Hydration status may have impacted individuals with minimal changes (15). Because DXA does not measure hydration status, DXA has assumptions of hydration built-in, while 3DO estimates were regressed to DXA. The current analysis further emphasized the importance of monitoring body composition, in addition to weight, to better understand changes in health status and performance factors.

In the current study, the change in 3DO FM was underestimated while FFM was overestimated in both females and males. These changes were not statistically significant in females but were significant in males. A potential reason for the male discrepancies could be that the current sample was beyond the adiposity range of the data used to train the original 3DO body composition models. The maximum percent fat of the training dataset was 38% (24), while the current population’s maximum was 47%. The training dataset lacked the shape variance of males with extreme adiposity and underestimated FM and %fat in this study. However, after adjusting for covariates, these discrepancies were no longer statistically significant. Using either the 3DO changes models with demographics or retraining the model with high adiposity participants could potentially address this issue.

Test-retest precision, also known as short-term precision, has an error that can be used to derive the LSC (2.77 x precision error) (37). Metrics with changes that surpass their LSC values were considered to be significantly greater than zero changes, with a 95% confidence. 3DO and DXA precision errors used for this study were taken from the Shape Up Adults group (24). To reduce the 3DO precision error, to be closer to that of DXA, and track body composition changes at the same significance, multiple 3DO scans could be taken and averaged at each time point. For total FM, the 3DO precision errors were 0.44 kg and 0.55 kg as well as 0.25 kg and 0.23 kg for DXA in males and females, respectively (24). Assuming a normal distribution of the precision errors, the precision of the averaged 3DO measure from multiple scans improves by the square root of the number of scans (e.g., the average of four scans would lower the precision error from 0.44 kg and 0.55 kg to 0.22 kg and 0.27 kg) (39). Unlike DXA, taking multiple scans with 3DO is fast and without additional radiation concerns. Dizziness can be a potential hazard, which is listed on the manufacturer warnings and our study consent form. However, our participants did not report dizziness even with multiple scans.

3DO VAT change compared to DXA VAT change had a low correlation, but this is likely to due to the compressed range of VAT change in our sample. In previous literature, DXA VAT has been shown to be highly correlated to gold standard, magnetic resonance imaging (MRI), cross-sectionally (R² range: 0.81 – 0.86) (40-42). However, DXA VAT often has proportional bias for individuals with high VAT and has not validated well longitudinally compared to MRI (42, 43). Some of the bias may be explained by technique differences as MRI and CT are analyzed by a cross-section of the abdomen, while DXA is estimated from an X-ray’s 2D area. Nevertheless, given the cross-sectional correlation of DXA VAT to MRI and cardiometabolic markers, DXA VAT can still be considered an accessible tool to characterize health risks that can prompt health initiatives (44, 45). Since 3DO body shape has been shown to be correlated and predictive of DXA VAT and cardiometabolic markers (21), 3DO may be considered an accessible tool that can be used frequently due to its lack of ionizing radiation concerns. However, users should be wary of DXA and 3DO’s ability to accurately monitor VAT changes given the limited amount of validation in the literature.

The 3DO vs DXA RMSEs for comparing baseline or follow-up scans were higher (worse) than the RMSEs for the change comparisons (Figures 3 and 4). This may be due to the individual systematic bias expressed in the comparison of one technique (3DO) to another (DXA) that is subtracted away when looking at changes in the measures, and the fact that the precision error of both techniques was much lower than this individual systemic bias found in inter-technique comparisons of measures (46).

According to the authors’ knowledge, one other study has reported 3DO’s ability to monitor body composition change. Tinsley et al. (25) compared the changes in proprietary body composition estimates from 3DO scanners (Fit3D ProScanner, Size Stream SS20 (Size Stream LLC, Cary, NC, USA), and Styku S100 (Styku Comp. Los Angeles, CA, USA)) with changes in 4-compartment (4C) model measures in 21 volunteers. Changes in 3DO FM underperformed (CCC, 0.22 to 0.40 for total FM) compared to the 4C model. CCC for total FM in this study was much higher (CCC; 0.91 and 0.79) for females and males, respectively, showing better agreement with DXA. However, there were major differences between the two studies. This study had a larger sample size (n=128), greater ethnic-racial diversity, larger body composition changes, publicly available equations, and a different body composition modality (DXA instead of the 4C model, given that data for the 4C model was not available). Larger studies are warranted to fully evaluate 3DO’s ability to monitor change using the scanners’ proprietary body composition outputs.

This study had several strengths including sample size, variety and lengths of interventions, and an ethnically diverse sample. A study with 128 participants is currently one of the larger body composition change studies. Although the majority of the participants were ethnically white, there was still a representation of other ethnic backgrounds. This study included a variety of diet, physical activity, and surgical interventions that ranged from 3 to 23 weeks and included participants with large as well as little to no body composition changes.

This study is not without limitations. Although the sample size was rather large for a longitudinal analysis, a representative test set could not be made to validate the adjusted and experimental linear models. In addition, results from this study may not be generalizable to populations with poor health status or children (< 18 years old). Future work would be required to address gaps in knowledge in different populations such as infants and children, the accuracy of change in specific subgroups (i.e., BMI, ethnicity, and age), and the assessment of more accessible 3DO technology.

CONCLUSION

This study indicated that 3DO body composition changes were highly correlated to DXA and show good feasibility to monitor changes over a variety of interventions. As the accessibility and popularity of 3DO continues to grow, more people will be able to use this technology to monitor their body composition in clinical and nonclinical settings. The findings of this study extend the 3DO literature, which has been limited to cross-sectional performance.

Author Contributions

MCW and JAS designed and conducted the research; MCW, LL, JB, IT, YEL, GM, and JAS were part of the data analysis; JMWW, NNK, CBE, BI, MCS, JS, BD, NMJ, RM, CV, DSL, EW, SBH, and JAS were in charge of their respective study recruitment and protocols; MCW and JAS drafted the manuscript and were responsible over the final content. All authors reviewed and approved the final manuscript.