New 3D body scan method beats traditional imaging for tracking body fat (2024)

New 3D body scan method beats traditional imaging for tracking body fat (1)By Tarun Sai LomteReviewed by Susha Cheriyedath, M.Sc.Oct 25 2024

A novel 3D body shape method promises accessible and accurate body composition predictions, potentially transforming how we monitor health over time and detect risks.

New 3D body scan method beats traditional imaging for tracking body fat (3)

Body meshes fitted to DXA. DXA image inputs (Row 1), initial fits using HKPD (Row 2) and optimised fits (Row 3). Study: Prediction of total and regional body composition from 3D body shape

In a recent study published in the journal npj Digital Medicine, researchers developed a novel method to predict body composition for three-dimensional (3D) body shapes. Body composition is linked to chronic disease risk. It can be assessed using computed tomography, dual-energy X-ray absorptiometry (DXA), and magnetic resonance imaging. However, due to ethical and practical constraints, these techniques are not readily available in epidemiological studies and clinical practice and are not easily accessible to the general public.

Conventional anthropometrics, such as waist-hip ratio, body mass index (BMI), and waist-hip circumferences, are used to infer body composition. Nevertheless, these methods do not differentiate between lean and fat mass and are inadequately accurate/convenient for longitudinal use, often requiring trained personnel and in-person visits. Thus, simple, accessible, inexpensive tools are needed to assess body composition accurately.

About the study

In the present study, researchers developed a novel method for body composition prediction using 3D body shape. They obtained DXA scans, metabolic health variables, and paired anthropometry data from the Fenland study established in 2005. The Fenland study involved 12,435 participants in Phase I and 7,795 in Phase II. Of these, 11,359 participants from Phase I and 6,102 from Phase II were included in the current study.

The team used 80% of Phase I data to train and derive 3D body shape composition models, and the remainder was used for validation. Phase II data were used as a test dataset for validation in a now older population. Moreover, a smartphone validation study was undertaken with 119 healthy adults, which, besides DXA scans, included air plethysmography and a mobile app capturing images. This sample was used to validate models derived from the Fenland study and assess the accuracy of 3D shapes obtained from smartphone images. Statistical validation metrics, including Pearson correlation coefficients and root-mean-square error (RMSE), were employed to measure the accuracy of these predictions.

Each three columns shows one participant who lost (Columns 1–3), maintained (Columns 4–6) or gained (Columns 7–9) fat mass, their DXA scans and fitted meshes for Fenland phase 1 (Row 1) and phase 2 (Row 2). Changes in body shape for the first and third participants are significant.

2D images of the front, back, right-side, and left-side profiles were taken using a purpose-built mobile app that constructs a 3D body mesh. The researchers fitted 3D body meshes to DXA silhouettes with paired anthropometry measures, and the fitted parameters were used for predicting body composition metrics. To fit a 3D mesh, DXA silhouettes were augmented with paired anthropometrics using the skinned multi-person linear (SMPL) model in a two-stage approach.

First, the hierarchical kinematic probability distributions (HKPD) method was used for initial pose and shape estimates. Next, an optimization method was developed to refine this initial guess. Optimized SMPL shape parameters were used to regress body composition metrics. A feed-forward neural network was constructed for regression, which used 10 SMPL shape parameters, height, weight, gender, and BMI as the input. The network outputs included total lean mass, total fat mass, etc. Further, the HKPD method generated SMPL avatars using multi-view information from smartphone images. A model was developed to predict regional and total body composition metrics using these methods. Its performance was evaluated using root-mean-square error values. The associations between predicted values and DXA measurements were assessed using Pearson correlation coefficients.

Findings

The smartphone validation study participants were younger, leaner, and lighter than those in the Fenland study. The researchers noted that the optimized meshes agreed with the DXA silhouette much better than the initial shape and pose estimates. In the Phase I sample of the Fenland study, correlation coefficients between DXA and predicted parameters were robust for all lean and fat mass variables. Similarly, correlation coefficients were strong for all variables in the Phase II sample.

In addition, comparable results were observed in the external validation sample. The Pearson correlation coefficients exceeded 0.86 for most metrics, indicating strong agreement between predicted and DXA values. Further, a comparison study was conducted on different regressor model inputs. One model, which used only height and weight as inputs, showed some predictive ability. Performance increased by including waist and hip circumferences, respectively. The final model, which used SMPL, height, and weight as inputs, showed substantial improvements in estimating body composition metrics. The model demonstrated a root-mean-square error (RMSE) of less than 3.5% for percentage body fat predictions, highlighting its accuracy.

In the Fenland study, 5,733 individuals participated in both phases, allowing for the evaluation of the model's ability to detect changes in body composition over an average of 6.7 years. The model detected changes for various fat mass metrics; lean mass changes were less well captured, mainly because lean mass remains essentially unchanged over time.

Conclusions

The researchers introduced a novel computer vision-based method fitting a 3D body mesh to a DXA silhouette with paired anthropometric data and generated a database of 3D body meshes. These meshes accurately predicted body composition metrics. Moreover, the model could detect longitudinal changes. However, the researchers noted that while the model was particularly effective at detecting changes in fat mass, its ability to track changes in lean mass was more limited, due to the stability of lean mass over time.

The team also illustrated that avatars generated from smartphone images could be used for body composition prediction. Overall, 3D body shapes generated from 2D images and relevant inference methods could be a viable alternative for clinical medical imaging. The study acknowledges the demographic limitations of the dataset, which predominantly included white European adults, suggesting further research in diverse populations for broader applicability.

Journal references:

New 3D body scan method beats traditional imaging for tracking body fat (2024)
Top Articles
Latest Posts
Recommended Articles
Article information

Author: Gregorio Kreiger

Last Updated:

Views: 6647

Rating: 4.7 / 5 (77 voted)

Reviews: 84% of readers found this page helpful

Author information

Name: Gregorio Kreiger

Birthday: 1994-12-18

Address: 89212 Tracey Ramp, Sunside, MT 08453-0951

Phone: +9014805370218

Job: Customer Designer

Hobby: Mountain biking, Orienteering, Hiking, Sewing, Backpacking, Mushroom hunting, Backpacking

Introduction: My name is Gregorio Kreiger, I am a tender, brainy, enthusiastic, combative, agreeable, gentle, gentle person who loves writing and wants to share my knowledge and understanding with you.