3D Shape-based Body Composition Prediction Model Using Machine Learning.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Anthropometry

Background:

• 3D body scanning technology is advancing, enabling large-scale anthropometric data collection.
• Limited correlation exists between 3D body shape data and health information like Body Fat Percentage (BFP).
• Accurate BFP estimation is crucial for biomedical research and health applications.

Purpose of the Study:

• To develop a novel prediction model for estimating Body Fat Percentage (BFP) using 3D body shape analysis.
• To introduce and utilize a "visual cue" derived from second-order shape descriptors to enhance BFP prediction.
• To establish a robust model validated against the gold standard Dual-energy X-ray Absorptiometry (DXA).

Main Methods:

• Established a baseline regression model for feature selection using zeroth-order shape descriptors.
• Incorporated a "visual cue" from second-order shape descriptors as a shape-prior to improve the baseline model.
• Trained and evaluated the model using Dual-energy X-ray Absorptiometry (DXA) as the ground truth for BFP measurement.

Main Results:

• The developed prediction model demonstrated improved accuracy in estimating BFP from 3D body shapes.
• The model incorporating "visual cues" significantly enhanced the baseline prediction performance.
• The novel model achieved, on average, 20.28% higher prediction accuracy compared to the BOD POD device.

Conclusions:

• The proposed model effectively estimates BFP using 3D body shape data, offering a non-invasive and accurate alternative.
• The "visual cue" concept provides a valuable approach to leverage shape information for improved health metric prediction.
• This technology has the potential to advance personalized medicine and health monitoring through enhanced body composition analysis.