Voxel-based Deep Regression for Enhanced Body Composition Estimation from 3D Body Scans.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Data Science

Background:

• Accurate body composition estimation is vital for personalized healthcare.
• Current methods like DXA and CT have accessibility and safety limitations.
• Manual feature engineering in body composition analysis is labor-intensive.

Purpose of the Study:

• To develop an automated, end-to-end learning pipeline for body composition analysis.
• To replace traditional feature extraction with a voxel-demographic approach.
• To improve the accuracy and scalability of clinical body composition assessment.

Main Methods:

• Utilized end-to-end learning from voxel maps and demographic data.
• Replaced handcrafted feature engineering with automated pattern learning.
• Validated the approach on real-world 3D scan datasets.

Main Results:

• Achieved promising Root Mean Square Error (RMSE) performance.
• Demonstrated effectiveness across multiple regional and total body composition values.
• Showcased superior performance compared to traditional part-based feature descriptors.

Conclusions:

• The proposed voxel-demographic approach offers a significant advancement in body composition analysis.
• This automated methodology enhances both accuracy and scalability for clinical applications.
• Represents a paradigm shift towards automated pattern learning in healthcare.