mmWave radar predicts body fat percentage with 3.2% error
Body composition from a quick scan through clothes, no undressing needed.
A new paper on arXiv presents a non-intrusive body composition assessment method using millimeter wave (mmWave) radar, commonly deployed in security scanning. The researchers generated synthetic mmWave-like point clouds from clinical CT/MRI data and parametric human models to train a multi-task learning model. The approach requires subjects to stand fully clothed during a brief scan, preserving privacy while capturing full-body shape data. This sidesteps the radiation exposure and high cost of CT/MRI, enabling frequent monitoring.
In a pilot cohort with real mmWave scans and bioimpedance reference, the model predicted visceral adipose tissue (VAT) with a mean absolute error of 1.0L and body fat percentage (BFP) within 3.2%. These results demonstrate that mmWave radar can estimate clinically relevant body composition metrics through clothing, potentially replacing BMI for personalized health tracking. The method is fast, non-intrusive, and suitable for routine use in gyms, clinics, or even home settings.
- mmWave radar scans through clothing in a standing posture, requiring no disrobing or contact.
- Model predicts VAT with 1.0L mean absolute error and BFP with 3.2% error against bioimpedance measurements.
- Training uses synthetic point clouds derived from CT/MRI scans and parametric human models to overcome limited real mmWave medical data.
Why It Matters
Enables frequent, privacy-safe body composition tracking for personalized medicine without radiation or inconvenience.