Non-linear Image-Based Regression of Body Segment Parameters

Biomechanical analysis of human movement often requires accurate estimation of body segment parameters (BSP). These values are segmental inertial properties, including mass, center of mass and moments of inertia. They can be measured directly on living subjects using techniques such as Magnetic Resonance Imaging (MRI), Computed Tomography (CT) and gamma-mass scanning. Despite their accuracy, these methods involve high radiation and require expensive scanners that are not always readily available to biomechanics researchers. Another popular way to estimate BSP is by studying regression equations on experimental data, commonly from cadaveric studies. These approaches, however, have been criticized for the limited cadaveric data. We propose a novel in vivo regression method for computing BSP using non-linear image-based techniques. Our method was first facilitated with X-ray images of sample subjects acquired from Dual-energy X-ray Absorptiometry (DXA), where the radiation dose is approximately 1/10(th) that of a standard chest X-ray. A feature-based image transformation was then applied to predict a mass distribution image for the new subject, while he was not required to undergo DXA scanning. The subject's BSP values were subsequently computed using the mass distribution obtained from the predicted image. Cross-validation of moments of inertia among population samples shows that our method has mean percentage errors of 7.5% for limbs and 7.1% for head and torso, while the corresponding errors are 9.3% and 15% in cadaver-based non-linear regression method. It suggests that our image-based regression approach is promising for estimating BSP on living subjects. It is not limited by ranges of cadaveric data or differences between living and dead tissues.