Estimation of spinal joint centers from external back profile and anatomical landmarks.

Defining a subject-specific model of the human body is required for motion analysis in many fields, such as in ergonomics and clinical applications. However, locating internal joint centers from external characteristics of the body still remains a challenging issue, in particular for the spine. Current methods mostly require a set of rarely accessible (3D back or trunk surface) or operator dependent inputs (large number of palpated landmarks and landmarks-based anthropometrics). Therefore, there is a need to provide an alternative way to estimate joint centers only using a limited number of easily palpable landmarks and the external back profile. Two methods were proposed to predict the spinal joint centers: one using only 6 anatomical landmarks (ALs) (2 PSIS, T8, C7, IJ and PX) and one using both 6 ALs and the external back profile. Regressions were established using the X-ray based 3D reconstructions of 80 subjects and evaluated on 13 additional subjects of variable anthropometry. The predicted location of joint centers showed an average error 9.7 mm (±5.0) in the sagittal plane for all joints when using the external back profile. Similar results were obtained without using the external back profile, 9.5 mm (±5.0). Compared to other existing methods, the proposed methods offered a more accurate prediction with a smaller number of palpated points. Additional methods have to be developed for considering postures other than standing, such as a sitting position.

A principal component analysis of the relationship between the external body shape and internal skeleton for the upper body.

Recent progress in 3D scanning technologies allows easy access to 3D human body envelope. To create personalized human models with an articulated linkage for realistic re-posturing and motion analyses, an accurate estimation of internal skeleton points, including joint centers, from the external envelope is required. For this research project, 3D reconstructions of both internal skeleton and external envelope from low dose biplanar X-rays of 40 male adults were obtained. Using principal component analysis technique (PCA), a low-dimensional dataset was used to predict internal points of the upper body from the trunk envelope. A least squares method was used to find PC scores that fit the PCA-based model to the envelope of a new subject. To validate the proposed approach, estimated internal points were evaluated using a leave-one-out (LOO) procedure, i.e. successively considering each individual from our dataset as an extra-subject. In addition, different methods were proposed to reduce the variability in data and improve the performance of the PCA-based prediction. The best method was considered as the one providing the smallest errors between estimated and reference internal points with an average error of 8.3mm anterior-posteriorly, 6.7mm laterally and 6.5mm vertically. As the proposed approach relies on few or no bony landmarks, it could be easily applicable and generalizable to surface scans from any devices. Combined with automatic body scanning techniques, this study could potentially constitute a new step towards automatic generation of external/internal subject-specific manikins.