General multivariate linear modeling of mandible surface using SurfStat

Abstract

We present a novel approach for modeling mandible shape variations in Temporomandibular Joint Osteoarthritis (TMJ OA) patients. We have employed weighted spherical harmonic (SPHARM) representation to parameterize and normalize mandible surfaces. This representation is fed to multivariate linear models which account for nuisance covariates such as age and mandible size. Multivariate linear models are implemented using SurfStat package and using this implementation one can avoid the complexity of specifying design matrices. In several multivariate shape models, the Hotelling's T-square has been used as a test statistic. In Hotelling's T-square statistic, we can test the equality of vector means without considering redundant covariates. Thus we have used SurfStat package in which Hotelling's T-square framework is generalized to incorporate additional covariates. Our proposed methodology has been applied for investigating Mandibular condyle shape variations in 19 TMJ OA subjects. Promising results have been demonstrated in lesion localization which is an important step in surgical planning and treatment.