Multimodal Integrative Dimension Reduction and Statistical Modeling with Applications to Temporomandibular Joint (TMJ) Morphometry and Biomechanics

点击翻译按钮获取中文摘要

Disorders associated with the temporomandibular joint (TMJ) and surrounding tissues affect 5-12% of Americans, cost $4 billion/year and, for reasons not fully known, are nearly twice as prevalent among females as males. Despite debilitating effects of these disorders, the functional morphology and biomechanics of the TMJ are not well understood due to its unique and complex anatomy. In particular, the biomechanics of the TMJ, how the muscular and skeletal features interact to affect function, and quantifying sex differences in TMJ characteristics are important scientific research questions and crucial for developing safe and effective clinical treatments. A recent, innovative study in TMJ research generated detailed data describing muscle morphometry and skeletal structure, but statistical analysis is challenging because methods do not exist that effectively integrate high dimensional, multimodal multivariate data and target effects of intrinsic heterogeneity due to sex, age and other factors. Such complex data now pervade craniofacial research, as computed tomography, magnetic resonance imaging, electromyography, genotyping and other technologies jointly generate diverse information that must be efficiently integrated to elucidate craniofacial development, abnormalities, and treatment strategies. There is hence a pressing need for statistical methods that effectively integrate these high-dimensional multimodal data and simultaneously enable discovery of features that explain heterogeneity and, therefore, can guide therapeutics. This project will develop new statistical models and dimension reduction methods to address the challenges in the TMJ morphometry data and in multimodal data arising from dental and craniofacial studies more generally. Three interconnected aims are: (1) Develop two new statistical models for simultaneous dimension reduction and feature selection that will identify the linear combinations of TMJ muscle attachment and skull measurements most differentially correlated for males and females; (2) Extend the two new models from multivariate to tensor-variate (i.e. multidimensional array data) to exploit the full structure of the 3D images and muscle attachments; and (3) analyze the TMJ morphometry data using the newly developed statistical methods and develop and disseminate user-friendly software. Successful completion of the proposed aims will provide new insight about musculoskeletal associations in the TMJ and, importantly, how they vary by sex and other covariates, that can inform future research and clinical treatment. The work will fill important gaps in analysis techniques by producing a series of new statistical models, dimension reduction methods and computational tools for analyzing conditional relationships among multimodal multivariate data, and make these methods widely available.