Integrative Predictors of Temporomandibular Osteoarthritis

点击翻译按钮获取中文摘要

ABSTRACT This application proposes the development of efficient web-based data management, mining, and analytics, to integrate and analyze clinical, biological, and high dimensional imaging data from TMJ OA patients. Based on our published results, we hypothesize that patterns of condylar bone structure, clinical symptoms, and biological mediators are unrecognized indicators of the severity of progression of TMJ OA. Efficiently capturing, curating, managing, integrating and analyzing this data in a manner that maximizes its value and accessibility is critical for the scientific advances and benefits that such comprehensive TMJ OA patient information may enable. High dimensional databases are increasingly difficult to process using on-hand database management tools or traditional processing applications, creating a continuing demand for innovative approaches. Toward this end, the DCBIA at the Univ. of Michigan has partnered with the University of North Carolina, the University of Texas MD Anderson Cancer Center and Kitware Inc. Through high-dimensional quantitative characterization of individuals with TMJ OA, at molecular, clinical and imaging levels, we will identify phenotypes at risk for more severe prognosis, as well as targets for future therapies. The proposed web-based system, the Data Storage for Computation and Integration (DSCI), will remotely compute machine learning, image analysis, and advanced statistics from prospectively collected longitudinal data on patients with TMJ OA. Due to its ubiquitous design in the web, DSCI software installation will no longer be required. Our long-term goal is to create software and data repository for Osteoarthritis of the TMJ. Such repository requires maintaining the data in a distributed computational environment to allow contributions to the database from multi-clinical centers and to share trained models for TMJ classification. In years 4 and 5 of the proposed work, the dissemination and training of clinicians at the Schools of Dentistry at the University of North Carol, Univ. of Minnesota and Oregon Health Sciences will allow expansion of the proposed studies. In Aim 1, we will test state-of-the-art neural network structures to develop a combined software module that will include the most efficient and accurate neural network architecture and advanced statistics to mine imaging, clinical and biological TMJ OA markers identified at baseline. In Aim 2, we propose to develop novel data analytics tools, evaluating the performance of various machine learning and statistical predictive models, including customized- Gaussian Process Regression, extreme boosted trees, Multivariate Varying Coefficient Model, Lasso, Ridge and Elastic net, Random Forest, pdfCluster, decision tree, and support vector machine. Such automated solutions will leverage emerging computing technologies to determine risk indicators for OA progression in longitudinal cohorts of TMJ health and disease.