Integrative Predictors of Temporomandibular Degenerative Joint Disease

ABSTRACT Temporomandibular Disorders (TMDs) represent a significant burden on musculoskeletal health, accounting for an estimated $4 billion in healthcare costs annually. Pediatric Temporomandibular Degenerative Joint Disease (TM DJD) is especially concerning, with risks including growth abnormalities, severe joint damage, and potential blindness. Addressing the critical gaps in our understanding of pediatric TM DJD's progression, severity, and management is imperative. Our research proposes innovative, evidence-based strategies to improve early diagnosis and personalized care, utilizing cutting- edge multimodal imaging, artificial intelligence (AI), and machine learning. Our strategy unfolds across two synergistic aims, each poised to significantly advance TMJ DJD diagnosis, prognosis, and patient-specific treatment. Aim 1 employs advanced analytics, combining multimodal Cone-Beam CT (CBCT) and Magnetic Resonance Imaging (MRI) registration, with articular disc quantitative markers. This aim introduces a novel application of symptom phenotyping via mobile app tracking and biological marker assessments, integrating AI algorithms to predict early disc and condylar changes in children—a critical step toward preemptive treatment strategies. Expanding on Aim 1's foundation, Aim 2 leverages longitudinal datasets of pediatric and adult TM DJD patients to develop a groundbreaking diagnostic and prognostic toolkit. This includes pioneering privileged information learning techniques to merge advanced quantitative CBCT and MRI imaging features with clinical data, filling crucial gaps in real-world data sources. A key innovation is the application of EMERSE, the Electronic Medical Record Search Engine, for comprehensive clinical notes data abstraction, compared to large language models’ accuracy and reliability for information retrieval. Our comprehensive approach evaluates the performance of feature selection, machine learning, and statistical models, refining our preliminary Ensemble via Hierarchical Predictions through Nested model. We commit to rigorously testing our models' usability, feasibility, and acceptability in clinical environments through a tri-phase assessment strategy, focusing on practical implementation, expert clinician calibration, and integration into community dentistry for broader application. Aligned with the NIH Heal Initiative, the NIDCR Temporomandibular Disorder Collaborative for IMproving PAtient-Centered Translational Research, our proposed work also fulfills objectives within the NIH Strategic Plan for Data Science. Our interdisciplinary team combines clinical experts at the University of Michigan and the University of the Pacific with the computational expertise of partners in bioinformatics, natural language processing, statistical modeling, machine learning, and software engineering from the Michigan Department of Learning Health Systems, the University of North Carolina, and Isomics Inc. Leveraging a decade of collective experience and an established, robust infrastructure for quantifying multimodal data, our research network is uniquely positioned to ensure the sustainability, accessibility, and utility of our research outcomes for the TMD research and clinical communities. This collaboration is set to enhance clinical decision- making and identify targeted treatment decisions for specific risk groups in pediatric TM DJD care.