ABSTRACT In the US, approximately 50,000 oral and pharyngeal cancers (OPCs) are diagnosed annually (10/100,000 incidence). Further, oral epithelial dysplasia (OED) is about 15 times more common than OPC. Patients diagnosed with OED are known to be at risk for malignant transformation (MT), and those treated for oral squamous cell carcinoma (OSCC) are known to be at elevated risk for cancer recurrence (CR). There is little consensus about the optimal clinical surveillance pathways for these patients. Individuals with a history of OSCC and potentially malignant oral lesions (PMOLs) harboring OED/OSCC can have widely variable clinical presentation that overlaps with oral lesions of no malignant potential. Thus, clinicians may be reluctant to perform serial scalpel biopsies on these patients. Commercially available diagnostic adjuncts lack adequate clinical validation across the lesion disease spectrum. When OSCC or high-grade OED is diagnosed early, there is an opportunity to provide appropriate timely treatment, and patient outcomes can improve dramatically. Thus, there is a compelling need for new highly effective non-invasive precision oral lesion diagnostic technologies that can be tailored for the needs of individual patients. This multi-institution prospective cohort study seeks to utilize and optimize first Point-of-Care Oral Cytopathology Tool (POCOCT), a microfluidics ensemble and single cell image-based data acquisition system employing artificial intelligence with interpretation of >100 image features including nuclear F-actin for precision oral lesion diagnostics to be completed. Portable diagnostic tools and embedded algorithms will be optimized for secondary and tertiary care settings for the first time. In this R01 study, POCOCT-derived OSCC CR and OED MT models will be developed to elucidate population and patient-specific dynamic changes in numerical index that yield key information related to CR and risk of MT. While past efforts focused on a single time point, this same multimodal chip-based approach will be used to sample repeatedly during surveillance to identify the value of speed of change to MT and CR. The overarching goals of this R01 study are: (1) to determine whether cytological signatures, when examined serially over time, can lead to better risk prediction for CR, (2) to determine if the same signatures can lead to earlier detection of local recurrence than the traditional clinical pathway, and (3) to further optimize the POCOCT for precision lesion diagnostics of MT and CR using newly identified biomarkers, including nuclear F-actin, and rare cell phenotypes identified by deep learning. This R01 will leverage unique NIDCR-Grand Opportunity databases for a new paradigm of precision diagnostics. High risk patients will be longitudinally monitored in secondary and tertiary care settings at intervals, and their risk trajectory will be established over time using personalized multivariate cytological signatures as well a...