PROJECT SUMMARY
Oral cavity and oropharyngeal (OC/OPC) cancers afflict more than 53,000 individuals in the United States
annually. Despite advancements in oncologic therapies, the majority of patients will experience significant toxicity
burden during and after therapy, including moderate-severe xerostomia, reduced mouth opening (i.e. trismus),
periodontal disease, and osteoradionecrosis. To date, acute and chronic orodental complications are largely
managed by clinicians and dentists based on empirical knowledge, with wide inter-provider management
variability influenced by provider experience and available clinical information which is often incomplete,
incorrect, or nonexistent. To further complicate long-term care of OC/OPC survivors, there is no standardized
method for communicating with dentists the extent and intensity of radiation doses delivered to tooth bearing
areas which is vital information for accurate assessment of risks related to dental procedures. Therefore,
development of a standardized radiotherapy dental information tool and data-driven, algorithmic toxicity risk
prediction models for enhanced communication and personalized medicine for OC/OPC survivors remains an
unmet public health need. In response to NIDCR’s NOT-DE-20-006, we herein propose a rigorous and
reproducible application of informatics and computational methods and approaches for the development of
machine learning “ML/AI based optimization of clinical procedures for precision dental care”, “novel and robust
data analysis algorithms to tackle causal mechanisms of action for onset and progression of disease” related to
posttherapy orodental complications, and “computational modeling for treatment planning and assessment of
treatment outcomes.” In Specific Aim 1, we will train and validate a deep learning contouring (DLC) neural
network for automatic delineation of tooth-bearing regions. Our collaborator, Dr. van Dijk, has previous
experience with DLC design and application for auto-delineation of non-dental head and neck organs at risk
(OAR). Her research, published in a peer-reviewed journal showing an equal or significantly improved OAR
automatic delineation using DLC over atlas-based contouring, will serve as a reproducible model for our
proposed project. Using DLC-based mandibular and dental OAR delineation (SA 1), we will develop a novel
“radiation odontogram” which will generate automated and accurate summative radiotherapy dose distribution
mapping reports for effective datatransmission and communication among providers (SA 2). Accurate prognosis
and management of high-morbidity high-prevalence post-therapy orodental sequelae will be enabled through
the development of a statistically robust machine-learning based model of toxicity risk predictions that
incorporates patient- and provide-generated data(Aim 3). In summary, the RADMAP proposal fosters innovative
informatics and computational modeling approaches to address existing challenges in multidisciplinary
communication and precision dental care for OC/OPC survivors, with practice-changing implications in the
clinical setting and for oral, dental, and craniofacial research.
