Using data to understand disease and advance more precise treatments.
Interdisciplinary teams of dental researchers, data scientists and engineers collaborate to study the mechanisms driving ailments ranging from oral cancer to chronic pain.
By applying AI tools to large, complex datasets, they uncover patterns that would otherwise be difficult to detect. This team science approach supports earlier detection, more consistent diagnosis and more targeted treatment decisions.
The Neurobiology of Pain and Aging
Advanced methodologies & molecular breakthroughs
Pain research led by UFCD clinicians and peers across the UF Health system focuses on developing integrated biopsychosocial signatures.
Rather than studying single biomarkers, our teams synthesize artificial intelligence with neuroimaging and molecular epigenetics.
This approach captures a complete profile of a patient’s biology, psychology and social environment to guide specific treatments.
What this research entails:
Applying AI to clinical brain imaging
Brain age prediction models typically require high-resolution research-grade MRIs. To apply these algorithms to standard hospital data, UF researchers use SynthSR, or Synthetic Super-Resolution, a publicly available AI tool that digitally enhances low-resolution clinical MRIs into high-resolution images.
The research team compiled 6,281 clinical MRIs from 1,559 patients across 15 UF Health System facilities and retrained the DeepBrainNet algorithm to calculate “biological brain age” from routine hospital scans. This allows researchers to search large archives of clinical images for early signs of cognitive decline.
The epigenome and Alzheimer’s disease
Epidemiological data indicates that high-impact chronic pain correlates with a higher risk of cognitive decline and Alzheimer’s disease. Researchers use epigenetic clocks like DNAmGrimAge, an algorithm that predicts biological age and health lifespan based on chemical changes to DNA, to map the molecular mechanics behind this correlation.
High-impact pain alters DNA methylation, the process by which genes are turned on or off without changing the DNA code itself. Pathway analyses show that these epigenetic changes mimic the early molecular stages of Alzheimer’s disease, establishing the epigenome as a diagnostic target for comorbid pain and dementia.
Preclinical histology and deep learning
Traditional microscopic evaluations, or histologies, of osteoarthritis use grading scales that measure overall disease severity but omit precise locations of joint damage.
UF scientists use convolutional neural networks, a class of deep-learning AI models capable of complex image segmentation and classification, to analyze and map microscopic joint changes in laboratory models. Automating these segmentations captures the spatial relationships between cartilage loss, bone thickening (subchondral bone sclerosis), and bone spurs (osteophyte formation), providing higher-resolution data to evaluate osteoarthritis therapies.
This level of detail is essential for identifying the precise mechanisms driving joint pain and accelerating our ability to evaluate the effectiveness of next-generation, disease-modifying therapies.
Working with large-scale clinical data
Some of the most useful clinical data in electronic dental records are rendered unusable because they exist as “unstructured free text.”
That is, human-written text that does not follow a pre-defined data model, schema or rigidly organized format. Because it lacks fixed fields (like a spreadsheet row), computers cannot easily search, sort or analyze it, making large-scale analysis difficult.
Manual Review
Time-intensive and difficult to scale
Manual review is painfully slow and limiting. In one case, reviewing 500 charts took UFCD researchers nearly a year and yielded only 40 complete cases for analysis.
At full scale, that approach was not workable: reviewing hundreds of thousands of clinical notes by hand would have taken decades (estimated at roughly 27 years).
NLP-assisted extraction
Scalable insights powered by AI
To address this barrier, UFCD researchers developed an orthodontic natural language processing (ONLP) model to extract diagnostic and treatment-planning information directly from free text.
The model can identify clinical concepts such as malocclusion type, crowding and overjet, even when notes contained inconsistent terminology or misspellings.
Using a dataset of more than 27,000 patients, including 7,693 orthodontic cases, the team converted electronic dental records into structured variables that supported large-scale analysis and machine learning.
The model achieved 91% accuracy in extracting clinically relevant data.
The resulting dataset made it possible to analyze treatment patterns at a scale that had not been practical, supporting research in interceptive orthodontics and informing the development of a clinical decision support tool.
Biomedical ontologies
Connecting data across systems
Even when clinical data is well-structured, it often remains siloed in incompatible systems with disparate terminology.
To resolve this disconnect, UFCD researchers develop biomedical ontologies. More than standardized vocabularies, ontologies are formal, machine-readable digital frameworks that serve as “semantic bridges” linking complex information across a broad range of disciplines.
By creating this common scientific language, researchers across medical fields and specialties can reliably share and analyze diverse health data across previously disconnected systems.
Transforming fragmented clinical data into a connected research resource requires human insight that AI simply can’t replace.
At UFCD, this consequential work is led by William “Bill” Duncan, Ph.D., the college’s first faculty member dedicated to artificial intelligence. His research focuses on building these foundational ontologies to serve as a universal digital language.
Explore Duncan’s ontology-driven research:
- “Structuring, reuse and analysis of electronic dental data using the Oral Health and Disease Ontology” (August 2020)
- “Improving the Quality and Utility of Electronic Health Record Data through Ontologies” (March 2023)
- “Representing dental caries and dysbiosis within the oral microbiome in the Oral Health and Disease Ontology” (April 2026)
Early Detection of Oral Cancer Risk
Predicting the progression of pre-cancerous oral lesions
UFCD researchers are combining clinical, microscopic and molecular data to understand how pre-cancerous oral lesions turn into cancer.
This approach builds on their extensive foundation in oral pathology and biomarker discovery.
To support large-scale, AI-driven analysis, researchers have digitized pathology slides and established standardized diagnostic criteria for high-risk lesions, such as proliferative verrucous leukoplakia (PVL). Additionally, UFCD clinical trials have tested innovative infrared imaging devices to evaluate the heat diffusion of suspicious lesions before a biopsy is even performed.
At the molecular level, researchers have identified the CIP2A protein as a key signal for early malignant changes. UF teams developed highly sensitive, point-of-care biosensors to detect this marker in both saliva and tissue. These devices generate precise, quantitative data that is ideal for computational modeling.
Currently, UFCD is using semantic technologies to unify these diverse data sources into a single framework. Because our researchers have already successfully deployed complex machine learning to predict risks in other diseases and head and neck cancers, extending these predictive AI models to oral cancer precursors is highly feasible.
Research foundations
- Standardizing diagnosis of high-risk oral lesions
Proliferative Verrucous Leukoplakia: An Expert Consensus Guideline for Standardized Assessment and Reporting - Detecting early molecular signals of malignant transformation
Sensitive Detection of Oral Leukoplakia: Analyzing P90 Biomarkers in Saliva and Tissue - Applying machine learning to clinical risk prediction
The Florida Scoring System for stratifying children with suspected Sjögren’s disease
Diagnostic precision in childhood autoimmunity
Identifying hidden markers in pediatric Sjögren’s disease
Childhood Sjögren’s disease (cSjD) is a rare and often underdiagnosed condition. Unlike adults, pediatric patients frequently lack standard biological markers, such as anti-SSA autoantibodies, and instead present with atypical challenges including glandular swelling, extreme fatigue, joint pain and neurological issues. As a result, traditional adult diagnostic tests often fail, leading to agonizing delays in diagnosis and care.
To address this disparity, researchers in the Cha Laboratory at UFCD developed the Florida Scoring System.
Using machine learning models — including artificial neural networks, random forests and gradient-boosted decision trees — the team analyzed hundreds of clinical and laboratory clues to identify patterns not captured by conventional diagnostic methods.
The models defined three distinct patient classes, most notably identifying a “hidden” high-risk group (Class II). These highly vulnerable children suffer severe systemic symptoms but consistently test negative on standard exams. The Florida Scoring System also supports gentler diagnostic tools, demonstrating that non-invasive salivary gland ultrasounds can reliably replace or complement minor salivary gland lip biopsies.
Predictive Analytics for Population Health
Clarifying oral health disparities and guiding resource allocation
UFCD researchers are applying AI-driven predictive modeling to state and national public health datasets to identify where oral health disparities are greatest among Florida’s aging population.
Led by prosthodontics researcher Merve Benli, D.D.S., Ph.D., this work extends clinical care to the population level.
In a study presented at the 2026 Spring Synergy symposium, Benli’s AI models integrated demographic, socioeconomic and dietary data to map geographic hotspots of complete tooth loss across Florida.
The models revealed that rural and high-poverty communities have the highest unmet prosthodontic needs. More importantly, the AI predicted exactly how targeted care in these areas could improve lives, estimating a 32% gain in chewing function and a 26% improvement in healthy dietary patterns if older adults received optimized prosthodontic interventions.
These insights offer a scalable, evidence-based approach to resource allocation, supporting precision public health strategies that funnel care exactly where it will have the greatest human impact.