Automated Characterization of Arterial Calcification in Dental Cone Beam Computed Tomography as Predictors of Cardiovascular Disease

SUMMARY: The utilization of cone beam computed tomography (CBCT) in dentistry has seen a significant rise, driven by its superior imaging capabilities that offer detailed 3D views of dental structures and adjacent anatomy. This allows for more precise interventions in orthodontics, implantology, and endodontics. However, an important aspect of CBCT's expanding role is the identification of incidental findings, particularly vascular calcifications, which can be indicative of systemic conditions such as cardiovascular disease (CVD). Detecting these calcifications early is crucial because they are associated with an increased risk of cardiovascular events like heart attacks and strokes. While many dentists utilize CBCT in their daily practice, this finding may go unnoticed due to lack of experience and familiarity with the complex craniofacial anatomy. Furthermore, currently, there are no established quantifiable metrics for prediction of CVD based on the imaging appearance, severity, and extent of vascular calcifications from CBCT. By recognizing these incidental findings, dental professionals can play a pivotal role in the early referral of patients to medical care, potentially mitigating the risks of severe cardiovascular outcomes and enhancing overall patient health. The proposed research involves the development of an automated tool to detect and characterize arterial calcifications in dental CBCT. A major challenge in dental image segmentation is the limited availability of high quality and large volume of training datasets. Additionally, the segmentation target (the arterial calcification region) is extremely small (< 1% of the entire image volume), rendering existing approaches suboptimal. Unlike traditional deep learning segmentation techniques that rely on vast amounts of training data, and where the segmentation target is not small, this work proposes to develop a framework that leverages an anatomy-driven self-pretraining paradigm to segment the calcifications reliably and robustly even when training data is limited and the target region substantially small. Though vascular calcifications have been shown to be associated with risk of CVD, the imaging presentation of these calcified regions in CBCT is relatively understudied. Having a quantifiable metric for prediction of CVD based on the severity and extent of vascular calcifications from CBCT data can assist the clinician with an objective risk assessment. This work is based on the hypothesis that morphometric and textural features from these regions have significant predictive value; this will be studied using novel radiomic analysis. Leveraging cutting edge computational approaches, this project will be the first to study CBCT extracted subtle “sub-visual” computational imaging features in addition to traditionally investigated clinical risk factors to strengthen the risk prediction models’ predictive capabilities. Successful completion of the project will result in establishing specific measurable imaging features that provide risk stratification better than conventional clinical metrics. The team will develop these computational imaging and machine learning tools using a set of N=1000 CBCT scans obtained from Stony Brook School of Dental Medicine.