Deep Learning Approach to Measure Alveolar Bone Loss After COVID-19

Severity of periodontal disease may be determined by measurement of alveolar crestal height (ACH) on dental bitewing radiographs; however, the prevailing method of assessment is through visualization which is time consuming and not a direct measure. The primary objective of this manuscript is to create and validate a deep learning technique for precise evaluation of alveolar bone loss in bitewing radiographs. Additionally, surveys were conducted with dental professionals to determine accuracy of visualized measures of ACH for severe periodontal disease versus the deep learning program and to determine the acceptability of utility of the program among diverse dental professionals. Lastly, the deep learning program was utilized in research to evaluate the role of COVID on periodontal disease through longitudinal measures of bitewing radiograph ACH from patients during the: "pre-pandemic" (Feb 2017 - Feb 2020) and "post-pandemic" (Feb 2020 - Feb 2023) periods. The pre-pandemic group had a mean percentage loss of ACH of -1.74 + 16.5%, representing a gain in alveolar bone. In contrast, the post-pandemic group had a gain in ACH of 2.46 + 14.6%, representing a loss in alveolar bone. There remained a trend for greater annualized percent change in ACH in the post-pandemic vs pre-pandemic group (1.33 + 11.9% vs -0.94 + 12.5%, p=0.07), after accounting for differences in duration between xrays. Overall, this study demonstrates the successful training and validation of a deep learning program for ACH measurement as well as its utility and acceptability among dental professionals for clinical and research.