(PR03) The Accuracy of Using Computer Aided Detection Software (CADs) in Identifying Apical Radiolucencies in Panoramic Radiographs

Abstract: The detection of apical radiolucencies is important in the diagnosis and treatment of pulpal disease and persistent apical periodontitis. One promising method for adjunctive detection of apical radiolucencies is artificial intelligence through computer aided diagnostic systems (CADs). Computer aided diagnostic systems are used in radiology to complement the practitioner’s evaluation in order to form a final decision. This study evaluated the accuracy of general dentists detecting apical radiolucencies in panoramic radiographs, with and without the help of CADs. 100 consecutive panoramic radiographs were acquired from 2019-2020 from UNC Adams School of Dentistry. Ground truth was established by a consensus panel of three board certified Oral and Maxillofacial Radiologists. Images were uploaded to Denti.AI, a deep learning tool using convoluted neural networks. 10 general dentist readers annotated the presence of apical radiolucencies in panoramic radiographs first without the help of CADs, and then with the help of CADs. Standard metrics of ROC, ROC/AUC with CI, sensitivity, and specificity were evaluated for the readers. AFROC AUC increased by 0.029 (4%) in total readers compared to unaided read with no statistical significance (p=0.063). ROC AUC increased by 0.30 (3.71%) with no statistical significance (p=0.067). Sensitivity increased by 0.061 (7.86%) with statistical significance (0.003). Specificity decreased by 0.04 with no statistical significance (p=0.061). Using computer aided detection software can improve dentists’ ability to detect more cases.