(PR108) Automatic Periapical Lesion Segmentation Using Artificial Intelligence Model on Cone Beam Computed Tomography Images

Objective: The purpose of the study was to develop and evaluate the performance of an artificial intelligence model for automatic periapical lesion segmentation on cone beam computed tomography (CBCT) images. Materials and

Methods: In total, 300 CBCT images were included in this study. DICOM files of CBCT images were converted to axial slice frame images in JPEG format. Annotation was made using CranioCatch annotation software (CranioCatch Inc., Eskişehir-Türkiye) for periapical lesions. For model development, Mask R-CNN architecture and the ResNet 101 model were used as transfer learning methods. The success metrics of the model were calculated with the confusion matrix method.

Results: The evaluation using the periapical lesion test data set resulted in 180 images with 236 labels. Evaluation of the diagnostic performance of the AI model to segmented periapical lesions showed values of 203 True Positive (TP), 12 False Positive (FP), and 33 False Negative (FN). Sensitivity, precision, and F1 score values were calculated as 0.8601, 0.9441, and 0.9002, respectively. The Area Under Curve (AUC) value was 0.8277.

Conclusion: The performance of the artificial intelligence algorithm developed by using the Mask R-CNN ResNet 101 model in detecting periapical lesions on CBCT images has promising results. Artificial Intelligence can play a role as a decision support system for dentists and specialists in the diagnosis and treatment planning of periapical lesions on CBCT images.