Application of Artificial Intelligence for Detection of Dental Carious Lesions in Digital Dental Radiographs: An

Aim: An artificial intelligence (AI) model has been introduced to recognize and classify attributes present on the digital intraoral periapical radiographs (IOPA) radiographic images and aid in identification of carious lesions in efficient manner. Materials and methods: An object detection-based CNN model (YOLOv4) has been modified and trained for real-time object (caries) localization and classification (based on depth) using a labeled dataset. During the training phase, the training dataset images with bounding box information [center coordinates (x, y), width, height (w, h) values, and class labels such as E, D1, D2 and D3] were prepped. The training was performed in Darknet-53 YOLOv4 GPU environment on Google Colab. Dentists with more than 5 years of clinical experience were recruited as observers. The statistical analysis was done using SPSS version 25. Interexaminer reliability was checked using Fleiss' Kappa test. The sensitivity, specificity and area under receiver operating characteristic curves (AUC) of AI model and human observers (MHO) were calculated and compared. Results: The AI model demonstrated lower sensitivity (80.0%) for caries detection than MHO (90.9%) Conclusion: The performance metrics showed promising outcomes in terms of detecting caries. The proposed model has achieved an accuracy of 84.66% and precision of 93.62% in diagnosing carious lesions. However, the performance of the model in carious depth classification was not satisfactory. The way forward is to collect more periapical images for training of AI model. Within the limitations of this study, it is concluded that the YOLO-based CNNs can be used in the diagnosis of caries in periapical radiographs. How to cite this article: Sikka N, Goel A, Arya L,