Research Information System
INTERNATIONAL JOURNAL OF IMAGING SYSTEMS AND TECHNOLOGY, vol.34, no.1, 2024 (SCI-Expanded, Scopus)
This retrospective study is aimed at developing a web-based artificial intelligence (AI) software (DiagnoCat) for periodontal bone loss detection on panoramic radiographs and evaluating the model's performance by comparing it with clinicians' results. Separate models are trained for tooth and periodontal bone loss detection. The first model's objective was to detect teeth, segmenting their masks, and to define their numbering and developed with Mask R-CNN using pretrained ResNet-101 as a backbone. The second model was based on Cascade R-CNN architecture and used for bone loss prediction. Around 100 radiographs are evaluated by three clinicians regarding tooth identification and periodontal bone loss, separately. Ground truth is determined by the consensus and model's performance is evaluated with kappa, precision, recall, and F-score statistics. For tooth conditions, the overall F-score, accuracy, and Cohen's kappa coefficients were found to be 0.948, 0.977, and 0.933 for the binary, and 0.992, 0.988, and 0.961 for the multiclass results. For bone loss detection, the overall F-score, accuracy, and Cohen's kappa coefficients were found to be 0.985, 0.980, and 0.956 for the binary, and 0.996, 0.993, and 0.974 for the multiclass results. The results of this study suggest that the use of a web-based AI software (DiagnoCat) can be beneficial in detecting periodontal bone loss on panoramic radiographs.