Success of Artificial Intelligence System in Determining Alveolar Bone Loss From Dental Panoramic Radiography Images

Abstract

This study aims to detect alveolar bone loss from dental panoramic radiography images by using an artificial intelligence (AI) system. Materials and Methods:A total of 2276 panoramic radiography images were evaluated. Of these, 1137 were of bone loss cases and 1139 were of periodontally healthy cases. This dataset is divided into training (n = 1856), validation (n = 210), and testing (n = 210) sets. All images were resized to 1472x718 pixels before training. A random sequence was created using the open-source Python programming language and OpenCV, NumPy, Pandas, and Matplotlib libraries. A pretrained Google Net Inception v3 convolutional neural network (CNN) was used for preprocessing, and the datasets were trained using transfer learning. The diagnostic performance was evaluated using a confusion matrix in terms of the sensitivity, specificity, precision, accuracy, and F1 score.Results: Of 105 cases with bone loss, the CNN system detected 99 with sensitivity, specificity, precision, accuracy, and F1 score of 0.94, 0.88, 0.89, 0.91, and 0.91, respectively. Conclusions:The CNN system successfully determines periodontal bone loss. Therefore, it can be used to facilitate diagnosis and treatment planning by oral physicians in the future.

Development and validation of an artificial intelligence software for periodontal bone loss in panoramic imaging

“…The model is trained using transfer learning with the Tensor Flow with a total of 2276 images. The sensitivity, specificity, precision, accuracy, and F1 score for the model are reported to be 0.94, 0.89, 0.89, 0.91, and 0.92, respectively 19 . In our study, the performance of the AI software is evaluated with binary and multiclass scores.…”

“…The sensitivity, specificity, precision, accuracy, and F1 score for the model are reported to be 0.94, 0.89, 0.89, 0.91, and 0.92, respectively. 19 In our study, the performance of the AI software is evaluated with binary and multiclass scores. 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.…”

“…AI techniques can be used in developing systems for the diagnosis of periodontitis. Several studies on the diagnosis of periodontal disease in panoramic images have been reported, and the CNN approach is commonly adopted in the developed models 7,15–19 . This study aims to evaluate the performance of a web‐based AI software (DiagnoCat) developed for tooth numbering and diagnosis of periodontal defects on digital panoramic radiographs, by comparing the results of the AI software with the hybrid results of human observers.…”

“…Several studies on the diagnosis of periodontal disease in panoramic images have been reported, and the CNN approach is commonly adopted in the developed models. 7,[15][16][17][18][19] This study aims to evaluate the performance of a web-based AI software (DiagnoCat) developed for tooth numbering and diagnosis of periodontal defects on digital panoramic radiographs, by comparing the results of the AI software with the hybrid results of human observers. The null hypothesis of this research is that the results of the AI system are comparable to the consensus determined by the voting of human observers for detecting periodontal bone loss in panoramic radiographs.…”

Applying artificial intelligence to detect and analyse oral and maxillofacial bone loss—A scoping review

“…Deep‐learning methods significantly improve the visual quality of images but do not necessarily improve the classifier's performance. The CNN system detected 99 bone loss cases out of 105, with sensitivity, specificity, precision, accuracy and F1 scores of 94%, 88%, 89%, 91% and 91%, respectively [63].…”