Comparative evaluation of images of alveolar bone loss using panoramic images and artificial intelligence

Aim: The present study aimed to employ a VGG-16 convolutional neural network (CNN) system to determine alveolar bone loss and periodontal disease/health status from dental panoramic radiography images.

Materials and Methods: This study was conducted with a dataset of panoramic images obtained from an institution. The training dataset contained 1874 panoramic images, of which 953 were of bone loss cases and 921 were of periodontally healthy cases. The presence/absence of resorption at the bone crest was recorded in consideration of the distance between the enamel–cementum junctions of the teeth and the alveolar bone crest. Radiographs showing bone resorption with a horizontal/vertical shape or bone defects were included in the bone loss group. Images with artefacts, image distortion, and blur were excluded. All images in the dataset were resized to 1472 x 718 pixels, followed by preprocessing for arbitrary sequence formation using python language along with OpenCV, NumPy, Pandas, and Matplotlib libraries to generate an image dataset. The dataset was divided into a testing set and validation set. The validation dataset was used to validate the model. The Feature Extraction approach strategy with 10–20 epochs and learning rate 1e-4 (0.0001) for trainable layers were used. Training and validation datasets were used to predict and generate optimal weight factors for this CNN.

Results: Of 100 bone loss cases, the CNN system evaluated 92 correctly and 8 incorrectly. Further, of 100 periodontally healthy cases, it evaluated 89 correctly and 11incorrectly. The sensitivity, specificity, precision, accuracy, and F1 score were 0.8317, 0.8683, 0.8918, 0.8927, and 0.8615, respectively.

Conclusion: The CNN model was successful in assessing the alveolar bone loss and periodontal disease status using panoramic radiographs. This study resulted in effective image segmentation which can more accurately predict grading of periodontal bone loss, avoiding user interpretation for overlapping on periodontal classification.