Automatic detection of radiographic alveolar bone levels in bitewing and periapical intraoral radiographs using deep-learning technology: a pilot study

Objective

This pilot study aims to evaluate the diagnostic performance of a deep-learning platform, employing convolutional neural networks (Denti.AI Technology), in detecting radiographic alveolar bone levels in bitewing and periapical digital intraoral radiographs.

Study Design

The pretrained deep-learning model was tested using 22 periapical radiographs and 17 bitewing radiographs. The reference standard for the study was a consensus of 3 board-certified dental specialists. The panel followed the American Academy of Periodontology guidelines to identify radiographic bone loss on the basis of specific criteria for bitewing and periapical radiographs. Performance metrics including sensitivity, specificity, predictive values, accuracy, and mean absolute error (MAE) of the deep learning detecting bone levels were calculated.

Results

Standalone performance analysis revealed a sensitivity of 76%, specificity of 86%, positive predictive value of 83%, negative predictive value of 80%, accuracy of 81%, and MAE of 0.046 for periapical radiographs. Bitewing radiographs exhibited a sensitivity of 65%, specificity of 90%, positive predictive value of 88%, negative predictive value of 70%, accuracy of 76%, and MAE of 0.499.

Conclusion

This pilot study highlights the potential of deep-learning technology, using convolutional neural networks, for automatic detection of radiographic alveolar bone levels in bitewing and periapical intraoral radiographs. The achieved diagnostic performance metrics indicate promising uses for this technology.