Automated deep learning for classification of dental implant radiographs using a large multi-center dataset

This study aimed to evaluate the accuracy of automated deep learning (DL) algorithm for identifying and classifying various types of dental implant systems (DIS) using a large-scale multicenter dataset. Dental implant radiographs of pos-implant surgery were collected from five college dental hospitals and 10 private dental clinics, and validated by the National Information Society Agency and the Korean Academy of Oral and Maxillofacial Implantology. The dataset contained a total of 156,965 panoramic and periapical radiographic images and comprised 10 manufacturers and 27 different types of DIS. The accuracy, precision, recall, F1 score, and confusion matrix were calculated to evaluate the classification performance of the automated DL algorithm. The performance metrics of the automated DL based on accuracy, precision, recall, and F1 score for 116,756 panoramic and 40,209 periapical radiographic images were 88.53%, 85.70%, 82.30%, and 84.00%, respectively. Using only panoramic images, the DL algorithm achieved 87.89% accuracy, 85.20% precision, 81.10% recall, and 83.10% F1 score, whereas the corresponding values using only periapical images achieved 86.87% accuracy, 84.40% precision, 81.70% recall, and 83.00% F1 score, respectively. Within the study limitations, automated DL shows a reliable classification accuracy based on large-scale and comprehensive datasets. Moreover, we observed no statistically significant difference in accuracy performance between the panoramic and periapical images. The clinical feasibility of the automated DL algorithm requires further confirmation using additional clinical datasets.

Automated summary

This study evaluated the accuracy of an automated deep learning algorithm for classifying dental implant systems using a large-scale dataset. The algorithm achieved high accuracy, precision, recall, and F1 score for both panoramic and periapical radiographic images. There was no significant difference in accuracy between the two types of images. The clinical feasibility of the algorithm requires further confirmation with additional datasets.