Deep learning in panoramic radiography: a scoping review for the diagnosis of maxillofacial cysts and tumors.

OBJECTIVE

To evaluate the architecture, performance metrics, and clinical applicability of Deep Learning (DL) models as a diagnostic support tool in panoramic radiographs (PR) for detecting, segmenting, and classifying maxillofacial cysts and tumors.

STUDY

DESIGN

A scoping review was conducted following the PRISMA-ScR guidelines. Electronic databases were searched without restrictions. Studies applying DL for maxillofacial cysts or tumors in PR were included. Extracted data were organized into three domains: model construction, performance evaluation, and clinical applicability.

RESULTS

Twenty-eight studies were included. A wide variety of DL models were identified, with U-Net, YOLO, and DetectNet being the most frequently used. Reported evaluation metrics included sensitivity, specificity, precision, Dice coefficient, intersection over union (IoU), and mean average precision (mAP), showing high performance values. Some studies compared DL models with specialists, reporting similar or superior accuracy and shorter execution times. Most DL models were applied to radicular cysts, dentigerous cysts, and odontogenic keratocysts located in the maxilla and/or mandible.

CONCLUSION

DL models demonstrate potential as diagnostic support tools for interpreting PR of maxillofacial cysts and tumors. This review provides an integrated overview of their construction, evaluation, and applicability, contributing to the development of computer-aided diagnosis systems in oral imaging.