Machine learning in the field of dentistry

Abstract

Oral disease is very common, both old and young will be troubled by it. For more severe and complex cases, dentists often use dental radiographs to diagnose and plan treatment. However, the use of dental radiographs for auxiliary diagnosis can only rely on the experience of doctors, and dental radiographs for the naked eye is complicated, long-time work will lead to human fatigue and misjudgment.

The development of machine learning makes object detection achieve an efficient and high accuracy performance. Therefore, machine learning related technologies have been applied in the field of dentistry. The RetinaNet chosen in this project is an one-stage algorithm which achieves the accuracy comparable to two-stage algorithm while retaining the advantages of one-stage algorithm with few memory consumption and fast processing speed.

The main purpose of this project is to demonstrate RetinaNet can be applied to dentistry by constructing a RetinaNet for the detection of dental crowns in dental radiographs, and to use a trained model to detect dental crowns. The algorithm was built by Python, and finally verified to achieve high precision in dental dataset. {mAP}_{80} reaches 89.80%, which is higher than the accuracy of the network on VOC2007 dataset.