A perspective on the diagnosis of cracked tooth: imaging modalities evolve to AI-based analysis

Abstract

Despite numerous clinical trials and pre-clinical developments, the diagnosis of cracked tooth, especially in the early stages, remains a challenge. Cracked tooth syndrome is often accompanied by dramatic painful responses from occlusion and temperature stimulation, which has become one of the leading causes for tooth loss in adults. Current clinical diagnostical approaches for cracked tooth have been widely investigated based on X-rays, optical light, ultrasound wave, etc. Advances in artificial intelligence (AI) development have unlocked the possibility of detecting the crack in a more intellectual and automotive way. This may lead to the possibility of further enhancement of the diagnostic accuracy for cracked tooth disease. In this review, various medical imaging technologies for diagnosing cracked tooth are overviewed. In particular, the imaging modality, effect and the advantages of each diagnostic technique are discussed. What's more, AI-based crack detection and classification methods, especially the convolutional neural network (CNN)-based algorithms, including image classification (AlexNet), object detection (YOLO, Faster-RCNN), semantic segmentation (U-Net, Segnet) are comprehensively reviewed. Finally, the future perspectives and challenges in the diagnosis of the cracked tooth are lighted.

Keywords: Artificial intelligence; Image processing; Review of oral diagnosis; Survey of crack detection.

Fig. 1

Illustrations of the methods for the diagnosis of cracked tooth

Fig. 2

Illustration of the measurement of oral X-ray radiograph

Fig. 3

Illustration of the measurement of CBCT

Fig. 4

Illustration of the measurement of OCT

Fig. 5

Comparison study of four crack detection methods: (a–c) images from trans-illumination detection; (d–f) images from intraoral radiography; (g–i) images from CBCT; (j–l) images from OCT. The red line indicates the cross section of CBCT and the OCT scan line. Red and blue arrows indicate crack lines. The blue circle indicates a false-positive crack in trans-illumination detection. The image resource was cited from reference [53]. Copyright © 2017. Korean Academy of Periodontology publishing

Fig. 6

Block diagram to illustrate the framework of the methods for detection of cracked tooth

Fig. 7

Representative frameworks of convolutional neural network for crack detection and segmentation

Fig. 8

Typical results of object detection (YOLO) [120] and semantic segmentation (U-Net) [136] under different situations. The anchor boxes in the figure (a–d) represent the area of crack. a Testing results of normal crack in the smooth pavement. b Testing results of normal crack in the rough ground. c Tiny crack in the smooth pavement. d Normal crack in the stained ground. For semantic segmentation methods, e, g represent the original images under shadow and rough pavement, respectively. f, h Results of semantic segmentation of the crack. Copyright © Elsevier publishing (2019)