Diagnosis of in vivo vertical root fracture using deep learning on cone-beam CT images

Abstract

Objectives: Evaluating the diagnostic efficiency of deep learning models to diagnose vertical root fracture in vivo on cone-beam CT (CBCT) images.

Materials and methods: The CBCT images of 276 teeth (138 VRF teeth and 138 non-VRF teeth) were enrolled and analyzed retrospectively. The diagnostic results of these teeth were confirmed by two chief radiologists. There were two experimental groups: auto-selection group and manual selection group. A total of 552 regions of interest of teeth were cropped in manual selection group and 1118 regions of interest of teeth were cropped in auto-selection group. Three deep learning networks (ResNet50, VGG19 and DenseNet169) were used for diagnosis (3:1 for training and testing). The diagnostic efficiencies (accuracy, sensitivity, specificity, and area under the curve (AUC)) of three networks were calculated in two experiment groups. Meanwhile, 552 teeth images in manual selection group were diagnosed by a radiologist. The diagnostic efficiencies of the three deep learning network models in two experiment groups and the radiologist were calculated.

Results: In manual selection group, ResNet50 presented highest accuracy and sensitivity for diagnosing VRF teeth. The accuracy, sensitivity, specificity and AUC was 97.8%, 97.0%, 98.5%, and 0.99, the radiologist presented accuracy, sensitivity, and specificity as 95.3%, 96.4 and 94.2%. In auto-selection group, ResNet50 presented highest accuracy and sensitivity for diagnosing VRF teeth, the accuracy, sensitivity, specificity and AUC was 91.4%, 92.1%, 90.7% and 0.96.

Conclusion: In manual selection group, ResNet50 presented higher diagnostic efficiency in diagnosis of in vivo VRF teeth than VGG19, DensenNet169 and radiologist with 2 years of experience. In auto-selection group, Resnet50 also presented higher diagnostic efficiency in diagnosis of in vivo VRF teeth than VGG19 and DensenNet169. This makes it a promising auxiliary diagnostic technique to screen for VRF teeth.

Keywords: Artificial intelligence, cone-beam computed tomography, deep learning; Neural networks (computer); Root fractures.

Fig. 1

The workflow of the deep learning framework. Firstly, the same tooth on dentition images were manually selected in manual selection group and auto-selected using tooth selection model in auto-selection group. The images in two groups were then preprocessed in the same way and used as datasets to three CNN models. Finally, the three CNN models output the diagnostic result of manual selection group and auto-selection group

Fig. 2

The schematic diagram of tooth selection model. A shows the original dentition images. B shows the dentition images got Gaussian blurred. The detail in image got reduced. C shows binary dentition images. the shape of dentition got extracted. D shows the moving line has been extracted and placed on the original dentition image in corresponding position. E shows the identification boxes has been placed on the dentition every 60–80 pixels. F is the cropped image original image along the outline of identification box

Fig. 3

ROC curve of three CNN models in two experimental groups. ResNet50 presented the highest AUC in both manual selection group and auto-selection group with AUC of 0.99 and 0.96, respectively

Fig. 4

Teeth in dataset with complex symptoms. A1, B1 and C1 are VRF teeth. A2, B2 and C2 are non-VRF teeth. A1 shows an arch low-density area (bone loss) at one side of the fracture on the CBCT image. A2 also shows an arch low-density area (bone loss) at the lingual side of distal root on the CBCT image. However, this tooth is a non-VRF tooth. B1 and B2 show a low-density area around the mesial root on the CBCT image. However, B1 is VRF tooth and B2 is non-VRF tooth. C1 shows a subtle fracture. C2 shows a tooth with horizontal bone loss. Low-density area is large and around the tooth. All teeth above were correctly diagnosed in manual selection group