Journal Article

Deep learning in cone-beam computed tomography image segmentation for the diagnosis and treatment of acute pulpitis

Xie, Zhiqiang, Zhang, Xiaoyan, Zhu, Xiaoyan

2022

Overview

To evaluate the effect of deep learning model on cone beam (CB) CT image analysis of patients with acute pulpitis. The improved principle of maximum entropy and minimum energy method (PME-MEM’) was proposed to preprocess CBCT images. The conditional generative adversarial network (cGAN) model of deep learning was adopted to segment images. In this study, 80 cases of acute pulpitis in our hospital were selected as the research objects. CT images of the patients were collected and pretreated with PME-MEM. The denoising effects of different Gaussian noise treatments were compared and analyzed, and cGAN model was used to segment different parts of teeth in the image. The treatment plan was made according to the processed CT images, and patients were rolled into two groups according to the treatment methods, with 40 cases in each group. The modified group received one-off root canal treatment, and the traditional group received multiple root canal treatments. The postoperative treatment effects of the patients were observed. The results showed that the PME-MEM’ had a better denoising effect on CBCT images relative to the original PME-MEM. The deep learning cGAN model can realize the segmentation of caries, enamel, dentin, dental pulp, crown, restoration, and root canal in CBCT images. The clinical treatment results showed that the treatment time and postoperative pain score of the modified group were considerably reduced versus those of the traditional group (P < 0.05). The postoperative comfort score and satisfaction with treatment results increased greatly (P < 0.05). In short, deep learning can be used to segment the target position in CBCT images of patients. Combined with one-off root canal therapy, the therapeutic effect was ideal for patients with acute pulpitis.

• Publisher: Springer Nature B.V
• Subject: Algorithms / Computed tomography / Deep learning / Dentistry / Disease / Endodontics / Energy methods / Entropy / Generative adversarial networks / Health services / Image analysis / Image segmentation / Machine learning / Maximum entropy / Medical imaging / Noise reduction / Patients / Random noise