2021 International Conference on Medical Imaging Physics and Engineering and the 10th National Annual Meeting of Medical Imaging Physics (ICMIPE 2021)

54 / 2021-10-14 19:06:46

Application of Deep Learning in Diagnosis of Ameloblastoma and Odontogenic Keratocyst Based on Panoramas

deep learning,Convolution neural network,panoramas,ameloblastoma,odontogenic keratocyst

Abstract Objective: ameloblastoma (AM) and odontogenic keratocyst (OKC)are common odontogenic tumors of the jaw. Their radiological findings are similar, but the biological behaviors of these two types of tumors are different, and there are obvious differences in clinical treatment principles. In order to help oral and maxillofacial surgeons plan appropriate treatment faster and more accurately, we select the method of deep learning to compare the performance of deep learning neural convolutional network with diagnosis results produced by oral and oral radiologists.

Methods: 1000 digital panoramic images of ameloblastoma and odontogenic keratocyst were collected retrospectively from the imaging department of Stomatology Hospital of Peking University. RESNET, VGG, EfficientNet and other deep learning neural network models were used to identify ameloblastoma and odontogenic keratocyst. To overcome this difficulty, the transfer learning method has been applied to neural convolutional networks. The training data comprised 400 ameloblastoma images and 400 OKC images. The test data comprised 100

ameloblastoma images and 100 OKC images. At the same time, 200 panoramic images of the test set were identified by 6 oral radiology doctors and students. The diagnosis results by neural convolutional network models and 6 oral radiology doctors and medical students were compared by Chi square test. When P value < 0.05, there was significant difference between them.

Results: the accuracy of neural network model was 0.825-0.875, Chi square test showed that P > 0.05, there was no significant difference in diagnostic accuracy among neural network models. The accuracy of oral radiologists was 69.3%. The average value of the diagnosis results of deep learning neural convolutional network and oral radiologists were tested by Chi square test, P < 0.05. The accuracy of deep learning neural diagnosis network is significantly higher than that of oral radiologists.

Conclusion: the deep learning neural network based on panoramas can make a more accurate differential diagnosis between ameloblastoma and odontogenic keratocyst. The accuracy is greatly improved compared with oral radiologists.