Deep Learning Based Diagnosis Of Oral And Maxillofacial Surgery Diseases

With the rapid development of information technology,a large amount of data has been accumulated in various fields.In the medical field,more and more datasets have been collected,especially medical image datasets.In recent years,deep learning has made great breakthrough in the field of natural image processing and achieved good results in many tasks.Therefore,more and more researchers have applied deep learning to medical image processing.However,most existing medical image research still focuses on two-dimensional image research,and most of the studies concentrate on the human internal organs and eye area.In view of the challenge brought to doctors by three-dimensional oral Cone Beam Computed Tomography(CBCT)data,this thesis studies the diagnosis of oral and maxillofacial surgery diseases based on deep learning.To the best of our knowledge,this is the first work to apply deep learning to oral and maxillofacial surgery diseases.The diagnosis consists of three aspects:patient classification,lesion segmentation and tooth segmentation.For patient classification,a deep learning based algorithm called Deep Diagno-sis of Oral and Maxillofacial diseases(DDOM)for disease classification of oral and maxillofacial surgery is proposed.Different from existing algorithms,which can only classify 2d images,DDOM can process 3d oral CBCT data of patients and classify them at the patient level.The results from a real dataset of 2500 oral CBCT data show that the proposed algorithm performs better than most professional physicians.For lesion segmentation,a semi-supervised oral lesion segmentation algorithm called Adversarial Synergistic Network(ASNet)is proposed.Unlike existing algo-rithms that can only use labeled data or utilize unlabeled data through self-training methods,ASNet can utilize unlabeled data through co-training and adversarial learn-ing.Results from a real dataset of 691 oral CBCT images show that the proposed algorithm outperforms the other algorithms.For tooth segmentation,a tooth segmentation algorithm based on deep learning called Tooth Segmentation Network(TSNet)is proposed.Compared with existing tooth segmentation algorithms,TSNet uses convolutional neural network and adver-sarial network.Results from a real dataset of 100 oral CBCT images show that the proposed algorithm achieves the best performance.While completing this thesis,we collected and labeled a dataset containing 2500 patients together with the professional doctors in the hospital.The dataset includes millions of CBCT images.As far as we know,there is no such public dataset in the world.At the same time,the research problems in this thesis are formed in the process of deep discussion with professional doctors,and come from the real needs of profes-sional doctors.The three tasks in this thesis can assist doctors comprehensively in the three aspects of patient diagnosis,lesion location and surgical planning.