Study On Imaging Of Oral Tissues Based On Optical Coherence Tomography,and Their Classification And Identification

Oral cavity is located in the key position of the human respiratory tract and digestive tract.About 90% of oral cancer is squamous cell carcinoma,and its 5-year survival rate is still less than 50%.Early detection and accurate surgical resection of oral cancer are important means for improving the survival rate and prognosis of patients.However,conventional screening methods,such as visual observation and palpation,depend on surgeons’ clinical experience,which has a possibility of misdiagnosis.As the gold standard for the diagnosis of oral cancer,histopathological examination is timeconsuming and not suitable for some high-risk patients.The existing clinical imaging technologies,such as magnetic resonance imaging,X-ray computed tomography,and ultrasound,have low resolution,so they are unable to carry out an early diagnosis for small lesions.Since it was proposed in the 1990 s,optical coherence tomography(OCT)has been developed rapidly because of its advantages of high resolution,non-invasion,and realtime imaging.In order to address the above issues of the early screening,diagnosis,and intraoperative resection of oral cancer,we used a homemade swept-source OCT system to study on imaging of oral tissues,classification,and identification in this dissertation.The specific research contents are as follows:1.OCT imaging of different oral tissues was performed and OCT images were compared with the corresponding histopathological images.Aiming at the difficulties of early screening of oral cancer and the problems of existing auxiliary imaging methods in clinical diagnosis,the application of OCT in oral disease imaging was studied.The homemade swept-source OCT system was used to image the intraoperative resected tissues including 19 types of oral diseases.By comparing with the histopathological images,OCT image features were described and analyzed.The morphological features of OCT images of different diseases were determined,and the OCT image library of oral diseases was established.This study verified the feasibility of applying OCT to the screening and auxiliary diagnosis of oral tissue diseases.2.The research on OCT image recognition and edge detection based on a quantitative model was carried out.In order to assist clinicians in determining the oral lesion area and accurate lesion resection during surgery,a quantitative analysis method based on the optical attenuation model was proposed.By extracting the attenuation coefficient of an optical signal along the depth direction in the OCT image,the attenuation coefficient threshold for distinguishing oral cancer and normal tissue was determined,which realized the identification of oral cancer tissue with an accuracy of91.15%.Also,tumor boundary was visualized in three-dimensional OCT images.In addition,through the identification of oral salivary gland tumors,the effectiveness of the quantitative model was verified,and it was proved that the quantitative model is feasible to assist clinicians in boundary detection and visualization in oral cancer resection.3.The research on the classification and identification of oral cancer based on the texture features in OCT images was carried out.Aiming at the problem that there are obstacles for clinicians to use OCT images as auxiliary media for oral cancer identification and clinical decision-making without prior knowledge,an automatic classification and identification algorithm of OCT images of oral cancer tissues based on texture features was studied.The high classification accuracy of distinguishing oral cancer,precancerous lesions,and normal tissues was realized.The impact of different texture features on the classification accuracy was also evaluated.It was verified that the effectiveness of the algorithms for automatic classification and identification of OCT images of different oral tissues and proved the feasibility of using the machine learning methods based on texture features to judge the pathology of medical staff in the process of oral cancer screening and diagnosis.4.The research on the classification and identification of OCT images of oral cancer based on deep learning was carried out.When the amount of images is large,the deep learning methods can play a great role in classification and identification.OCT image recognition of oral cancer based on deep learning was carried out and the visualization of the deep learning network in oral lesions recognition was studied.The performance of the neural networks and machine learning classifiers was evaluated.The high identification accuracy of OCT images of oral cancer was realized.The internal basis of deep learning network recognition of oral cancer was further visualized through feature aggregation.It was verified that the effectiveness of the deep learning algorithm in identifying OCT images of oral cancer and proved that the application value of the combination of OCT and artificial intelligence to solve specific oral clinical problems.