Research On Deep Learning Based Image Recognition Technology For Dentistry

Oral health has received more and more attention in recent years.Oral leukoplakia,oral lichen planus,oral cancer and recurrent oral ulcers are typical oral mucosal diseases.Oral leukoplakia and oral lichen planus are precancerous lesions which have the possibility of being transformed into oral cancer as well as the recurrent oral ulcers which become malignant if they are not cured for a long time,so that it is such important to have a timely diagnosis for these diseases.At present,the identification of oral mucosal diseases mainly relies on the subjective judgment of doctors’ clinical experience,which leads to low accuracy of disease identification rate and high workload of doctors.To solve the above problems,this paper aims to study the deep learning-based image recognition technology for oral medicine.The researches in this paper are as follows.Firstly,to address the problems of small number of oral disease data sets and low recognition accuracy,we use various data enhancement algorithms for data augmentation and propose an image recognition algorithm for oral disease based on migration learning and Efficient Net.The structure of the network is based on the Efficient Net B0 network;the Focal Loss is used as the loss function,the model undergoes migration learning twice,migrating the parameters trained on Image Net to the ISIC2018 skin dataset first,saving the parameters after the training is completed,and then migrating to the oral dataset;optimizing the fine-tuning strategy,using Adam combined with SGD optimization algorithm together to fine-tune the network model,so that the overall stability and generalization ability of the network is improved and the recognition accuracy of the network is increased.Secondly,an oral mucosal disease recognition method based on multi-level feature fusion is proposed to further improve the recognition accuracy of the network model.The Efficient Net model is used to make the extraction of deep-level features,and HSV,histogram of oriented gradient（HOG）and gray-level co-occurrence matrix（GLCM）are used to extract the color,shape and the shallowfeatures texture of oral diseases,respectively.The features with greater relationship to the target value are selected by random forest（RF）algorithm to reduce the dimensionality of the features and avoid overfitting of the model,and the traditional classifier such as support vector machine（SVM）is used for final classification and recognition,which can improve the recognition of oral diseases.accuracy.Finally,the proposed deep learning algorithm is combined with the actual diagnosis scenario of oral mucosal diseases,and the visualization GUI interface is developed based on Py Qt and Opencv to combine the trained convolutional neural network model with the interface.According to the specific needs of doctors in diagnosing,the fully functional oral mucosal disease diagnosis software is designed.The software includes a disease diagnosis function to diagnose the input oral disease images,an image cropping function to extract the region of interest（ROI）of oral mucosal disease images to improve the accuracy of diagnosis,an image zooming and panning function to facilitate the dentist to view the details of the images,an image contrast function to make the color change of the images more obvious and facilitate the doctor to locate the location of the lesions.The research results show that the proposed algorithm can effectively solve the problems of many false diagnoses and low accuracy rate in oral disease recognition.Meanwhile,the proposed method can achieve accuracy 92.89%,sensitivity 89.91%,specificity 96.06% and AUC value0.9809,respectively.The developed software for oral mucosal diseases can assist doctors in diagnosis and quickly locate the location of oral lesions,which improves the efficiency and accuracy of doctors,accelerates the treatment progress of patients and reduces the economic loss caused by diseases.