Nasopharyngeal Carcinoma Classification Diagnosis And Distant Prediction Based On Deep Learning

Nasopharyngeal carcinoma is a malignant tumor with a high mortality rate in China and it is of great significance to achieve its rapid diagnosis.It is important for the patients' health that the early diagnosis of nasopharyngeal carcinoma and the occurrence of distant metastasis after treatment.The diagnosis of nasopharyngeal carcinoma is usually diagnosed according to the results of pathological images.The analysis of pathological image is accurate so it has high authority,but the amount of data is large and it takes a lot of manpower and material resources to analyze it.In recent years,the rapid development of deep learning has made the diagnosis of nasopharyngeal carcinoma relatively simple in pathology.In the paper,the pathological image data of nasopharyngeal carcinoma labeled by doctors are taken as the research object,and uses the method that based on the deep convolution neural network to make the diagnosis and analysis of the nasopharyngeal carcinoma,and identifies the region of the cancer in the pathological data image.According to the above algorithm,the paper uses it to label some data of distant metastasis of nasopharyngeal carcinoma,and merging data with doctors'annotation as dataset to predict the distant metastasis of nasopharyngeal carcinoma after treatment.The experimental results show that the algorithm has high academic significance and use value.The innovative research work of this paper mainly includes the following two aspects.A model based on deep convolution neural network for automatic detection and diagnosis of pathological regions of nasopharyngeal carcinoma is constructed.The pathological image data features of different resolution are extracted by VGG-16 and Inception-V3 model,and the features are fused at the end of network through the fusion model so that the cancer area is identified and labeled.In view of the distant metastasis of nasopharyngeal carcinoma,the data set is labeled on the basis of the above experiments.In the process of data processing,transfer learning is used to extract features.Combine it with traditional machine learning methods to analyze and determine whether patients have distant metastasis.