Research On The Methods Of Cancer Diagnosis Based On Priori Guidance And Deep Learning Model

With the development of society,malignant tumor disease has gradually threatened the humans' life and health.Gastric cancer and lung cancer are the most concerned malignant tumors.Whether it is gastric cancer or lung cancer,timely detection,timely diagnosis are important for the cancer treatment.For the early diagnosis of gastric cancer,lymph node metastasis can be used as an important indicator to determine the treatment and evaluation of prognosis from medical image and clinical science to lymph nodes and tumors with many other features,and how to analyze these characteristics and implement lymph node transfer of diagnosis will be an important issue that we need to study.For the diagnosis of lung cancer,thoracic CT scan provides the possibility for the early diagnosis of lung cancer.And diagnosis of benign and malignant pulmonary nodules is an important basis for the early diagnosis of lung cancer,based on this,this paper concentrates on marker-based data acquisition strategy and pulmonary nodules benign and malignant diagnosis of two aspects to make research.This article mainly completed the following work.The diagnosis of lymph node metastasis of gastric cancer based on a priori characteristic learning was proposed.Firstly,this study analyzed the clinical data,and the data preprocessing method was presented.On the basis of data preprocessing,a method of diagnosis of lymph node metastasis based on a priori feature learning was proposed.This method firstly analyzes the correlation coefficient between lymph node features,tumor features and patient characteristics.for the relevant feature groups,the feature selection is based on the prior sparse model,then the selected feature subsets are combined.Finally the characteristics of sparse model were selected,and the lymph node classification was diagnosed by the selected characteristics.The experimental results were compared and analyzed.This feature optimizes the selection of features that are effective for the diagnosis of lymph nodes,which reduces the number of features,both for diagnostic and medical examinations and the second is to improve the diagnostic accuracy based on the removal of redundant features and useless features rate.To obtain benign and malignant data of lung nodules,this paper use the method based on nodular geometry window to obtain nodular data.Firstly,based on the pulmonary nodule image database,the radiographic method was used to obtain the labeled region of each node of the radiologist.On this basis,the marker area of the four radiologists and the method based on TPM were used.Aiming at the number of different nodules,for less misdiagnosis rate and improve work efficiency,a best method of localization of pulmonary nodules based on nodular geometry window is proposed First,all sections of all nodules were statistically determined to determine the size of the nodal data acquisition window,then based on the nodal window size method to locate the required nodal data frame,and access to pulmonary nodules data,to obtain following diagnosis of pulmonary nodules benign experimental data.For benign and malignant diagnosis of pulmonary nodules,this paper presents a method of benign and malignant pulmonary nodules diagnosis based on curvilinear convolution neural network.Basing on the essential characteristics of pulmonary nodules analysis,the base that benign and malignant differences on pulmonary nodules is its edge burr and other internal texture structures.First,through the analysis of benign and malignant nodules to determine the reasonable curve wave scale transformation and angle transformation interval,generate a curve wave redundancy dictionary,and then use the curve wave redundancy dictionary to initialize the convolution of the neural network filter,After initializing the network filter,the network is constructed with the initial depth learning model,and the training set of benign and malignant nodes is trained by the deep convolution network.Finally,this study needs to diagnose the test data for classification diagnosis.