The Prediction Of Cancer Based On Dictionary Learning And Sparse Representation Classification

Cancer is one of the serious threats to human health.The 2017 International Gastric Cancer Congress pointed out that the incidence of gastric cancer in China has accounted for more than 40% of the global incidence,and most of them have developed into advanced gastric cancer,the patient's 5-year survival rate is less than 30%.The lung cancer is the first killer of many cancers,in our country and the world,its incidence and mortality of malignant tumors are the highest.Therefore,the early diagnosis of cancer,will play a vital role on treatment programs,it can improve the prognosis of cancer patients.In the clinical diagnosis,the metastatic state of lymph nodes define the stage of gastric cancer,determine whether the tumor has metastasized,pulmonary nodules of benign and malignant diagnosis is the important basis of stage judgment of lung cancer,based on this The main work is as follows::(1)We propose a method for predicting gastric cancer based on K-SVD dictionary learning and sparse representation classification.For the different stages of gastric cancer patients,we classify gastric cancer patients,so we convert the prediction of gastric cancer staging into classification problem Gastric cancer data are not sparse,and in order to approximate the sparse domain of the sample,the K-SVD dictionary learning method is used to study the sample characteristics.Before learning the training samples,we divided the training samples into non-transferred samples and transferred samples,according to the gastric cancer staging of the patients.By learning the two types of samples,we obtained two dictionaries that could represent two types of patients.The atoms in the dictionary represent the test sample data,and the classification of gastric cancer is predicted according to the principle of minimum error.In this way,the sample data of gastric cancer patients through the dictionary learning,achieve the characteristics the transformation,in the new feature space,gastric cancer staging prediction achieve better predictive effect than the original space.(2)Aiming at the determination of pulmonary nodule(PN)in CT images of lungs,we propose a method for obtaining adaptive window of pulmonary nodules.On this basis,we propose a method based on K-SVD dictionary learning and sparse representation Classification for determining the best scale window of pulmonary nodules.First,according to the radiologist's mark,to obtain a single nodule region,then,consider the mark of four radiologists in the same slice,get the comprehensive pulmonary nodule region in the same slice;and then,with the proposed method,obtain the adaptive window of the pulmonary nodule,these window sizes are not exactly same.Then,obtain the samples of pulmonary nodules under different scale windows,and the best scale window of pulmonary nodule was obtained validated by K-SVD dictionary learning.(3)In the best scale window obtained in Chapter 3,obtain the pulmonary nodules data,we propose a qualitative diagnosis method of pulmonary nodules based on ridge wave hyper-complete redundancy dictionary and sparse representation classification.Because the dimension of dictionary obtained by the K-SVD dictionary is fixed,although it is redundant,but not super complete,the dictionary obtained based on Ridgelet is not only redundant,but also super complete,while the direction of the pulmonary nodules and high-dimensional singularity approximation.First,construct the over-complete redundancy dictionary based on Ridgelet;then,classify the training samples,search the atoms that can represent benign and malignant samples in the super-complete redundancy dictionary,compose benign and malignant sub-dictionary;Finally,the atoms in the sub-dictionary represent the test samples,predict the diagnosis of pulmonary nodules based on the principle of minimum error.The experimental results show that the diagnosis effect based on Ridgelet over-complete redundancy dictionary and sparse representation classification is better than that of K-SVD dictionary learning.At the same time,we compare result of the different values of the parameters L and find that the diagnosis of pulmonary nodules is the best when the value of L is 8.