Research On Tumor Classification Methods Based On Sparse Representation

In recent decades,tumor disease has become one of the diseases that most difficult to cure that human faced,which does harm to the human's life and health seriously and it has not found the effective treatment yet.To treat tumor disease,the first thing is to carry out a correct diagnosis of tumor subtypes,which is the so-called tumor classification in terms of bioinformatics.Previously,the traditional method of tumor classification is mainly by using the way of microscopy histopathologic examination and it can get a good result in common tumor diagnosis.However,when there are two or even more tumor subtypes and they have very similar features in tektology,which is very common in the real situation,then the result of the traditional method is barely satisfactory.Therefore,the paper mainly studies the methods of tumor subtypes classification,including the below two points:(1)In view of the shortage of the existing tumor classification methods based on sparse expression showed in the final classification strategy,the paper proposes a new model that considering both the reconstruction power of the dictionary and the discrimination ability of the sparse coding coefficients at the same time,it can further improve the classification accuracy theoretically.Using the dictionary,we can approximate the test samples,and the pre-existing classification methods based on dictionary learning generally use the residual of the reconstruction sample as the final classification decision;Sparse coding coefficients as another constraint,which make the used dictionary“atoms”as few as possible but enough when approximate the testing sample.The model is assessed on six commonly used tumor dataset and a series of experiments show that the proposed new method improve the classification accuracy in most cases.(2)Considering the good classification and feature identification performance of D-KSVD method showed in data processing,this paper considers apply it to the field of tumor classification based on gene expression profile.The essence of D-K.SVD is introduce a liner regression as a punishment on the basis of K-SVD method in order to improve the classification accuracy further.D-KSVD also combines the reconstruction power of the dictionary for the test sample and sparse constraint of the sparse coding coefficients,introducing the classification error term based on the K-SVD method,which means consider the discrimination ability of the dictionary during the dictionary construction in the meantime.We test the classification performance of the method in the area of tumor classification on six datasets,the results show that it does have good classification performance.