Diagnosis Of Liver Disease Based On Sparse Representation

Since liver ultrasonic figure is the important basis of liver diseases diagnosis in clinical practice,the figure plays an important role in liver diseases diagnosis.But the figures have some drawbacks like uneven echo,edge blurring and other factors under the ultrasonic environment,the diagnose accuracy of liver diseases can be influenced.Traditional feature extraction algorithm cannot describe the feature of liver diseases well.Besides,traditional recognition methods have lower recognition rate.Aiming at the above problems,algorithms of liver diseases recognition based on sparse representation are deeply discussed in this paper,and some new solutions are put forward.The main aspects are as follows:(1)We studied multi feature extraction algorithms such as Local Binary Pattern(LBP),Histograms of Oriented Gradients(HOG),Gabor Transformation,Gray Level Co-occurrence Matrix(GLCM).We learned about characters of different algorithms and chose algorithms which are suitable to describe features of liver ultrasonic figures.In addition,we used the feature fusion method to fuse the different features and utilized the Random Forest method to select features in which we can get the dimensions to describe figure character well.We used the improved model of Support Vector Machine,and introduced the total variation method to solve it.We can get the most useful features and the most important dimensions simultaneously.Both of the two methods can we not only decrease dimensions to accelerate the computing speed,but also describe figures well to guarantee the higher accuracy.(2)We optimized the performance of dictionary,and designed a dictionary with good represent ability.Using the method based on SGK algorithm to train the dictionary not only has robust stability,but also has higher efficiency.Optimize the dictionary,train the dictionary to get the least error to improve the accuracy.(3)A method combining Support Vector Machine or sparse representation is proposed.Feature vectors or the matrix get through sparse representation algorithm are fed into SVM or use the sparse representation theory to classify.Experimental results show that using sparse representation method can get better result.