Studies Of Automatic Diagnosis Methods Based On Sparse Representation And Contrast-enhanced Ultrasound Images Of Liver Tumors

Liver is an important metabolism organ in the human body, and human’s lifewould be seriously endangered once liver cancer occurs. Therefore, timely andaccurate diagnosis of liver tumors is particularly important for patients. Currently,mainly through the liver perfusion in patients, clinical doctors could judge the stateof liver tumors by observing contrast–enhanced ultrasound images of liver tumorsand then combining morphology characteristics of time-intensity curve inperfusion process with clinical experience. However, there are some limitations incurrent clinical diagnosis method of liver tumors by observing the process of liverperfusion, for example, the result of diagnosis is largely dependent on the doctor’spersonal experience, which could easily lead to misjudgment, in addition, differentexperiences of different doctors may result in great random results, which are notreproducible. To solve this problem, this paper proposes two kinds of automaticdiagnosis methods of liver tumors based on sparse representation and contrast–enhanced ultrasound, and expects to find an automatic diagnosis method to meet theneeds of clinical diagnosis, thus assists physicians to diagnose liver tumors moreintuitively, more conveniently and more accurately in clinical practice.In this paper, many theoretical and experimental studies have been carried outto study automatic diagnostic methods based on CEUS images of liver tumors.Theoretical research: Firstly, a feature extraction method based on sparserepresentation and PCA is proposed and applied in SVM automatic diagnosis ofliver tumors, and then a kind of SVM classification method based on CEUS imagesof liver tumors and the feature extraction method based on sparse representation isproposed, so that this method could optimize the automatic diagnosis method basedon SVM and CEUS images of liver tumors. Secondly, a kind of classificationmethod based on sparse representation and CEUS images of liver tumors isproposed, and then, classification decision-making principle and sparse coefficientreconstruction method, which are more suitable for the CEUS automatic diagnosisof liver tumors are determined through comparative analysis. Finally, the bestautomatic diagnosis method could be confirmed to assist physicians in clinicaldiagnosis of liver tumors.Experimental study: First, qualitative analysis of different characteristics ofTIC of tissues in different areas in CEUS images of liver, such as: the normal areaand the tumor area and the blood vessel, are carried out in order to determine theclassification basis to achieve automatic diagnosis; Secondly, through sampling the TIC in different areas in CEUS images of liver tumors, six groups of TIC featurevector data sets are constructed out respectively as an experimental basis, whichhave different numbers of sampling points, samples and sample distribution;Additionally, the two methods can be respectively used to achieve classification anddiagnosis of liver tumors based on the six sets of experimental data, and theexperimental results show that the SVM classification method based on CEUSimages of liver tumors and the feature extraction method based on sparserepresentation can effectively optimize the results of diagnosis comparing with theSVM automatic diagnosis method, moreover, the classification method based onsparse representation and CEUS images of liver tumors achieves the expected effect,which is able to fully meet the actual needs of the clinical diagnosis of liver tumorsautomatically.