The Ultrasound Image Based Quantitative Evaluation On The Effects Of Radiofrequency Treatment For Liver Cancer

Ultrasound images have the advantages of low-cost, real-time, non-invasive, no-radiation. They can be used as evaluation of the effects of radiofrequency ablation treatment for liver caner. At present, doctors primarily directly observed the changes of the surrounding vascular tissues, size and echo characteristics of RFA region, assisted by such means as biopsy and random visiting to evaluate the effect of radiofrequency ablation treatment. Due to the lack of quantitative standards and rely on doctors' subjective evaluation, misdiagnosis occurs from time to time, and leads to delays in the treatment. In response to the status, researchers used computer-assisted image analysis technology and achieved some success. However, these studies focused on several common liver diseases, particular on classification of fatty liver disease in order to assist doctors to diagnose, but there has not yet any outcome about using digital image processing technology to inspect the recovery process after radiofrequency ablation and giving out the evaluation of the effect, especially the quantitative evaluation.To achieve the quantitative evaluation system, based on the ultrasound images got from animal experiment, Gray Level Co-Occurrence Matrix, Gray Tour Matrix, Complex Curve, Pyramid Wavelet Decomposition and Multi-resolution Fractal Brownian Motion are used to draw the texture features of several kinds of liver ultrasound images, experiment results and analysis are given out. The result shows out the feasibility that a number of image texture analyses can be used in medical ultrasound images.In this paper, Gabor transform is used to extract texture features of the images, and feature-distance of each stage is calculated. Then a quantitative evaluation system is established using the cube of feature-distance to weight the 2-norm of image matrix. To avoid the shortcomings such as non-monotonic changes and non-consecutive numerical range, this paper presents an improved method. In this paper, Texture Feature Coding Method is used to extract the texture feature of images of different stages, and then the texture feature is inputted to train Radial Basis Function Neural Network. Based on pathology test results, the fitting recovery curve that is in monotonous continuous changes over time axis is divided in several numerical ranges, which will describe the status of RFA treatment region numerically and precisely. Test samples are used in this paper to test the 2 sets of quantitative evaluation, and both can reach the accuracy of 90%. So the established quantitative evaluation system has potential application value to help assist doctors more accurately assess the effects of radiofrequency ablation of liver cancer.