Computer Assistant Diagnosis Based On Texture Features Of Ultrasonic Liver Images

Conventionally, the diagnosis of B-scan liver images is done by doctors directly observing and analyzing images, then judging the station of liver. This diagnostic method depends on doctor’s experience a lot. Besides, when the doctor can’t make a conclusion, there is lack of a objective reference. Computer assistant diagnose is a technology that uses computer to analyze, learn and classify the images derived from medical imaging, providing a diagnostic reference for doctor. The purpose of this article is to design a computer assistant diagnosing (CAD) system for B-scan images, which mainly diagnoses fatty and cirrhosis.To choose the suspected region of the B-scan images, this article makes the part of manual choosing region of interest (ROI), and proposes some requirements for choosing ROI, based on the distribution of liver. Because of the imaging mechanism, there are noises in the B-scan images, mainly speckle noise. This article supplies median filter, wiener filter and wavelet threshold de-noising methods, and chooses the best one to de-noise the ROI, by making contrast of the performance of these methods. To enhance the texture difference, contrast gray-level stretch, sharp, and enhancement based on low pass filter, and choose the best one to enhance the ROI. This article chooses the methods of GLCM, CLDS and Fourier power spectrum to extract texture features, and combines those features as a feature group. Then train the feature group by the method of neural networks to build the recognizing system, and finally accomplish the design of the CAD for B-scan liver images. Put the samples used for testing into the CAD, which leads to a good recognition rate.