Study On Liver Ultrasound Image Lesion Segmentation And Recognition

Computer-aided diagnosis system has merged as a "second reader" for analyzing medical images using computational algorithms. And the radiologist can take the CAD outcome as a second opinion and make a conclusive diagnosis. Therefore, study on medical image processing is of great significance. In this paper, we focus on liver ultrasound image segmentation and recognition. The main work is as follows:(1) A modified maximally stable extremal region (MSER) method is proposed for the segmentation of ultrasound liver images. Firstly, the feature regions including liver lesions are extracted using the modified MSER detector. Next, the edges of the liver lesions are detected from the binary images and a merging process is designed to refine the contour of the liver lesion. The last step is the segmentation of the liver lesion according to the refined contour. Unlike the MSER algorithm, the improved MSER detector merely needs dozens of gray levels rather than 256 possible gray levels ranging from 0 to 255. The segmentation results of ultrasound liver images demonstrate that there is a significant correlation between the liver lesions selected by a medical expert and the liver lesions segmented by the proposed method. A comparison of the proposed method and other segmented methods shows that the proposed method can detect a more accurate contour of liver lesion images.(2) We investigated the behavior of 22 co-occurrence statistics and 11 run-length statistics combined with multi-scale sizes of region of interest (ROI) to classify liver ultrasound images. Experimental results show that the classification accuracy rate is up to 95.86% for the ROI size of 90×90 pixels when 26 best feature subsets are applied.