Characterization of Benign and Malignant Liver lesions using Ultrasonography, and Texture Analysis

Abstract

This is a descriptive analytic study of characterization of benign and malignant liver lesion using ultrasonography and texture analysis, aimed to characterize liver lesions such as cysts, hydated cyst, hepatocelullar carcinoma, and metastases using ultrasonography and to differentiate between them along with normal liver in ultrasound images by texture analysis and classification methods, using image processing program,( interactive data language IDL), by created an images in a tiff format in order to used as IDL variables. Three features from FOS which were mean, entropy and energy computed from the intensity function of the image and were calculated for each region of interest through all images using 3x3 window size and data prepared for SW-LD analysis, and linear discriminate analysis. The study carried out during the period from 2016 to 2019 in (Ibn Siena Teaching Hospital and DR Osman Abdalwahab private clinic) in Khartoum, Sudan. A total of 260 patients, 156 were male and 104 were females, whom aged between 4 to 90 years old known case of liver lesions were performed . All patients examined by ultrasound scanning using ultrasound machine Xario Diagnostic system with 3.5 MHz curve-linear probe. The result showed that the most common liver lesions diagnosed by ultrasound were 112(43%)cases as metastases ,followed by 49(18.85%)cases as HCC , 43(16.5%)cases as haemangioma. while the others were, 22(8.5%), 16 (6.2%) , 12(4.6%),3(1.2%),2(0.8%) and 1(0.4%)of cases as cyst, abscesses, hydated cyst, lymphoma, adenoma, and intrahepatic choliangiocarcinoma respectively. The majority of liver lesions were appear hypo echoic in metastases followed by hyper echoic in haemangioma, and heterogeneous echo pattern in hepatocelullar carcinoma. For texture analysis, the study showed resulted there was a well concentration of features around the classes centre which gave a remarkable difference among the VIII classes with classification accuracy of each classes in which 92.2% of normal liver was correctly classified and 75.6%, 81.4%, 100.0%and 100.0%, of classification sensitivity for liver cyst, HCC, Hydated Cyst, and Liver metastases respectively with highest predictive overall classification accuracy of 89.1%. the study recommended that if there is indication of liver disorder by ultrasound screening ,the patents image must be analyzed by IDL program ,then the histopathology will be done only for confirmation , if it is not available, no need for it. Because the prediction of ultrasound and texture analysis tools together is very high The study concluded that the ultrasound is considered as first line of liver lesions diagnosis. In addition, the study concluded that the computer aided diagnostic system increased the accuracy of liver lesions classification. Texture analysis serves as second methods to perform more characterization of the lesion.