Image Analysis for the Detection and Diagnosis of Hepatocellular Carcinoma from Abdominal CT Images

Abstract

This paper describes the application of machine learning algorithms in detecting the liver lesion from abdominal CT images. Liver lesions are usually detected by different imaging modalities such as computed tomography (CT), ultrasonography (USG), or Magnetic resonance imaging (MRI). These lesions can be of benign or malignant in nature, both of which present similar appearance in the images. The segregation of the liver lesions into benign or malignant has been done with painful invasive techniques by the medical practitioners which create considerable difficulty to the patients. In this work the malignant liver lesion namely the hepatocellular carcinoma (HCC) is differentially identified from normal liver tissue with the help of supervised classification algorithms. The grainy nature of liver parenchyma is the prime factor in making the texture features for discriminating it from other tissues. Different textures features have been extracted from the Region of Interest (ROI) identified by an expert radiologist and are applied as an input to the classifiers for training and testing. The support vector machine (SVM) and naive Bayes’ classifier are supervised learning techniques. Thus our proposed Computer Aided Diagnostic (CAD) system can be easily operated by a medical practitioner in differentially detecting the HCC tissue from normal liver tissue with best classification accuracy which is more than 90%. Hence the painful invasive methods can be avoided to a great extent. The performance evaluation is done by calculating the accuracy, confusion matrix, sensitivity, and specificity. The receiver operating characteristics (ROC) also have been plotted.