A rapid approach for prediction of liver cirrhosis based on first order statistics

In the present work, a liver state index (LSI) based on first order statistics (FOS) is proposed for rapid classification of normal and cirrhotic liver segmented regions of interest (SROIs). Thirty four B-mode ultrasound images taken from 22 normal volunteers and 12 patients suffering from liver cirrhosis were collected from Department of Radiodiagnosis and Imaging, PGIMER, Chandigarh, India. The texture quantification is done by computing statistics based on intensity histogram. Firstly, Six statistical measures namely {average gray level (AGL), standard deviation (SD), smoothness, skewness, uniformity and entropy} are computed for 82 normal SROIs and 39 cirrhotic SROIs taken from 34 B-Mode ultrasound liver images. These 121 (82 + 39) instances of feature vectors are feed to a neural network (NN) classifier and the classification accuracy of 93.3% is obtained by using stratified 10 fold cross validation (10FCV) method. Secondly, correlation based feature selection (CFS) method is used for feature selection and it has been observed that the feature vectors consisting of only 03 features namely, {AGL, SD and uniformity} together with a NN classifier can provide a comparable classification accuracy of 92.5% using 10 FCV method. Since the features selected by CFS method are least correlated to each other and highly correlated to the class label, these features are combined in a way to propose a Liver State Index (LSI) which can be computed on the fly as soon as the radiologist marks the region of interest (ROI) and predict whether the SROI is normal or cirrhotic. The experimental results indicate that the proposed LSI can discriminate between normal and cirrhotic SROIs with 95% classification accuracy.