Abstracts Submitted: 510
Number of Users: 710
View Abstracts Submitted
Back to home Page
Liver is the most vital organ of the human body which carries out over 500 essential tasks including protein synthesis, chemical production for food digestion, detoxification etc. Liver diseases can be caused by various factors like alcohol use or viruses. The most common liver disorder encountered nowadays is Fatty Liver Disease (FLD) caused due to the accumulation of excessive fats. The progression of FLD leads to the swelling of liver which in its extreme form is the major cause of liver cirrhosis. Cirrhosis is an irreversible liver condition characterized by the scarring and degeneration of liver tissue. Biopsy is considered to be the gold standard for evaluating the severity of liver diseases however certain non-invasive imaging modalities like Magnetic Resonance Imaging (MRI), Computerized Tomography (CT), ultrasound (US) can be used for the diagnosis. Ultrasound being the least expensive modality has been widely used for detecting the liver diseases. The ultrasound images contain speckle noise which may obscure certain diagnostically important details of the image thus resulting in poor diagnosis of the abnormalities. To overcome this limitation, computer assisted systems were developed for the identification and characterization of liver diseases. In the present re-search work, a Computer Aided Diagnostic (CAD) model has been proposed to differentiate between fatty and cirrhotic liver based on the textural patterns exhibited by them on ultrasound images. From each ultrasound image multiple non-overlapping ROIs of size 40 × 40 pixels have been extracted. From the extracted ROIs, different statistical features have been extracted namely, First Order Statistics (FOS), Gray Level Co-occurrence Matrix (GLCM) features and Gray Level Run Length Matrix (GLRLM) features. The extracted features have been subjected to different classification models namely Decision Tree (DT), Random Forest (RF), Linear Model (LM), Support Vector Machine (SVM) and Neural Network (NN) for the differentiation between fatty and cirrhotic liver. The performance of each classification model has been evaluated on the basis of the classification accuracy obtained for five different training-testing partitions (50/50, 60/40, 70/30, 80/20 and 90/10). Result shows the highest accuracy of 84.37 % using the GLCM feature vector and SVM classification model at 90/10 data partition. Also, the results obtained indicate the usefulness of the proposed CAD model in routine clinical practice as an assistance tool for the radiologists to efficiently discriminate between the fatty liver images and cirrhotic liver images.
© Copyright 2017 All Rights Reserved