Prediction of Liver Cirrhosis Using Weighted Fisher Discriminant Ratio Algorithm Simarjot Kaur Randhawa Department of Electronics and Communication Engineering Dr. B.R. Ambedkar National Institute of Technology Jalandhar, India simar.randhawa@gmail.com Ramesh Kumar Sunkaria Department of Electronics and Communication Engineering Dr. B.R. Ambedkar National Institute of Technology Jalandhar, India sunkariark@gmail.com Anterpreet Kaur Bedi Department of Electronics and Communication Engineering Dr. B.R. Ambedkar National Institute of Technology Jalandhar, India anterpreetbedi27@gmail.com Abstract—Liver diseases are one of the significant public health issues in present scenario. Liver cirrhosis is one of the leading causes of deaths due to liver related diseases. In this work, classification of normal and cirrhotic liver is done using texture analysis. Various texture features are extracted for classification. Out of all the extracted features, seven best features are identified using fisher discriminant ratio. Further, weighted fisher discriminant ratio algorithm is designed using the selected features, such that maximum accuracy and sensitivity is achieved. Keywords—ultrasound, cirrhosis, feature extraction, classification, fisher discriminant ratio I. INTRODUCTION Medical imaging uses different techniques such as CT, ultrasound, and MRI for clinical diagnosis. Each technique has its own advantages and disadvantages. Ultrasound imaging is one of the popular approaches for imaging organs and soft tissues of our body as it is cost effective, non- invasive, portable and no harmful radiations are used. Ultrasound imaging uses high-frequency acoustic waves to have a view of internal body organs or tissues such as heart and blood vessels, liver, kidney, gallbladder, spleen, pancreas, uterus etc.It is widely for diagnosis of diseases as it is non-invasive, non-radioactive and economical. Liver is a vital organ of human body as it performs significant functions like excretion of bile, storage of minerals, breakdown of insulin etc. According to World Health Organization data published in April 2011, 2.31% of total deaths in India are caused by liver related diseases. Liver diseases are broadly classified as focal and diffused diseases. In diffused liver diseases, the abnormality is present throughout the liver tissue whereas in focal liver diseases, the abnormality is concentrated at single or multiple sites. Among diffused liver diseases, the liver cirrhosis is considered serious and is often caused by hepatitis and alcoholism. In cirrhosis, scarring of liver tissue occurs and the blood flow to the liver slows down. Variation of liver size and shape is observed depending upon severity of the liver cirrhosis. Cirrhosis mostly affects right lobe of the liver [1].Many techniques and classifiers have been employed by various researchers for classification of liver cirrhosis [2- 5].In this study, normal and cirrhotic liver are distinguished using various texture features and weighted fisher discriminant classifier. II. PREVIOUS WORK Many researchers have proposed computer aided systems for prediction of cirrhotic liver from normal liver as well as other conditions of liver. Classification of grading of liver cirrhosis has also been done by some researchers [6-7]. Lu et al. classified cirrhotic liver from normal liver by comparing their echo texture with echo texture of corresponding spleen [2]. Wan and Zhou calculated mean and energy from sub- bands of 2D wavelet packet transform (WPT) and 2D discrete wavelet transform (DWT) using db4 wavelet and classified normal and cirrhotic liver using SVM classifier [3].Virmani et al. calculated 75 laws’ texture features out of which 8 were selected using correlation and classification between cirrhotic and normal liver was done by using neural network and SVM classifiers [4].Lee et al. classified normal liver, cirrhotic liver, hepatoma by extracting M-band wavelet transform features and fractal features using proposed technique [8]. Wu et al. classified normal liver, cirrhotic liver and hepatoma by feature fusion (GLCM, wavelet transform, gabor wavelet transform) and used kNN, fuzzy kNN, SVM and probabilistic neural network (PNN) classifiers [9]. Virmani et al. proposed a classification system to distinguish normal, cirrhotic liver and cirrhosis evolved over hepatocellular carcinoma using multi resolution wavelet packet texture features and SVM classifier[10]. Wu et al. proposed heirarchial feature fusion system which used evolution-based algorithms for selecting dominant features for classification of normal liver, cirrhotic liver and hepatoma using kNN, fuzzy kNN and PNN classifiers [11]. Lee et al. proposed ensemble of classifiers for classification of liver tissue into normal, cirrhotic and hepatoma which used multi resolution analysis for extracting features from the images[12]. Kalyan et al. classified fatty liver, cirrhosis and hepatomegaly by extracting various texture features and compared their performance using mutilayer perceptron neural network classifier [13]. Virmani et al. distinguished cirrhotic liver from normal liver by calculating mean and standard deviation from sub-bands of 2D DWT, 2D WPT and 2D gabor wavelet transform (GWT) and then using SVM for classification [5].Acharya et al. proposed a system for automatic diagnosis of cirrhosis, fatty liver and normal liver using non-linear featues based on curvelet transform. They also proposed liver disease index for discrimination of the three classes of liver [14]. III. METHODOLOGY As we know, ultrasound is a non-invasive, cost effective imaging technique which doesn’t use harmful radiations. In this study, classification of normal and cirrhotic liver has been done using ultrasound images. The various stages for liver classification are: data collection, feature extraction, feature reduction and classification. A. Database Publically available database of ultrasound images have been used in this study [15, 16]. In total, 45 images are used out of which 25 are normal liver images and 20 are cirrhotic 2018 First International Conference on Secure Cyber Computing and Communication (ICSCCC) 184 978-1-5386-6373-8/18/$31.00 ©2018 IEEE