Loading...

Summer Research Fellowship Programme of India's Science Academies

Classification of non-alcoholic fatty liver disease into grades using CNN

Pranitha Peddi

VNR VJIET, Vignana Jyothi Nagar, Pragathi Nagar, Nizampet(S.O), Hyderabad, TS, India 500090

Mr Uday. B. Desai

Emeritus Professor, Department of Electrical Engineering, IIT Hyderabad, Hyderabad, India

Abstract

The problem that is aimed to be solved in this project revolves around a disease called "Nonalcoholic Fatty Liver Disease". This is a fairly common disease especially in the countries where health care is struggling to modernize and get equipped with technology, i.e., developing countries. This disease is known to get chronic when timely treatment is not provided. The above-mentioned liver disease is categorized into four grades namely–Grade 1, Grade2, Grade3 and Normal, with Grade3 being the most serious of all. This classification is done on the basis of the density of fat in the liver. This difference in density of fat, in turn, gives liver parenchyma its distinctive texture which when studied properly can help the doctors in initial diagnosis. The main complication that is to be addressed is the minute differences in the above-said texture of liver parenchyma. Although there is no doubting that the highly skilled clinicians can differentiate the textures, there is a high chance it is not always as accurate as we want it to be. The solution should also be computerised in order to reduce human error and related complications as wrong classification can lead to improper treatment which only makes the disease worse, instead of being treated in the desired way. Hence to improve the classification accuracy in diagnosing the fatty liver, we propose a convolution neural network-based computer-aided diagnosis algorithm for categorizing the ultrasound liver parenchyma texture into four classes which also involves transfer learning. The proposed algorithm is analyzed using 1000 texture images comprising of 250 images belonging to each class. Performance analysis shows that the proposed framework classifies the texture with an accuracy of 90.0% when 80% and 20% of data used for training and testing respectively.

Keywords: Grade1, Grade2, Grade3, Normal, Covolutional Neural Network, Transfer Learning

More
Written, reviewed, revised, proofed and published with