Enhancing Local Binary Patterns for higher accuracy in Fatty Liver classification using Deep Learning

Abstract

In deep learning, local binary patterns (LBP) are inefficient for the textural feature-based classification of the fatty liver because they lose some of the relevant features. The purpose of this study is to enhance classification accuracy. We analyze accuracy and processing time. The proposed system con-sists of a convolutional neural network with curvelet local binary pattern for feature extraction which improves accuracy and can also now determine the size of the fatty liver. Accuracy is measured using probability scores and processing time is measured with total execution time, using sample image groups from CT/MRI images. Results shows that the proposed solution has improved the classification accuracy to 98% from 94% on average and reduced the processing time to 0.313 seconds compared to the existing 0.561 seconds. Moreover, the proposed system has added a volume feature, a, green border represents the volume of the fatty liver. Overall, the proposed system has improving accuracy and processing time required for fatty liver detection whilst leaving desirable features of the best current solution intact.