Abstract—This paper proposes a method to classify ultrasound (US) images of normal and fatty human liver using pattern recognition tools. For classification 32 simple novel features, namely, anisotropy features, proposed by authors, are compared with traditional 200 GLCM features. The extracted features are selected by two methods: i) ranking (Welch’s test) and ii) meta heuristic (Particle Swarm Optimisation (PSO)). These selected features are fed into multilayer perceptron (MLP) classifier. It is shown that only 6 anisotropy features, selected by PSO when fed into MLP classifier yield 100% accuracy and the proposed algorithm is much less computational intensive compared to ones found in literature.
Index Terms—anisotropy features, feature selection, particle swarm optimization, fatty liver, ultrasound
Cite: Nivedita Neogi, Arunabha Adhikari, and Madhusudan Roy, "Fatty Liver Identification with Novel Anisotropy Features Selected by PSO," Journal of Image and Graphics, Vol. 6, No. 2, pp. 160-166, December 2018. doi: 10.18178/joig.6.2.160-166
Copyright © 2012-2021 Journal of Image and Graphics, All Rights Reserved