Nonalcoholic Fatty Liver Texture Characterization based on Transfer Deep Scattering Convolution Network and Ensemble Subspace KNN classifier

Bharath, R and Rajalakshmi, P (2019) Nonalcoholic Fatty Liver Texture Characterization based on Transfer Deep Scattering Convolution Network and Ensemble Subspace KNN classifier. In: URSI Asia-Pacific Radio Science Conference, AP-RASC, 9 -15 March 2019, New Delhi, India.

Full text not available from this repository. (Request a copy)

Abstract

Nonalcoholic Fatty Liver Disease (NAFLD) is highly prevalent and may progress to chronic diseases if left untreated. Early detection and diagnosis are crucial to prevent the complications associated with NAFLD. Fatty liver diagnosis is widely done through ultrasound scanning. Based on the density of fat, the liver is classified into four categories. The ultrasonic texture characteristics of liver parenchyma vary with the concentration of fat, and hence the radiographers use this as a property to classify the fatty liver. Classifying the nonalcoholic fatty liver is highly challenging to the radiographers due to the minute variations observed in the characteristics of the texture. To assist the radiographers in doing accurate diagnosis, we propose a novel computer-assisted novel algorithm based on compressed transfer scattering coefficients and ensemble subspace KNN classifier. The proposed algorithm classified the texture with an accuracy of 98.8% when tested on a data size of 1000 images, where each category consists of 250 images each.

[error in script]
IITH Creators:
IITH CreatorsORCiD
Rajalakshmi, PUNSPECIFIED
Item Type: Conference or Workshop Item (Paper)
Subjects: Electrical Engineering
Divisions: Department of Electrical Engineering
Depositing User: Team Library
Date Deposited: 22 Jul 2019 10:14
Last Modified: 22 Jul 2019 10:14
URI: http://raiith.iith.ac.in/id/eprint/5776
Publisher URL: http://doi.org/10.23919/URSIAP-RASC.2019.8738717
Related URLs:

Actions (login required)

View Item View Item
Statistics for RAIITH ePrint 5776 Statistics for this ePrint Item