Medical Image Classification with Hand-Designed or Machine-Designed Texture Descriptors: A Performance Evaluation

No Thumbnail Available

Date

Journal Title

Journal ISSN

Volume Title

Publisher

Springer Verlag

Abstract

Description

Accurate diagnosis and early detection of various disease conditions are key to improving living conditions in any community. The existing framework for medical image classification depends largely on advanced digital image processing and machine (deep) learning techniques for significant improvement. In this paper, the performance of traditional hand-designed texture descriptors within the image-based learning paradigm is evaluated in comparison with machine-designed descriptors (extracted from pre-trained Convolution Neural Networks). Performance is evaluated, with respect to speed, accuracy and storage requirements, based on four popular medical image datasets. The experiments reveal an increased accuracy with machine-designed descriptors in most cases, though at a higher computational cost. It is therefore necessary to consider other parameters for tradeoff depending on the application being considered. © Springer International Publishing AG, part of Springer Nature 2018.

Keywords

QA75 Electronic computers. Computer science

Citation

Collections

Endorsement

Review

Supplemented By

Referenced By