Multiclass histology image retrieval, classification using Riesz transform and local binary pattern features. Issue 6 (1st November 2020)

• Record Type: Journal Article
• Authors: Yazdi, Mehran; Erfankhah, Hamed
• Is Part Of: Computer methods in biomechanics and biomedical engineering. Volume 8:Issue 6(2020)
• Page Start: 595
• Page End: 607
• Publication Date: 2020-11-01
• Subjects: Histology -- monogenic signal -- local features -- content based image retrieval -- MLBP -- Riesz transform; Imaging systems in biology -- Periodicals; Imaging systems in medicine -- Periodicals; Biomechanics -- Data processing -- Periodicals; Biomedical engineering -- Periodicals; 616.0757
• DOI: 10.1080/21681163.2020.1761885
• Languages: English
• ISSNs: 2168-1163

Abstract

ABSTRACT: Structure and distribution of cells in histological images cause considerable variability among the tissue patterns and make the retrieving and classification as challenging tasks for this type of medical images. In this paper, we propose to extract four local features, namely local energetic information, local structural information, local geometric information, and local patterns to represent the huge texture variability of the histological images in a feature space. For extracting these features, the Riesz transform and Monogenic local binary patterns are used. Riesz transform is an extension of the Hilbert transform for 2-D signals. It provides a monogenic representation for an image where we can capture some local structure information. We used these local features for classification and retrieving purposes. We validated our proposed method on two multiclass archives of histological images, namely Kimiapath24 and Kather datasets. We achieved performance of more than 90% for classification and retrieval in two datasets.