Article: A pattern selection scheme for content-based histopathological image retrieval: an effective tool for feature dimensionality reduction Journal: International Journal of Bioinformatics Research and Applications (IJBRA) 2026 Vol.22 No.3 pp.344 - 369 Abstract: Local pattern features such as local binary patterns are reported to be highly appropriate for describing the tissue patterns of histopathology scans. Meanwhile, using a single descriptor to extract the features cannot, generally, achieve the expected performance. In that context, combining multiple descriptors is commonly used in content-based histopathological image retrieval (CBHIR) to improve the retrieval performance. However, the feature combination usually leads to high dimensional feature spaces. To overcome this challenge, we propose in this paper a pattern selection scheme based on sine cosine algorithm (SCA) that allows selecting the most discriminant patterns from some combined local pattern features, and therefore reducing their dimension. Experimental results on KIMIA Path960 and Kather-5K histopathology datasets show that our proposed approach can effectively reduce the feature space dimensionality up to 67% while improving the retrieval performance. Moreover, our proposed methods achieve better retrieval performance than other handcrafted-based and deep learning-based CBHIR methods. Inderscience Publishers - linking academia, business and industry through research

Title: A pattern selection scheme for content-based histopathological image retrieval: an effective tool for feature dimensionality reduction

Authors: Rahima Boukerma; Bachir Boucheham; Salah Bougueroua

Addresses: Department of Computer Science, Laboratoire de Recherche en Electronique de Skikda (LRES), University of 20 Août 1955-Skikda, DZ 21000 Skikda, Algeria ' Department of Computer Science, Laboratoire de Recherche en Electronique de Skikda (LRES), University of 20 Août 1955-Skikda, DZ 21000 Skikda, Algeria ' Department of Computer Science, Laboratoire de Recherche en Electronique de Skikda (LRES), University of 20 Août 1955-Skikda, DZ 21000 Skikda, Algeria

Abstract: Local pattern features such as local binary patterns are reported to be highly appropriate for describing the tissue patterns of histopathology scans. Meanwhile, using a single descriptor to extract the features cannot, generally, achieve the expected performance. In that context, combining multiple descriptors is commonly used in content-based histopathological image retrieval (CBHIR) to improve the retrieval performance. However, the feature combination usually leads to high dimensional feature spaces. To overcome this challenge, we propose in this paper a pattern selection scheme based on sine cosine algorithm (SCA) that allows selecting the most discriminant patterns from some combined local pattern features, and therefore reducing their dimension. Experimental results on KIMIA Path960 and Kather-5K histopathology datasets show that our proposed approach can effectively reduce the feature space dimensionality up to 67% while improving the retrieval performance. Moreover, our proposed methods achieve better retrieval performance than other handcrafted-based and deep learning-based CBHIR methods.

Keywords: content-based histopathological image retrieval; CBHIR; dimensionality reduction; feature selection; FS; local binary pattern; LBP; OCLBP; pattern selection.

DOI: 10.1504/IJBRA.2026.154825

International Journal of Bioinformatics Research and Applications, 2026 Vol.22 No.3, pp.344 - 369

Received: 02 Jul 2024
Accepted: 16 Apr 2025
Published online: 15 Jul 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: A pattern selection scheme for content-based histopathological image retrieval: an effective tool for feature dimensionality reduction

Keep up-to-date