LBR-TBLR: an innovative local binary pattern approach for breast cancer diagnosis

Abstract

The local Binary Pattern (LBP) method is widely recognized in image processing and pattern recognition due to its non-parametric nature, computational efficiency, and robustness to changes in illumination. In this study, we propose a new LBP approach for the problem space of breast cancer. Within this framework, we investigate the performance of considering the LBP filter as a 3x3x3 volume and examining it from different perspectives instead of converting RGB images to grayscale or applying independent LBP to each channel. In this method, which we refer to as LBP-TBLR, we adopt a three-dimensional perspective by treating each color channel as an independent dimension. Accordingly, the performance of applying LBP to each channel and surface of the LBP volume is experimentally examined. We explore various combinations and amalgamations of LBP feature maps to discover their impact on classification performance. Our experiments demonstrate that merging different LBP features significantly affects classification performance. We observed that combining feature maps obtained from different color channels of the LBP volume with the original model input improves model performance. Additionally, a more notable finding is the superior performance of combinations of LBP feature maps obtained by considering specific perspectives on the LBP volume. Specifically, combinations like lbpTop, lbpRight, lbpButton, and lbpLeft improve classification performance by properly capturing directional variations. Moreover, the Lbp-TBLR method, created by combining these feature maps, exhibits higher accuracy and performance compared to other combinations. The proposed Lbp-TBLR method resulted in a 15.63% performance gain over the previous model. The effectiveness of the proposed method has been thoroughly examined both quantitatively and qualitatively within the scope of this study, demonstrating superior performance. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2026.