978-1-5386-8369-9/18/$31.00 ©2018 IEEE Simple Landscapes Analysis for Relevant Regions Detection in Breast Carcinoma Histopathological Images Xiao Jian Tan School of Mechatronic Engineering University Malaysia Perlis (UniMAP), 02600 Arau Perlis, Malaysia. xj_0506@hotmail.com Mohd Yusoff Mashor School of Mechatronic Engineering University Malaysia Perlis (UniMAP), 02600 Arau Perlis, Malaysia. yusoff@unimap.edu.my Nazahah Mustafa School of Mechatronic Engineering University Malaysia Perlis (UniMAP), 02600 Arau Perlis, Malaysia. nazahah@unimap.edu.my Wei Chern Ang Clinical Research Centre Hospital Tuanku Fauziah, 01000 Kangar Perlis, Malaysia. wei.ang.1990@gmail.com Khairul Shakir Ab Rahman Department of Pathology Hospital Tuanku Fauziah, 01000 Kangar Perlis, Malaysia. ksyakir@gmail.com Abstract— Breast carcinoma represents a huge global health problem among women in both developed and developing countries. It is estimated that over 508,000 women worldwide died in 2011 due to breast carcinoma. Nottingham Histological Grading (NHG) system is recognized as the gold standard to provide overall grade for breast carcinoma. One of the breast carcinoma criteria considered in the grading system is tubule formation. The assessment of tubule formation starts with visual inspection on breast histopathological image using 10x magnification. However, not all regions in the image provide meaningful information. Histopathological image with score 3 in tubule formation usually has a small tubule size. Thus, a visual inspection at a higher magnification is required. A continuous inspection at a higher magnification is time consuming. By eliminating the irrelevant regions in the histopathological image, histopathologist can focus on the relevant region for further examination. This study proposed a simple method to detect relevant region on the breast histopathological images using landscape analysis. The proposed method was tested using three groups of histopathological images: Group 1: relevant and irrelevant regions, Group 2: relevant regions only and Group 3: irrelevant regions only. The proposed method is found to be effective in eliminating irrelevant regions as the overall accuracy for Groups 1, 2 and 3 are 86.6%, 100.0% and 100.0%, respectively. Keywords— breast carcinoma; histopathological image; landscapes analysis; relevant region I. INTRODUCTION Nottingham Histological Grading (NHG) system is recognized as the gold standard to provide overall grade for breast carcinoma [1]. Tubule formation is one of the three critical factors that is stated in the NHG system. The other two critical factors are mitotic count and nucleus pleomorphism [2, 3]. In recent years, pathology laboratories have undergone transformation where digital workflow has been introduced as standard practice [4]. The introduction of whole slide imaging (WSI) scanner allows a high throughput slide digitalization with relatively low cost [5]. The application of WSI scanner is fully automated. Slide digitalization is recognized as a part of the standard practice in the pathology laboratory. The analogue histopathological slides obtained from surgical biopsy are converted to the digital slides using WSI scanner. Quantitative and qualitative analyses could be performed on the digital slides by implementing various image processing algorithms [5]. In the assessment of tubule formation, tumor regions that provide meaningful information which indicate the degree of differentiation in tumor cells are referred as relevant regions, whereas, the non-tumor regions and background are referred as irrelevant regions. Standard practice assessment of tubule formation starts with visual inspection at 10x magnification on a histopathological image. However, not all regions in the histopathological image provides meaningful information (ie., relevant regions). Histopathological image with score 3 in tubule formation (obtained from NHG system) usually has a small tubule size. Histopathologist may require a visual inspection at a higher magnification (e.g., 20x to 40x magnification). A continuous visual inspection at a high magnification is time consuming [6]. A histopathological image could be formed by as high as 700,000 pixels. By eliminating the irrelevant regions in the histopathological image, histopathologist can focus on the relevant region for further examination. In Figure 1, images (a-d) and (e-h) show examples of relevant regions and irrelevant regions respectively found in the histopathological image. Study to eliminate irrelevant regions from histopathological images of breast carcinoma for breast carcinoma grading using image processing technique is very few. [7-9] proposed pixel-wise labeling approaches which is suitable to be implemented in small size images. This is a good approach but not practical for a large size image. Implementing pixel-wise labeling approach on a large size image may slow down the overall computation time of the system. Therefore, this paper proposed a simple landscapes analysis that offers a fast and accurate detection of relevant region in breast histopathological images. The organization of the paper is as follows: Section II provides details description on the proposed method, Section III provides a full description in experimental results and the conclusion is given in Section IV. Fundamental Research Grant Scheme: FRGS/1/2016/SKK06/UNIMAP/02/3