Abstract—Estrogen receptor (ER) status has proven to be a significant factor for the prediction of clinical response to hormonal therapy, in patients with breast cancer. In clinical practice, assessment of ER positive status relies on the subjective identification of the expressed nuclei in the specimens. The aim of this study was the development of a computer-aided image analysis system, employing an unsupervised segmentation algorithm based on the L*a*b color space transformation. Kendall’s coefficient of concordance showed an adequate level of agreement (Kendall’s W=0.79) between the clinical evaluation of the physician and the objective quantification of the automatic computer-aided system. Computer-assisted determination of ER status may be used as a second opinion tool in routine assessment of immunohistochemical sections. I. INTRODUCTION REAST carcinoma is the most common malignancy among females with an increasing tendency [1]-[3]. Estrogens act on breast tissue (normal and malignant), by binding to estrogen receptors (ER), and cause cell proliferation. Although this process is important for normal breast development, it includes an inherent risk of developing cancer cells. Patients with breast cancer cells that express ER in their nuclei (ER+ status) undergo different therapeutic management and treatment from those that breast cancer cells do not possess ER (ER- status) [4], [5]. It has been shown that the ER status is a significant biologic factor for the prediction of clinical response to hormonal therapy [6]. Recently, evaluation of ER status is performed by means of immunohistochemistry (IHC) [7], [8]. In clinical practice, the histopathologist chooses the IHC stained sections to be assessed, based on the diagnostic assessment of hematoxylin and eosin stained slides [4]. Assessment of ER positive status relies on the subjective identification of the percentage positive (stained) nuclei, under microscopic Manuscript received June 30, 2006. This work was supported by the Greek State Scholarship Foundation (IKY). S. Kostopoulos, A. Daskalakis and G. Nikiforidis are with the Medical Image Processing and Analysis Group, Department of Medical Physics, School of Medicine, University of Patras, 26500 Rio, Greece (2610-997745; e-mail: skostopoulos@upatras.gr). D. Cavouras is with Medical Image ans Signal Processing Laboratory, Department of Medical Instruments Technology, Technological Educational Institute of Athens, 12210 Athens, Greece (e-mail: cavouras@teiath.gr). P. Ravazoula is with the Department of Pathology, University Hospital of Patras, 26500 Rio, Greece. review. Due to the intrinsic inter- and intra-observer variability, an objective quantification is required for the accurate determination of the ER status [9]. Computer-aided image analysis systems have been proposed for the objective quantification of ER status as second opinion tools. Previous studies are mainly concerned with the employment of commercially developed image analysis systems [9]-[14] that mostly employ global thresholding techniques. Schnorrenberg et al. [15] have proposed a promising approach for the accurate detection of ER status, utilizing an in-build algorithm for the detection and classification of individual nuclei, however, requiring user interaction. The present study is focused on the development of an automatic computer-aided image analysis system for the objective assessment of ER positive status of breast cancer with no user interaction. For this, a specific color transformation followed by unsupervised clustering were developed to separate positively stained (brown) from negatively stained (blue) nuclei and from background tissue and, thus, to automatically assess the percentage of positive nuclei (ER+ status) present in the IHC stained image. Results were compared against the physician’s objective evaluation. II. MATERIAL AND METHODS Twenty nine immunohistochemically stained specimens of breast cancer were collected by an experienced histopathologist (P.R) from the Department of Pathology of the University Hospital of Patras, Greece. For each specimen, the ER expression was semi-quantitative assessed (P.R), based on a clinical scoring protocol [16]. According to this protocol, the percentage of the number of positive stained nuclei to the total number of positive and negative nuclei was visually inspected (Table I), from a representative region, where a large number of positive nuclei existed. Brown and blue nuclei were regarded as positive and negative stained respectively. Five percent was Image Analysis System For Assessing The Estrogen Receptor’s Positive Status In Breast Tissue Carcinomas Spiros Kostopoulos, Dionisis Cavouras, Antonis Daskalakis, Panagiota Ravazoula and George Nikiforidis B TABLE I THE SCORING SYSTEM USED IN THIS STUDY Score Proportion of positive nuclear staining 1 6-33% 2 34-66% 3 67-100%