Indian Journal of Science and Technology

Abstract

Objectives: Abnormalities of protein receptors in the cell induce cancer. Detection of protein receptors such as Estrogen Receptor (ER)/Progesterone Receptor (PR) helps in hormone treatment, which improves the prognosis factor. Methods: Immunohistochemistry stained breast cytology images are used for finding the protein receptors. Separate stains are used for finding each receptor status. The presence of receptors is identified based on the brown color present in the nucleus. Brown color extracted through the channel separation, thresholding and relevant features are obtained from Gray Level Co-occurrence Matrix (GLCM). Based on these feature values an Artificial Neural Network (ANN) will classify the scores. Findings: Manual procedure for ER/PR scoring is based on the value of HSCORE, which is calculated by counting the brown colored nuclei and its intensity levels by the pathologist. This is a subjective procedure and has the risk of human fatigue errors. The medical expert decides the treatment plan based on the scores. Here we developed a new technique by which the manual scoring process could be imitated using the optimal set of features through an Artificial Neural Network (ANN), and obtained a result of 95.52 percent. Application: This could be a step towards the automation of Immunohistochemistry images and help in the survival of the patients. Hormone treatments are costlier procedure because of, the large amount of data to be processed manually.
Keywords: Estrogen Receptors, Feature Extraction, Immunohistochemistry, Progesterone Receptors