Nuclei Detection in HER2-SISH Histopathology Images

Automatic quantification of cell nuclei in silver-enhanced in situ hybridization (SISH) images can be of great help to pathologists to examine HER2 status based on HER2 and CEN17 biomarkers. This paper proposed an image processing-based method for nuclei detection in HER2-SISH images. We first extracted sections of the foreground image using a combination of local thresholding, morphological filtering, and expanding regions based on intensity. Then the marker-controlled watershed is applied for separating the clustered nuclei in the foreground regions. A set of nuclei marked by our collaborating pathologists on SISH-stained breast cancer images are used to measure the effectiveness of the proposed approach. HER2-SISH histo-scoring is highly dependent on the accurately identified nuclei, hence the importance of the proposed detection method. Experimental results shows very promising detection performance, with high concordance against the pathologists’ marking.

I. Introduction

In many diseases, including malignancies, morphological abnormalities in the cell nuclei are thought to be a significant warning indication [1]. These changes can also offer therapeutically useful information during diagnosis [2]. The standard procedure entails manual examination, and diagnosis and decision making carried out by pathologists are based on specific morphological aspects of the nuclei. Other than being tedious and time-consuming, manual examination can also suffer from problems like limited sensitivity, specificity, and reproducibility. Nuclei segmentation is the most fundamental but crucial step in histopathology prognosis because the subsequent classification or scoring is highly dependent on the accuracy of the segmented nuclei. This fact emphasises the urgent need to develop and improve speedy and automated histology image processing systems [3].