Automatically diagnosing HER2 amplification status for breast cancer patients using large FISH images

Fluorescence in situ hybridization (FISH) is a technique that prepares acceptable results for molecular imaging biomarkers to precisely and dependably detect and diagnose disorders which are sign of cancers. Since contemporary manual FISH signal analysis is low-effective and inconsistent, it is an attractive research area to develop automated FISH image scanning systems and computer-aided diagnosis (CAD) schemes. The gene expression of epidermal growth factor receptors 2 (HER2) is highly related to results of patients with probable breast cancer. Although FISH technology outperforms other methods, yet it has so many drawbacks. Traditional approaches on FISH analysis are performed manually by clinician. This lets the results are highly dependent to human eye. Also FISH test colors constitutes of dark blue and black regions, it is reasonable that human eye will fail to distinguish between colors. Therefore, the success of computer vision algorithms compared to human eye in analyzing gene expression rate in FISH images will be discussed in this study. Another objective of this study is to expand a CAD program to evaluate HER2 status using acquired images that have MIRAX format. Different large FISH images were chosen for this study from pathology laboratory from Acibadem Maslak hospital. The proposed CAD scheme first applies pre-processing median and gaussian filters. An adaptive thresholding method followed by a watershed segmentation algorithm is employed to segment cells of interest areas. Furthermore, analyzable cells are found and non-detectable cells because of cell overlapping or image staining are discarded. The scheme then splits the detected analyzable region of interest into two red and green color spaces which is also followed by application of a scanning algorithm to detect the CEP17 green and HER2/neu red FISH signals separately. Finally, the proposed method calculates the ratio between independent green and red FISH signals of all analyzable cells identified on the image. The results express that the tool has the ability to automatically express HER2 status using very large FISH images. The results of the computer aided tool would lead to a more effective method in specifying HER2 state of probable patients.