| Abstrakt: |
Breast cancer (BC) is the most prevalent type of cancer among women. However, it can be cured if detected in early stages and treated in proper way. As pathologists play the lead role in the whole process of diagnosis and prognosis, the process might include inaccuracy due to external factors such as interobserver and human error. Besides, in most places of the world there is lack of experienced pathologists and cancer experts. Therefore, computer programs can help extensively to compensate for these setbacks. Computer programs can be trained through Machine Learning (ML) algorithms in order to perform the scoring process on BC medical images such as histopathological and immunohistochemistry (IHC) images. IHC staining is a technique in which a set of biomarkers are used to detect the positive reaction to a certain type of treatment. By using whole slide imaging (WSI) for digitizing the IHC samples, the images can be prepared in digital form to be used as the entry to the computer programs. To provide an automated scoring program, researchers take advantage of image processing techniques combined with ML algorithms for segmentation and classification of positive cells based on the stain color and intensity. Since image quality highly rely on the laboratory equipment, operator's experience and other factors such as stain intensity, the programs may utilize image enhancement methods for more accurate segmentation and classification results. In this paper some of these methods are investigated based on preprocessing, segmentation, and classification steps for four biomarkers i.e., ER, PR, Her2, and Ki67. The results show that lack of IHC datasets and high computational time as well as complexity of implementation are the main obstacles that need to be investigated by researchers. [ABSTRACT FROM AUTHOR] |
