Automated characterization and counting of Ki-67 protein for breast cancer prognosis: A quantitative immunohistochemistry approach. (February 2017)

• Record Type: Journal Article
• Authors: Mungle, Tushar; Tewary, Suman; Arun, Indu; Basak, Bijan; Agarwal, Sanjit; Ahmed, Rosina; Chatterjee, Sanjoy; Maity, Asok Kumar; Chakraborty, Chandan
• Is Part Of: Computer methods and programs in biomedicine. Volume 139(2017)
• Page Start: 149
• Page End: 161
• Publication Date: 2017-02
• Subjects: Breast cancer -- Immunohistochemistry -- Ki-67 protein -- Fuzzy C-means -- k-means -- Proliferation index; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2016.11.002
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: Hybrid clustering using fuzzy C -means and k -means to segment Ki-67 nuclei. Characterization of Ki-67 protein for assessing proliferation index. Validation of methodology from the ground truth information. Qualitative assessment of the proposed methodology using F-measure. Abstract: Ki-67 protein expression plays an important role in predicting the proliferative status of tumour cells and deciding the future course of therapy in breast cancer. Immunohistochemical (IHC) determination of Ki-67 score or labelling index, by estimating the fraction of Ki67 positively stained tumour cells, is the most widely practiced method to assess tumour proliferation (Dowsett et al. 2011). Accurate manual counting of these cells (specifically nuclei) due to complex and dense distribution of cells, therefore, becomes critical and presents a major challenge to pathologists. In this paper, we suggest a hybrid clustering algorithm to quantify the proliferative index of breast cancer cells based on automated counting of Ki-67 nuclei. The proposed methodology initially pre-processes the IHC images of Ki-67 stained slides of breast cancer. The RGB images are converted to grey, L*a*b*, HSI, YCbCr, YIQ and XYZ colour space. All the stained cells are then characterized by two stage segmentation process. Fuzzy C-means quantifies all the stained cells as one cluster. The blue channel of the first stage output is given as input to k-means algorithm, which provides separate cluster for Ki-67 positive and negative cells. The count of positive and negative nuclei is used to calculate the F-measure for each colour space. A comparative study of our work with the expert opinion is studied to evaluate the error rate. The positive and negative nuclei detection results for all colour spaces are compared with the ground truth for validation and F-measure is calculated. The F-measure for L*a*b* colour space (0.8847) provides the best statistical result as compared to grey, HSI, YCbCr, YIQ and XYZ colour space. Further, a study is carried out to count nuclei manually and automatically from the proposed algorithm with an average error rate of 6.84% which is significant. The study provides an automated count of positive and negative nuclei using L*a*b*colour space and hybrid segmentation technique. Computerized evaluation of proliferation index can aid pathologist in assessing breast cancer severity. The proposed methodology, further, has the potential advantage of saving time and assisting in decision making over the present manual procedure and could evolve as an assistive pathological decision support system.