Automated Quantification of Ki-67 on Gastric Epithelial Tissue based on Cell Nuclei Area Ratio

Abstract

The objective was to develop an automated algorithm for the estimation of a protein (Ki-67) index based on cell nuclei area ratio of gastric epithelial tissue cells; for this purpose, digital histopathology images were used. An expert manually annotated each region of interest of the images. A proportion of Ki-67 positive and negative cells within that region was used to obtain the color distribution of the corresponding pixels. The histogram of each color distribution was modeled as a Gaussian and, later, thresholded for segmentation and classification. Finally, the Ki-67 index was estimated as the ratio between the segmented positive area of the nuclei divided by the total area of the positive and negative nuclei. The automated method has a strong correlation of 0.725 and a root mean square error of 0.293 when compared to the manual method, which gives certainty that the automated method can be used to analyze the proliferation rate. Furthermore, compared to manual classification, the presented method automatically classifies every image in the same Ki-67 category: low, intermediate, and high. Despite the small sample size, the utility of the presented method was demonstrated. However, the low number of scored images did not allow for thoroughly sampling the ranges of pixel values and intensities observed by pathologists, which will be addressed in future work.

Keywords

• digital pathology
• digital image processing
• immunohistochemical analysis
• pattern recognition
• nuclei segmentation
• ki-67 quantification
• gastric cancer
• gastric cells