Iterative randomized irregular circular algorithm for proliferation rate estimation in brain tumor Ki-67 histology images

• This CAD system calculates proliferation rate estimation (PRE) automatically.
• A novel Iterative Randomized Irregular Circular Algorithm (IRIC) has been proposed.
• The Brain Tumor Ki-67 Histology Images are taken from UKM Medical Centre.
• Prior to a random set region of interest, IRIC counts blue and brown cells and PRE.
• IRIC outperforms Circular Hough Transform about 98% and of F-measurement rate.

Proliferation rate estimation (PRE) is clinically performed from Ki-67 histopathology images. As brain tumor tissues are very complex, accurate PRE determination requires manual cell counting that is tedious, time consuming and inherently inaccurate due to inter-personal variations. Therefore, pathologists usually determine the PRE based on their experience and visualization without actual counting. Automating PRE can substantially increase the efficiency and accuracy of pathologists’ determination of PRE. In addition, developing a deterministic and reproducible proliferation rate value is crucial to reduce inter-observer variability. In this paper, a PRE Computer Aided Diagnosis (PRECAD) system has been developed to automate the determination of PRE from Ki-67 histopathology microscopic images for brain tumors. The process involves six steps: color space transformation, customized color modification, nuclei segmentation based on K-Means clustering, preprocessing the extracted cells, counting based on an iterative structured circle detection (IRIC) algorithm, and finally, calculating the PRE value. The proposed IRIC algorithm is able to detect irregular and overlapping cells by introducing dynamic initialization to the basic RCD method, dividing the entire image into partitions based on 8-neighbor connected components. We initiated a new selection method for determining a best circle candidate that yields a reduced probability of incorrectly detecting circles, and proposed a new technique for detecting irregular cells via a dynamic number of iterations that guarantees finding all the cells in a selected partition. Using the same innovations mentioned above, our proposed IRIC algorithm can also be used to detect irregular and  two or more overlapping cells. The proposed PRECAD system achieved high accuracy, as determined by quantitative analysis of precision, recall and F-measurement test values of 97.8%, 98.3% and 98% for blue cells and 98.7%, 98% and 98.4% for brown cells, respectively. Thus, our proposed PRECAD system is as reliable as a pathologist for estimating the proliferation rate, while also featuring inherent reproducibility.