Microscopic Biopsy Image Segmentation Using Hybrid Color K-Means Approach

Introduction

Segmentation of microscopic biopsy images is typical task due to its complex nature. The cellular components of the microscopic biopsy images are segmented on the basis of intensity, color, shape, and texture features. The microscopic biopsy has several segmentation issues and challenges like high quality segmentation with low computational cost. There are many approaches which are reported in the literature for the segmentation of microscopic biopsy images (Pal, 1993), but most of the segmentation algorithms are time consuming, and even some times not accurate up to the mark. Some of the popular segmentation approaches are adaptive thresholding (Pal and Pal, 1993), watershed based approach (Beucher et. al., 1979), color k-means (Ray and Turi, 1999), fuzzy c-means (Lim et. al., 1990) and texture based segmentation (Lorigo et. al., 1998), and every approach has its own advantages and drawbacks. Thresholding approaches are roughly categorized into local, global, and adaptive thresholding approaches. The adaptive thresholding is the most popular approach for creating the binary images. The computational complexity of thresholding approaches is low but its losses the colour information of microscopic biopsy images. Therefore, it does not use alone for the segmentation of microscopic biopsy images. The watershed segmentation algorithms is an example of region based segmentation approach. It is popular segmentation from the field of mathematical morphology and proposed by Boucher et al. (1979), the watershed can be in intuitively thought as a landscape or topographic relief which is flooded by water, and watersheds are the dividing lines of the domains of attraction of rain falling over the region. The intensity value is represented by the height of each point. The gradient of the original image is taken as input of the watershed transform. However, catchment basin boundaries are located at high gradient points (Pal and Pal, 1993). The watershed transform has good properties that make it useful for many different image segmentation applications: it is simple, easy to implement and intuitive. It can also be parallelized (Beucher et. al., 1979), and always produces a complete division of the image. The major drawback of the watershed segmentation includes over segmentation and sensitive to the false edges. Some works reported in literature for the segmentation of microscopic biopsy images are described as follows; a novel synergistic boundary and region- based active contour model was presented by authors, to demonstrate for level set formulation with automated initialization based on watershed. Digitized histopathology images of breast, and prostate biopsy specimens are demonstrated as an application of these synergistic active contour models using multiple level sets, to segment nuclear and glandular structures (Mouelhia et. al., 2013). The qualitative and quantitative evaluation were tested on 100 prostate and 14 breast cancer histology images for the task of detecting and segmenting nuclei, and lymphocytes reveals that the model easily outperforms two state of the art segmentation schemes, and on average is able to resolve up to 91% of overlapping/occluded structures in the images. The watershed based algorithm, and concave vertex graph approaches were proposed for touching nuclei to perform accurate quantification of the different stains. Two datasets of breast cancer cell images containing different level of malignancy were tested on these segmentation algorithms. The segmentation accuracy in term of cancer nuclei number is over than 97%, reaching an improvement of 3–4% over earlier methods for the complete image database.