Seed Detection algorithm using multi-thresholding measures on histopathological images

Abstract

The recent advancing of computational resources, led to a significant improvement in histopathological image analysis. These improvements helped  to diagnosis various diseases and dive into cellular level of the tissue for accurate prognosis. Therefore, an automated algorithm is proposed to enhance diagnostic accuracy and efficiency. This paper proposes a detection technique to detect the cells nuclei on histopathological images that are stained by Hematoxylin and Eosin (H&E). The proposed technique applies multiple thresholds on the grayscale image version of the H&E-stained image and from each resulted binary image, several centroids are extracted for each disconnected foreground region. Three measures such as area, centroid location, and circularity ratio have been used to determine the selection of nuclei seed. The technique assigns certainty weights based on threshold values, enhancing the reliability of detected seeds. Comparisons with existing methods, like the generalized Laplacian of Gaussian (gLoG) technique, demonstrate the proposed method’s efficiency and accuracy, providing a robust foundation for further segmentation processes.