A New Method for Microscopy Image Segmentation Using Multi-scale Line Detection

Abstract

Image segmentation plays a crucial role in many biomedical imaging applications by automating and facilitating delineating of anatomical structures and different regions of interest. The objective of microscopic image segmentation is to accurately identify the boundaries of cells, cell nuclei, or histological structures in stained tissue images with various markers. Over the last few years, numerous techniques for segmenting microscopy images have been developed. Despite the advancements in segmentation techniques for microscopy images, there are still significant challenges when dealing with variations, high levels of noise, and variations in image features. (e.g., nucleus shape, cell size, concavity points between nuclei) and complexity of method parameter space. This paper introduces a new method for distinguishing nuclei and other biological structures in microscopy images. The proposed method exploits the multi scales line detection method, which presents an effective method for automatically extracting blood vessels from retinal images to detect boundaries of cells, cell nuclei and the concave points that split the contours into segments. The effectiveness of the method has been evaluated both qualitatively and quantitatively and compared with four other segmentation algorithms - the Chan & Vese model and three thresholding techniques, namely Otsu, Kapur, and Kittler method. These evaluations were conducted using a variety of publicly available datasets from the Broad-Bioimage Benchmark. The comparison results were analyzed using the Jaccard similarity index (JI) and the Dice coefficient (DSC) measure set agreement.