关键词: 染色体实例分割, 染色体核型分析, 医学图像处理, 语义分割, 实例分割, 深度学习

Abstract:

Chromosome analysis is a basic method of cytogenetic research, which is widely used in genetic disease screening and prenatal diagnosis. It can effectively avoid the birth of children with severe defects, and has positive significance for eugenics.Chromosome segmentation is the most critical step in chromosome karyotype analysis, whose goal is to segment overlapping or touching chromosome clusters for metaphase cell images into corresponding instances. Chromosome segmentation is the basis of subsequent chromosome classification, karyotype arrangement, clinical diagnosis and treatment. As the non-rigid intrinsic nature of chromosomes, many chromosome instances overlap and touch with each other, making chromosome instance segmentation difficult. Recently, deep learning technologies have been advanced in many computer vision tasks, and a variety of deep learning-based methods have been proposed to address the issue of chromosome instance segmentation and achieve advanced results. This work first formalizes the problem statement of the current issue, summarizes its challenges of chromosome segmentation. And then, this paper concludes existing datasets and evaluation metrics. Afterward, this article surveys existing deep learning-based chromosome instance segmentation methods, including works based on semantic segmentation networks and studies based on instance segmentation networks. Finally, this paper concludes existing works for the issue of chromosome instance segmentation based on deep learning techniques and gives prospects for future research on deep learning-based chromosome instance segmentation.

Key words: chromosome instance segmentation, chromosome karyotype analysis, medical image processing, semantic segmentation, instance segmentation, deep learning