深度学习在染色体分割中的应用综述

Survey of Application of Deep Learning in Chromosome Segmentation

染色体分析是细胞遗传学研究的基本方法,被广泛地应用在遗传疾病筛查和产前诊断中,能有效地避免重度缺陷患儿的出生,对优生优育有着积极意义。染色体分割是染色体核型分析中最为关键的一步,其目标是将染色体实例从细胞分裂中期的显微镜图像中分割出来。在实际染色体分割应用中,由于染色体实例之间极其容易发生重叠和交叉的现象,给染色体分割带来巨大的挑战。随着深度学习技术在图像分割领域的快速发展,基于深度学习技术的算法和模型被广泛地应用于染色体分割任务中。分析了目前染色体分割领域的研究问题和挑战,并总结了现有的数据集和评价指标。重点综述基于深度学习技术在染色体分割领域中的研究,包括基于语义分割网络的重叠染色体分割的相关研究和基于实例分割网络的染色体实例分割的相关研究。对深度学习技术在染色体实例分割领域的研究现状进行总结和展望。

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.