基于网格重构学习的染色体分类模型  

A Grid Reconstruction Learning Model for Chromosome Classification

作　　者：张林[1,2] 易先鹏 王广杰 范心宇 刘辉 王雪松[1,2] ZHANG Lin;YI Xian-Peng;WANG Guang-Jie;FAN Xin-Yu;LIU Hui;WANG Xue-Song(Engineering Research Center of Intelligent Control for Underground Space,Ministry of Education,China University of Mining and Technology,Xuzhou 221116;School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116)

机构地区：[1]中国矿业大学地下空间智能控制教育部工程研究中心,徐州221116 [2]中国矿业大学信息与控制工程学院,徐州221116

出　　处：《自动化学报》2024年第10期2013-2021,共9页Acta Automatica Sinica

基　　金：国家自然科学基金(61971422,31871337)资助。

摘　　要：染色体的分类是核型分析的重要任务之一.因其柔软易弯曲,且类间差异小、类内差异大等特点,其精准分类仍然是一个具有挑战性的难题.对此,提出一种基于网格重构学习(Grid reconstruction learning,GRiCoL)的染色体分类模型.该模型首先将染色体图像网格化,提取局部分类特征;然后通过重构网络对全局特征进行二次提取;最后完成分类.相比于现有几种先进方法,GRiCoL同时兼顾局部和全局特征提取更有效的分类特征,有效改善染色体弯曲导致的分类性能下降,参数规模合理.通过基于G带、荧光原位杂交(Fluorescence in situ hybridization,FISH)、Q带染色体公开数据集的实验表明:GRiCoL能够更好地弱化染色体弯曲带来的影响,在不同数据集上的分类准确度均优于现有分类方法.Chromosome classification is one of the key tasks of karyotype analysis.However,due to chromosomes are flexible hence exhibit less difference between different types while significant difference within same type,accurate classification of chromosome remains a challenging issue.In this paper,a chromosome classification model based on grid reconstruction learning(GRiCoL)is proposed.To weaken the impact of the bendy state,chromosome images are first grid-enabled for feature extraction separately.Subsequentially,global features are extracted for the second time by reconstruction network,which is followed by classification.Compared with the state-of-the-art methods,the proposed GRiCoL can get more efficient discriminable features based on both local and global features,therefore can overcome the adverse effects of bandy form of chromosome with reasonable parameter scale.Experiments on public G band,fluorescence in situ hybridization(FISH)as well as Q band chromosome datasets show that GRiCoL can extract discriminative features that weaken the bending of chromosomes more efficiently,meanwhile,higher performance was obtained as compared to current classification algorithms.

关 键 词：核型分析 染色体分类 特征重构 网格化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] Q343.2[自动化与计算机技术—计算机科学与技术]