南京大学学报(自然科学版) ›› 2020, Vol. 56 ›› Issue (1): 116–124.doi: 10.13232/j.cnki.jnju.2020.01.013

• • 上一篇    下一篇

基于卷积神经网络和几何优化的统计染色体核型分析方法

李康1,谢宁1(),李旭2,谭凯1   

  1. 1. 电子科技大学计算机科学与工程学院，成都，611731
    2. 电子科技大学格拉斯哥学院，成都，611731
  • 收稿日期:2019-09-17 出版日期:2020-01-30 发布日期:2020-01-10
  • 通讯作者: 谢宁 E-mail:seanxiening@gmail.com
  • 基金资助:
    国家自然科学基金(61602088);中央高校基本科研业务费基础研究项目(Y03019023601008011)

Statistical Karyotype analysis using CNN and geometric optimization

Kang Li1,Ning Xie1(),Xu Li2,Kai Tan1   

  1. 1. School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu，611731, China
    2. Glasgow College, University of Electronic Science and Technology of China, Chengdu, 611731, China
  • Received:2019-09-17 Online:2020-01-30 Published:2020-01-10
  • Contact: Ning Xie E-mail:seanxiening@gmail.com

RichHTML

1029

PDF (PC)

1464

摘要：

染色体核型分析是细胞遗传学研究的主要技术之一,在现代医学治疗和诊断中有重要的作用.通常在染色体核型分析的过程中，首先需要在染色体中期图像中分割出单条染色体，然后再对染色体逐一进行分析、比较、排序和分类.由于传统的基于几何及基于统计的分割和分类的辅助工具精度低，辅助作用有限，因此在实际工作中仍然需要医生花费大量的时间和精力进行人工核型分析.为此提出一种基于卷积神经网络和几何优化的染色体核型分析新方法，利用Mask R?CNN（Region?Convolutional Neural Networks）从染色体中期图像中分割出染色体，并训练一个新型多输入的卷积神经网络对分割后的单条染色体进行分类；还提出一种全新的基于局部特征的染色体分割数据合成方法对分割数据集进行扩充.此外，为了保证分类训练数据的一致性，提出一种基于中线的染色体伸直几何优化算法.实验结果表明提出的方法在自动核型分析中表现优秀.

关键词: 深度学习, 核型分析, 医疗图像处理, 几何优化

Abstract:

Karyotype analysis is one of the main techniques of cytogenetics through medical image processing,which plays an important role in modern medical diagnosis and treatment. The process of human karyotype analysis contains two key components. Firstly,chromosomes are segmented from metaphase chromosome digital images taken under a microscope. Then，chromosomes are analyzed,compared,ordered and classified one by one carefully. Under this procedure,the operation on segmentation and classification is cumbersomely time consuming,where traditional geometric or statistical methods only have limited effect due to low accuracy. Thus,in most conditions,human effort is still heavily required to monitor the workflow and correct the errors. In this paper,we present an integrated workflow to segment out and classify chromosomes automatically using a combination of Convolutional Neural Networks (CNN) and geometric optimization. We investigate Mask R?CNN (Region?CNN) to segment out chromosomes from metaphase chromosome images and train a CNN to classify the sub?images.To improve the performance of the segmentation network,we adapt a new local feature?based approach to synthesize images on the annotated data. Furthermore,we develop a geometric algorithm to straighten the chromosomes before classification to ensure the consistency on the training data. Experimental results demonstrate that our approach has better performance on automatic karyotype analysis.

Key words: deep learning, karyotype analysis, medical image processing, geometry optimization

中图分类号: 

  • TP31

图1

染色体中期图像(a)和对应的核型分析图片(b)"

图2

系统工作流程，包括三个阶段:染色体分割、几何优化和染色体分类"

图3

一种简单有效的染色体实例分割的像素级数据标注合成方法"

图4

(a) (i)~(v)通过德劳内三角形获取染色体中线， (vi)找到切割点，(vii)切割并旋转上下臂；(b) (i)为提取算法获取的中轴,(ii)和(iii)为scikit?image[18]的中轴提取算法"

表1

分割网络在不同数据集上训练的模型在测试集上的测试结果(%)"

数据集 测试集1（ A P ） 测试集2（ A P ） 测试集1（ A P 50 ） 测试集2（ A P 50 ）
343张手工标注图片 52.059 44.488 90.590 90.010
343张合成图片 47.563 31.805 88.194 76.967
1000张合成图片 53.476 35.450 91.657 83.855
手工标注和合成图片比例为1∶1 57.827 41.882 94.030 91.931
手工标注和合成图片比例为3∶7 57.841 43.248 94.563 91.673
手工标注和合成图片比例为1∶4 59.998 44.794 95.644 91.662

图5

分割网络生成的掩码"

表2

单个CNN和组合CNN的准确率比较"

方法 准确率

Global?only

Local?only

Straighten?only

Global & Local

All?combined

0.9445

0.9200

0.8200

0.9455

0.9570

表3

我们的方法与一些常见模型的准确率比较"

方法 准确率
AlexNet 0.8975
VGG?16 0.9023

DenseNet

ResNet?50

Sharma et al[1]

Our model

0.9414

0.9445

0.9012

0.9570

表4

分类模型在每一类上的表现"

染色体类别 1 2 3 4 5 6 7 8 9 10 11
准确率 0.9941 0.9737 0.9862 0.9370 0.9611 0.9677 0.9615 0.9191 0.9251 0.9401 0.9715
召回率 0.9861 0.9927 0.9887 0.9554 0.9295 0.9841 0.9619 0.9337 0.9246 0.9446 0.9729
F1分数 0.9901 0.9831 0.9875 0.9461 0.9450 0.9758 0.9615 0.9267 0.9248 0.9423 0.9426
染色体类别 12 13 14 15 16 17 18 19 20 21 22
准确率 0.9681 0.9681 0.9322 0.9490 0.9639 0.9693 0.9643 0.9426 0.9644 0.9425 0.9291
召回率 0.9678 0.9813 0.9397 0.9406 0.9639 0.9726 0.9469 0.9571 0.9571 0.9560 0.9301
F1分数 0.9644 0.9747 0.9360 0.9448 0.9639 0.9711 0.9556 0.9316 0.9608 0.9492 0.9296
染色体类别 X Y AVG
准确率 0.9493 0.9200 0.9567
召回率 0.9407 0.9327 0.9552
F1分数 0.9450 0.9263 0.9552
1 Sharma M , Saha O , Sriraman A ,et al . Crowdsourcing for chromosome segmentation and deep classification∥Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops. Honolulu,HI,USA：IEEE,2017：34-41.
2 He K M , Gkioxari G , Dollár P ,et al . Mask R?CNN∥Proceedings of the 2017 IEEE International
Conference on Computer Vision. Venice,Italy：IEEE,2017：2961-2969.
3 Pham D L , Xu C Y , Prince J L . Current methods in medical image segmentation. Annual Review of Biomedical Engineering,2000,2：315-337.
4 Charters G C , Graham J . Trainable grey?level models for disentangling overlapping chromosomes. Pattern Recognition,1999,32(8)：1335-1349.
5 Ji L . Fully automatic chromosome segmentation. Cytometry,1994,17(3)：196-208.
6 BenTaieb A , Hamarneh G . Topology aware fully convolutional networks for histology gland segmen?tation∥International Conference on Medical Image Computing and Computer?Assisted Intervention. Springer Berlin Heidelberg,2016：460-468.
7 Long J , Shelhamer E , Darrell T . Fully convolutional networks for semantic segmentation∥Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston,MA,USA：IEEE,2015：3431-3440.
8 Ronneberger O , Fischer P , Brox T . U?net：convolutional networks for biomedical image segmen?tation∥International Conference on Medical Image Computing and Computer?assisted Intervention. Springer Berlin Heidelberg,2015：234-241.
9 Li Y , Qi H Z , Dai J F ,et al . Fully convolutional instance?aware semantic segmentation∥Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu,HI,USA：IEEE,2017：2359-2367.
10 Pinheiro P O , Lin T Y , Collobert R ,et al . Learning to refine object segments∥European Conference on Computer Vision. Springer Berlin Heidelberg,2016：75-91.
11 Lahiri A , Ayush K , Biswas P K ,et al . Generative adversarial learning for reducing manual annotation in semantic segmentation on large scale miscroscopy images：automated vessel segmentation in retinal fundus image as test case∥Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops. Honolulu,HI,USA：IEEE,2017：42-48.
12 Dwibedi D , Misra I , Cut Hebert M. ,paste and learn：surprisingly easy synthesis for instance detection∥Proceedings of the 2017 IEEE International Conference on Computer Vision. Venice,Italy：IEEE,2017：1301-1310.
13 Ritter G , Gallegos M T . Outliers in statistical pattern recognition and an application to automatic chromosome classification. Pattern Recognition Letters,1997,18(6)：525-539.
14 Lerner B , Guterman H , Dinstein I ,et al . Medial axis transform?based features and a neural network for human chromosome classification. Pattern Recognition,1995,28(11)：1673-1683.
15 Markou C , Maramis C , Delopoulos A ,et al . Automatic chromosome classification using support vector machines.Technicial Report. Aristotle University of Thessaloniki,2012.
16 Swati S , Gupta G , Yadav M ,et al . Siamese networks for chromosome classification∥Proceedings of the 2017 IEEE International Conference on Computer Vision. Venice,Italy：IEEE,2017：72-81.
17 Wu Y R , Yue Y S , Tan X ,et al . End?to?end chromosome karyotyping with data augmentation using gan∥2018 25th IEEE International Conference on Image Processing (ICIP). Athens,Greece：IEEE,2018：2456-2460.
18 Van Der Walt S , Sch?nberger J L , Nunez?Iglesias J ,et al . Scikit?image：image processing in python. PeerJ,2014,2(2)：e453.
19 Javan?Roshtkhari M , Setarehdan S K . A new approach to automatic classification of the curved chromosomes∥2007 5th International Symposium on Image and Signal Processing and Analysis. Istanbul,Turkey：IEEE,2007：19-24.
20 Shen W , Zhou M , Yang F ,et al . Multi?scale convolutional neural networks for lung nodule classification∥International Conference on Informa?tion Processing in Medical Imaging. Springer Berlin Heidelberg,2015：588-599.
21 He K M , Zhang X Y , Ren S Q ,et al . Deep residual learning for image recognition∥Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas,NV,USA：IEEE,2016：770-778.
22 Abdulla W . Mask R?CNN for object detection and instance segmentation on keras and tensorflow. https：∥github.com/matterport/Mask_RCNN. GitHub repository,2017.
[1] 朱伟,张帅,辛晓燕,李文飞,王骏,张建,王炜. 结合区域检测和注意力机制的胸片自动定位与识别[J]. 南京大学学报(自然科学版), 2020, 56(4): 591-600.
[2] 韩普,刘亦卓,李晓艳. 基于深度学习和多特征融合的中文电子病历实体识别研究[J]. 南京大学学报(自然科学版), 2019, 55(6): 942-951.
[3] 张家精,夏巽鹏,陈金兰,倪友聪. 基于张量分解和深度学习的混合推荐算法[J]. 南京大学学报(自然科学版), 2019, 55(6): 952-959.
[4] 钟琪,冯亚琴,王蔚. 跨语言语料库的语音情感识别对比研究[J]. 南京大学学报(自然科学版), 2019, 55(5): 765-773.
[5] 王蔚, 胡婷婷, 冯亚琴. 基于深度学习的自然与表演语音情感识别[J]. 南京大学学报(自然科学版), 2019, 55(4): 660-666.
[6] 张鹏，黄毅，阮雅端，陈启美*. 基于稀疏特征的交通流视频检测算法[J]. 南京大学学报(自然科学版), 2015, 51(2): 264-270.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
[1] 叶绿萌1，樊少芬2，常　鸣1，司徒淑娉3，王雪梅4*. 珠三角地区秋季臭氧生成敏感性时空变化模拟研究[J]. 南京大学学报(自然科学版), 2016, 52(6): 977 .
[2] 阚　威, 李　云. 基于LSTM的脑电情绪识别模型[J]. 南京大学学报(自然科学版), 2019, 55(1): 110 -116 .
[3] 王飞永, 彭建兵, 卢全中, 黄强兵, 孟振江, 乔建伟. 渭河盆地地裂缝同生机制研究[J]. 南京大学学报(自然科学版), 2019, 55(3): 339 -348 .
[4] 王蔚, 胡婷婷, 冯亚琴. 基于深度学习的自然与表演语音情感识别[J]. 南京大学学报(自然科学版), 2019, 55(4): 660 -666 .
[5] 韩普,刘亦卓,李晓艳. 基于深度学习和多特征融合的中文电子病历实体识别研究[J]. 南京大学学报(自然科学版), 2019, 55(6): 942 -951 .
[6] 张家精,夏巽鹏,陈金兰,倪友聪. 基于张量分解和深度学习的混合推荐算法[J]. 南京大学学报(自然科学版), 2019, 55(6): 952 -959 .
[7] 王敏,谢生,毛陆虹,刘一波,杜永超. 一种宽调谐范围的毫米波LC压控振荡器设计[J]. 南京大学学报(自然科学版), 2020, 56(1): 151 -157 .

版权所有 © 2021 《南京大学学报(自然科学)》

地址：江苏省南京市鼓楼区汉口路22号，南京大学学报编辑部，210093

电话：025-83592704 , 传真：025-83592704        技术支持：北京玛格泰克科技发展有限公司