南京大学学报(自然科学版) ›› 2021, Vol. 57 ›› Issue (5): 801–809.doi: 10.13232/j.cnki.jnju.2021.05.010

• • 上一篇    

基于图同构网络的自闭症功能磁共振影像诊断算法

张礼(), 王嘉瑞   

  1. 南京林业大学信息科学技术学院，南京，210016
  • 收稿日期:2021-06-16 出版日期:2021-09-29 发布日期:2021-09-29
  • 通讯作者: 张礼 E-mail:lizhang@njfu.edu.cn
  • 作者简介:E⁃mail：lizhang@njfu.edu.cn
  • 基金资助:
    国家自然科学青年科学基金(61802193);江苏省自然科学基金(BK20170934);南京林业大学青年科技创新基金(CX2017031)

Diagnosing autism spectrum disorder from functional MRI using graph isomorphic network

Li Zhang(), Jiarui Wang   

  1. College of Computer Science and Technology，Nanjing Forestry University, Nanjing，210016，China
  • Received:2021-06-16 Online:2021-09-29 Published:2021-09-29
  • Contact: Li Zhang E-mail:lizhang@njfu.edu.cn

RichHTML

134

PDF (PC)

771

摘要：

图表示法通常用于个人或者总体级别上对结构化数据进行建模分析，已成功应用于网络分析、交通预测、推荐系统等领域.随着成像设备的发展和普及，从神经影像中学习脑的连接特性，开展基于脑网络的疾病诊断（自闭症、阿斯海默症等）受到广泛关注.图表示法可用于对一组大脑区域之间的结构或功能连接进行建模，揭示与大脑发育和疾病有关的模式，然而评估基于图结构的脑连接网络之间相似性并非易事.传统的深度学习方法无法适用图结构，会丢弃有益于图分类任务的信息，因此提出一个基于图同构网络的自闭症功能磁共振影像的诊断算法.该模型包含四层同构层，每层通过空间领域卷积学习得到脑功能连接网络的特征表示.为了考虑脑功能连接网络中节点的医学意义，将节点特征通过展平方式转换为图特征.在自闭症ABIDE数据库上对提出的方法进行验证，与图卷积网络和深度神经网络相比，实验结果证明提出的方法是有效的，明显提升了自闭症诊断准确性.

关键词: 自闭症, 脑功能连接网络, 图同构网络, 功能磁共振影像

Abstract:

Graph representations are usually used to model and analyze structured data on an individual or population level，and are successfully applied in network analysis，traffic forecasting，recommendation systems and other fields. With the development of imaging equipment，learning connectivity characteristics of brain from neuroimaging data has received widespread attention for brain disorders diagnosis (such as Autism Spectrum Disorder (ASD)，Alzheimer's Disease，etc.) based on brain network. Graph representations are able to model the structural or functional connections between a group of brain regions，and to reveal patterns related to brain development and disease. However，evaluating the similarity between theses brain connectivity networks is non?trivial. Traditional deep learning methods cannot adapt to graph structures and discard some useful information for graph classification tasks. Therefore，we propose a model based on graph isomorphic network (GIN) to ASD diagnosis using fMRI. This model contains four isomorphic layers，each of which obtains the feature representation of the brain functional connectivity network through spatial?based convolutions. To account for the medical meaning of nodes in the brain network，the node features are transformed into graph features by a flatten layer. Compared with Graph Convolutional Network (GCN) and Deep Neural Network (DNN)，the experimental results on ABIDE database show that our proposed method is effective and significantly improves the accuracy of ASD diagnosis.

Key words: Autism Spectrum Disorder (ASD), brain functional connectivity, Graph Isomorphism Network (GIN), fMRI

中图分类号: 

  • TP391

图1

GIN层中节点更新过程"

图2

READOUT函数对节点图特征影响的示例：(A)相同颜色表示图中节点无明确意义；(B)不同颜色表示节点具有不同意义，圆点中数值表示节点的激活值"

图3

GIN模型的框架结构图"

表1

ABIDE数据集的统计信息"

SiteASDNC
AgeSex (M/F)AgeSex (M/F)
ALL16.86±7.36382/8617.70±8.93350/57
NYU14.95±7.0763/915.67±6.1872/26
UM13.84±2.2939/915.30±3.5344/15
UCLA13.33±2.5534/213.17±1.7633/6

表2

不同方法在全中心数据上的分类结果"

ACCSENSPEBAC
GIN74.20%±2.92%77.59%±6.95%72.49%±5.32%75.04%±2.66%
GCN71.39%±2.62%73.58%±6.37%70.74%±7.46%72.16%±2.18%
DNN71.65%±4.23%72.29%±6.13%70.63%±3.39%71.46%±3.91%

图4

不同方法在全中心数据上的AUC"

图5

不同方法在三个中心数据上的AUC"

表3

不同方法在三个分中心数据的分类精度"

UCLAUMNYU
GIN76.00%74.76%76.47%
GCN73.33%73.83%72.35%
DNN73.33%75.70%72.94%

表4

不同GIN层数的分类精度"

GIN层数3456
分类精度70.88%±3.25%74.20%±2.92%72.41%±2.92%72.80%±3.25%

表5

使用不同READOUT函数的分类精度"

函数FLATTENAVGSUMMAX
分类精度74.20%±2.92%67.31%±4.09%67.56%±3.54%66.16%±3.30%

图6

不同阈值下GIN模型的分类精度"

1 Lord C，Brugha T S，Charman T，et al. Autism spectrum disorder. Nature Reviews Disease Primers，2020，6(1)：5.
2 陈顺森，白学军，张日昇. 自闭症谱系障碍的症状、诊断与干预. 心理科学进展，2011，19(1)：60-72. （Chen S S，Bai X J，Zhang R S. The symptom，
diagnosis and treatment for autism spectrum disorder. Advances in Psychological Science，2011，19(1)：60-72.
3 邓明昱，劳世艳. 自闭症谱系障碍的临床研究新进展(DSM?5新标准). 中国健康心理学杂志，2016，
24(4)：481-490. （Deng M L，Lao S Y. New
progress of clinical research to autistic spectrum
(DSM? disorder5Update). China Journal of Health Psychology，2016，24(4)：481-490.
4 Shen D G，Wu G R，Suk H I. Deep learning in medical image analysis. Annual Review of Biomedical Engineering，2017(19)：221-248.
5 Litjens G，Kooi T，Bejnordi B E，et al. A survey on deep learning in medical image analysis. Medical Image Analysis，2017(42)：60-88.
6 Kumar E S，Bindu C S. Medical image analysis using deep learning：A systematic literature review∥The 2nd International Conference on Emerging Technologies in Computer Engineering. Springer Berlin Heidelberg，2019：81-97.
7 Zhang L，Wang M L，Liu M X，et al. A survey on deep learning for neuroimaging?based brain disorder analysis. Frontiers in Neuroscience，2020(14)：630.
8 Kong Y Z，Gao J L，Xu Y P，et al. Classification of autism spectrum disorder by combining brain connectivity and deep neural network classifier. Neurocomputing，2019(324)：63-68.
9 Guo X Y，Dominick K C，Minai A A，et al. Diagnosing autism spectrum disorder from brain resting?state functional connectivity patterns using a deep neural network with a novel feature selection method. Frontiers in Neuroscience，2017(11)：460.
10 Choi H. Functional connectivity patterns of autism spectrum disorder identified by deep feature learning. 2017,arXiv：.
11 Eslami T，Mirjalili V，Fong A，et al. ASD?DiagNet：A hybrid learning approach for detection of autism spectrum disorder using fMRI data. Frontiers in Neuroscience，2019(13)：70.
12 Li X X，Dvornek N C，Papademetris X，et al. 2?channel convolutional 3D deep neural network (2CC3D) for fMRI analysis：ASD classi?cation and feature learning∥2018 IEEE 15th International Symposium on Biomedical Imaging. Washington，DC,USA：IEEE，2018：1252-1255.
13 Khosla M，Jamison K，Kuceyeski A，et al. 3D convolutional neural networks for classi?cation of functional connectomes∥Stoyanov D. Deep learning in medical image analysis and multimodal learning for clinical decision support. Springer Berlin Heidelberg，2018：137-145.
14 Zhang S，Tong H H，Xu J J，et al. Graph convolutional networks：A comprehensive review. Computational Social Networks，2019，6(1)：11.
15 Wu Z H，Pan S R，Chen F W，et al. A comprehensive survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems，2021，32(1)：4-24.
16 Zhang Z W，Peng C，Zhu W W. Deep learning on graphs：A survey. IEEE Transactions on Knowledge and Data Engineering，2020，doi：10.1109/TKDE. 2020.2981333.
17 Ktena S I，Parisot S，Ferrante E，et al. Metric learning with spectral graph convolutions on brain connectivity networks. NeuroImage，2018(169)：431-442.
18 Parisot S，Ktena S I，Ferrante E，et al. Spectral graph convolutions for population?based disease prediction∥20th International Conference on Medical Image Computing and Computer:Assisted Intervention. Springer Berlin Heidelberg，2017：177-185.
19 Yao D R，Liu M X，Wang M L，et al. Triplet graph convolutional network for multi?scale analysis of functional connectivity using functional MRI∥The 1st International Workshop on Graph Learning in Medical Imaging. Springer Berlin Heidelberg，2017：70-78.
20 Anirudh R，Thiagarajan J J. Bootstrapping graph convolutional neural networks for autism spectrum disorder classi?cation∥2019 IEEE International Conference on Acoustics，Speech and Signal Processing. Brighton，UK：IEEE，2019：3197-3201.
21 You J X，Gomes?Selman J，Ying R，et al. Identity?aware graph neural networks. 2021,arXiv:.
22 Kim B H，Ye J C. Understanding graph isomorphism network for rs?fMRI functional connectivity analysis. Frontiers in Neuroscience，2020(14)：630.
23 Xu K L，Hu W H，Leskovec J，et al. How powerful are graph neural networks?2018,arXiv:.
24 Kipf T N，Welling M. Semi?supervised classification with graph convolutional networks. 2016,arXiv：.
25 Hamilton W L，Rex Y，Leskovec J. Inductive representation learning on large graphs∥Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook，NY，USA：Curran Associates Inc.，2017：1025-1035.
26 Di Martino A，Yan C G，Li Q，et al. The autism brain imaging data exchange：Towards a large?scale evaluation of the intrinsic brain architecture in autism. Molecular Psychiatry，2014，19(6)：659-667.
27 Craddock C，Sikka S，Cheung B，et al. Towards automated analysis of connectomes：The configurable pipeline for the analysis of connectomes (C?PAC). Frontiers in Neuroinformatics，2013，doi:10.3389/conf.fninf.2013.09.00042.
28 Zanin M，Sousa P，Papo D，et al. Optimizing functional network representation of multivariate time series. Scientific reports，2012(2)：630.
[1] 张强, 杨吉斌, 张雄伟, 曹铁勇, 梅鹏程. 基于生成对抗网络的音频目标分类对抗[J]. 南京大学学报(自然科学版), 2021, 57(5): 793-800.
[2] 李娜, 段友祥, 孙歧峰, 沈楠. 一种基于样本点距离突变的聚类方法[J]. 南京大学学报(自然科学版), 2021, 57(5): 775-784.
[3] 倪斌, 陆晓蕾, 童逸琦, 马涛, 曾志贤. 胶囊神经网络在期刊文本分类中的应用[J]. 南京大学学报(自然科学版), 2021, 57(5): 750-756.
[4] 陈磊, 孙权森, 王凡海. 基于深度对抗网络和局部模糊探测的目标运动去模糊[J]. 南京大学学报(自然科学版), 2021, 57(5): 735-749.
[5] 卢锦亮, 吴广潮, 冯夫健, 王林. 基于联合轨迹特征的徘徊行为识别方法[J]. 南京大学学报(自然科学版), 2021, 57(5): 724-734.
[6] 薛峰, 李凡, 李爽, 李华锋. 基于域分离和对抗学习的跨域行人重识别[J]. 南京大学学报(自然科学版), 2021, 57(5): 715-723.
[7] 贾霄, 郭顺心, 赵红. 基于图像属性的零样本分类方法综述[J]. 南京大学学报(自然科学版), 2021, 57(4): 531-543.
[8] 吴礼福, 徐行. 融合韵律与动态倒谱特征的语音疲劳度检测[J]. 南京大学学报(自然科学版), 2021, 57(4): 709-714.
[9] 颜志良, 丰智鹏, 刘丹, 王会青. 一种混合深度神经网络的赖氨酸乙酰化位点预测方法[J]. 南京大学学报(自然科学版), 2021, 57(4): 627-640.
[10] 王发旺, 陈睿, 伏云发. 基于DBN和RF的跨被试情绪识别研究[J]. 南京大学学报(自然科学版), 2021, 57(4): 617-626.
[11] 段建设, 崔超然, 宋广乐, 马乐乐, 马玉玲, 尹义龙. 基于多尺度注意力融合的知识追踪方法[J]. 南京大学学报(自然科学版), 2021, 57(4): 591-598.
[12] 杜淑颖, 侯海薇, 丁世飞. 基于多层次特征的深度集成聚类算法[J]. 南京大学学报(自然科学版), 2021, 57(4): 575-581.
[13] 黄鹤, 吴琨, 宋京, 王会峰, 茹锋, 郭璐. 融合全局与区域大气光值图的暗通道图像去雾方法[J]. 南京大学学报(自然科学版), 2021, 57(4): 551-565.
[14] 乔宇鑫, 葛洪伟. 自适应样本加权的多视图聚类算法[J]. 南京大学学报(自然科学版), 2021, 57(4): 544-550.
[15] 王晓琳, 赵磊, 张维, 伏云发. 基于HHT和核函数选择的情绪特征提取与识别[J]. 南京大学学报(自然科学版), 2021, 57(3): 502-511.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!

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