Explainable Classification of Brain Networks via Contrast Subgraphs

KDD '20: The 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining Virtual Event CA USA July, 2020, pp. 3308-3318, 2020.

Cited by: 0|Bibtex|Views45|DOI:https://doi.org/10.1145/3394486.3403383
EI
Other Links: arxiv.org|dl.acm.org|academic.microsoft.com|dblp.uni-trier.de
Weibo:
We evaluate the effectiveness of contrast subgraphs in classifying brain networks according to the target classes Typically Developed and Autism Spectrum Disorder

Abstract:

Mining human-brain networks to discover patterns that can be used to discriminate between healthy individuals and patients affected by some neurological disorder, is a fundamental task in neuro-science. Learning simple and interpretable models is as important as mere classification accuracy. In this paper we introduce a novel approach for...More

Code:

Data:

0
Introduction
  • The development of magnetic resonance imaging (MRI) techniques has paved the way to connectomics [32], i.e., modeling the brain as a network, allowing to tackle interesting neuroscience research questions as graph-analysis problems [6].

    A connectome is a map describing neural connections between brain regions of interest (ROIs), either by observing anatomic fiber density, or by computing pairwise correlations between time series of activity associated to ROIs.
  • The development of magnetic resonance imaging (MRI) techniques has paved the way to connectomics [32], i.e., modeling the brain as a network, allowing to tackle interesting neuroscience research questions as graph-analysis problems [6].
  • Given fMRI scans of patients affected by a mental disorder and scans of healthy individuals, the goal is to discover patterns in the corresponding connectomes that explain differences in the brain mechanism of the two groups
Highlights
  • The development of magnetic resonance imaging (MRI) techniques has paved the way to connectomics [32], i.e., modeling the brain as a network, allowing to tackle interesting neuroscience research questions as graph-analysis problems [6].

    A connectome is a map describing neural connections between brain regions of interest (ROIs), either by observing anatomic fiber density, or by computing pairwise correlations between time series of activity associated to regions of interest
  • Given functional magnetic resonance imaging scans of patients affected by a mental disorder and scans of healthy individuals, the goal is to discover patterns in the corresponding connectomes that explain differences in the brain mechanism of the two groups
  • The results are shown in the scatter plot of Figure 6, where we show each graph Gi with value ∥Gi [S]−GTD[S]∥1 on the x-axis and value ∥Gi [S] − GASD[S]∥1 on the y-axis Here, G[S] denotes the subgraph induced by S in G
  • We evaluate the effectiveness of contrast subgraphs in classifying brain networks according to the target classes Typically Developed and Autism Spectrum Disorder
  • Learning models that are able to discriminate brains of patients affected by a mental disorder from those of healthy individuals, is attracting a lot of interest
  • The framework we propose, based on the notion of contrast subgraph, satisfies both of these conditions
Methods
  • The aim of the experimental evaluation is to show how contrast subgraphs can be exploited profitably for finding discriminative patterns between two groups of brain networks.
  • (1) Can the contrast-subgraph approach be used to identify structural differences between two groups of brain-networks, which are not easy to detect through standard analysis?.
  • Since the quantity of signals detected by any MRI is huge, there is usually a process of aggregating voxels, so as to reduce the data dimensionality
  • Since such signals are heavily subject to noise caused by different confounding factors, a plethora of preprocessing strategies, either from the signal-processing side or from the network-analysis side have been proposed.
  • The dataset contains neuroimaging data of 1112 different patients, 573 Typically Developed (TD) and 539 suffering from Autism Spectrum Disorder (ASD)
Conclusion
  • CONCLUSIONS AND FUTURE WORK

    Learning models that are able to discriminate brains of patients affected by a mental disorder from those of healthy individuals, is attracting a lot of interest.
  • In this paper the authors approach the task of brain-network classification with a two-fold goal: to achieve good accuracy, but most importantly, to identify discriminant brain patterns that lead a model to classify an individual.
  • The framework the authors propose, based on the notion of contrast subgraph, satisfies both of these conditions.
  • It outperforms several state-of-the-art competitors and returns some very intuitive explanations, which can be shared with experts from the neuroscience field.
  • Contrast subgraphs are exceptionally easy to compute, both in terms of runtime and memory
Summary
  • Introduction:

    The development of magnetic resonance imaging (MRI) techniques has paved the way to connectomics [32], i.e., modeling the brain as a network, allowing to tackle interesting neuroscience research questions as graph-analysis problems [6].

    A connectome is a map describing neural connections between brain regions of interest (ROIs), either by observing anatomic fiber density, or by computing pairwise correlations between time series of activity associated to ROIs.
  • The development of magnetic resonance imaging (MRI) techniques has paved the way to connectomics [32], i.e., modeling the brain as a network, allowing to tackle interesting neuroscience research questions as graph-analysis problems [6].
  • Given fMRI scans of patients affected by a mental disorder and scans of healthy individuals, the goal is to discover patterns in the corresponding connectomes that explain differences in the brain mechanism of the two groups
  • Methods:

    The aim of the experimental evaluation is to show how contrast subgraphs can be exploited profitably for finding discriminative patterns between two groups of brain networks.
  • (1) Can the contrast-subgraph approach be used to identify structural differences between two groups of brain-networks, which are not easy to detect through standard analysis?.
  • Since the quantity of signals detected by any MRI is huge, there is usually a process of aggregating voxels, so as to reduce the data dimensionality
  • Since such signals are heavily subject to noise caused by different confounding factors, a plethora of preprocessing strategies, either from the signal-processing side or from the network-analysis side have been proposed.
  • The dataset contains neuroimaging data of 1112 different patients, 573 Typically Developed (TD) and 539 suffering from Autism Spectrum Disorder (ASD)
  • Conclusion:

    CONCLUSIONS AND FUTURE WORK

    Learning models that are able to discriminate brains of patients affected by a mental disorder from those of healthy individuals, is attracting a lot of interest.
  • In this paper the authors approach the task of brain-network classification with a two-fold goal: to achieve good accuracy, but most importantly, to identify discriminant brain patterns that lead a model to classify an individual.
  • The framework the authors propose, based on the notion of contrast subgraph, satisfies both of these conditions.
  • It outperforms several state-of-the-art competitors and returns some very intuitive explanations, which can be shared with experts from the neuroscience field.
  • Contrast subgraphs are exceptionally easy to compute, both in terms of runtime and memory
Tables
  • Table1: Datasets used in the experiments
  • Table2: Results of the experiments performed over the datasets described in Table 1. Each value represents the average accuracy, along with its relative standard deviation
Download tables as Excel
Related work
  • Graph classification. Graph classification, i.e., the task of building a model able to predict the target class for unseen graphs accurately, is receiving increasing attention as witnessed by the many approaches proposed in the last few years in the literature.

    Shervashidze et al [31], propose a graph kernel based on the Weisfeiler-Lehman test of graph isomorphism. Narayanan et al [27] propose graph2vec, a method that considers rooted subgraphs as the components that define a specific graph and performs an embedding under this assumption. Adhikari et al [1] propose sub2vec, whose aim is to learn subgraph embeddings, preserving structural and neighborhood properties. In our experimental comparison in
Funding
  • Francesco Bonchi acknowledges support from Intesa Sanpaolo Innovation Center
  • Aristides Gionis is supported by three Academy of Finland projects (286211, 313927, 317085), the ERC Advanced Grant REBOUND (834862), the EC H2020 RIA project “SoBigData++” (871042), and the Wallenberg AI, Autonomous Systems and Software Program (WASP)
Reference
  • B. Adhikari, Y. Zhang, N. Ramakrishnan, and B. A. Prakash. Sub2vec: Feature learning for subgraphs. In PAKDD 2018.
    Google ScholarLocate open access versionFindings
  • Y. Asahiro, K. Iwama, H. Tamaki, and T. Tokuyama. Greedily finding a dense subgraph. Journal of Algorithms, 34(2):203–221, 2000.
    Google ScholarLocate open access versionFindings
  • O. D. Balalau, F. Bonchi, T. H. Chan, F. Gullo, and M. Sozio. Finding subgraphs with maximum total density and limited overlap. In WSDM 2015.
    Google ScholarLocate open access versionFindings
  • B. Biswal, F. Z. Yetkin, V. M. Haughton, and J. S. Hyde. Functional connectivity in the motor cortex of resting human brain using echo-planar mri. Magnetic Resonance in Medicine, 34(4):537–541, oct 1995.
    Google ScholarLocate open access versionFindings
  • B. Boden, S. Günnemann, H. Hoffmann, and T. Seidl. Mining coherent subgraphs in multi-layer graphs with edge labels. In KDD 2012.
    Google ScholarLocate open access versionFindings
  • E. Bullmore and O. Sporns. Complex brain networks: graph theoretical analysis of structural and functional systems. Nature reviews neuroscience, 10(3):186–198, 2009.
    Google ScholarLocate open access versionFindings
  • J. Cadena, A. K. Vullikanti, and C. C. Aggarwal. On dense subgraphs in signed network streams. In ICDM 2016.
    Google ScholarLocate open access versionFindings
  • L. Chu, S. Wang, S. Liu, Q. Huang, and J. Pei. ALID: scalable dominant cluster detection. PVLDB, 8(8):826–837, 2015.
    Google ScholarLocate open access versionFindings
  • C. Craddock, Y. Benhajali, C. Chu, F. Chouinard, A. Evans, A. Jakab, B. S. Khundrakpam, J. D. Lewis, Q. Li, M. Milham, C. Yan, and P. Bellec. The neuro bureau preprocessing initiative: open sharing of preprocessed neuroimaging data and derivatives. Frontiers in Neuroinformatics, (41), 2013.
    Google ScholarLocate open access versionFindings
  • Y. Du, Z. Fu, and V. D. Calhoun. Classification and prediction of brain disorders using functional connectivity: Promising but challenging. Frontiers in Neuroscience, 12:525, 2018.
    Google ScholarLocate open access versionFindings
  • L. T. Eyler, K. Pierce, and E. Courchesne. A failure of left temporal cortex to specialize for language is an early emerging and fundamental property of autism. Brain, 135(3):949–960, Feb. 2012.
    Google ScholarLocate open access versionFindings
  • E. Galimberti, F. Bonchi, and F. Gullo. Core decomposition and densest subgraph in multilayer networks. In CIKM 2017.
    Google ScholarLocate open access versionFindings
  • S. J. Gilbert, G. Bird, R. Brindley, C. D. Frith, and P. W. Burgess. Atypical recruitment of medial prefrontal cortex in autism spectrum disorders: An fMRI study of two executive function tasks. Neuropsychologia, 46(9):2281–2291, July 2008.
    Google ScholarLocate open access versionFindings
  • A. Gionis and C. E. Tsourakakis. Dense subgraph discovery: Kdd 2015 tutorial. In KDD 2015.
    Google ScholarLocate open access versionFindings
  • A. V. Goldberg. Finding a maximum density subgraph. Technical report, University of California at Berkeley, 1984.
    Google ScholarFindings
  • L. Gutiérrez-Gómez and J.-C. Delvenne. Unsupervised network embeddings with node identity awareness. Applied Network Science, 4(1), oct 2019.
    Google ScholarLocate open access versionFindings
  • D. Jiang and J. Pei. Mining frequent cross-graph quasi-cliques. TKDD, 2(4):16, 2009.
    Google ScholarLocate open access versionFindings
  • N. Jin and W. Wang. Lts: Discriminative subgraph mining by learning from search history. In ICDE 2011.
    Google ScholarLocate open access versionFindings
  • A. J. Khan, A. Nair, C. L. Keown, M. C. Datko, A. J. Lincoln, and R.-A. Müller. Cerebro-cerebellar resting-state functional connectivity in children and adolescents with autism spectrum disorder. Biological Psychiatry, 78(9):625 – 634, 2015.
    Google ScholarLocate open access versionFindings
  • N. M. Kleinhans, R.-A. Müller, D. N. Cohen, and E. Courchesne. Atypical functional lateralization of language in autism spectrum disorders. Brain Research, 1221:115 – 125, 2008.
    Google ScholarLocate open access versionFindings
  • E. W. Lang, A. M. Tomé, I. R. Keck, J. M. Górriz-Sáez, and C. G. Puntonet. Brain connectivity analysis: A short survey. Computational Intelligence and Neuroscience, 2012:1–21, 2012.
    Google ScholarLocate open access versionFindings
  • J. B. Lee, R. Rossi, and X. Kong. Graph classification using structural attention. In KDD 2018.
    Google ScholarLocate open access versionFindings
  • A. Lord, D. Horn, M. Breakspear, and M. Walter. Changes in community structure of resting state functional connectivity in unipolar depression. PLOS ONE, 7(8):1– 15, 08 2012.
    Google ScholarLocate open access versionFindings
  • A. D. Martino, C. Kelly, R. Grzadzinski, X.-N. Zuo, M. Mennes, M. A. Mairena, C. Lord, F. X. Castellanos, and M. P. Milham. Aberrant striatal functional connectivity in children with autism. Biological Psychiatry, 69(9):847 – 856, 2011.
    Google ScholarLocate open access versionFindings
  • L. Meng and J. Xiang. Brain network analysis and classification based on convolutional neural network. Frontiers in Computational Neuroscience, 12, dec 2018.
    Google ScholarLocate open access versionFindings
  • M. F. Misman, A. A. Samah, F. A. Ezudin, H. A. Majid, Z. A. Shah, H. Hashim, and M. F. Harun. Classification of adults with autism spectrum disorder using deep neural network. In 2019 1st International Conference on Artificial Intelligence and Data Sciences (AiDAS), pages 29–34, Sep. 2019.
    Google ScholarLocate open access versionFindings
  • A. Narayanan, M. Chandramohan, R. Venkatesan, L. Chen, Y. Liu, and S. Jaiswal. graph2vec: Learning distributed representations of graphs, 2017.
    Google ScholarFindings
  • J. A. Nielsen, B. A. Zielinski, P. Fletcher, A. L. Alexander, N. Lange, E. D. Bigler, J. E. Lainhart, and J. S. Anderson. Abnormal lateralization of functional connectivity between language and default mode regions in autism. Molecular Autism, 5(1):8, 2014.
    Google ScholarLocate open access versionFindings
  • J. Pei, D. Jiang, and A. Zhang. On mining cross-graph quasi-cliques. In KDD 2005.
    Google ScholarLocate open access versionFindings
  • M. Rubinov, S. A. Knock, C. J. Stam, S. Micheloyannis, A. W. Harris, L. M. Williams, and M. Breakspear. Small-world properties of nonlinear brain activity in schizophrenia. Human Brain Mapping, 30(2):403–416, 2009.
    Google ScholarLocate open access versionFindings
  • N. Shervashidze, P. Schweitzer, E. J. v. Leeuwen, K. Mehlhorn, and K. M. Borgwardt. Weisfeiler-lehman graph kernels. Journal of Machine Learning Research, 12(Sep):2539–2561, 2011.
    Google ScholarLocate open access versionFindings
  • O. Sporns, G. Tononi, and R. Kötter. The human connectome: a structural description of the human brain. PLoS Comput Biol, 1(4), 2005.
    Google ScholarLocate open access versionFindings
  • M. Thoma, H. Cheng, A. Gretton, J. Han, H.-P. Kriegel, A. Smola, L. Song, P. S. Yu, X. Yan, and K. M. Borgwardt. Discriminative frequent subgraph mining with optimality guarantees. Statistical Analysis and Data Mining: The ASA Data Science Journal, 3(5):302–318, 2010.
    Google ScholarLocate open access versionFindings
  • R. M. H. Ting and J. Bailey. Mining minimal contrast subgraph patterns. In Proceedings of the 2006 SIAM International Conference on Data Mining, pages 639–643, 2006.
    Google ScholarLocate open access versionFindings
  • C. Tsourakakis. Streaming Graph Partitioning in the Planted Partition Model. In COSN 2015.
    Google ScholarLocate open access versionFindings
  • C. Tsourakakis, F. Bonchi, A. Gionis, F. Gullo, and M. Tsiarli. Denser than the densest subgraph: extracting optimal quasi-cliques with quality guarantees. In KDD 2013.
    Google ScholarLocate open access versionFindings
  • N. Tzourio-Mazoyer, B. Landeau, D. Papathanassiou, F. Crivello, O. Etard, N. Delcroix, B. Mazoyer, and M. Joliot. Automated anatomical labeling of activations in spm using a macroscopic anatomical parcellation of the mni mri single-subject brain. NeuroImage, 15(1):273 – 289, 2002.
    Google ScholarLocate open access versionFindings
  • S. Wang, L. He, B. Cao, C.-T. Lu, P. S. Yu, and A. B. Ragin. Structural deep brain network mining. In KDD 2017.
    Google ScholarLocate open access versionFindings
  • J. Wu, J. He, and J. Xu. Demo-net: Degree-specific graph neural networks for node and graph classification. In KDD 2019.
    Google ScholarLocate open access versionFindings
  • Y. Wu, X. Zhu, L. Li, W. Fan, R. Jin, and X. Zhang. Mining dual networks: Models, algorithms, and applications. TKDD, 10(4):40:1–40:37, 2016.
    Google ScholarLocate open access versionFindings
  • X. Yan, H. Cheng, J. Han, and P. S. Yu. Mining significant graph patterns by leap search. In SIGMOD 2008.
    Google ScholarLocate open access versionFindings
  • Y. Yan, J. Zhu, M. Duda, E. Solarz, C. Sripada, and D. Koutra. Groupinn: Groupingbased interpretable neural network for classification of limited, noisy brain data. In KDD 2019.
    Google ScholarLocate open access versionFindings
  • P. Yanardag and S. Vishwanathan. Deep graph kernels. In KDD 2015.
    Google ScholarLocate open access versionFindings
  • Y. Yang, L. Chu, Y. Zhang, Z. Wang, J. Pei, and E. Chen. Mining density contrast subgraphs. In ICDE 2018. All our code and datasets are made available at: https://github.com/tlancian/contrast-subgraph
    Findings
  • CS-P1 Source: https://github.com/tlancian/contrast-subgraph Parameters tuned:
    Findings
  • CS-P2 Source: https://github.com/tlancian/contrast-subgraph Parameters tuned:
    Findings
  • [27] Source: https://github.com/MLDroid/graph2vec_tf Parameters tuned:
    Findings
  • [1] Source: https://goo.gl/Ef4q8g
    Findings
  • [16] Source: https://github.com/leoguti85/GraphEmbs
    Findings
  • [31] Source: http://mlcb.is.tuebingen.mpg.de/Mitarbeiter/Nino/WL Parameters tuned:
    Findings
Full Text
Your rating :
0

 

Tags
Comments