Discriminative log-euclidean kernels for learning on brain networks

Jonathan Young; Du Lei; Andrea Mechelli

doi:10.1007/978-3-319-67159-8_4

Discriminative log-euclidean kernels for learning on brain networks

Jonathan Young^*, Du Lei, Andrea Mechelli

^*Corresponding author for this work

King's College London

Research output: Chapter in Book/Report/Conference proceeding › Other chapter contribution › peer-review

3 Citations (Scopus)

Abstract

The increasing availability of functional Magnetic Resonance Imaging (fMRI) has led to a number of studies of brain networks with the aim of developing computer aided diagnosis of disease. Typically these are based on a statistical or machine learning method operating on connectivity networks, or features derived from them. This work presents a novel kernel method allowing classification tasks on connectivity networks represented as symmetric positive definite (SPD) matrices. It defines a kernel based on geodesic distances measured on the Riemannian manifold of SPD matrices, and automatically adjusts the eigenvalues of the matrices to improve accuracy. This is coupled with a Gaussian Process (GP) classifier, and used to discriminate healthy controls from Schizophrenia patients. The new kernel offers superior classification accuracy to previous kernels, and the adjusted eigenvalues allow discovery of clinically meaningful differences in connectivity between patients and controls.

Original language	English
Title of host publication	Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings
Publisher	Springer Verlag
Pages	25-34
Number of pages	10
Volume	10511 LNCS
ISBN (Print)	9783319671581
DOIs	https://doi.org/10.1007/978-3-319-67159-8_4
Publication status	E-pub ahead of print - 2 Sept 2017
Event	1st International Workshop on Connectomics in NeuroImaging, CNI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017 - Quebec City, Canada Duration: 14 Sept 2017 → 14 Sept 2017

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	10511 LNCS
ISSN (Print)	03029743
ISSN (Electronic)	16113349

Conference

Conference	1st International Workshop on Connectomics in NeuroImaging, CNI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017
Country/Territory	Canada
City	Quebec City
Period	14/09/2017 → 14/09/2017

Access to Document

10.1007/978-3-319-67159-8_4

Cite this

Young, J., Lei, D., & Mechelli, A. (2017). Discriminative log-euclidean kernels for learning on brain networks. In Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings (Vol. 10511 LNCS, pp. 25-34). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10511 LNCS). Springer Verlag. Advance online publication. https://doi.org/10.1007/978-3-319-67159-8_4

Young, Jonathan ; Lei, Du ; Mechelli, Andrea. / Discriminative log-euclidean kernels for learning on brain networks. Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings. Vol. 10511 LNCS Springer Verlag, 2017. pp. 25-34 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inbook{ef8ab9429bc04bbc86dbf1994a3ce055,

title = "Discriminative log-euclidean kernels for learning on brain networks",

abstract = "The increasing availability of functional Magnetic Resonance Imaging (fMRI) has led to a number of studies of brain networks with the aim of developing computer aided diagnosis of disease. Typically these are based on a statistical or machine learning method operating on connectivity networks, or features derived from them. This work presents a novel kernel method allowing classification tasks on connectivity networks represented as symmetric positive definite (SPD) matrices. It defines a kernel based on geodesic distances measured on the Riemannian manifold of SPD matrices, and automatically adjusts the eigenvalues of the matrices to improve accuracy. This is coupled with a Gaussian Process (GP) classifier, and used to discriminate healthy controls from Schizophrenia patients. The new kernel offers superior classification accuracy to previous kernels, and the adjusted eigenvalues allow discovery of clinically meaningful differences in connectivity between patients and controls.",

author = "Jonathan Young and Du Lei and Andrea Mechelli",

year = "2017",

month = sep,

day = "2",

doi = "10.1007/978-3-319-67159-8_4",

language = "English",

isbn = "9783319671581",

volume = "10511 LNCS",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "25--34",

booktitle = "Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings",

address = "Germany",

note = "1st International Workshop on Connectomics in NeuroImaging, CNI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017 ; Conference date: 14-09-2017 Through 14-09-2017",

}

Young, J, Lei, D & Mechelli, A 2017, Discriminative log-euclidean kernels for learning on brain networks. in Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings. vol. 10511 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10511 LNCS, Springer Verlag, pp. 25-34, 1st International Workshop on Connectomics in NeuroImaging, CNI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017, Quebec City, Canada, 14/09/2017. https://doi.org/10.1007/978-3-319-67159-8_4

Discriminative log-euclidean kernels for learning on brain networks. / Young, Jonathan; Lei, Du; Mechelli, Andrea.
Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings. Vol. 10511 LNCS Springer Verlag, 2017. p. 25-34 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10511 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Other chapter contribution › peer-review

TY - CHAP

T1 - Discriminative log-euclidean kernels for learning on brain networks

AU - Young, Jonathan

AU - Lei, Du

AU - Mechelli, Andrea

PY - 2017/9/2

Y1 - 2017/9/2

N2 - The increasing availability of functional Magnetic Resonance Imaging (fMRI) has led to a number of studies of brain networks with the aim of developing computer aided diagnosis of disease. Typically these are based on a statistical or machine learning method operating on connectivity networks, or features derived from them. This work presents a novel kernel method allowing classification tasks on connectivity networks represented as symmetric positive definite (SPD) matrices. It defines a kernel based on geodesic distances measured on the Riemannian manifold of SPD matrices, and automatically adjusts the eigenvalues of the matrices to improve accuracy. This is coupled with a Gaussian Process (GP) classifier, and used to discriminate healthy controls from Schizophrenia patients. The new kernel offers superior classification accuracy to previous kernels, and the adjusted eigenvalues allow discovery of clinically meaningful differences in connectivity between patients and controls.

AB - The increasing availability of functional Magnetic Resonance Imaging (fMRI) has led to a number of studies of brain networks with the aim of developing computer aided diagnosis of disease. Typically these are based on a statistical or machine learning method operating on connectivity networks, or features derived from them. This work presents a novel kernel method allowing classification tasks on connectivity networks represented as symmetric positive definite (SPD) matrices. It defines a kernel based on geodesic distances measured on the Riemannian manifold of SPD matrices, and automatically adjusts the eigenvalues of the matrices to improve accuracy. This is coupled with a Gaussian Process (GP) classifier, and used to discriminate healthy controls from Schizophrenia patients. The new kernel offers superior classification accuracy to previous kernels, and the adjusted eigenvalues allow discovery of clinically meaningful differences in connectivity between patients and controls.

UR - http://www.scopus.com/inward/record.url?scp=85029461346&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-67159-8_4

DO - 10.1007/978-3-319-67159-8_4

M3 - Other chapter contribution

AN - SCOPUS:85029461346

SN - 9783319671581

VL - 10511 LNCS

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 25

EP - 34

BT - Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings

PB - Springer Verlag

T2 - 1st International Workshop on Connectomics in NeuroImaging, CNI 2017 held in conjunction with the 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017

Y2 - 14 September 2017 through 14 September 2017

ER -

Young J, Lei D, Mechelli A. Discriminative log-euclidean kernels for learning on brain networks. In Connectomics in NeuroImaging - 1st International Workshop, CNI 2017 Held in Conjunction with MICCAI 2017, Proceedings. Vol. 10511 LNCS. Springer Verlag. 2017. p. 25-34. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). Epub 2017 Sept 2. doi: 10.1007/978-3-319-67159-8_4

Discriminative log-euclidean kernels for learning on brain networks

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Cite this