On Training Deep 3D CNN Models with Dependent Samples in Neuroimaging

Yunyang Xiong, Hyun Woo Kim, Bhargav Tangirala, Ronak Mehta, Sterling C. Johnson, Vikas Singh

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

There is much interest in developing algorithms based on 3D convolutional neural networks (CNNs) for performing regression and classification with brain imaging data and more generally, with biomedical imaging data. A standard assumption in learning is that the training samples are independently drawn from the underlying distribution. In computer vision, where we have millions of training examples, this assumption is violated but the empirical performance may remain satisfactory. But in many biomedical studies with just a few hundred training examples, one often has multiple samples per participant and/or data may be curated by pooling datasets from a few different institutions. Here, the violation of the independent samples assumption turns out to be more significant, especially in small-to-medium sized datasets. Motivated by this need, we show how 3D CNNs can be modified to deal with dependent samples. We show that even with standard 3D CNNs, there is value in augmenting the network to exploit information regarding dependent samples. We present empirical results for predicting cognitive trajectories (slope and intercept) from morphometric change images derived from multiple time points. With terms which encode dependency between samples in the model, we get consistent improvements over a strong baseline which ignores such knowledge.

Original languageEnglish
Title of host publicationInformation Processing in Medical Imaging - 26th International Conference, IPMI 2019, Proceedings
EditorsSiqi Bao, Albert C.S. Chung, James C. Gee, Paul A. Yushkevich
PublisherSpringer Verlag
Pages99-111
Number of pages13
ISBN (Print)9783030203504
DOIs
Publication statusPublished - 2019 Jan 1
Event26th International Conference on Information Processing in Medical Imaging, IPMI 2019 - Hong Kong, China
Duration: 2019 Jun 22019 Jun 7

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11492 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference26th International Conference on Information Processing in Medical Imaging, IPMI 2019
CountryChina
CityHong Kong
Period19/6/219/6/7

Fingerprint

Neuroimaging
Neural Network Model
Neural networks
Dependent
Imaging techniques
Neural Networks
Computer vision
Brain
Biomedical Imaging
Trajectories
Pooling
Intercept
Training Samples
Computer Vision
Baseline
Slope
Regression
Imaging
Training
Trajectory

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Xiong, Y., Kim, H. W., Tangirala, B., Mehta, R., Johnson, S. C., & Singh, V. (2019). On Training Deep 3D CNN Models with Dependent Samples in Neuroimaging. In S. Bao, A. C. S. Chung, J. C. Gee, & P. A. Yushkevich (Eds.), Information Processing in Medical Imaging - 26th International Conference, IPMI 2019, Proceedings (pp. 99-111). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11492 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-20351-1_8

On Training Deep 3D CNN Models with Dependent Samples in Neuroimaging. / Xiong, Yunyang; Kim, Hyun Woo; Tangirala, Bhargav; Mehta, Ronak; Johnson, Sterling C.; Singh, Vikas.

Information Processing in Medical Imaging - 26th International Conference, IPMI 2019, Proceedings. ed. / Siqi Bao; Albert C.S. Chung; James C. Gee; Paul A. Yushkevich. Springer Verlag, 2019. p. 99-111 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11492 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Xiong, Y, Kim, HW, Tangirala, B, Mehta, R, Johnson, SC & Singh, V 2019, On Training Deep 3D CNN Models with Dependent Samples in Neuroimaging. in S Bao, ACS Chung, JC Gee & PA Yushkevich (eds), Information Processing in Medical Imaging - 26th International Conference, IPMI 2019, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11492 LNCS, Springer Verlag, pp. 99-111, 26th International Conference on Information Processing in Medical Imaging, IPMI 2019, Hong Kong, China, 19/6/2. https://doi.org/10.1007/978-3-030-20351-1_8
Xiong Y, Kim HW, Tangirala B, Mehta R, Johnson SC, Singh V. On Training Deep 3D CNN Models with Dependent Samples in Neuroimaging. In Bao S, Chung ACS, Gee JC, Yushkevich PA, editors, Information Processing in Medical Imaging - 26th International Conference, IPMI 2019, Proceedings. Springer Verlag. 2019. p. 99-111. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-20351-1_8
Xiong, Yunyang ; Kim, Hyun Woo ; Tangirala, Bhargav ; Mehta, Ronak ; Johnson, Sterling C. ; Singh, Vikas. / On Training Deep 3D CNN Models with Dependent Samples in Neuroimaging. Information Processing in Medical Imaging - 26th International Conference, IPMI 2019, Proceedings. editor / Siqi Bao ; Albert C.S. Chung ; James C. Gee ; Paul A. Yushkevich. Springer Verlag, 2019. pp. 99-111 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{fc37bb51da9f49eeb26d923459cb4719,
title = "On Training Deep 3D CNN Models with Dependent Samples in Neuroimaging",
abstract = "There is much interest in developing algorithms based on 3D convolutional neural networks (CNNs) for performing regression and classification with brain imaging data and more generally, with biomedical imaging data. A standard assumption in learning is that the training samples are independently drawn from the underlying distribution. In computer vision, where we have millions of training examples, this assumption is violated but the empirical performance may remain satisfactory. But in many biomedical studies with just a few hundred training examples, one often has multiple samples per participant and/or data may be curated by pooling datasets from a few different institutions. Here, the violation of the independent samples assumption turns out to be more significant, especially in small-to-medium sized datasets. Motivated by this need, we show how 3D CNNs can be modified to deal with dependent samples. We show that even with standard 3D CNNs, there is value in augmenting the network to exploit information regarding dependent samples. We present empirical results for predicting cognitive trajectories (slope and intercept) from morphometric change images derived from multiple time points. With terms which encode dependency between samples in the model, we get consistent improvements over a strong baseline which ignores such knowledge.",
author = "Yunyang Xiong and Kim, {Hyun Woo} and Bhargav Tangirala and Ronak Mehta and Johnson, {Sterling C.} and Vikas Singh",
year = "2019",
month = "1",
day = "1",
doi = "10.1007/978-3-030-20351-1_8",
language = "English",
isbn = "9783030203504",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "99--111",
editor = "Siqi Bao and Chung, {Albert C.S.} and Gee, {James C.} and Yushkevich, {Paul A.}",
booktitle = "Information Processing in Medical Imaging - 26th International Conference, IPMI 2019, Proceedings",

}

TY - GEN

T1 - On Training Deep 3D CNN Models with Dependent Samples in Neuroimaging

AU - Xiong, Yunyang

AU - Kim, Hyun Woo

AU - Tangirala, Bhargav

AU - Mehta, Ronak

AU - Johnson, Sterling C.

AU - Singh, Vikas

PY - 2019/1/1

Y1 - 2019/1/1

N2 - There is much interest in developing algorithms based on 3D convolutional neural networks (CNNs) for performing regression and classification with brain imaging data and more generally, with biomedical imaging data. A standard assumption in learning is that the training samples are independently drawn from the underlying distribution. In computer vision, where we have millions of training examples, this assumption is violated but the empirical performance may remain satisfactory. But in many biomedical studies with just a few hundred training examples, one often has multiple samples per participant and/or data may be curated by pooling datasets from a few different institutions. Here, the violation of the independent samples assumption turns out to be more significant, especially in small-to-medium sized datasets. Motivated by this need, we show how 3D CNNs can be modified to deal with dependent samples. We show that even with standard 3D CNNs, there is value in augmenting the network to exploit information regarding dependent samples. We present empirical results for predicting cognitive trajectories (slope and intercept) from morphometric change images derived from multiple time points. With terms which encode dependency between samples in the model, we get consistent improvements over a strong baseline which ignores such knowledge.

AB - There is much interest in developing algorithms based on 3D convolutional neural networks (CNNs) for performing regression and classification with brain imaging data and more generally, with biomedical imaging data. A standard assumption in learning is that the training samples are independently drawn from the underlying distribution. In computer vision, where we have millions of training examples, this assumption is violated but the empirical performance may remain satisfactory. But in many biomedical studies with just a few hundred training examples, one often has multiple samples per participant and/or data may be curated by pooling datasets from a few different institutions. Here, the violation of the independent samples assumption turns out to be more significant, especially in small-to-medium sized datasets. Motivated by this need, we show how 3D CNNs can be modified to deal with dependent samples. We show that even with standard 3D CNNs, there is value in augmenting the network to exploit information regarding dependent samples. We present empirical results for predicting cognitive trajectories (slope and intercept) from morphometric change images derived from multiple time points. With terms which encode dependency between samples in the model, we get consistent improvements over a strong baseline which ignores such knowledge.

UR - http://www.scopus.com/inward/record.url?scp=85066152007&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85066152007&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-20351-1_8

DO - 10.1007/978-3-030-20351-1_8

M3 - Conference contribution

AN - SCOPUS:85066152007

SN - 9783030203504

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

SP - 99

EP - 111

BT - Information Processing in Medical Imaging - 26th International Conference, IPMI 2019, Proceedings

A2 - Bao, Siqi

A2 - Chung, Albert C.S.

A2 - Gee, James C.

A2 - Yushkevich, Paul A.

PB - Springer Verlag

ER -