Using multi-voxel pattern analysis of fMRI data to decoding human visual cortex activations

Liye Wang, Xiaoying Tang, Bing Wang, Jinglong Wu, Tianyi Yan

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

Abstract

The potential for human neuroimaging to read-out the detailed contents of a person's mental state has yet to be fully explored. For fMRI decoding, it is important to choose an appropriate set of voxels (or features) as inputs to the decoder, since the presence of many irrelevant voxels could lead to poor generalization performance, a problem known as overfitting. We applied ARD-based sparse Bayesian algorithm to solve overfitting in fMRI classification. The simulated data demonstrated that sparse logistic regression can select effective features through the weight parameters for each class, and most of the selected features lied in the class they belong to. We observed that sparse logistic regression over pruned some effective features under the condition of 80 features, yet it had limited negative impact on the performance of prediction. The reason is perhaps that these over pruned features have similar values for all classes. On the other hand, it indicated that the selected features contained enough information for classification. Another interesting result is that as more irrelevant features were added to the training set, the performance of C-SVM dropped sharply. This demonstrated that sparse logistic regression can outperform C-SVM when irrelevant features are far more than relevant ones. Despite sparse sparse logistic regression wrongly selected some effective features, it had very limited impact on the prediction performance using both correct and wrong features.

Original languageEnglish
Title of host publication2012 ICME International Conference on Complex Medical Engineering, CME 2012 Proceedings
Pages793-798
Number of pages6
DOIs
Publication statusPublished - 2012
Event6th International Conference on Complex Medical Engineering, CME 2012 - Kobe, Japan
Duration: Jul 1 2012Jul 4 2012

Other

Other6th International Conference on Complex Medical Engineering, CME 2012
CountryJapan
CityKobe
Period7/1/127/4/12

Fingerprint

Decoding
Logistics
Chemical activation
Neuroimaging
Magnetic Resonance Imaging

ASJC Scopus subject areas

  • Biomedical Engineering

Cite this

Wang, L., Tang, X., Wang, B., Wu, J., & Yan, T. (2012). Using multi-voxel pattern analysis of fMRI data to decoding human visual cortex activations. In 2012 ICME International Conference on Complex Medical Engineering, CME 2012 Proceedings (pp. 793-798). [6275750] https://doi.org/10.1109/ICCME.2012.6275750

Using multi-voxel pattern analysis of fMRI data to decoding human visual cortex activations. / Wang, Liye; Tang, Xiaoying; Wang, Bing; Wu, Jinglong; Yan, Tianyi.

2012 ICME International Conference on Complex Medical Engineering, CME 2012 Proceedings. 2012. p. 793-798 6275750.

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

Wang, L, Tang, X, Wang, B, Wu, J & Yan, T 2012, Using multi-voxel pattern analysis of fMRI data to decoding human visual cortex activations. in 2012 ICME International Conference on Complex Medical Engineering, CME 2012 Proceedings., 6275750, pp. 793-798, 6th International Conference on Complex Medical Engineering, CME 2012, Kobe, Japan, 7/1/12. https://doi.org/10.1109/ICCME.2012.6275750
Wang L, Tang X, Wang B, Wu J, Yan T. Using multi-voxel pattern analysis of fMRI data to decoding human visual cortex activations. In 2012 ICME International Conference on Complex Medical Engineering, CME 2012 Proceedings. 2012. p. 793-798. 6275750 https://doi.org/10.1109/ICCME.2012.6275750
Wang, Liye ; Tang, Xiaoying ; Wang, Bing ; Wu, Jinglong ; Yan, Tianyi. / Using multi-voxel pattern analysis of fMRI data to decoding human visual cortex activations. 2012 ICME International Conference on Complex Medical Engineering, CME 2012 Proceedings. 2012. pp. 793-798
@inproceedings{82097e7d1ec344fb9e8becc42b9c21cc,
title = "Using multi-voxel pattern analysis of fMRI data to decoding human visual cortex activations",
abstract = "The potential for human neuroimaging to read-out the detailed contents of a person's mental state has yet to be fully explored. For fMRI decoding, it is important to choose an appropriate set of voxels (or features) as inputs to the decoder, since the presence of many irrelevant voxels could lead to poor generalization performance, a problem known as overfitting. We applied ARD-based sparse Bayesian algorithm to solve overfitting in fMRI classification. The simulated data demonstrated that sparse logistic regression can select effective features through the weight parameters for each class, and most of the selected features lied in the class they belong to. We observed that sparse logistic regression over pruned some effective features under the condition of 80 features, yet it had limited negative impact on the performance of prediction. The reason is perhaps that these over pruned features have similar values for all classes. On the other hand, it indicated that the selected features contained enough information for classification. Another interesting result is that as more irrelevant features were added to the training set, the performance of C-SVM dropped sharply. This demonstrated that sparse logistic regression can outperform C-SVM when irrelevant features are far more than relevant ones. Despite sparse sparse logistic regression wrongly selected some effective features, it had very limited impact on the prediction performance using both correct and wrong features.",
author = "Liye Wang and Xiaoying Tang and Bing Wang and Jinglong Wu and Tianyi Yan",
year = "2012",
doi = "10.1109/ICCME.2012.6275750",
language = "English",
isbn = "9781467316163",
pages = "793--798",
booktitle = "2012 ICME International Conference on Complex Medical Engineering, CME 2012 Proceedings",

}

TY - GEN

T1 - Using multi-voxel pattern analysis of fMRI data to decoding human visual cortex activations

AU - Wang, Liye

AU - Tang, Xiaoying

AU - Wang, Bing

AU - Wu, Jinglong

AU - Yan, Tianyi

PY - 2012

Y1 - 2012

N2 - The potential for human neuroimaging to read-out the detailed contents of a person's mental state has yet to be fully explored. For fMRI decoding, it is important to choose an appropriate set of voxels (or features) as inputs to the decoder, since the presence of many irrelevant voxels could lead to poor generalization performance, a problem known as overfitting. We applied ARD-based sparse Bayesian algorithm to solve overfitting in fMRI classification. The simulated data demonstrated that sparse logistic regression can select effective features through the weight parameters for each class, and most of the selected features lied in the class they belong to. We observed that sparse logistic regression over pruned some effective features under the condition of 80 features, yet it had limited negative impact on the performance of prediction. The reason is perhaps that these over pruned features have similar values for all classes. On the other hand, it indicated that the selected features contained enough information for classification. Another interesting result is that as more irrelevant features were added to the training set, the performance of C-SVM dropped sharply. This demonstrated that sparse logistic regression can outperform C-SVM when irrelevant features are far more than relevant ones. Despite sparse sparse logistic regression wrongly selected some effective features, it had very limited impact on the prediction performance using both correct and wrong features.

AB - The potential for human neuroimaging to read-out the detailed contents of a person's mental state has yet to be fully explored. For fMRI decoding, it is important to choose an appropriate set of voxels (or features) as inputs to the decoder, since the presence of many irrelevant voxels could lead to poor generalization performance, a problem known as overfitting. We applied ARD-based sparse Bayesian algorithm to solve overfitting in fMRI classification. The simulated data demonstrated that sparse logistic regression can select effective features through the weight parameters for each class, and most of the selected features lied in the class they belong to. We observed that sparse logistic regression over pruned some effective features under the condition of 80 features, yet it had limited negative impact on the performance of prediction. The reason is perhaps that these over pruned features have similar values for all classes. On the other hand, it indicated that the selected features contained enough information for classification. Another interesting result is that as more irrelevant features were added to the training set, the performance of C-SVM dropped sharply. This demonstrated that sparse logistic regression can outperform C-SVM when irrelevant features are far more than relevant ones. Despite sparse sparse logistic regression wrongly selected some effective features, it had very limited impact on the prediction performance using both correct and wrong features.

UR - http://www.scopus.com/inward/record.url?scp=84867635990&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84867635990&partnerID=8YFLogxK

U2 - 10.1109/ICCME.2012.6275750

DO - 10.1109/ICCME.2012.6275750

M3 - Conference contribution

AN - SCOPUS:84867635990

SN - 9781467316163

SP - 793

EP - 798

BT - 2012 ICME International Conference on Complex Medical Engineering, CME 2012 Proceedings

ER -