Related Experiment Video
Updated: Apr 11, 2026
Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
Randomized structural sparsity via constrained block subsampling for improved sensitivity of discriminative voxel
Yilun Wang1, Junjie Zheng2, Sheng Zhang3
1School of Mathematical Sciences, University of Electronic Science and Technology of China, Chengdu, Sichuan 611731 PR China; Key laboratory for Neuroinformation of Ministry of Education, School of Life Science and Technology, Center for Information in Biomedicine, University of Electronic Science and Technology of China, Chengdu, Sichuan 611054, PR China; Center for Applied Mathematics, Cornell University, Ithaca, NY 14853, USA.
This study introduces randomized structural sparsity for functional Magnetic Resonance Imaging (fMRI) voxel selection. The new method improves biomarker discovery by better identifying correlated brain voxels while controlling false positives and negatives.
Area of Science:
- Neuroimaging
- Biostatistics
- Machine Learning
Background:
- Voxel selection in functional Magnetic Resonance Imaging (fMRI) is challenging due to high dimensionality and limited samples.
- Existing methods like stability selection control false discoveries but miss correlated voxels, leading to high false negative rates.
Purpose of the Study:
- To develop a novel voxel selection method for fMRI data that identifies a more complete set of correlated discriminative voxels.
- To improve the interpretability of potential biomarkers discovered from fMRI data.
Main Methods:
- Proposed a new method called "randomized structural sparsity", a variant of stability selection.
- Incorporated structural sparsity to leverage the correlation property of discriminative features.
Main Results:
- The proposed randomized structural sparsity method demonstrated superior performance in controlling for false negatives.
- The method maintained the control of false positives, a key advantage of stability selection.
Conclusions:
- Randomized structural sparsity offers an improved approach for voxel selection in fMRI.
- This method enhances biomarker discovery by reducing missed relevant voxels and improving interpretability.
