Related Experiment Video
Updated: Dec 21, 2025
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
16.0K
Designing weighted correlation kernels in convolutional neural networks for functional connectivity based brain
Biao Jie1, Mingxia Liu2, Chunfeng Lian2
1School of Computer and Information, Anhui Normal University, Anhui 241003, China; Department of Radiology and BRIC, University of North Carolina at Chapel Hill, North Carolina 27599, USA.
Medical Image Analysis
|May 18, 2020
Summary
This study introduces a novel weighted correlation kernel and CNN framework to improve brain disease diagnosis using fMRI data. The method captures richer interaction information for more accurate identification of conditions like Alzheimer's disease.
Area of Science:
- Neuroimaging
- Machine Learning
- Medical Diagnostics
Background:
- Functional connectivity networks (FCNs) derived from fMRI are crucial for diagnosing brain disorders like Alzheimer's disease (AD) and mild cognitive impairment (MCI).
- Traditional methods using Pearson correlation coefficient (PCC) for FCN construction overlook time-varying information and high-level network features.
Purpose of the Study:
- To develop a novel weighted correlation kernel (wc-kernel) for enhanced FCN construction, capturing dynamic temporal contributions.
- To propose a wc-kernel based convolutional neural network (wck-CNN) framework for hierarchical feature learning from fMRI data for improved disease diagnosis.
Main Methods:
- A novel wc-kernel was developed to weigh brain region correlations based on data-driven contributions from different time points.
- A wck-CNN framework was designed, incorporating layers for dynamic FCN construction and sequential extraction of local, global, and temporal features.
Main Results:
- The proposed wc-kernel method effectively captures richer interaction information compared to standard PCC.
- The wck-CNN framework demonstrated efficacy in learning hierarchical features for disease classification using rs-fMRI data.
Conclusions:
- The novel wc-kernel and wck-CNN framework offer a promising approach for analyzing fMRI data in brain disease diagnosis.
- This method enhances the characterization of brain interactions, potentially leading to earlier and more accurate detection of conditions like AD and MCI.
