fMRI Data Analysis Using Deconvolution and Matrix Factorization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current BOLD fMRI techniques face challenges in accurately identifying brain networks due to low signal-to-noise ratio, structural noise, and partial volume effects, which lead to unreliable and reproducible results, especially in comparing atrophied or diseased brains to healthy ones.
Innovation Solution
A method that models fMRI data as the convolution of neural activation time courses and a haemodynamic filter, followed by deconvolution and non-negative matrix factorization to decompose the data into brain network components, reducing noise and spatial constraints, and providing reproducible functional networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the exposure time is lengthened to reduce noise, then the signal-to-noise ratio improves, but the time required for scanning increases and costs increase
Solution Approach 1:
The patent segments the fMRI data analysis into multiple components using independent component analysis (ICA), separating neural activity signals from noise and artifacts. This allows the use of shorter scanning times while maintaining signal quality through computational decomposition rather than temporal extension
Solution Approach 2:
The patent introduces an intermediary processing stage using ICA and machine learning algorithms that act as mediators between the raw fMRI data and the final brain network identification. This intermediary computational layer enhances signal extraction efficiency, allowing accurate network mapping without proportionally increasing scan duration
2Measurement precision
If spatial smoothing is applied to reduce noise, then the signal-to-noise ratio improves, but partial volume effects increase and spatial precision deteriorates
Solution Approach 1:
The patent extracts and removes noise components and artifacts from the fMRI data using independent component analysis and machine learning classification. By taking out unwanted signals separately from the neural activity components, the method achieves noise reduction without requiring spatial smoothing that would compromise spatial precision
Solution Approach 2:
The patent applies different processing quality levels to different spatial regions based on their characteristics. Through ICA decomposition, each brain network component is processed independently with appropriate spatial constraints, allowing high spatial precision in critical regions while maintaining overall signal quality through region-specific optimization
3Productivity
If machine learning techniques are used to decompose data, then the computational efficiency improves, but the model complexity increases
Solution Approach 1:
The patent employs self-service algorithms where the machine learning models are trained on representative data and then automatically apply the learned patterns to new fMRI datasets without requiring complex manual configuration. The ICA decomposition and classification algorithms self-optimize based on the input data characteristics, achieving high computational efficiency with manageable model complexity through automated adaptation
Data Source
AI summary
A system and method are provided for analysing a functional magnetic resonance imaging (MRI) scan representing a time-sequence, comprising t time points, of images comprising v voxels, where each scan can be represented by a data matrix X∈t×v. The method comprises modelling the scan data X as the convolution of neural activation time courses N∈<sup2>+</sup2>t×v and a haemodynamic filter Ψ, and performing an inverse operation to estimate N from X and Ψ. The method further compasses decomposing the neural activation time courses N into multiple brain network components by representing N as the product of a first matrix H defining, for each component, a spatial map of the voxels belonging to that component, and a second matrix W defining, for each component, the time sequence of activation of that component during the scan. In other implementations, the matrix decomposition may be performed before the deconvolution.