Motor-tasks fMRI BOLD activations in chronic stroke with residual hemiparesis in the upper extremity: a pre-neurofeedback baseline characterization

This study establishes a pre-neurofeedback baseline of motor-task fMRI BOLD activations in chronic stroke patients with upper extremity hemiparesis, revealing consistent cerebellar recruitment across participants that likely reflects compensatory mechanisms for motor control.

The Gist

The Big Picture: Fixing a Broken Brain Map

Imagine your brain is like a massive, high-tech city. When a stroke happens, it's like a sudden, massive power outage in one part of the city. Some roads are blocked, buildings are damaged, and the traffic (signals) can't get through.

For people who have had a stroke a long time ago (chronic stroke) but still have trouble moving their arms (hemiparesis), the city is trying to rebuild. The goal of this study was to take a "snapshot" of how this city is functioning before they start a special training program called neurofeedback.

Neurofeedback is like giving the city's traffic controllers a live video feed of their own traffic patterns, so they can learn to reroute traffic more efficiently. But before you can teach them, you need to know exactly what the traffic looks like right now. That's what this study did.

The Participants: A Diverse Group of "City Planners"

The researchers recruited 16 people who had survived a stroke at least six months ago.

• The Damage: Some had damage in just one side of their brain (like a fire in one neighborhood). Two people had damage on both sides (like fires in two different neighborhoods).
• The Goal: They were all going to undergo a 10-week training program to help their arms move better.
• The Test: Before the training started, these participants went into an MRI machine. While inside, they were asked to do two things:
  1. Actually move their hand (grasping or releasing an object).
  2. Imagine moving their hand (thinking about the motion without actually doing it).

The Challenge: Taking a Photo of a Damaged Building

Taking an MRI of a healthy brain is like taking a photo of a pristine house. Standard computer software knows exactly where the walls and windows should be, so it can line up the photo perfectly with a "master blueprint."

But when you scan a brain with a stroke, part of the house is missing. If you try to force a damaged house into a standard blueprint, the software gets confused and stretches or squishes the remaining parts, making the photo look blurry and useless.

The Solution: Two Different Repair Kits
The researchers realized they needed two different "repair kits" (preprocessing pipelines) to fix the photos:

1. The "Mirror" Kit: For most people (13 participants), the damage was only on one side. The software took the healthy side of their brain, mirrored it like a reflection in a pond, and filled in the missing holes. This allowed them to use a high-tech, precise alignment tool (DARTEL) to match everyone's brain to the master blueprint.
2. The "Old School" Kit: For the two people with damage on both sides, the mirror trick wouldn't work (you can't mirror a house that's burned down on both sides). For them, the researchers used an older, simpler method to align the images.

The Findings: What the "City" Was Doing

After cleaning up the photos and lining them up, the researchers looked for "hotspots" of activity (where the brain was lighting up).

1. Moving vs. Imagining

• Real Movement: When people actually moved their hands, their brains lit up like a Christmas tree. The activity was strong and widespread.
• Imagined Movement: When people just thought about moving, the lights were much dimmer. For some, the brain didn't light up at all in the areas responsible for movement. It was as if they were trying to drive a car but only pressing the gas pedal in their mind without turning the wheel.

2. The Surprise Star: The Cerebellum
The most consistent finding was activity in the cerebellum (a small structure at the back of the brain).

• The Analogy: Think of the main brain as the CEO making decisions, and the cerebellum as the project manager or the mechanic.
• What it means: In healthy people, the CEO does most of the heavy lifting. But in stroke survivors, the CEO's office is damaged. So, the project manager (cerebellum) has to step up and work overtime to coordinate the movement. The study found that this "mechanic" was working hard in almost everyone, suggesting the brain is using a backup system to compensate for the damage.

3. The Visual Connection
The brain also lit up in the occipital lobe (the back of the brain responsible for vision). This makes sense because the participants were watching a screen with instructions. It's like a driver looking at a GPS while trying to steer; both the eyes and the hands are working together.

Why This Matters

This study is like taking a "before" photo of a construction site.

• It proved that even with damaged brains, the "backup systems" (like the cerebellum) are active and trying to help.
• It showed that actually doing the movement is much easier for the brain to process than just imagining it.
• It established a baseline. Now that the researchers know what the brain looks like before the neurofeedback training, they can measure exactly how much the training helps in the future.

The Bottom Line

The researchers successfully navigated the messy reality of scanning damaged brains by using two different technical "repair kits." They found that while the brain's main command center is struggling, the support teams are working overtime. This baseline data is the crucial first step in helping these patients relearn how to move their arms through neurofeedback training.

Technical Summary

1. Problem Statement

Stroke is a leading cause of global disability, often resulting in chronic upper extremity hemiparesis. While neurofeedback training is an emerging therapeutic strategy to promote neural plasticity and motor recovery, there is a need to establish robust baseline characterizations of brain activation patterns in chronic stroke patients prior to intervention.

A specific technical challenge addressed in this study is the preprocessing of fMRI data in the presence of brain lesions. Standard fMRI pipelines often fail or introduce significant distortions when applied to brains with large or bilateral lesions, compromising spatial normalization and group-level analysis. The study aims to:

1. Develop and validate specialized fMRI preprocessing pipelines to handle varying lesion topographies (unilateral vs. bilateral).
2. Characterize baseline motor-task activations (motor-execution and motor-imagery) in a cohort of chronic stroke patients before they undergo neurofeedback training.

2. Methodology

Participants and Study Design

• Cohort: 16 participants with chronic stroke (≥6 months post-onset) and residual upper extremity hemiparesis.
• Demographics: Mean age 51; 69% male; 69% ischemic stroke, 31% hemorrhagic. Mean NIHSS score was 4 (mild-moderate severity); mean FMA-UE score was 24 (moderate-severe impairment).
• Intervention Context: Participants underwent a 10-week clinical intervention including neurofeedback, EEG, and fMRI. This study focuses on the baseline fMRI data collected in weeks 2 and 3 (prior to neurofeedback training).
• Tasks: During fMRI, participants performed four motor tasks:
  1. Motor-Execution Grasping (MEG)
  2. Motor-Execution Releasing (MER)
  3. Motor-Imagery Grasping (MIG)
  4. Motor-Imagery Releasing (MIR)
  • These were interleaved with idling/rest periods.

Data Acquisition

• Scanner: Siemens MAGNETOM Prisma 3T.
• Parameters: Gradient echo EPI, 2×2×2 mm voxels, TR=2090 ms, TE=30 ms.
• Anatomical: T1-weighted (MPRAGE) and T2-weighted (FLAIR) images were acquired for lesion masking.

Preprocessing Pipelines (Key Technical Contribution)

Due to lesion heterogeneity, two distinct pipelines were implemented using SPM (Statistical Parametric Mapping):

1. Pipeline A (Single Hemisphere/Small Secondary Lesions): Used for 13 participants.
   • Technique: Enantiomorphic lesion filling (mirroring voxels from the contralesional hemisphere) followed by DARTEL (Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra) for high-dimensional nonlinear normalization.
   • Advantage: Fully automated, reduces user-dependent variability, and provides superior inter-subject alignment.
2. Pipeline B (Large Bilateral Lesions): Used for 2 participants.
   • Technique: Manual origin reset at the anterior commissure, lesion mask inversion, and the "Old Normalization" algorithm (Ashburner, 1999).
   • Reasoning: DARTEL requires a relatively intact template; large bilateral lesions prevent the creation of a valid group template, necessitating a simpler normalization approach.

Quality Control (QC) and Analysis

• Visual Inspection: Checked for ghosting and artifacts using MRIcroGL.
• Automated QC: Used Artifact Detection Tools (ART) with conservative thresholds (5 SD for global signal, 1.5 mm motion) to identify and exclude outlier volumes.
• Statistical Modeling:
  • First-Level (Single Subject): General Linear Model (GLM) with regressors for all motor tasks, realignment parameters, and outliers. Contrasts included: Any Task > Baseline, Motor Execution > Baseline, and Motor Imagery > Baseline.
  • Second-Level (Group): One-sample t-test.
  • Correction: Family-Wise Error (FWE) for first-level; False Discovery Rate (FDR) and uncorrected ($p < 0.001$) for second-level.
  • Lesion Masking: Lesioned voxels were excluded from estimation to prevent signal contamination.

3. Key Results

First-Level Analysis (Single Subject)

• Motor Execution: Produced the strongest and most extensive activations. Significant clusters (FWE corrected) were consistently found in the ipsilesional parietal lobe, bilateral occipital lobes, and cerebellum.
• Motor Imagery: Produced weaker responses. Three participants showed no suprathreshold activation. Activations were more spatially restricted, primarily in the parietal lobe and occipital lobes. Notably, some participants showed occipital activation without parietal involvement, suggesting a reliance on visual processing rather than kinesthetic imagery.
• Pipeline Comparison: Participants processed via Pipeline B (bilateral lesions) showed activation patterns consistent with the rest of the cohort, indicating the alternative pipeline did not introduce systematic artifacts.

Second-Level Analysis (Group)

• Motor Execution: After FDR correction, a significant cluster was identified only in the cerebellum.
• Motor Imagery: No significant clusters survived correction; results showed isolated voxels at sulcal/gyral borders or ventricular edges, likely representing noise.
• Interpretation: The lack of significant group-level activation in the parietal/occipital lobes is attributed to high inter-subject variability in lesion location and the spatial distribution of activations.

4. Key Contributions

1. Dual-Preprocessing Strategy: The study successfully demonstrates a practical workflow for handling heterogeneous stroke lesions by applying two distinct SPM-based pipelines (DARTEL vs. Old Normalization) based on lesion volume and symmetry.
2. Baseline Characterization: Provides a detailed pre-neurofeedback map of motor-task activations, establishing that motor-execution elicits robust, widespread activation (including compensatory cerebellar recruitment), while motor-imagery yields weaker, more variable signals in this specific chronic cohort.
3. Rigorous QC Implementation: Highlights the necessity of conservative motion correction thresholds in clinical stroke populations to prevent spatial blurring of activation maps, even in task-based designs.
4. Lesion Masking Protocol: Validates the use of lesion masking during GLM estimation to ensure that statistical inferences are driven by healthy tissue activity.

5. Significance and Limitations

Significance:

• The findings confirm that the cerebellum is consistently recruited in chronic stroke patients during motor tasks, supporting the hypothesis of compensatory mechanisms for impaired cortical motor networks.
• The study provides a methodological framework for future neurofeedback studies, ensuring that baseline data is processed accurately despite lesion heterogeneity.
• The observation that motor-imagery yields weaker signals than execution in this cohort suggests that neurofeedback protocols targeting imagery may require specific optimization or that execution-based feedback might be more effective initially.

Limitations:

• Sample Size: The effective sample size for group-level analysis was reduced to 13 participants (excluding those with bilateral lesions and one with insufficient data), limiting statistical power.
• Pipeline Discrepancy: The use of two different normalization methods prevented the inclusion of the bilateral lesion group in the second-level analysis, though qualitative comparison showed consistency.
• Resting-State Omission: Resting-state fMRI data was collected but not analyzed in this paper, missing an opportunity to characterize functional connectivity changes at baseline.
• Future Directions: The authors suggest exploring automated tools like fMRIPrep for future studies to handle lesioned brains more uniformly.

In conclusion, this paper establishes a robust methodological baseline for studying motor recovery in chronic stroke, emphasizing the critical role of tailored preprocessing and the consistent involvement of the cerebellum in compensatory motor control.