Electromagnetic Noise Characterization and Suppression in Low-Field MRI Systems

This paper presents and validates a systematic protocol for identifying and suppressing electromagnetic interference in low-field MRI systems, enabling them to operate near their fundamental thermal noise limit and significantly improve signal-to-noise ratio and image quality.

The Gist

Imagine you are trying to listen to a friend whispering a secret in a crowded, noisy room. If the room is too loud, you can't hear them, no matter how good your ears are.

This paper is about building a special "listening room" for Low-Field MRI machines. These are smaller, cheaper, and more portable versions of the giant MRI scanners found in hospitals. Because they are smaller, they produce a weaker signal (the "whisper"). The problem is that the world around them is full of electromagnetic "noise" (the "crowd")—from power lines, computers, and even the person lying inside the machine.

The authors, a team of engineers and scientists, created a step-by-step recipe to silence that crowd so the machine can hear the whisper clearly.

Here is the breakdown of their work using simple analogies:

1. The Goal: Hearing the Whisper

In a perfect world, the only noise in an MRI machine should be the natural "static" of electricity itself (called thermal noise). Think of this as the quiet hum of a refrigerator in a silent house. It's the lowest possible noise floor.

• The Problem: Most low-field MRI machines are so noisy that they drown out the signal. It's like trying to hear that whisper while a rock band is playing next door.
• The Solution: The team wanted to get their machine so quiet that it was only about 1.5 times louder than that natural refrigerator hum.

2. The Recipe: Building the Machine Like a Puzzle

Instead of building the whole machine at once and hoping for the best, they built it piece by piece, testing the "noise level" after adding every single part.

Imagine you are building a high-end audio system. You wouldn't just plug everything in and hope it sounds good. You would:

1. Start with the basics: Connect just the amplifier and a dummy speaker. Is it quiet? Yes.
2. Add the switch: Plug in the button that turns the sound on and off. Does it get louder? If yes, fix the wiring.
3. Add the transmitter: Connect the part that sends the signal out. Still quiet? Good.
4. Add the big power amps: Turn on the heavy-duty motors. Uh oh, now it's buzzing!
5. The Fix: They realized the motors were acting like antennas, picking up noise from the wall outlet. They added shielding (like wrapping the wires in aluminum foil) and grounding (connecting the wires to the earth to drain the noise away).

They repeated this process until they added the human subject.

• The "Human Antenna" Problem: A person lying in the scanner can actually act like a giant antenna, picking up radio waves from the room and amplifying the noise.
• The Fix: They wrapped the person in a special conductive blanket (like a giant, grounded metal sheet) that acts as a Faraday cage, blocking the outside noise from reaching the machine.

3. The Results: From Static to Clear Sound

By following their recipe, they achieved amazing results:

• The Noise Level: They got the machine down to just 1.5 times the theoretical limit. That's incredibly quiet.
• The Images: Before this, the images were grainy and full of "zipper" lines (artifacts caused by noise). After silencing the noise, they got clear pictures of a human brain.
• The Proof: They showed that if you leave a computer box open or move a power supply too close to the wires, the noise spikes immediately. It's like leaving a window open in a storm; the wind (noise) rushes in.

4. Why This Matters

The authors argue that many people try to fix noisy MRI images using software (like AI) to "clean up" the picture after it's taken. They compare this to trying to fix a blurry photo by sharpening it in Photoshop. It helps a little, but it can't create detail that wasn't there.

Their approach is different: Fix the camera first.
If you build a quiet machine (the camera) using good engineering (shielding and grounding), the picture comes out clear naturally. This makes the machine reliable, safe, and ready to be used in hospitals, clinics, or even remote villages where there is no fancy shielding.

The Takeaway

This paper is a guidebook for silence. It teaches engineers how to build a low-cost MRI machine that is so well-shielded and grounded that it can hear the faintest signals from the human body, turning a noisy, grainy mess into a clear, diagnostic image. It proves that with the right physical setup, you don't need magic software to get great results; you just need to stop the noise from getting in.

Technical Summary

1. Problem Statement

Low-field MRI systems (operating in the 1–10 MHz range) offer significant advantages in cost, portability, and safety, making them ideal for point-of-care and resource-limited settings. However, their primary limitation is a low Signal-to-Noise Ratio (SNR).

• The Challenge: At these frequencies, the intrinsic noise floor is dominated by thermal (Johnson-Nyquist) noise from the RF coil resistance. To achieve clinical utility, all extrinsic electromagnetic interference (EMI) from cabling, active electronics, gradient drivers, and the human subject must be suppressed to allow the system to operate near this fundamental thermal limit.
• The Gap: Existing literature often quantifies noise reduction in relative terms (e.g., "noise reduced by X%") or within image space, lacking a standardized benchmark against the absolute theoretical thermal noise floor. This makes it difficult to determine if a system is truly optimized or if residual EMI is degrading performance.

2. Methodology

The authors propose a systematic, stepwise protocol for characterizing and suppressing EMI, validated on a custom-built low-field MRI scanner (NextMRI project) controlled by the open-source MaRCoS framework.

A. Theoretical Baseline

The study establishes a reference thermal noise floor using the Johnson-Nyquist equation:
$$v_{out} = G \cdot \sqrt{k_B T R \Delta f}$$
Where $G$ is the amplifier gain, $k_B$ is Boltzmann's constant, $T$ is temperature, $R$ is impedance (50 $\Omega$), and $\Delta f$ is bandwidth. For their setup, this theoretical limit is approximately 18 $\mu$V.

B. The Four-Step Protocol

The methodology involves incremental system assembly and noise measurement at each stage:

1. Establish Baseline: Calculate the theoretical thermal noise limit based on hardware specifications.
2. Measure with 50 $\Omega$ Resistor: Connect a 50 $\Omega$ terminator to the Low-Noise Amplifier (LNA) input. Measure noise while incrementally adding components:
   • Minimal Rx chain (LNA, ADC).
   • Tx/Rx switch.
   • Transmit (Tx) chain.
   • RF Power Amplifier (RFPA) powered on.
   • Gradient cables connected.
   • Gradient Power Amplifier (GPA) powered on.
   • Auxiliary subsystems (e.g., auto-tuning).
   • Action: If noise exceeds ~1.5$\times$ the thermal limit, isolate faulty components, check shielding, or switch to battery/isolated power.
3. Add RF Coil: Replace the 50 $\Omega$ resistor with the actual RF coil (tuned/matched to < -20 dB return loss) and repeat the incremental assembly steps.
4. In Vivo Evaluation: Repeat the process with a human subject inside the scanner to evaluate subject-coupled EMI.

C. Validation Techniques

• Noise Measurements: Performed using a 50 kHz acquisition bandwidth centered at the Larmor frequency. RMS voltage is measured and normalized against the thermal baseline.
• Imaging: 3D RARE (Rapid Acquisition with Relaxation Enhancement) sequences were used to visualize the direct impact of noise on image quality.
• Controlled Experiments: Specific hardware configurations (e.g., removing shields, opening control boxes, varying distance to power supplies) were tested to identify specific EMI pathways.

3. Key Contributions

• Standardized Protocol: A practical, reproducible framework for diagnosing noise sources in DIY and custom low-field MRI systems, moving beyond relative comparisons to absolute thermal benchmarks.
• Quantitative Benchmarking: Demonstrates that low-field systems can operate within 1.5$\times$ the theoretical thermal noise limit even with a human subject present.
• Identification of Critical EMI Pathways:
  • Gradient Systems: Identified as the most challenging noise source, often coupling via self-resonance in gradient structures.
  • Shielding: Proved that the outer shield (around the magnet) is critical for blocking EMI from gradient wiring, while the inner shield protects the RF coil.
  • Subject Coupling: Highlighted that unshielded subjects act as antennas, potentially raising noise by two orders of magnitude.
• Physical vs. Algorithmic Suppression: Argues that physical EMI suppression (shielding, grounding, cabling) provides a robust, predictable baseline that is superior to relying solely on AI/Deep Learning post-processing, which may fail on unseen noise patterns or compromise signal fidelity.

4. Key Results

• Noise Levels:
  • Minimal Chain: Achieved ~1.26$\times$ the thermal limit.
  • Full Assembly (No Subject): ~1.53$\times$ the thermal limit.
  • In Vivo (With Subject): ~1.51$\times$ the thermal limit (when using a grounded conductive blanket).
• Impact of Shielding (Table II):
  • Removing shielding from a single RF cable increased noise to 3.4$\times$ the thermal limit.
  • Opening control boxes (MaRCoS and TM) simultaneously increased noise to 2.1$\times$.
  • Disconnecting the outer shield while gradients were active caused noise to spike to 14.8$\times$ the thermal limit.
  • Placing a switched-mode power supply in direct contact with gradient cables increased noise to 5.4$\times$.
• Image Quality:
  • Images acquired with proper noise suppression (grounded blanket) showed clear anatomical detail.
  • Images with compromised shielding exhibited "zipper" artifacts and severe degradation, directly correlating with the measured noise spikes.

5. Significance

This work provides a critical roadmap for the low-field MRI community to transition from experimental prototypes to clinically viable systems.

• Clinical Translation: By enabling operation near the fundamental thermal limit, the protocol ensures that SNR is maximized, making low-field MRI viable for point-of-care and pediatric applications.
• Reproducibility: The adoption of absolute thermal noise metrics allows for unbiased comparison between different low-field platforms and noise suppression techniques.
• Design Guidance: The findings emphasize that rigorous RF engineering (grounding, shielding, cable management) is a prerequisite for high-quality imaging, serving as a necessary foundation before applying advanced algorithmic noise cancellation.
• Global Impact: The protocol has already been successfully applied to boost the performance of a low-field installation in Uganda, demonstrating its utility in diverse and challenging electromagnetic environments.