Continuous Ventricular Volumetric Quantification in Patients with Arrhythmias using Real-Time 3D CMR-MOTUS

Imagine your heart is a drummer in a band. For most people, the drummer keeps a perfect, steady beat: thump-thump, thump-thump, thump-thump. This makes it easy for a photographer (the MRI machine) to take a clear picture of the drum. They just snap a photo every time the drum hits, and if they miss a beat, they can just take a few more photos and average them out to get a perfect picture.

But for some patients, the drummer has a glitch. They throw in a sudden, extra hard hit or skip a beat entirely. This is called an arrhythmia (specifically, Premature Ventricular Contractions or PVCs).

The Problem: The "Blurry Group Photo"

Traditional MRI scans work like taking a group photo of a crowd. They assume everyone is standing still and doing the same thing. If the crowd is moving chaotically (like a heart with arrhythmia), the traditional camera tries to take a "long exposure" photo, averaging all the movements together.

The result? A blurry, messy photo where the heart looks like a ghost. You can't see the specific moments when the heart skipped a beat or pumped weakly. You lose the most important information: how the heart actually behaves when it's struggling.

The Solution: The "Super-Fast 3D Movie Camera"

This paper introduces a new way to film the heart using a technology called CMR-MOTUS. Think of this not as a still camera, but as a super-fast, high-definition 3D movie camera that never stops recording.

Here is how it works, using a few simple analogies:

1. The "Free-Running" Recorder
Instead of waiting for the heart to tell the camera when to snap a picture (which is hard when the heart is irregular), this method just records everything continuously. It doesn't care if the heart beats fast, slow, or skips a beat. It captures the entire performance, 20 times every second.

2. The "Magic Elastic Sheet" (Motion Fields)
This is the clever part. The computer doesn't just try to take a picture of a moving heart. Instead, it builds a 3D map of how the heart moves.

Imagine you have a photo of a balloon.
Now, imagine you have a sheet of elastic rubber that tells you exactly how to stretch that balloon to make it look like it's expanding and contracting.
The computer calculates this "elastic sheet" (called a motion field) for every single millisecond.

3. The "One-Time Cut, Infinite Copies"
Once the computer has the "elastic sheet" (the motion map) and one clear, sharp photo of the heart (the reference image), it can stretch and squash that one photo to create a perfect movie of the heart beating.

It doesn't need to take a new photo for every heartbeat.
It just uses the math of the "elastic sheet" to show you exactly how the heart looked at that specific moment.

What Did They Discover?

The researchers tested this on a rubber heart phantom (a fake heart in a machine) and then on real humans.

The Healthy Heart: The movie showed a steady, rhythmic beat. The volume of blood pumped was consistent, just like a well-oiled machine.
The Arrhythmic Heart: This is where the magic happened. The traditional method would have just shown a blurry average. But this new method showed a bimodal distribution (a fancy way of saying "two distinct groups").
- It showed the "normal" beats where the heart pumped well.
- Crucially, it also showed the "bad" beats (the PVCs) where the heart barely pumped any blood at all.

Why Does This Matter?

Think of it like checking a car's fuel efficiency.

Old Method: You drive for an hour, average the speed, and say, "The car gets 30 miles per gallon." You miss the fact that the car stalled three times during that hour.
New Method: You see the exact moment the car stalled, how long it took to restart, and how much fuel was wasted during that stall.

For patients with arrhythmias, this new method reveals the true cost of their irregular heartbeats. It shows that while their heart might look "okay" on average, the individual bad beats are actually very dangerous and inefficient.

The Bottom Line

This paper proves that we can now film the heart in 3D, in real-time, without needing the heart to behave itself. It turns a blurry, confusing mess into a clear, detailed movie that shows exactly how the heart struggles and recovers, beat by beat. This could help doctors choose better treatments and predict who is at higher risk, simply by watching the "movie" of the heart rather than just looking at a blurry "group photo."

Here is a detailed technical summary of the paper "Continuous Ventricular Volumetric Quantification in Patients with Arrhythmias using Real-Time 3D CMR-MOTUS."

1. Problem Statement

Conventional Cardiovascular Magnetic Resonance (CMR) cine sequences rely on binning reconstructions that average data from multiple heartbeats to achieve high signal-to-noise ratio (SNR) and spatiotemporal resolution. This approach assumes that cardiac cycles are nearly identical.

The Limitation: In patients with arrhythmias (specifically Premature Ventricular Contractions or PVCs), beat-to-beat variations in morphology and function violate this assumption. Binning leads to motion artifacts, loss of clinically relevant functional information, and the inability to capture the true hemodynamic impact of irregular beats.
Current Alternatives: While 2D real-time MRI can capture individual beats, acquiring a stack of 2D slices is sub-optimal for mapping complex 3D cardiac dynamics during arrhythmia, as irregular beats may be captured on some slices but not others. Furthermore, existing 3D real-time methods often rely on deep learning (implicit regularization) where motion fields are not guaranteed to be motion-static, limiting their interpretability for downstream clinical analysis.

2. Methodology

The authors propose an extension of the CMR-MOTUS framework to 3D, enabling continuous, free-running volumetric quantification without breath-holds or ECG gating.

A. Acquisition Protocol

Trajectory: Uses an Ordered Pseudo RAdial (OPRA) variable-density Cartesian sampling trajectory. This minimizes eddy current artifacts compared to non-Cartesian methods and allows for efficient undersampling.
Sequence: 3D spoiled gradient echo (GRE) or balanced steady-state free precession (bSSFP) sequences.
Protocol: Free-running (continuous acquisition) with free breathing. No ECG gating is used.
Subjects: Validated on a cardiac motion phantom, 4 healthy volunteers, and 4 patients with PVCs.

B. Reconstruction Framework (CMR-MOTUS 3D)

The core innovation is the joint reconstruction of a motion-corrected reference image ( $q$ ) and real-time 3D motion fields ( $D_t$ ) by solving a minimization problem.

Explicit Low-Rank Factorization: To manage the computational cost of 3D data at ~20 Hz, the motion fields are parameterized as dense displacement vectors using an explicit low-rank factorization ( $D = \Phi\Psi^T$ ). This avoids computationally expensive Singular Value Decompositions (SVD) during optimization.
Regularization:
- Data Consistency: Uses an $\ell_1$ norm for in vivo data to provide robustness against inflow artifacts (blood entering the volume) and an $\ell_2$ norm for phantom data.
- Spatial Regularization: Isotropic Total Variation (TV) is applied to motion fields to ensure spatial smoothness while preserving sliding motion (e.g., between the heart and lungs).
- Temporal Regularization: A Butterworth lowpass filter (4 Hz cutoff) is applied to temporal basis components to enforce smoothness.
Optimization: Implemented in Python/PyTorch using Stochastic Gradient Descent (SGD) on GPUs. The process involves alternating minimization for the reference image and motion fields, typically converging within 120 iterations.

C. Volumetric Quantification Workflow

Segmentation: A single manual segmentation is performed on the reconstructed motion-corrected reference image ( $q$ ).
Propagation: This segmentation mask is propagated through all time frames using the reconstructed real-time motion fields ( $D_t$ ).
Calculation: Beat-to-beat End-Diastolic Volume (EDV) and End-Systolic Volume (ESV) are automatically detected from the volume curves, allowing for the calculation of Ejection Fraction (EF) for every individual cardiac cycle.

3. Key Contributions

3D Real-Time Motion Field Reconstruction: Successfully extended the CMR-MOTUS framework to 3D, achieving ~20 Hz temporal resolution without ECG gating.
Interpretable Motion Fields: Unlike deep learning-based "black box" approaches, this method uses explicit low-rank models and regularization to produce physically interpretable motion fields.
Single-Segmentation Propagation: Demonstrated that a single manual segmentation on a reference image can be accurately propagated through an entire arrhythmic dataset to quantify beat-to-beat volumes.
Free-Running 3D Imaging: Proved that 3D isotropic imaging can capture complex arrhythmic dynamics that 2D stacks miss, providing a complete 3D view of the heart's hemodynamics.

4. Results

Phantom Validation:
- The reconstructed EF (22.1 ± 0.6%) showed excellent agreement with the ground truth (21.9%) derived from static acquisitions.
- Spatiotemporal plots confirmed stable, periodic motion patterns consistent with the phantom's mechanical operation.
Healthy Volunteers:
- Mean EF values closely matched standard 2D clinical measurements.
- Beat-to-beat EF distributions were narrow, reflecting normal physiological consistency.
PVC Patients:
- Bimodal Distributions: The method revealed bimodal EF distributions in patients with PVCs. The lower mode corresponded to PVC episodes with substantially reduced ejection fractions, while the higher mode represented normal sinus beats.
- ECG Correlation: Simultaneously acquired ECG signals confirmed that volume irregularities and reduced EFs temporally aligned with PVC events.
- Comparison to Standard 2D: Standard 2D methods (which often discard arrhythmic beats or average them) showed poor agreement with the proposed method's mean EF in PVC patients. This is because the proposed method includes the "bad" beats in the average, revealing a true reduction in overall cardiac function that standard methods miss. Subject 6 (who had no PVCs during the scan) showed good agreement between the two methods, validating the discrepancy in other patients is due to arrhythmia capture.

5. Significance and Clinical Implications

True Hemodynamic Assessment: The method quantifies the true hemodynamic impact of arrhythmias, which is often underestimated by conventional binning techniques that average out or discard irregular beats.
Treatment Monitoring: It offers a new metric for monitoring treatment efficacy in arrhythmia patients, not just by counting PVC frequency, but by assessing the improvement in EF distribution characteristics and the reduction of beat-to-beat variability.
Functional Heterogeneity: It reveals functional heterogeneity (e.g., desynchrony, regional strain) that is obscured in standard single-beat or averaged measurements.
Future Potential: The extracted motion fields provide a foundation for deriving advanced metrics like strain, strain rate, and regional wall motion analysis from a single free-running acquisition, potentially replacing multiple standard CMR sequences.

Limitations: The current reconstruction time (~2 hours per dataset) limits immediate clinical workflow integration, though algorithmic optimizations are planned. The study cohort was small (4 patients), and different MRI systems were used for different groups, introducing potential confounding variables.