A Cohort-Based Global Sensitivity Benchmark of MRI-Derived Whole-Heart Electromechanical Models in Healthy Hearts

This study establishes a global sensitivity benchmark for patient-specific four-chamber electromechanical heart models across five healthy subjects, demonstrating that hemodynamic loading conditions consistently dominate whole-heart function outputs regardless of inter-individual anatomical variations.

The Gist

Imagine your heart not just as a biological pump, but as a highly sophisticated, custom-built digital twin. This is a computer simulation so detailed that it mimics your specific heart's shape, electrical signals, and muscle movements.

Scientists have been building these digital twins to understand heart disease and plan surgeries. But there's a problem: these models are incredibly complex, containing hundreds of "knobs and dials" (parameters) that control everything from how fast electricity moves to how stiff the muscle is.

The Big Question: If you turn one of these knobs, does it change how the whole heart works? And does the answer depend on the specific shape of the person's heart, or is it the same for everyone?

This paper is like a massive stress test for five different healthy hearts. The researchers built a digital twin for five real men, ran thousands of simulations, and asked: "Which knobs matter the most?"

Here is the breakdown of what they found, using some everyday analogies:

1. The "Traffic Jam" vs. The "Engine"

Think of your heart as a car engine.

• The Engine Parts: These are the heart's internal mechanics—how fast the electrical spark travels, how strong the muscle fibers are, and how the cells handle calcium.
• The Traffic Conditions: These are the hemodynamic loads—the resistance in your blood vessels and the pressure pushing blood back into the heart.

The Discovery: The researchers found that for a healthy heart at rest, the traffic conditions matter way more than the engine parts.

No matter how different the five men's hearts looked (some were larger, some smaller, some had different shapes), the most important factors were always:

• Systemic and Pulmonary Resistance: How hard it is for blood to flow through the body's pipes (like a traffic jam on the highway).
• Filling Pressures: How much pressure is pushing blood into the heart chambers (like the water pressure in a garden hose).

The Analogy: Imagine you are driving a car. It doesn't matter if you have a Ferrari engine or a Honda Civic engine; if you are stuck in gridlock traffic (high resistance) or if the gas station is far away (low pressure), your ability to get to your destination (pump blood) is determined by the traffic, not the engine. In a healthy heart, the "traffic" (blood pressure and vessel resistance) dictates the performance, not the tiny details of the muscle fibers.

2. The "Teamwork" Between the Upper and Lower Rooms

The heart has four rooms: two upper (atria) and two lower (ventricles). The researchers looked at how these rooms talk to each other. They found a very specific, rhythmic pattern of teamwork that stayed the same for all five men:

• The Pressure (The Squeeze): The pressure inside the upper rooms (atria) is almost entirely controlled by the global traffic conditions (the rest of the body's blood system). It's like the water pressure in a house; it depends on the main city supply, not on the specific pipes inside your bathroom.
• The Filling (The Pour): How much blood fills the upper rooms is heavily influenced by the lower rooms (ventricles). When the lower rooms squeeze, they pull blood from the upper rooms. It's like a vacuum cleaner; the suction (ventricle) determines how much dust (blood) gets pulled into the bag (atrium).
• The Emptying (The Release): How well the upper rooms empty themselves is mostly controlled by their own internal muscles. This is the one part where the "engine" matters most. The upper rooms need their own strength to push the blood out.

The Analogy: Think of the heart as a relay race.

1. The Start (Pressure): The race is started by the crowd cheering (the body's blood pressure).
2. The Handoff (Filling): The baton is passed based on how fast the runner behind you is pulling you (the ventricles).
3. The Finish (Emptying): How you sprint to the finish line depends on your own leg strength (the atrial muscle).

3. Why This Matters

Before this study, scientists worried that because every human heart is shaped differently, they would have to build a completely new set of rules for every single patient. They thought, "Maybe for Person A, the muscle stiffness is the most important thing, but for Person B, it's the electrical speed."

The Good News: This study proves that the rules are universal.
Even though the five men had different heart sizes and shapes, the "most important knobs" were the same for all of them. The blood pressure and vessel resistance always ruled the show.

The Takeaway

This paper provides a blueprint for building better digital hearts.

• For Doctors: If you want to simulate a healthy heart, don't waste time tweaking every tiny electrical detail first. Focus on getting the blood pressure and vessel resistance right. That's where the magic happens.
• For the Future: This creates a "benchmark" (a standard reference point). Now, when scientists build digital twins for sick patients, they can compare them against this healthy baseline to see exactly what has gone wrong.

In short: In a healthy heart, the environment (blood pressure and vessels) is the boss, and the heart muscle is just the worker following orders. Understanding this helps us build better computer models to save lives.

Technical Summary

1. Problem Statement

Patient-specific four-chamber electromechanical models are increasingly used as "digital twins" to investigate cardiac physiology and disease mechanisms. However, these models contain a vast number of parameters (spanning cellular electrophysiology, calcium dynamics, tissue mechanics, and hemodynamics), making it difficult to determine which parameters most significantly influence whole-heart function.

• Gap in Knowledge: Previous Global Sensitivity Analyses (GSA) of cardiac models have typically been limited to a single heart. It remains unclear whether the ranking of influential parameters is consistent across individuals with different anatomical geometries or if results are specific to a single subject's structure.
• Objective: This study aims to establish a benchmark by performing a cohort-based GSA on five healthy subjects to determine if parameter influence rankings are robust to inter-individual anatomical variations and to elucidate the mechanisms of atrioventricular coupling.

2. Methodology

A. Cohort and Data Acquisition

• Subjects: Five healthy male subjects (ages 16–63) with varying Body Mass Index (BMI) and body surface area (BSA).
• Imaging: Cine Magnetic Resonance Imaging (MRI) was acquired using a 3T scanner (Siemens Prisma) with steady-state free precession (SSFP) sequences.
• Anatomical Variability: End-diastolic volumes for the Left Ventricle (LV), Right Ventricle (RV), Left Atrium (LA), and Right Atrium (RA) were extracted, showing significant inter-subject variability (e.g., LV volumes ranged from ~159 to 203 cm³).

B. Model Construction Pipeline

The authors utilized a reproducible, image-based workflow to construct four-chamber electromechanical models:

1. Segmentation & Reconstruction: A hybrid deep-learning (Transformer-CNN) and semi-manual pipeline was used to segment chambers and vessels. A Label-Completion U-Net (LC-UNet) reconstructed missing anatomical regions not fully captured in MRI slices.
2. Mesh Generation: Unstructured tetrahedral meshes were generated with a mean edge length of ~1 mm. Mesh quality was verified using the Scaled Jacobian metric (median SJ > 0.73), ensuring suitability for large-deformation mechanics.
3. Fiber Assignment: Rule-based approaches defined myocardial fiber orientations (transmural rotation in ventricles; longitudinal/circumferential patterns in atria).
4. Electromechanical Framework:
   • Electrophysiology: ToR-ORd model (ventricles) and Courtemanche model (atria) coupled with a reaction–eikonal formulation for activation times.
   • Mechanics: Transversely isotropic hyperelastic tissue (Guccione strain-energy function) coupled with the Land active tension model.
   • Boundary Conditions: A closed-loop 0D circulation model (CircAdapt) provided preload/afterload. Pericardial constraints were modeled via spatially varying springs.
5. Parameter Selection: A subset of 46 input parameters was selected from the full multiscale framework, covering mechanics, CircAdapt hemodynamics, and electrophysiology.
6. Output Metrics: 20 output metrics were defined, including pressure, volume, and rate of pressure change ($dP/dt$) for all four chambers.

C. Experimental Design & Analysis

• Sampling: Latin Hypercube Sampling (LHS) was used to generate 500 simulation scenarios per subject (2,500 total simulations) across the parameter space.
• Surrogate Modeling: Due to the computational cost of full electromechanical simulations, Gaussian Process Emulators (GPEs) were trained on the simulation data. Emulator accuracy was validated via 5-fold cross-validation ($R^2$ and Independent Standard Error).
• History Matching: To ensure physiological plausibility, iterative history matching was applied to exclude parameter sets that did not match target QRS and P-wave durations.
• Global Sensitivity Analysis (GSA): Sobol variance-based indices (Total Sobol indices, $S_T$) were computed to rank parameters by their influence on outputs.
• Coupling Analysis: A specific post-processing analysis grouped parameters into three categories to assess atrioventricular coupling:
  1. Ventricular-specific parameters.
  2. Atrial-specific parameters.
  3. Global/Hemodynamic parameters (resistances, pressures, valves).

3. Key Results

A. Dominance of Hemodynamic Loading

Across all five anatomically distinct subjects, the sensitivity rankings were highly consistent. The most influential parameters for global cardiac function (pressures and volumes) were:

• Systemic and Pulmonary vascular resistances ($R_{sys}$, $R_{pulm}$).
• Ventricular end-diastolic pressures ($EDP_{lv}$, $EDP_{rv}$).
• Venous reference pressure ($V_{ePref}$).
• Conclusion: In healthy resting hearts, organ-scale function is governed primarily by hemodynamic boundary conditions (preload/afterload) rather than intrinsic tissue properties or electrophysiological variations. Electrophysiological parameters primarily influenced activation timing rather than pressure-volume relationships.

B. Structured Atrioventricular Coupling

The analysis revealed a phase-dependent pattern in how parameters drive atrial outputs:

1. Atrial Pressures: Dominated (>90%) by global hemodynamic parameters. Intrinsic atrial or ventricular tissue properties had minimal influence on peak atrial pressures.
2. Atrial Filling Volumes (End-Diastolic): Showed substantial influence from ventricular-specific parameters (~40–55%), reflecting the mechanical coupling where ventricular relaxation and filling pressures drive atrial preload.
3. Atrial Emptying (End-Systolic Volumes): Primarily determined by intrinsic atrial parameters (~60–65%), indicating that the active contraction and stiffness of the atrial myocardium control the emptying phase.

C. Robustness to Anatomy

Despite significant differences in chamber volumes and body sizes among the five subjects, the ranking of dominant parameters remained unchanged. This suggests that for healthy hearts, the relative importance of model parameters is robust to anatomical variability.

4. Key Contributions

1. Cohort-Based Benchmark: This is the first study to perform a systematic GSA across a cohort of five healthy, anatomically distinct hearts, moving beyond single-subject analyses.
2. Open Data & Framework: The authors provide a repeatable workflow and have made the finite-element meshed geometries of the five subjects publicly available (via Zenodo), establishing a benchmark dataset for the community.
3. Mechanistic Insight: The study quantifies the specific contributions of ventricular, atrial, and global parameters to atrial function, revealing a structured, phase-dependent coupling mechanism.
4. Calibration Guidance: By identifying that hemodynamic loading parameters are the primary drivers of pressure/volume outputs, the study provides clear guidance for calibrating future cardiac digital twins, suggesting that boundary conditions should be prioritized over complex tissue property tuning for healthy resting states.

5. Significance and Limitations

• Significance: The findings validate that patient-specific modeling in healthy hearts can rely on consistent parameter hierarchies, simplifying the calibration process for digital twins. It highlights that in healthy physiology, the cardiovascular system's boundary conditions are the "master regulators" of whole-heart performance.
• Limitations:
  • The cohort consisted only of healthy males; sex-specific differences were not evaluated (though consistency with a previous female heart-failure model suggests broad applicability).
  • Models represent healthy resting states; sensitivity patterns may shift in pathological conditions (e.g., severe fibrosis or arrhythmias).
  • The model lacks neurohumoral feedback (e.g., baroreflex), which would dynamically adjust parameters in vivo.
  • Electrophysiological parameters were constrained by history matching to ECG targets, which may have reduced their apparent influence on hemodynamic outputs.

In summary, this paper establishes a robust, reproducible framework for whole-heart modeling and demonstrates that in healthy hearts, hemodynamic loading conditions are the dominant determinants of cardiac function, with a distinct, phase-dependent interplay between atrial and ventricular mechanics.