📄 medicine

Monte Carlo Propagation for Confidence Scoring of Ventricular Tachycardia Ablation Targets in Patient-Specific Cardiac Models

This study demonstrates that applying Monte Carlo Propagation to patient-specific cardiac models significantly improves the accuracy and confidence of ventricular tachycardia ablation target predictions compared to traditional deterministic methods, thereby enabling more reliable, risk-stratified clinical decision-making.

Original authors: Sudipta Barua

Published 2026-08-24

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Sudipta Barua

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The human heart is a tireless pump, but its rhythm depends on a precise electrical system that tells the muscle when to squeeze. When this system falters, the heart can race dangerously fast, a condition known as ventricular tachycardia. This often happens after a heart attack, where damaged tissue leaves behind a scar. While healthy heart muscle conducts electricity quickly, this scarred tissue slows the signal down, sometimes causing the electrical current to loop back on itself in a self-sustaining circuit. To stop this, doctors perform a procedure called catheter ablation, where they use heat or cold to create new, controlled scars that block these rogue loops. However, finding the exact spot to treat is difficult. Current methods rely on computer models that simulate the heart's electricity based on medical images, but these simulations are rigid. They treat the heart as a perfect, known object, ignoring the fact that medical images are never perfectly clear and that scars have fuzzy, irregular edges. If a model misses a critical loop or targets the wrong spot, the patient may need repeated, risky procedures.

A researcher named Sudipta Barua has developed a new way to handle this uncertainty, moving away from rigid predictions to a more flexible approach that acknowledges what we do not know. Instead of running a single simulation to find the best place to ablate, the study used a method called Monte Carlo propagation. Imagine trying to predict the path of a river through a landscape where the map is slightly blurry; instead of drawing one line, you draw many possible lines based on different interpretations of that blur. In this study, the researcher took MRI scans of nine patients who had suffered heart attacks and created digital models of their hearts. Rather than accepting the scar boundaries as fixed facts, the computer generated ten slightly different versions of each patient's scar, representing the natural variation in how a doctor might interpret the image or how the tissue might actually look at a microscopic level. For each of these ten versions, the computer simulated how electricity would travel through the heart.

By comparing the results of these ten simulations, the system could identify which parts of the heart were consistently involved in the dangerous electrical loops across all versions. These consistent areas were marked as high-confidence targets for treatment. Areas that appeared as targets in only one or two versions were flagged as uncertain. This process allowed the model to assign a confidence score to every potential treatment spot, distinguishing between a location that is almost certainly a problem and one that might just be an artifact of the image quality. The results showed a significant improvement over the standard, rigid method. When looking at the top ten most likely targets, the new method was correct 96.7 percent of the time, whereas the traditional method was correct only 56.7 percent of the time. This means the new approach was able to filter out nearly 37 percent of the false alarms that the old method would have suggested, preventing unnecessary treatment.

The study also revealed that the new system could adapt to the specific needs of a patient. Because the model provided a range of confidence scores, doctors could choose how aggressive they wanted to be. A cautious doctor could choose to treat only the spots with the highest confidence, while a doctor needing to cover more ground could include spots with moderate confidence. In contrast, the traditional method offered only a single, fixed list of targets with no way to adjust for risk. The research demonstrated that this uncertainty-aware approach works well even with the limited clarity of standard medical images, identifying stable targets in about one-third of the candidate spots. While the study was a simulation and has not yet been tested on patients in a hospital setting, it proves that adding a layer of statistical reasoning to heart models can make them far more reliable. By turning a binary guess into a spectrum of probability, this work offers a clearer path for doctors to make safer, more precise decisions when treating life-threatening heart rhythms.

Technical Summary: Monte Carlo Propagation for Confidence Scoring of Ventricular Tachycardia Ablation Targets

Problem Statement
Ventricular tachycardia (VT) is a life-threatening arrhythmia often caused by reentrant circuits within scarred myocardial tissue following a myocardial infarction. Current clinical practice relies on catheter ablation to disrupt these circuits, guided by electroanatomic maps and computational models. However, existing patient-specific cardiac models are predominantly deterministic, relying on single simulation runs that ignore the inherent uncertainties in medical imaging and tissue properties. This deterministic approach can lead to missed critical ablation sites, the generation of unnecessary targets, and high rates of VT recurrence (23–49% at one year). Furthermore, standard models lack a mechanism to quantify the reliability of their predictions, potentially misleading clinicians regarding the confidence of specific ablation targets.

Methodology
The study proposes a computational pipeline that integrates Monte Carlo Propagation (MCP) with a clinically standard Eikonal Cardiac Model to quantify uncertainty in ablation target prediction. The workflow processes data from the EMIDEC dataset (Evaluation of Myocardial Infarction from Delayed-Enhancement Cardiac MRI) through the following stages:

Data Processing and Mesh Generation:
- Input: 3D Delayed-Enhancement MRI (DE-MRI) volumes and expert-annotated segmentations (background, LV cavity, normal myocardium, scar, microvascular obstruction).
- Preprocessing: Binary masks undergo morphological cleaning (binary closing) to remove artifacts.
- Meshing: Surface meshes are extracted via the marching cubes algorithm, cleaned, smoothed, and decimated using PyVista. Volumetric tetrahedral meshes are generated using GMSH (1.2–2.5 mm element size) and converted to FEniCS format.
- Scar Mapping: Scar intensity is projected from the MRI segmentation to the mesh nodes, creating a graded intensity field (0.0 to 1.0) to represent healthy tissue, border zones, and dense scar.
Uncertainty Quantification (Monte Carlo Propagation):
- To address epistemic uncertainty (e.g., MRI resolution limits, contrast diffusion, inter-observer variability), the study generates an ensemble of $N=10$ plausible scar realizations.
- These realizations are created using stochastic morphological operations (randomized dilation/erosion) and randomized intensity assignments within defined ranges for core scars and border zones.
- This ensemble captures the continuous uncertainty distribution of the anatomical substrate.
Electrophysiological Simulation:
- The Eikonal equation ( $|\nabla T(x)| = 1/c(x)$ ) is solved for each realization using the Fast Marching Method (scikit-fmm).
- Conduction velocity (CV) is spatially heterogeneous: 0.7 m/s for healthy tissue, 0.25 m/s for dense scar, and interpolated values for border zones.
- This approach prioritizes computational efficiency (seconds per run) over the biophysical detail of ionic models, suitable for ensemble generation.
Confidence Scoring and Target Identification:
- Reentry Detection: A multi-criteria scoring system identifies reentry substrates based on slow conduction (<0.5 m/s), border zone tissue, and late activation (>70th percentile).
- Confidence Metric: For each node, a confidence score $C(x)$ is calculated as the fraction of the 10 realizations in which the node participates in a detected reentry circuit or isthmus.
- Classification: Nodes are classified as high-confidence (consensus across realizations), uncertain, or low-confidence. Adaptive percentile-based thresholds (e.g., 75th percentile) are used to define target sets.

Key Contributions

Uncertainty-Aware Modeling: This work introduces one of the first applications of uncertainty quantification specifically for VT ablation targeting, moving beyond binary deterministic predictions to probabilistic confidence scoring.
Pipeline Integration: A fully automated pipeline combining MRI segmentation, mesh generation (GMSH/FEniCS), and ensemble-based Eikonal simulation.
Risk-Stratified Decision Support: The framework enables clinicians to select ablation strategies based on risk tolerance (Conservative, Moderate, Aggressive) by adjusting confidence thresholds, rather than relying on a single fixed target set.

Results
The study evaluated the MCP framework against a deterministic baseline across nine post-infarction patients:

Precision Improvement: At the Top-10 threshold, MCP achieved a mean precision of 0.967, compared to 0.567 for the deterministic model—a 40.0 percentage point improvement. Eight of nine patients achieved perfect precision (1.00) with MCP at Top-10.
Target Stability: The model identified 21.0 ± 8.7 stable high-confidence targets per patient, representing a 33.7% stability rate among candidate nodes.
False Positive Reduction: The framework flagged 36.8% of deterministic-only predictions as low-confidence false positives. Deterministic-only targets had a mean confidence score of 0.168, whereas stable targets had a score of 0.650.
Discrimination Performance: The Receiver Operating Characteristic (ROC) analysis yielded an AUC of 0.966 for MCP, significantly outperforming the deterministic baseline (AUC 0.500), demonstrating the model's ability to distinguish high- from low-confidence targets.
Adaptive Behavior: The false positive flagging rate varied inversely with scar concentration (0% to 58.3%), indicating the model correctly signals caution in cases with ambiguous or diffuse substrates.

Significance and Claims
The author claims this study establishes a proof-of-concept for uncertainty-aware cardiac modeling. The significance lies in the shift from rigid, binary recommendations to probabilistic guidance that acknowledges the limitations of imaging data.

Clinical Utility: By providing confidence-weighted predictions, the framework supports risk-stratified clinical decision-making, allowing clinicians to tailor ablation extent based on patient-specific risk profiles and data quality.
Efficiency: The reduction in false positives and improved precision suggests a potential for reduced procedural times and fewer unnecessary ablations.
Limitations and Future Work: The author modestly notes that validation relies on consensus across realizations rather than actual clinical VT termination outcomes. They acknowledge the Eikonal model's inability to simulate complex functional reentry (compared to Monodomain models) and the current focus on anatomical uncertainty rather than tissue conductivity or fiber orientation. However, they assert that explicit uncertainty quantification is a necessary step toward the responsible clinical translation of model-based therapies.

Technical Summary: Monte Carlo Propagation for Confidence Scoring of Ventricular Tachycardia Ablation Targets

More like this