Abstract
Delayed-enhancement magnetic resonance imaging (DE-MRI) is an effective technique for detecting left atrial (LA) fibrosis both pre and postradiofrequency ablation for the treatment of atrial fibrillation. Fixed thresholding models are frequently utilized clinically to segment and quantify scar in DE-MRI due to their simplicity. These methods fail to provide a standardized quantification due to interobserver variability. Quantification of scar can be used as an endpoint in clinical studies and therefore standardization is important. In this paper, we propose a segmentation algorithm for LA fibrosis quantification and investigate its performance. The algorithm was validated using numerical phantoms and 15 clinical data sets from patients undergoing LA ablation. We demonstrate that the approach produces good concordance with expert manual delineations. The method offers a standardized quantification technique for evaluation and interpretation of DE-MRI scans.
Introduction
Atrial Fibrillation (AF) remains a prevalent cardiac rhythm disorder, significantly elevating the risk of stroke and mortality. Radiofrequency Catheter Ablation (RFCA) is a primary intervention aimed at creating scar tissue to isolate abnormal electrical signals. Evaluating the efficacy of RFCA requires precise quantification of these scars via Delayed-Enhancement MRI (DE-MRI). However, the thin walls of the left atrium (LA) and low signal-to-noise ratios (SNR) make manual segmentation highly subjective and prone to inter-observer variability.
Methodology
The proposed approach utilizes a semi-automated pipeline based on graph-cut energy minimization to standardize scar detection.
Probabilistic Modeling and Priors
The algorithm avoids fixed thresholds by employing a four-class Gaussian Mixture Model (GMM) to represent non-scar tissues, such as the blood pool and healthy myocardium. It incorporates a "scar prior" based on the ratio of scar intensity to the blood pool (SC-BP), a metric learned from training data that enhances robustness against varying scanner settings and patient-specific contrast levels.
Adaptive Feedback and Spatial Constraints
A unique "adaptive feedback" mechanism iteratively refines the scar model based on previous segmentation results until convergence. To prevent noise-induced errors, a Lorentzian error norm is applied to ensure spatial smoothness, effectively filtering out isolated high-intensity pixels that do not represent true fibrosis.
Clinical Validation and Performance
The algorithm's performance was evaluated using 100 numerical phantoms and 15 clinical datasets from post-ablation patients. Clinical "ground truth" was established via the STAPLE algorithm, aggregating manual delineations from three experts to create a consensus map.
The method achieved a median Dice score of $\ge$ 80 on synthetic data and demonstrated high correlation with expert consensus in clinical settings. Notably, the total processing time is less than one minute on a standard PC, offering a significant speed advantage over manual methods.
Discussion and Conclusion
By focusing on normalized intensity ratios (SC-BP) rather than absolute values, the proposed method mitigates the "leakage" issues common in Full Width at Half Maximum (FWHM) or standard deviation-based thresholding. The removal of manual "remote region" selection eliminates the primary source of inter-observer variability, ensuring reproducible results across different clinicians.
This computational framework provides a standardized tool for evaluating ablation success and planning follow-up procedures. Future work will focus on multi-center validation and the development of 3D "gap maps" to identify incomplete ablation lines.
This summary is based on research published in the IEEE Journal of Translational Engineering in Health and Medicine.
Summary updated