Elevated fibrosis burden as assessed by MRI predicts cryoballoon ablation failure

Abstract

Introduction: Late-gadolinium enhancement magnetic resonance (LGE-MRI) imaging is increasingly used in management of atrial fibrillation (AFib) patients. Here, we assess the usefulness of LGE-MRI-based fibrosis quantification to predict arrhythmia recurrence in patients undergoing cryoballoon ablation. Our secondary goal was to compare two widely used fibrosis quantification methods.

Methods: In 102 AF patients undergoing LGE-MRI and cryoballoon ablation (mean age 62 years; 64% male; 59% paroxysmal AFib), atrial fibrosis was quantified using the pixel intensity histogram (PIH) and image intensity ratio (IIR) methods. PIH segmentations were completed by a third-party provider as part of the standard of care at our hospital; Image intensity ratio (IIR) segmentations of the same scans were carried out in our lab using a commercially available software package. Fibrosis burdens and spatial distributions for the two methods were compared. Patients were followed prospectively for recurrent arrhythmia following ablation.

Results: Average PIH fibrosis was 15.6 ± 5.8% of the left atrial (LA) volume. Depending on threshold (IIR_thr ), the average IIR fibrosis (% of LA wall surface area) ranged from 5.0 ± 7.2% (IIR_thr = 1.2) to 37.4 ± 10.9% (IIR_thr = 0.97). An IIR_thr of 1.03 demonstrated the greatest agreement between the methods, but spatial overlap of fibrotic areas delineated by the two methods was modest (Sorenson Dice coefficient: 0.49). Fourty-two patients (41.2%) had recurrent arrhythmia. PIH fibrosis successfully predicted recurrence (HR 1.07; p = .02) over a follow-up period of 362 ± 149 days; regardless of IIR_thr , IIR fibrosis did not predict recurrence.

Conclusions: PIH-based volumetric assessment of atrial fibrosis was modestly predictive of arrhythmia recurrence following cryoballoon ablation in this cohort. IIR-based fibrosis was not predictive of recurrence for any of the IIR_thr values tested, and the overlap in designated areas of fibrosis between the PIH and IIR methods was modest. Caution must therefore be exercised when interpreting LA fibrosis from LGE-MRI, since the values and spatial pattern are methodology-dependent.

Keywords: LGE-MRI; atrial fibrillation; atrial fibrosis; cryoballoon ablation.

Fig. 1:. Visual summary of the main project elements.

NRR = non-rigid registration, which was used to deform the 3D models reconstructed from segmented IIR images to match the geometry of PIH models to facilitate quantitative comparison of fibrosis spatial patterns.

Fig. 2:. Comparison of LA geometry in 3D models reconstructed from IIR and PIH segmentations of the same images.

(A) Data plotted in x-y format with simple linear regression line, as described in text. (B) Same data in truncated violin plot, showing medians (dashed lines) and upper/lower quartiles (dotted lines); Wilcoxon nonparametric test for paired data points (p = 0.9549). (C) Box-and-whisker plots showing point-by-point registration error between IIR and PIH segmentations for each model in the whole cohort.

Fig. 3:. Comparison of fibrosis spatial patterns in models reconstructed from PIH and IIR image segmentations.

Columns 2 and 3 show the IIR-derived pattern before and after non-rigid registration (NRR). Regions of spatial overlap are shown in column 4; corresponding Sørensen-Dice Coefficient (SDC) values are shown. In rows A-C, different values of IIR_thr are used to delineate fibrosis in IIR-based models. See also maps showing IIR-derived models for all IIR_thr values for these cases in Supplemental Fig. 2.

Fig. 4:. Quantitative comparison of PIH- vs. IIR-based fibrosis across all models.

(A) Fibrosis surface area for PIH models (column 1) vs. IIR models with different IIR_thr values; n = 102 in all columns; Dunn’s multiple comparisons test P<0.0001 for all pairs except as labeled. (B) Sørensen-Dice Coefficient (SDC) for quantification of spatial overlap between fibrosis patterns between PIH models and IIR models with IIR_thr values as shown; n = 102 in all columns; Dunn’s multiple comparisons test P<0.0001 for all pairs except as labeled. (C) Patient-by-patient plot showing SDC values for the optimal IIR_thr value only for all 102 models. (D) x-y plot showing the relationship between the IIR fibrosis in the optimal-IIR_thr model vs. PIH fibrosis for the same individual; same color-coding as in (C). See similar x-y plots for each individual IIR_thr value in Supplemental Fig. 3.

Fig. 5:. Kaplan Meier survival curves demonstrating time to recurrent atrial arrhythmia in three groups of atrial fibrosis quantified using the PIH method and grouped by previously published stages.

Only one patient had PIH fibrosis >30% and did experience recurrence; this individual is not shown. The log-rank test of equality of survivor functions between stages demonstrated a statistically significant difference in arrhythmia recurrence between the groups (p=0.01).