Automated generation of ablation lesion masks: a unison of electro and optic flow mapping for persistent AF virtual cohorts

Figures

• 

  Figure 1. Study pipeline overview. (A) Patient cohort generation: (A1) Bi-atrial anatomies (n=1,000) were generated using a CT-based statistical shape model. (A2,3) Functional twinning was achieved by mapping LGE-MRI-derived fibrosis distributions and calibrating tissue. Fibre orientation was assigned using a population-averaged human fibre atlas and cellular heterogeneity was introduced with respect to anatomical regions. (A4,5) AF induction: Multiple AF episodes were initiated using varied initial conditions. PVI lesions were applied after 5 s of AF evolution. Recordings from simulations that sustained AF post-PVI (n=250) were selected to form the PeAF cohort. (B) Consensus EOF mapping. (B1,2) PS maps for the post-PVI AF recordings of LA and RA were calculated and binarized. (B3) Consensus PS maps were computed using 5 × binarized PS maps per case. (B4,5) HS optical flow for the consensus PS regions was calculated and centroids of high value curl clusters were computed for each AF evolution. B6,7) Weighted average of curl cluster centroids were computed to estimate EOF ablation targets. (C) EOF-based ablation planning: (C1,2) Curl cluster centroids were projected onto 3D LA and RA…

• 

  Figure 2. Conceptual framework: EOF mapping utilizing a single AF map for generating phase map and average optical flow curl map as inputs for generating an EOF map. Created in BioRender. Jaffery, O. (2025) https://BioRender.com/4ui5k7d.

• 

  Figure 3. Electro-optical flow lesion generation algorithm: (I) Re-entry simulated on anatomical LA model. (II) Our pipeline estimates ablation targets using phase mapping (II)(Ai–Aiii) and optical flow (II)(Bi–Bii). The overlap demonstrates how phase mapping identifies substrate locations that causes re-entry to anchor, while the precise anchor location is pinpointed within the region outlined by PS using HS optical flow. III) An EOF map estimates optimal ablation lesion out of large regions identified by 4 × PS (serial 1–4). Within each region spatial coordinates of the largest optical curl cluster is identified as EOF ablation target. An example is shown inside serial 1 on (IV) where ablation target is connected to the PVI inert boundary using an ablation line forming an EOF lesion. Created in BioRender. Jaffery, O. (2025) https://BioRender.com/4ui5k7d.

• 

  Figure 4. (A) Spiral re-entry AF evolution for 3D biatrial anatomy. universal atrial coordinates (UACs) allow analysis on a standard 2D unit square. Using UACs the AF evolution is projected to (I) 2D right atrium (RA) and (II) 2D left atrium (LA). Evolution equation transforms Cartesian coordinates (X, Y) into a parametric variable t for generating Archimedean spirals as initial conditions. Radial distance (Dspiral), angular position (ω), and periodicity (κ) control the spatial structure of the spiral. (B) Varied AF episodes: variables X and Y represent spatial coordinates, while xcen, ycen define the centre of the spiral. The coordinates are shifted to (Xshift,Yshift) to allow variations in initial conditions. (C) Ensemble of five representative AF evolutions at t=450 ms. Created in BioRender. Jaffery, O. (2025) https://BioRender.com/4ui5k7d.

• 

  Figure 5. A consensus-based electro-optic flow ablation strategy for persistent AF. Created in BioRender. Jaffery, O. (2025) https://BioRender.com/4ui5k7d.

• 

  Figure 6. Burst pacing protocol for ascertaining post-ablation inducibility and ablated tissue surface area for the tested ablation strategies. Created in BioRender. Jaffery, O. (2025) https://BioRender.com/4ui5k7d.

• 

  Figure 7. Myocardium preserving benefit of a consensus based EOF ablation strategy: PVI and empiric ablations result in sustained AF in 6/10 and 3/10 cases (A and B). Ablating all PS regions resulted in AF non-inducibility; however, a large tissue area (over 50%) was ablated (C). Using a consensus-based strategy maintained AF non-inducibility through identifying consensus based PS regions common in all cases (D). Electro-optic flow ablation using single map (E) renders AF not inducible through targeted EOF lesions as compared to PS based ablations occupying lesser surface area in comparison. As compared to single map PS ablations, consensus-based EOF reduces the ablated tissue area by 50% while rendering AF non-inducible from any of the burst pacing locations (F). Created in BioRender. Jaffery, O. (2025) https://BioRender.com/4ui5k7d.

• 

  Figure 8. Pre- and post-ablation comparison of EGM cycle length and EGM fractionation: Pre-ablation EGMs at recording points (R1–R4) demonstrating fibrillatory activity region on the LA roof with fractionated EGMs (A1) and high average optical curl (B1). Ablation lesions delivered based on EOF mapping (C2) removed fractionation in EGMs (A2) and reduced the curl of average optical flow (B2). Regions of agreement between EGMs (A1, A2) and optical curl (B1, B2) show the effectiveness of EOF as a multimodal metric for identifying non-PVI ablation targets. Created in BioRender. Jaffery, O. (2025) https://BioRender.com/4ui5k7d.

• 

  Figure 9. (A) Comparison of ablation outcomes in the cohort of 250 PeAF patients: Average percent inducibility of each ablation type, along with the corresponding percent tissue area ablated for tested ablation strategies to illustrate the relative benefit of each non-PVI ablation strategy. (B) AF non-inducibility to ablated tissue area ratio computed to calculate AEI for the tested ablation strategies. A representative ablation mesh for each ablation type is shown at the end of each bar. Consensus-based EOF resulted in the lowest amount of tissue area ablated with AF rendering non-inducible in the highest percentage of cases. Created in BioRender. Jaffery, O. (2025) https://BioRender.com/4ui5k7d.