Automatic Localization of Stable Highest Dominant Frequency Area in AF Patients Based on Spatial Aggregation

Figures

• 

  Figure 1. Flowchart of atrial electrogram data processing. (A) Schematic diagram of spatial aggregation. (B) Schematic diagram of time-domain sparse sampling. (C) The amplitude and spectrum of signal when calculating DF. (D) Generating DF distribution model. (E) The result of local neighborhood spatial filtering. (F) The result of HDF extraction.

• 

  Figure 2. Binary search tree for time-domain sparse sampling.

• 

  Figure 3. Schematic diagram of the structure of local neighborhood space filtering.

• 

  Figure 4. LA-EGM Signal Preprocessing. (A) Original LA-EGM signal; (B) LA-EGM signal filtered by Botteron; (C) spectrum ranging from 0 to 30 Hz obtained through Welch.

• 

  Figure 5. Original 3D DF distribution model. (A) Anterior–Posterior. (B) Posterior–Anterior. LSPV: Left Superior Pulmonary Vein. RSPV: Right Superior Pulmonary Vein. LAA: Left Atrial Appendage. MV: Mitral Valve.

• 

  Figure 6. 3D DF distribution model after local neighborhood spatial filtering. (A) Anterior–Posterior. (B) Posterior–Anterior. Purple arrows indicate potential re-entry paths.

• 

  Figure 7. HDF area of the 3D DF model (red region). (A) Anterior–Posterior. (B) Posterior–Anterior.

• 

  Figure 8. The comparison between the HDF regions calculated by this paper and the actual surgical ablation targets. (A) The HDF area of the 3D DF model of PeAF patients (left side is Anterior–Posterior, right side is Posterior–Anterior). (B) The intraoperative ablation planning for PeAF patients (left side is Anterior–Posterior, right side is Posterior–Anterior). (C) The distribution of ablation targets for PeAF patients (left side is Anterior–Posterior, right side is Posterior–Anterior).

• 

  Figure 9. DF models generated with different voxel size (upper side is Anterior-Posterior, lower side is Posterior-Anterior). Gray arrows indicate the regions with significant differences. (A) Model generated with voxel size of 1 mm. (B) Model generated with voxel size of 2 mm. (C) Model generated with voxel size of 3 mm. (D) Model generated with voxel size of 4 mm.

• 

  Figure 10. Comparison of 3D DF distribution models generated without using the spatial aggregation method. Gray arrows indicate the locations of abnormal DF sites. (A) Model generated with spatial aggregation. (B) Model generated without spatial aggregation.

• 

  Figure 11. F1 Score under Different OI Thresholds.