Table 1.
Summary of the previously published methods for atrial scarring segmentation and our proposed method
| References | Subjects (number) | Cardiac anatomy segmentation (modality) | Atrial scarring segmentation | Evaluation of atrial scarring segmentation (results: mean ± std) |
|---|---|---|---|---|
| Oakes et al.10 | Human (81) | Manual segmentation of LA wall (LGE MRI) | 2–4 SD | Atrial scarring percentage (8 ± 4, 21 ± 6, 50 ± 15)a |
| Knowles et al.23 | Human (7) | Semi‐automatic thresholding and region growing (MRA) | Maximum intensity projection | Atrial scarring percentage (31 ± 10)b |
| Perry et al. (2012)68 | Human (34) | Manual segmentation of LA wall (LGE MRI) | k‐means Clustering | Dice (81 ± 11, ground truth by manually selected thresholds) |
| Ravanelli et al.15 | Human (10) | Manual segmentation of LA and PV in 3D (MRA) | 4 SD | Dice (60 ± 21 ground truth by a semi‐automatic approach)c |
| Karim et al.25 | Human (15) | Statistical shape model with manual correction (b‐SSFP) | Graph cuts | Dice, ROC, and total scar volumed |
| Tao et al.24 | Human (46) | Automatic atlas‐based method with level set refinement (MRA) | Maximum intensity projection | Qualitative visualization (N/A) |
| Proposed method | Human (37) | Fully automated multiatlas whole heart segmentation (b‐SSFP) | Super‐pixel and SVM | Multiple quantitative metrics (Dice: 79 ± 5) |
a
Results (%) for mild (n = 43), moderate (n = 30), and extensive (n = 8) enhancement cases.
b
Moderate and extensive enhancement cases.
c
The Dice score was calculated for an automated atrial scarring segmentation. The method was also evaluated using Bland–Altman analysis of the atrial scarring percentage (after skeletonization) obtained from LGE MRI and EAM.
d
Multiple Dice scores were calculated for various experimental settings, and they were reported by plotting the median Dice scores (around 80) with the minimum and the maximum.