Skip to main content

Table 3 Internal validation: segmentation performance on images with artifacts

From: Cardiovascular magnetic resonance images with susceptibility artifacts: artificial intelligence with spatial-attention for ventricular volumes and mass assessment

 

LV ED

LV ES

DSC

HD (mm)

Recall

Precision

DSC

HD (mm)

Recall

Precision

CNN vs GT

0.93 (0.91–0.95)⁎

7.8 (4.7–11.8)

0.92 (0.86–0.94)⁎

0.95 (0.93–0.98)⁎

0.91 (0.84–0.93)⁎

6.0 (4.6–9.5)⁎

0.88 (0.84–0.94)⁎

0.94 (0.83–0.95)⁎

Circle vs GT

0.43 (0.25–0.86)†

10.6 (2.7–20.0)

0.34 (0.18–0.80)†

0.61 (0.41–0.90)†

0.63 (0.14–0.83)†

11.1 (4.1–20.3)†

0.56 (0.08–0.77)†

0.74 (0.30–0.88)†

O1 vs O2

0.93 (0.82–0.95)

5.2 (3.5–10.2)

0.91 (0.78–0.97)

0.93 (0.88–0.95)

0.92 (0.85–0.96)

6.1 (3.4–8.1)

0.94 (0.85–0.97)

0.91 (0.84–0.93)

 

RV ED

RV ES

DSC

HD (mm)

Recall

Precision

DSC

HD (mm)

Recall

Precision

CNN vs GT

0.87 (0.84–0.91)⁎

10.5 (6.8–14.3)⁎

0.85 (0.81–0.90)⁎

0.91 (0.86–0.93)⁎†

0.83 (0.73–0.90)⁎

9.3 (6.7–13.9)⁎†

0.81 (0.70–0.89)⁎

0.88 (0.80–0.92)⁎†

Circle vs GT

0.59 (0.21–0.78)†

25.5 (15.7–47.3)

0.45 (0.14–0.72)†

0.74 (0.42–0.85)†

0.50 (0.19–0.82)†

21.4 (7.9–38.2)

0.43 (0.13–0.78)†

0.69 (0.31–0.86)

O1 vs O2

0.85 (0.70–0.90)

13.0 (6.5–20.6)

0.86 (0.70–0.95)

0.82 (0.73–0.86)

0.76 (0.70–0.85)

14.1 (9.5–24.3)

0.85 (0.65–0.93)

0.72 (0.65–0.79)

 

LVM ED

LVM ES

 

DSC

HD (mm)

Recall

Precision

DSC

HD (mm)

Recall

Precision

CNN vs GT

0.77 (0.71–0.82)⁎†

6.5 (5.2–9.0)⁎

0.82 (0.79–0.87)⁎†

0.70 (0.67–0.79)⁎†

0.82 (0.73–0.84)⁎

7.1 (5.1–11.4)

0.84 (0.76–0.91)⁎†

0.75 (0.70–0.81)⁎

Circle vs GT

0.41 (0.15–0.70)†

14.8 (6.8–30.4)

0.31 (0.11–0.68)†

0.50 (0.29–0.74)†

0.59 (0.10–0.79)†

10.7 (5.5–25.4)†

0.51 (0.07–0.79)†

0.59 (0.26–0.83)†

O1 vs O2

0.71 (0.62–0.78)

6.4 (3.7–11.8)

0.75 (0.68–0.79)

0.70 (0.61–0.80)

0.76 (0.66–0.81)

8.2 (4.3–9.9)

0.72 (0.62–0.85)

0.76 (0.71–0.84)

  1. Values are reported as median (interquartile range)
  2. Segmentation performance on images with artifacts for the proposed CNN and for the commercial software (Circle) compared to the manual gold standard (GT). Also, the results of the comparison between two observers (O1 vs O2) is reported
  3. Abbreviations as in Table 2. ⁎p < 0.05 CNN vs. Circle; †p < 0.05 vs. inter-observer variability