Table 1 LV-scar detection performance by our method in terms of positive predictive value (PPV), sensitivity (Se), specificity (Sp), F1 score (F1), and 95% confidence interval around accuracy.

From: Explainable artificial intelligence identifies and localizes left ventricular scar in hypertrophic cardiomyopathy using 12-Lead electrocardiogram

Region

John Hopkins Dataset

# of patients = 500 (5-fold)

UCSF Dataset

# of patients = 248 (left-out test)

PPV

Se

Sp

F1

95% CI Acc.

PPV

Se

Sp

F1

95% CI Acc.

Basal

94

87

90

90

86.2–87.4

88

85

82

87

81.3–82.7

Mid

91

87

85

89

86.4–88.3

85

85

83

85

86.1–87.5

Apical

89

85

91

87

85.6–86.9

88

82

92

85

84.6–85.7

Global LV Scar determination

\(\:basal\:\varvec{o}\varvec{r}\:mid\:\varvec{o}\varvec{r}\:apical\)

90

95

80

92

86.9–88.0

88

91

78

89

84.5–86.1

  1. \(\:Sp=\frac{TN}{TN+FP}\), \(\:Se=\frac{TP}{TP+FN}\:\), \(\:PPV=\frac{TP}{TP+FP}\), \(\:F1=\frac{TP}{TP\:+\:\frac{1}{2}(FP+FN)}\), where TP (True Positives) denotes LV scar correctly identified by XplainScar. TN (true negatives) denotes patients without LV scar on MRI who are not identified to have LV scar by XplainScar; FP (false positives) denotes patients who do not have LV scar by MRI but are misclassified by XplainScar to have LV scar. FN (false negatives) denotes patients who were incorrectly labeled by XplainScar to not have LV-scar.
Back to article page