Table 6 Performance of PVP multi-classifier in confounded test group.

From: Artificial intelligence improves detection and classification of pulmonary venous hypertension related to left ventricular diastolic dysfunction by chest radiography

HGT Assignments

PVH Detection

[3-Class Output Normalized Accuracies]

HGT

Assignments

PVH-Severity Ranking

[4-Class Output Normalized Accuracies]

Normal

PVH

Stages 1 & 2+

Group 1 PH

Normal

PVH

Stage 1

PVH

Stage 2+

Group 1

PH

Normal

0.40

(4/10)

0.60

(6/10)

0.00

(0/10)

Normal

0.40

(4/10)

0.30

(3/10)

0.30

(3/10)

0.00

(0/10)

PVH

Stages 1 & 2+

0.05

(1/20)

0.85

(17/20)

0.10

(2/20)

PVH

Stage 1

0.00

(0/10)

0.40

(4/10)

0.40

(4/10)

0.20

(2/10)

PVH

Stage 2+

0.10

(1/10)

0.20

(2/10)

0.70

(7/10)

0.00

(0/10)

Group 1

PH

0.00

(0/10)

0.20

(2/10)

0.80

(8/10)

Group 1

PH

0.00

(0/10)

0.00

(0/10)

0.20

(2/10)

0.80

(8/10)

 

Overall Accuracy = 0.73 (29/40)

Overall Balanced Accuracy = 0.68

 

Overall Accuracy = 0.58 (23/40)

Overall Balanced Accuracy = 0.58

  1. HGT: Human Ground Truth, PH: Pulmonary Hypertension, PVH: Pulmonary Venous Hypertension, PVP: Pulmonary Vasculature Pattern, PVPI: Pulmonary Vasculature Pattern Identifier.
Back to article page