Table 2 Table of evaluation results for each model for view classification

From: Deep echocardiography: data-efficient supervised and semi-supervised deep learning towards automated diagnosis of cardiac disease

Model

Train Accuracy

Test Accuracy

Training Time/Epoch

CNN

95.39%

92.05% (+/−0.72)

919 s

CNN with Segmentation

95.18%

93.64% (+/−0.61)

903 s

Resnet50*

92.61%

91.36% (+/−1.43)

7482 s

VGG16*

98.15%

83.67% (+/−2.68)

4559 s

   

(second stage)

Ensemble

—

94.40% (+/−0.52)

—

  1. Highest performing model by overall test accuracy is an ensemble of three CNN models with field of view segmentation, while transfer learning from Resnet50 and VGG16 models yielded lower test accuracies compared to a single CNN model.
Back to article page