Table 1 Performance for discriminating other respiratory sounds from wheezing.

From: An explainable and accurate transformer-based deep learning model for wheeze classification utilizing real-world pediatric data

Models

Selected hyper-parameters with grid search

Accuracy

AUC

Precision

Recall

F1-score

Primary study results(ResNet34 + CBAM)

Epoch: 120 / Batch Size: 32 / Learning Rate: 1e-3

0.912

0.891

0.944

0.810

0.872

AST model using primary study data

Epoch: 100 / Batch Size: 16 / Learning Rate: 1e-4

0.930

0.944

0.840

1.000

0.913

ResNet34 + CBAM of follow-up study data

Epoch: 120 / Batch Size: 32 / Learning Rate: 1e-3

0.836

0.758

0.742

0.590

0.657

AST model using follow-up study data

Epoch: 150 / Batch Size: 10 / Learning Rate: 5e-6

0.911

0.866

0.882

0.769

0.822

  1. Abbreviations: AUC, area under the curve; CBAM, convolutional block attention module; ResNet, residual network; AST, audio spectrogram transformer.
Back to article page