Table 5 Diagnostic performance of baseline and proposed methods in terms of AUC. We evaluated the patient-level mean diagnostic performance for detection of stridor or snoring. The mean, standard deviation, and 95% confidence interval were obtained from 10-fold Monte Carlo validation.

From: Automatic stridor detection using small training set via patch-wise few-shot learning for diagnosis of multiple system atrophy

Metric

Method

No. training samples

4

6

8

Macro-avg AUC (95% CI)

Baseline

0.7961 ± 0.08 (0.747, 0.846)

0.8921 ± 0.04 (0.867, 0.917)

0.9203 ± 0.03 (0.902, 0.939)

Proposed (Ver. 1)

0.8941 ± 0.04 (0.869, 0.919)

0.9340 ± 0.04 (0.909, 0.959)

0.9526 ± 0.02 (0.940, 0.965)

Proposed (Ver. 2)

0.9151 ± 0.06 (0.878, 0.952)

0.9541 ± 0.03 (0.935, 0.973)

0.9597 ± 0.02 (0.947, 0.972)

Micro-avg AUC (95% CI)

Baseline

0.8152 ± 0.07 (0.772, 0.859)

0.8604 ± 0.08 (0.811, 0.910)

0.9105 ± 0.06 (0.873, 0.948)

Proposed (Ver. 1)

0.8833 ± 0.04 (0.858, 0.908)

0.9323 ± 0.04 (0.908, 0.957)

0.9411 ± 0.03 (0.923, 0.960)

Proposed (Ver. 2)

0.8974 ± 0.09 (0.842, 0.953)

0.9541 ± 0.03 (0.935, 0.973)

0.9578 ± 0.02 (0.945, 0.970)

  1. The highest values in the results for each number of training samples are shown in bold.
Back to article page