Table 1 Experiment results
From: Classification of lung sounds using convolutional neural networks
Training accuracy | Test accuracy | Training precision | Test precision | Training recall | Test recall | Training sensitivity | Test sensitivity | Training specificity | Test specificity | |
|---|---|---|---|---|---|---|---|---|---|---|
Classification of healthy versus pathological respiratory sounds | ||||||||||
CNN (spectrogram) | 87% | 86% | 90% | 86% | 89% | 86% | 89% | 86% | 95% | 86% |
SVM (MFCC) | 91% | 86% | 94% | 89% | 87% | 87% | 87% | 87% | 87% | 82% |
Classification of respiratory sounds labeled with a singular type | ||||||||||
CNN (spectrogram) | 90% | 76% | 94% | 79% | 86% | 74% | 86% | 74% | NA | NA |
SVM (MFCC) | 99% | 75% | 99% | 75% | 99% | 99% | 99% | 99% | NA | NA |
Classification of respiratory sounds labeled with only as type rale, rhonchus, and normal | ||||||||||
CNN (spectrogram) | 87% | 80% | 88% | 79% | 85% | 79% | 85% | 79% | NA | NA |
SVM (MFCC) | 89% | 80% | 89% | 80% | 89% | 89% | 89% | 89% | NA | NA |
Classification of respiratory sounds with all labels | ||||||||||
CNN (spectrogram) | 74% | 62% | 80% | 73% | 66% | 56% | 66% | 56% | NA | NA |
SVM (MFCC) | 78% | 62% | 78% | 62% | 78% | 78% | 78% | 78% | NA | NA |