Table 1 Summary of dataset characteristics.
From: TriSpectraKAN: a novel approach for COPD detection via lung sound analysis
Category | ICBHI 2017 challenge database | Chest wall lung sound database (CWLSD) | Respiratory database@TR (RD@TR) |
|---|---|---|---|
Dataset origin | Created by research teams in ICBHI 2017 respiratory sound database25 | Publicly available repository from King Abdullah University Hospital19 | Publicly available database with recordings from different COPD severity levels26 |
Number of recordings | 920 annotated audio recordings, varying lengths (10 s to 90 s) | 336 lung sound signals collected from 112 subjects | 12-channel recordings from subjects with different COPD severity levels |
Patient data | From 126 patients of all age groups | 112 subjects with a wide spectrum of respiratory diseases | Subjects with different COPD severity levels, ranging from COPD0 to COPD4 |
Recording duration | 5.5 h of audio recordings | 10 to 50 s per audio signal | Minimum 17 seconds of recording |
Diagnosis (classes) | URTI, COPD, asthma, LRTI, bronchiectasis, pneumonia, bronchiolitis | Asthma, COPD, bronchiolitis obliterans, heart failure, pulmonary fibrosis, etc. | Different COPD severity levels |
Respiratory cycles | 6898 cycles, with 1864 with crackles, 886 wheezes, and 506 having both | Not specified | Not specified |
Noise simulation | Contains clean and noisy recordings; noisy recordings simulate real-life conditions | Not specified | 85% of ambient noise is reduced from the lung sound recordings. |
File formats | .wav files | .wav files and annotation .txt files | Not specified |
Sampling rate | Not specified | 4 kHz using Littmann 3200 digital stethoscope | 4 kHz using Littmann 3200 digital stethoscope |
