Table 1 Comparison of Lung Sound Classification Models.
From: Deep Learning-Driven Early Diagnosis of Respiratory Diseases using CNN-RNN Fusion on Lung Sound Data
Author | Model Type | Dataset Used | Disease(s) Targeted | Accuracy (%) | F1-Score | Remarks |
|---|---|---|---|---|---|---|
Proposed Model | CNN + RNN (LSTM Fusion) | ICBHI, Coswara | Healthy, Pneumonia, Asthma, COPD, COVID-19 | 94.0 | 0.94 | Multi-disease classification with data augmentation and weighted mapping |
Kim et al. (2025) 25 | Transformer (AST Model) | Pediatric (2019–2020) | Wheezes, Other Respiratory Sounds | 91.1 | 0.82 | High accuracy in wheeze detection, uses Score-CAM for explainability, focused on pediatric data |
Wang et al. (2025) 26 | Transformer (REDT) | HF_Lung_V1 | Inspiration, Expiration, Continuous/Discontinuous Adventitious Sounds | 90.5% (Insp), 77.3% (Exp), 78.9% (Cont), 59.4% (Disc) | Not Mentioned | Event-based transformer for respiratory phase and adventitious sound detection, excels in detailed event localization |
Luay et al. (2021) 14 | Deep CNN Ensemble | ICBHI | Pediatric Asthma | 93.1 | 0.92 | Focused on asthma using ensemble CNN architecture |
Demir et al. (2020) 22 | RNN Ensemble | ICBHI | Respiratory Sound Classes | 94.6 | 0.93 | High temporal modeling, but limited disease annotation |
Borwankar et al. (2022) 17 | Multi-layer CNN | Custom Dataset | Respiratory Pathologies | 91.7 | 0.91 | Did not include COVID-19 or temporal modeling |
Ijaz et al. (2022) 16 | CNN | Custom Dataset | Abnormal Lung Sounds | 94.2 | 0.93 | No fusion with temporal analysis (RNN absent) |
Goyal et al. (2024) 11 | CNN-Based Classifier |  ~ 1500 recordings | Pneumonia |  > 90 | 0.90 | Dataset focused on pneumonia only |
Chen et al. (2018) 23 | CNN | Lung Sound Corpus | Crackles and Wheezes | 92.7 | 0.92 | No multi-class classification |
Pahar et al. (2022) 28 | CNN + STFT features | R.A.L.E Dataset | Wheeze Detection | 90.3 | 0.89 | Limited to binary classification |
