Table 2 Machine learning in computerized respiratory sound analysis systems
From: Classification of lung sounds using convolutional neural networks
Author | Subjects | Classified items | Feature extraction method | Classification method | Accuracy |
|---|---|---|---|---|---|
Forkheim 1995 [39] | Not specified | Wheeze and normal | Raw signal data, FFT | MLP | The training sets 1 and 2 were 93 and 96% |
Kahya 1997 [16] | 69n | Normal or abnormal | AR model | k-NN | 69.59% |
Rietveld 1999 [38] | 60n | Normal and asthma | FT | MLP | 43% |
Oud 2000 [26] | 10n | Asthmatic patients | Spectral analysis | k-NN | 60 to 90% |
Waitman 2000 [25] | 17p, 17c | Normal or abnormal | FT | MLP | 73% |
Bahoura 2003 [27] | 24n | Wheeze | MFCC, FFT, LPC, WPD, SBC | VQ | 75.80 and 77.50% |
Baydar 2003 [28] | 20n | Normal or abnormal | Periodogram, Welch, Yule-Walker, Burg | Nearest mean classifier | 72% in expiration and 69% in inspiration |
Kandaswamy 2004 [44] | Not specified | Lung sounds | WT, STFT | MLP | 94.02% |
Folland 2004 [45] | Not specified | Lung sounds | Spectral computation parametric model, generation linear normalization | MLP, RBFN, CPNN ANN | 97.8% |
Güler 2005 [47] | 129n | Normal, wheeze, and crackles | Welch | MLP, GANN | ANN 81–91%, GANN 83–93% |
Martinez-Hernandez 2005 [29] | 19n | Normal or abnormal | Multivariate AR model | MLP | 87.68% |
Kahya 2006 [5 | 20p, 20c | Rale | WT | k-NN | 46% |
Lu 2008 [42] | Not specified | Fine and coarse crackles | GMM | GMM VQ | 95.1% |
Alsmadi 2008 [31] | 42n | Lung sounds | AR model | k-NN and minimum distance classifier | 96% |
Riella 2009 [40] | Not specified | Wheeze | FFT, STFT | MLP | 92.86% |
Riella 2010 [46] | Not specified | Lung sounds | DWT | RBFNN | 92.36% |
Yamamoto 2010 [48] | 114n | Normal or abnormal | Raw data | HMM | 84.2% |
Charleston-Villalobos 2011 [32] | 27n | Normal or abnormal | AR model | MLP | 75 and 93% |
Yamashita 2011 [33] | 168n | Normal or emphysema | Segmentation | HMM | 87.4 and 88.7% |
Feng 2011 [34] | 21n | Normal or abnormal | Temporal–spectral dominance spectrogram | k-NN | 92.4% |
Serbes 2011 [35] | 26n | Crackles | WT, DWT | SVM | 97.20% |
Flietstra 2011 [24] | 257n | Pneumonia and CHF | Manual crackle analysis | SVM | Pneumonia 86% and CHF 82% |
Hashemi 2011 [41] | 140p, 140s | Wheeze | WT | MLP | 89.28% |
Aras 2015 [36] | 27 pathological, 21 normal s | Rale, rhonchus, and normal | MFCC LFCC | k-NN | The datasets 1 and 2 were 96 and 100% |
Chen 2015 [37] | 20p | Rale, rhonchus, wheeze, and normal | MFCC | k-NN | 93.2% |