Table 2.

Respiratory sounds analysis related papers.

Reference	Year	Topic	Data and Cohort	Recording Device	ML Models Used	Data Processing Methods	KPIs
[43]	2018	Symptom Identification: Wheeze	Private-255 breathing cycles, 50 patients	Smartphone	SVM	Bag-of-Words To Features	Acc: 75.21%
[44]	2020	Disease Identification: Bronchitis, Pneumonia	Private-739 recordings	Various Microphones	K-NN	EMD, MFCCs, GTCC	Acc: 99%
[45]	2021	Disease Identification: Bronchial Asthma	Private-952 recordings	High-end Microphones	NN, RF	Spectral Bandwidth, Spectral Centroid, ZCR, Spectral Roll-Off, Chromacity	Sens: 89.3% Spec: 86% Acc: 88% Youden’s Index: 0.753
[46]	2019	Disease Identification: COPD	Private-55 recordings	Stethoscope	Fine Gaussian SVM	Statistical Features, MFCCs	Acc: 100%
[47]	2018	Disease Identification: Asthma, COPD	Private-80 normal, 80 COPD, and 80 asthma recordings	Stethoscope	ANN	PSD Extracted Features, Feature Selection (ANOVA)	Acc: 60% Spec: 54.2%
[48]	2022	Disease Identification: COVID-19	Coswara-120 recordings from COVID-19 patients, 120 recordings from Healthy patients	Various Microphones	Neural Network	Statistical and CNN-BiLSTM Extracted Features	Acc: 100% (shallow recordings), 88.89% (deep recordings)
[49]	2021	Disease Identification: COVID-19	COVID-19 Sounds-141 recordings	High-end Microphones	VGGish	Spectral Centroid, MFCCs, Roll-off Frequency, ZCR	ROC-AUC: 80% Prec: 69% Recall: 69%
[50]	2022	Symptom Identification: Wheeze, Crackle	Respiratory Sounds Database (RSDB) and private-943 recordings	Various Microphones	ResNet	Padding, STFT, Spectrum Correlation, Log-Mel Spectrograms, Normalization	Sens: 76.33% Spec: 78.86%
[51]	2021	Symptom Identification: Wheeze, Crackle	RSDB-920 recordings	Various Microphones	ANN, SVM, RF	Time Statistics and Frequency features	Acc: NN: 73%, RF: 73%, SVM: 78.3%
[52]	2019	Symptom Identification: Wheeze, Crackle	Private-21 normal samples, 12 wheezes, and 35 crackles	Stethoscope	SVM	CWT, Gaussian Filter, Average Power, Stacked Autoencoder	Acc: 86.51%
[53]	2019	Symptom Identification: Wheeze, Crackle	RDSB’s stethoscope recordings-834 recordings	Stethoscope	ResNet	Optimized S-transform	Sens: 96.27% Spec: 100% Acc: 98.79%
[54]	2022	Symptom Identification: Wheeze, Crackle	RDSB-920 recordings	Various Microphones	VGG-16	Fluid-Solid Modeling, Recording Simulation, Downsampling, Feature Extraction	Sens: 28%, Spec: 81%
[55]	2020	Disease Identification: Bronchiectasis, Bronchiolitis, COPD, Pneumonia, URTI, Healthy	RDSB-920 recordings	Various Microphones	RF	Resampling, Windows, Filtering, EMD, Features	Acc: 88%, Prec: 91%, Recall: 87%, Spec: 97%
[47]	2020	Symptom Identification: Wheeze, Crackle	Private-705 lung sounds (240 crackle, 260 rhonchi, and 205 normal)	Stethoscope	SVM, NN, K-Nearest Neighbors (K-NN)	CWT	Acc: 90.71%, Sens: 91.19%, Spec: 95.20%
[56]	2021	Symptom Identification: Crackle, Normal, Stridor, Wheeze	Private-600 recordings	Stethoscope	SVM, K-NN	Filtering, Amplification, Dimensionality Reduction, MFCCs, NLM Filter	Acc: SVM: 92%, K-NN: 97%
[57]	2021	Symptom Identification: Wheeze, Crackle	RSDB-920 recordings	Various Microphones	2D CNN	RMS Norm, Peak Norm, EBU Norm, Data Augmentation	Acc: 88%
[58]	2019	Symptom Identification: Wheeze, Crackle	Private-384 recordings	Stethoscope	VGGish-BiGRU	Spectrograms	Acc: 87.41%
[59]	2017	Symptom Identification: Wheeze, Crackle	Private-60 recordings	Stethoscope	Gaussian Mixture Model	MFCCs	Acc: 98.4%
[60]	2017	Symptom Identification: Wheeze, Crackle	Private-recordings containing 11 crackles, 3 wheezes, 4 stridors, 2 squawks, 2 rhonchi, and 29 normal sounds	Digital stethoscope	MLP	EMD, IMF, Spectrum, Feature Extraction	Acc: Crackles 92.16%, Wheeze 95%, Stridor 95.77%, Squawk 99.14%, Normal 88.36%, AVG 94.82%
[61]	2021	Symptom Identification: Wheeze, Crackle	RSDB-920 recordings	Various Microphones	VGG-16	Resampling, Windows, Filtering, Mel spectrogram (Mel, Harmonic, Percussive, Derivative)	Acc: Wheeze 89.00%, Rhonchi 68.00%, Crackles 90.00%
[62]	2020	Symptom Identification: Wheeze, Crackle	RSDB-920 recordings	Various Microphones	ResNet	Resampling, Windows, Filtering, Data Augmentation, Mel-spectrogram, Device Specific Features	80/40 Split 4 class (per device): Spec: 83.3%, Sens: 53.7%, Score: 68.5%
[63]	2021	Symptom Identification: Crackle, Wheeze – Disease Identification: Asthma, Cystic Fibrosis	Private-Recordings from 95 patients	Various Microphones	N/A	N/A	85% agreement (k = 0.35 (95% CI 0.26-0.44)) between conventional and smartphone auscultation Features
[64]	2021	Symptom Identification: Wheeze, Crackle, Other	RSDB-920 recordings	Various Microphones	LDA, SVM with Radial Basis Function (SVMrbf), Random Undersampling Boosted trees (RUSBoost), CNNs.	Spectrogram, Mel-spectrogram, Scalogram, Feature Extraction	Acc: 99.6%
[65]	2022	Symptom Identification: Wheeze, Crackle, Normal	RSDB-920 recordings	Various Microphones	Hybrid CNN-LSTM	Feature Extraction	Sens: 52.78% Spec: 84.26% F1: 68.52% Acc: 76.39%

Note. ML models: SVM = Support Vector Machine; K-NN = K-Nearest Neighbors; RF = Random Forest; ANN = Artificial Neural Network; NN = Neural Network; CNN = Convolutional Neural Network; MLP = Multilayer Perceptron; RUSBoost = Random Undersampling Boosted trees; LSTM = Long Short-Term Memory; LDA = Linear Discriminant Analysis. Data Processing Methods: EMD = Empirical Mode Decomposition; MFFC = Mel-Frequency Cepstral Coefficient; GTCC = Gamatone Cepstral Coefficient; ZCR = Zero-Crossing Rate; PSD = Power Spectral Density; STFT = Short Time Fourier Transform; CWT = Continous Wavelet Transform; S-Tranform = Stockwell Transform; NLM = Non-Local Means; RMS = Root Mean Square. Metrics: Acc = Accuracy; Sens = Sensitivity; Spec = Specificity AUC = Area Under Curve; IoU = Intersection over Union; ROC Curve = Receiver Operating Characteristic Curve; AROC = Area under the ROC curve.