| [21] |
✖ |
✔ |
Linear Predictive Cepstral Coefficients (LPCC) |
Multilayer Perceptron Classifier |
Accuracy of 99.22% was obtained |
| [22] |
✖ |
✔ |
Mel Spectrograms with clipped black (zero energy) regions |
RespireNet framework consisting of ResNet-34 trained on concatenation-based augmented samples of respiratory cycles along with device- specific optimizations. |
Achieved a sensitivity of 0.54 and specificity of 0.83 in classifying wheezes (W), crackles (C), both wheezes and crackles (B) and healthy/normal (N) respiratory cycles. |
| [23] |
✖ |
✔ |
Mel Frequency Cepstral Coefficients (MFCCs) and Power Spectrum Density (PSD) |
For breath detector, models such as KNN, Random Forest and Logistic Regression were proposed. |
All models achieved a precision of 0.98 and recall of 0.99, 0.98 and 0.99 respectively. |
| For anomaly detection engine, models such as Logistic Regression, SVM, ANN, Random Forest and KNN were used. |
SVM and Logistic Regression achieved a precision and recall of 0.93, 0.94 and 0.91, 0.91 respectively, All other models achieved a precision and recall of 0.92 and 0.91 respectively. |
| [24] |
✖ |
✔ |
Short Time Fourier Transformation (STFT) |
Pretrained Deep Convolutional Network + SVM Classifier |
Accuracy of 65.5% was obtained |
| Pretrained Deep Convolutional Network fined tuned for classification |
Accuracy of 63.09% was obtained |
| [25] |
✔ |
✔ |
Mel Frequency Cepstral Coefficients (MFCCs) |
Recurrent Neural Networks like LSTM, GRU, BiGRU and BiLSTM |
Sensitivity and Specificity of 64% and 82% were obtained respectively |
| [26] |
✖ |
✔ |
Mel Frequency Cepstral Coefficients (MFCCs). |
Noise Masking Recurrent Neural Network (NMRNN) |
Sensitivity and Specificity of 56% and 73.6% were obtained respectively for end to end classification |
| [27] |
✖ |
✔ |
Mel Frequency Cepstral Coefficients (MFCCs). |
Hidden Markov Models in combination with Gaussian mixture models |
Best Score achieved in second evaluation phase of ICHBI was 39.56 |
| [19] |
✔ |
✖ |
Abnormal lung sounds, presence of breathlessness, peak meter readings and family history |
Logistic Regression with L1 Regularization |
Achieved 0.95 AUC score in separating COPD and asthma patients from other categories of diseases and 0.97 AUC score in distinguishing COPD and Asthma |
