A portable Raspberry Pi-based system for diagnosis of heart valve diseases using automatic segmentation and artificial neural networks

Figures & data

Figure 1. The block diagram of the proposed diseases classification approach

Figure 2. A visual illustration of the approach used for determining the period (T)

Figure 3. The flow chart of the proposed automatic segmentations algorithm

Figure 4. Illustrating the importance of satisfying the two main criteria for extracting the 4-sec segment from the SEE. These two criteria are not considered in (a)–(c) but considered in (d)–(f). (a), (d) Shannon Energy Envelope (SEE) of PCG signal; (b), (e) 4-sec segment from the SEE; (c), (f) determining the period (T)

Figure 5. Illustrating the importance of satisfying the two main criteria for defining the second window (W2(t)) within S(t). (a) determining the period (T), where the two main criteria are ignored; (b) determining the period (T), where the two main criteria are considered

Figure 6. Segmentation algorithm results for PCG signals for (1) normal case and (2) abnormal case: (a1), (a2) preprocessed PCG signal; (b1), (b2) the murmur-attenuated signal; (c1), (c2) Shannon Energy Envelope (SSE); (d1), (d2) 4 sec segment from the SSE; (e1), (e2) determining the period (T); (f1), (f2) a single-segmented cardiac cycle

Figure 7. OBW illustrated on PSD curve of the single cycle of normal PCG signal

Figure 8. Wavelet approximation coefficient (A5) and Wavelet detail coefficients (D5, D4, D3) resulting from decomposing the single cycle of normal PCG signal with ‘db6ʹ wavelet

Figure 9. The feedforward backpropagation ANN used in classification

Table 1. The performance of several feedforward backpropagation ANN models

Feedforward backpropagation ANN model	Activation function in the hidden layer	Activation function in the output layer	Classification accuracy %
1	Logsig	Tansig	91.66
2	Tansig	Logsig	93.33
3	Logsig	Pureline	90
4	Tansig	Pureline	93.33
5	Logsig	Logsig	95
6	Tansig	Tansig	86.66

Table 2. Performance comparison between the adopted backpropagation

ANN architecture	Classification accuracy %
The adopted backpropagation ANN (with logsig functions) (nine neurons in the hidden layer and four neurons in the output layer)	95
Radial basis ANN (nine neurons in the radial basis layer and four neurons in the output linear layer)	83.33
LVQ ANN (11 neurons in the competitive layer and 9 neurons in the output linear layer)	90
Probabilistic ANN (140 neurons in the pattern layer and 9 neurons in the competitive layer)	95

Figure 10. The diagnostic hardware system

Figure 11. [Courtesy of Medical Electronics Lab], A demo of a clinical test process using the diagnostic hardware system

Figure 12. The block diagram of PCG signal acquisition module

Figure 13. Connecting the microphone to the chest piece

Figure 14. Processing and displaying unit

Figure 15. The system’s GUI designed in the software of Raspberry Pi

Table 3. Obtained performance metrics of the diagnostic hardware system

Accuracy	Sensitivity	Specificity
96%	95.23%	100%

Table 4. Comparison between the proposed system and previous systems

Reference	Used signal	No. of diagnosed heart conditions	Segmentation method	Diagnosis method
(Grzegorczyk et al., 2016)	PCG	Two valvular heart cases	Based on hidden markov model (needs a long processing time)	Pre-recorded signal analysis
(Emre & Uguz, 2011)	PCG	Three valvular heart cases	Manually	Pre-recorded signal analysis
(Noman et al., 2018)	PCG	Four valvular heart cases	Based on wavelet decomposition (needs a long processing time)	Pre-recorded signal analysis
(Suboh et al., 2008)	PCG	Five valvular heart cases	Manually	Pre-recorded signal analysis
(Alfarhan et al., 2018)	Electrocardiogram (ECG)	Five nonvalvular cases (arrhythmias)	Based on a fixed time window (needs a short processing time)	Real-time patient examination
Proposed system	PCG	Nine valvular heart cases	Using a proposed algorithm in the time domain (needs a short processing time)	Real-time patient examination

Grzegorczyk, I., Solinski, M., Lepek, M., Perka, A., Rosinski, J., Rymko, J., Stepien, K., & Gieraltowski, J. (2016). PCG classification using a neural network approach. Computing in Cardiology Conference. Vancouver: IEEE. https://doi.org/10.22489/CinC.2016.323-252

 Google Scholar

Emre, G., & Uguz, H. (2011). Classification of heart sounds based on the least squares support vector machine. International Journal of Innovative Computing, Information & Control, 7(12), 7131–7144.

 Web of Science ®Google Scholar

Noman, F., Ghani, U., Saba, T., Rehman, A., & Iqbal, S. (2018). Microscopic abnormality classification of cardiac murmurs using ANFIS and HMM. Microscopy Research and Technique, 81(5), 449–457. https://doi.org/10.1002/jemt.22998

 PubMed Web of Science ®Google Scholar

Suboh, M., Mashor, M., Saad, A., & Mohamed, M. (2008). Classification of heart valve diseases using correlation analysis. 5th International Conference on Electrical Engineering, Computing Science and Automatic Control (pp. 173–176). IEEE https://doi.org/10.1109/ICEEE.2008.4723402.

 Google Scholar

Alfarhan, K., Mashor, M., Saad, A., & Omar, M. (2018). Wireless heart abnormality monitoring kit based on Raspberry Pi. Journal of Biomimetics, Biomaterials and Biomedical Engineering, 35, 96–108. https://doi.org/10.4028/www.scientific.net/JBBBE.35.96

 Web of Science ®Google Scholar