Table 1.
Pulmonary arterial hypertension: Subjects #1–11 with pulmonary arterial hypertension (mean pulmonary arterial pressure ≥ 25 mmHg).
Table 2.
Subjects #12–22 with normal pulmonary arterial pressures (mean pulmonary arterial pressure <25mmHg).
Table 3.
Pulmonary Vascular Hemodynamic data.
Subjects #1–11 with Pulmonary arterial hypertension (mean pulmonary arterial pressure ≥25 mmHg).
Table 4.
Pulmonary Vascular Hemodynamic data.
Subjects #12–22 with normal pulmonary arterial pressures (mean pulmonary arterial pressure <25mmHg).
Table 5.
Systemic Vascular Hemodynamic and Electrocardiographic data.
Subjects #1–11 with pulmonary arterial hypertension (mean pulmonary arterial pressure ≥25 mmHg).
Table 6.
Systemic Vascular Hemodynamic and Electrocardiographic data.
Subjects #12–22 with normal pulmonary arterial pressures (mean pulmonary arterial pressure <25 mmHg).
Table 7.
Comparison of clinical and hemodynamic data between subjects with pulmonary arterial hypertension (mean PAp ≥25 mmHg) and normal pulmonary arterial pressure (mean PAp <25 mmHg).
Fig 1.
Demarcation of the 1st (S1) and 2nd (S2) heart sounds.
The normalized amplitude (y-axis) is plotted against time in seconds (x-axis). Time zero second depicts the annotated peaks for S1 and S2 events.
Fig 2.
Power spectrum of ‘db6’ wavelet for details (top) and approximations (bottom) at scales a = 2j, j = 1, .., 6.
Note, the sampling frequency of the heart sounds is 4000 Hz.
Fig 3.
Behavior of the ‘db6’ wavelet dealing with different morphologies of S1 and S2.
(a) Wavelet details for heart sounds with low S1 amplitude measured at the second intercostal space for a subject with mean PAp < 25 mmHg, (b) Wavelet approximations for the same heart sounds used in (a), (c) Wavelet details for heart sounds with low S2 amplitude measured at apex for a subject with mean PAp ≥ 25 mmHg, and (d) Wavelet approximations for the same heart sounds used in (c).
Fig 4.
Flowcharts for five methods to detect S1 and S2 waves in heart sounds.
(a) Method I, (b) Method II, (c) Method III, (d) Method IV, (e) Method V.
Fig 5.
(a) Original heart sound signal from a subject with mean pulmonary arterial pressure of 20 mmHg (b) second-order Shannon energy of D5 wavelet in Method I (c) second-order Shannon energy of D6 wavelet in Method II (d) third-order Shannon energy in Method III (e) wavelet approximation A6 in Method IV (f) generating blocks of interest in Method V.
Fig 6.
Power spectrum of S1 and S2 segments compared to the power spectrum of wavelet details used in all methods at the second intercostal space (left) and apex (right).
A total of 284 heart beats used in this analysis for subjects with mean PAp > = and < 25 mmHg. Note, the sampling frequency of the heart sounds is 4000 Hz.
Table 8.
The function that detects the first heart sound (S1) and the second heart sound (S2) waves has five inputs: the heart sound signal (HSsignal), event-related durations W1, W2, anticipated block width (BlockSize), and the offset (β). Daubechies 'db6' wavelet is used for filtering the signal and the wavelet detail D6 represents the heart sounds in the analysis.
Table 9.
A rigorous optimization over all parameters of Method V: event-related durations W1, W2, anticipated block width (BlockSize), and the offset (β).
All possible combinations of parameters (46,376 iterations) have been investigated and sorted in descending order according to their overall accuracy. The data used in this training phase was heart sounds measured at apex for all subjects with mean PAp ≥ 25 mmHg. The overall accuracy is the average value of SE and +P.
Table 10.
Comparison of the first (S1) and second heart sound (S2) detection algorithms.
To evaluate the performance of the detectors, two statistical measures were used: SE = 100×(TP/(TP+FN))and +P = 100×(TP/(TP+FP)), where TP is the number of true positives (S1/S2 detected as S1/S2), FN is the number of false negatives (S1/S2 has not been detected), and FP is the number of false positives (non-S1/S2 detected as S1/S2).
Fig 7.
Methods performance in detecting first (S1) (left column) and second S2 (right column) heart sound waves.
The black circle represents the annotated S1/S2 wave, and the green star represents the detected S1/S2 wave using each algorithm. If the black circle is empty it means a false negative, while the red circle means a false positive.
Fig 8.
Performance at different signal-to-noise ratio (SNR) levels.
It is clear that the overall accuracy of Method V increases when the SNR increases compared to the other methods.