PK traditional diagnostic method, ADI peak and starting point detection algorithm is more reliable!

HRV is a tool that researchers use extensively to aid clinical diagnosis and measure health-related bioinformation, such as sleep stages, stress states, and fatigue. HRV is a tool that researchers use extensively to aid clinical diagnosis and measure health-related bioinformation, such as sleep stages, stress states, and fatigue. Given the technical requirements of ECG measurements, it is not always possible to obtain this signal at accident/disaster sites, battlefields, or in areas where electrical interference may be introduced into the electrocardiogram.

Pulse rate variability extracted from the photoplethysmographic signal may be used as an alternative to HRV. The PPG signal is obtained by illuminating the human skin with an LED and then measuring the changes in the intensity of the reflected light caused by blood flow with a photodiode.

In addition, PPG can also provide information related to the cardiovascular system, such as heart rate, arterial pressure, stiffness index, pulse transit time, pulse wave velocity, cardiac output, arterial compliance, and peripheral resistance. However, the performance of PPG-based algorithms is degraded by poor blood perfusion, ambient light, and most importantly, motion artifacts (MA). Many signal processing techniques have been proposed to eliminate MA noise, including ADI's motion rejection and frequency tracking algorithms, which are implemented by using a triaxial accelerometer placed close to the PPG sensor.

It is critical to accurately extract important points such as systolic peak, onset, and dicrotic notch from the PPG waveform for PRV analysis. The onset of the PPG waveform is caused by the start of blood output from the heart to the aorta, while the dicrotic notch indicates the end of ejection or closure of the aortic valve. The lack of a reliable detection algorithm for PPG signals has prevented researchers from fully performing PRV analysis using PPG, at least to a certain extent. Some previous studies on PRV have ignored the reference points, and some have reportedly used manual methods or empirical methods to detect the systolic peak, while others have been based on unvalidated time window algorithms to obtain the pulse peak.

This article proposes a robust peak and onset detection algorithm using a delineation method originally proposed for arterial blood pressure (ABP) waveforms. It is important to note that PPG signals from wrist-worn devices contain many motion artifacts, baseline fluctuations, reflected waves, and other noises that can affect the behavior of the detection algorithm. Therefore, the data is preprocessed before being fed into the beat-to-beat extraction model. The automatic delineator used in this work is a hybrid approach that uses different signals from the raw PPG preprocessing and the first-order derivative of the signal to extract the peak and onset. We use a large database collected through the ADI watch platform to provide synchronized PPG and ECG signals. Regarding memory size, this algorithm requires less memory and is available as an embedded algorithm in the ADI watch platform. The algorithm is validated using coverage, sensitivity, positive detection rate, and RMS of successive differences and compared with the beat-to-beat results from the ECG signal.

In this section, we present in detail the proposed beat-to-beat algorithm for wrist PPG signals, which consists of (i) preprocessing and (ii) high-resolution beat-to-beat extraction modules. The block diagram of the algorithm is shown in Figure 1.

Fig. 1. Flowchart of the proposed beat-to-beat extraction algorithm, including (i) preprocessing and (ii) high-resolution B2B extraction.

It is well known that the PPG signal is susceptible to poor blood perfusion of peripheral tissues and motion artifacts.18 In order to minimize the impact of these factors and prevent them from interfering with subsequent PPG analysis and beat-to-beat estimation, a preprocessing stage is required. This step includes:

• Framing and Windowing

• Bandpass filtering (0.4 Hz to 4 Hz)

• Automatic gain control (AGC), used to limit the signal amplitude

• Signal Amplitude

The TPPG input data is processed using a T0 second window, and further blocks of data are processed by moving windows with mT ₀ (m=3/4) overlap. A bandpass filter is then required to remove high frequency components (such as power) and low frequency components (such as changes in capillary density and venous blood volume, temperature changes, etc.) of the PPG signal.

Figure 2. PPG diagram.

Figures 2a and 2b show the PPG signal before and after filtering. The filter cutoff frequencies are 0.4Hz and 4Hz. The fundamental frequency of HR ranges from 0.4Hz to 3Hz. Therefore, using a higher range for beat-to-beat estimation can include harmonics that emphasize the beat number. A median filter is used to remove sudden spikes in the filtered signal. The AGC module then limits the signal level to ±V volts to verify the selected peak by confirming the amplitude of the signal at a later stage. The PPG measurement process for HRV is long in duration and inevitably introduces another artifact, such as baseline wander. Therefore, a low-pass finite impulse response (FIR) filter is used to smooth the PPG sample array within the frame (as shown in Figure 2c) to remove the baseline wander noise and obtain a smoother signal suitable for the delineation module.

The beat-to-beat extraction algorithm consists of the following modules:

• Interpolation

• Depiction

• High-resolution beat-to-beat extraction

• Signal quality indicators

The output of the preprocessing module is fed into the interpolation module to improve the accuracy of the beat-to-beat extraction algorithm. Given a G segment from t ₀ to t _τ in the first frame with beat-to-beat intervals b ₀ and b _τ- τ, we linearly interpolate the beat-to-beat interval values ​​using n points between the endpoints and then extract high-resolution beat-to-beat interval values ​​(e.g., 1 ms resolution) from b _{0 and bτ.τ.}

Next, the delineation module relies on signal morphology and rhythm information to extract the peak and onset. Therefore, for beat-to-beat detection, not only the systolic peak is required, but also the onset and dicrotic notch should be reported. The proposed delineator is theoretically similar to the delineators shown in two papers, “ An adaptive delineator for photoplethysmography waveforms ” and “ On an automatic delineator for arterial blood pressure waveforms ”, which adapt to the wrist PPG signal using a pair of inflection and zero-crossing points from the first-order derivative of the signal. Figure 2d depicts the inflection and zero-crossing points for PPG characterization. For the zero-crossing points, the signal is processed through a zero phase distortion filter to minimize the start and end transients by matching the initial conditions. This is to ensure that the time domain characteristics are still preserved after filtering.

Note that the onset from the derivative of the PPG waveform corresponds to the zero crossing before the maximum inflection point, while the systolic peak is associated with the zero crossing after this inflection point. The signal quality metric used in this beat-to-beat algorithm is clarity and indicates the pitch range of the signal. This metric was originally proposed in the article "A smarter way to find pitch" by Philip McLeod and Geoff Wyvill and uses the normalized squared difference function (a type of autocorrelation function) to find periodicity in the signal. We use this metric to determine when the beat-to-beat algorithm can reliably report peaks and onsets.

The results of our PPG beat-to-beat algorithm were compared with those of the Pan-Tompkins algorithm, which is a well-established ECG peak detection algorithm. The data collected was used to evaluate the algorithm using the ADI Vital Signs Monitoring (VSM) watch platform. The ADI VSM iOS app was used to interface with the watch via a Bluetooth® connection. The ADI watch contains a PPG sensor that collects PPG signals from the subject’s wrist. In addition, ECG signals are collected on the ADI watch. There are three ECG electrodes attached to the subject’s chest area. Wires from these electrodes are connected to the ADI watch to process these signals and record them simultaneously with the PPG signals. This platform provides synchronized PPG and ECG signals. Figure 3a shows the ADI watch used for data collection, while Figure 3b shows the iOS app interface and sample signals acquired from the platform.

Figure 3. ADI platforms and tools.

Before calculating the beat-to-beat metrics, it is important to perform an outlier removal process to identify missing/extra peaks in the Pan-Tompkins algorithm output and our PPG beat-to-beat algorithm output. Ignoring the missing/extra peaks may result in abnormal beat durations and inaccurate results. Missing/extra peaks in the ECG signal are determined by examining the durations of consecutive beats provided by the Pan-Tompkins algorithm. Any ECG peak that changes the beat duration by more than 20% is flagged as an outlier. After removing these ECG peaks, the missing/extra peaks in the PPG signal are determined by correlating each ECG peak with a peak in the PPG signal. A PPG peak is correlated with an ECG peak if it is within the temporal proximity of an ECG peak. When a PPG peak cannot be determined or when too many peaks are determined within the temporal proximity of an ECG peak, they are identified as outliers. Abnormal beat durations that may result from these missing/extra PPG beats are ignored as outliers during the calculation of the metrics.

Several metrics are calculated using the beat-to-beat values ​​derived from our proposed algorithm and the Pan-Tompkins algorithm. These metrics include:

(i) Coverage (Equation 1);

(ii) Sensitivity (Se) Equation 2);

(iii) positive detection rate (P+) Equation 3);

(iv) Root mean square of successive differences (RMSSD) Equation 4).

Figure 4 shows a visual representation of some of the values ​​used for the indicator calculation.

Figure 4. Shows the ECG and PPG signals with the IBI, and the peak and onset points of the raw PPG signal analyzed by the beat-to-beat algorithm.

Where TP (True Positive) indicates the number of heartbeats correctly identified by the PPG B2B algorithm, FP (False Positive) indicates the number of PPG heartbeats that do not correspond to the actual heartbeats in the ECG, and FN (False Negative) indicates the number of heartbeats missed by the PPG beat-to-beat algorithm. The interbeat interval (IBI) is the time interval between consecutive ECG peaks, PPG peaks, or PPG onsets.

To evaluate our algorithm, we collected both PPG and ECG signals from each subject. We collected data from a wide range of subjects of different ages, skin colors, and body shapes. This is to ensure that our evaluation results are applicable to all populations. Data was collected from 27 subjects (males and females with different skin colors), each for 2 minutes and 30 seconds. The subjects were required to stand for the first half of the time and sit for the second half. Table 1 shows the average values ​​of each metric obtained by the beat-to-beat algorithm. As shown in the table, the coverage, sensitivity, and positive detection rate of wrist data are all higher than 83% compared to the results from ECG signals, and the RMSSD average difference is less than 20 ms.

Table 1. Beat-to-Beat Metrics Results

This article proposes a robust peak and onset detection algorithm for PRV analysis of wrist PPG signals. The algorithm uses multiple preprocessing stages and proposes a hybrid delineation algorithm to detect the fiducials of wrist PPG signals. The proposed algorithm was tested using the ADI multi-sensory watch as an evaluation platform. The results showed strong correlation and consistency with ECG HRV. Future work will focus on applying motion suppression algorithms and handling the issue of missing heart beats in PRV analysis.

Give you a careful heart, please click "Watching"