The PPG signal is an optical measurement of the arterial pulse wave (Charlton et al 2019), i.e. the wave generated when blood is ejected from the heart, temporarily increasing arterial pressure and causing vessel expansion and contraction (Alastruey et al 2023). Consequently, the PPG signal is influenced by a range of physiological systems, such as the heart, including heart rate, heart rhythm and the nature of ejection (Charlton et al 2022b), the blood vessels, including vessel stiffness, diameter and blood pressure (Charlton et al 2022b), the microvasculature, including peripheral compliance and resistance (Charlton et al 2022b), the autonomic nervous system, which influences heart rate variability (Gil et al 2010) and the respiratory system, which impacts the pulse wave through changes in intrathoracic pressure (Charlton et al 2017b). Thus, there is potential to extract much physiological information from the PPG signal.

The PPG signal can be acquired using a range of sensors and devices. PPG sensors consist of a light source such as a LED, and a light sensor, such as a photodiode (Sun and Thakor 2015). The light source illuminates a region with vasculature, such as the fingertip, and the light sensor measures how much light is either transmitted through or reflected from the tissue. In pulse oximeters, the PPG is typically acquired in transmission mode using a fingerclip probe (Nitzan et al 2020), while in consumer devices such as smartwatches, fitness trackers, and earbuds (i.e. hearables), it is typically acquired in reflection mode (Charlton and Marozas 2022). The amount of light received by sensors fluctuates with each heartbeat; usually, the amount of absorbed light increases during systole, when blood volume is greatest, and then decreases during diastole, when blood volume returns to its initial level (Allen 2007). This produces a pulse wave bearing several features, which can serve as physiological biomarkers (see figure 2) (Charlton et al 2022b).

Several factors can affect the morphology and quality of the PPG signal (Charlton et al 2022). First, PPG signals recorded during movement are often contaminated by motion artifacts (Park et al 2022). Second, PPG sensors must have good contact with the skin to obtain high-quality signals (Sun and Thakor 2015). Third, LED and photodiode positioning (Khan et al 2019), and acquiring multiple PPG signals (Charlton and Marozas 2022) can impact signal quality. Fourth, the wavelength of light emitted by the PPG sensor influences signal quality, with green wavelengths often preferred for reflectance mode PPG (Bashkatov et al 2005). Fifth, the contact pressure exerted by the sensor on the skin affects signal morphology (Chandrasekhar et al 2020). Sixth, the level of skin perfusion affects signal quality (Kyriacou et al 2002, Park et al 2022); low perfusion levels are usually associated with low wrist temperatures and with diseases such as Raynaud’s syndrome. Finally, the PPG measurement site (e.g. upper wrist) can affect signal morphology (Rajala et al 2018), quality (Prinable et al 2017) and susceptibility to motion artifact (Charlton et al 2022). Alternative sites, such as the arm, ear, chest, or face, may offer advantages in motion artifact reduction and in heart rate estimation accuracy (Charlton et al 2022). In addition, flexible and adhesive sensor attachments are emerging, promising improved contact and signal quality (Khan et al 2019).

At present, photoplethysmography is most commonly applied for heart rate monitoring in smartwatches (Temko 2017), and for oxygen saturation monitoring in pulse oximeters (Nitzan et al 2020). Smartwatches, fitness trackers and hearables are widely used, with an estimated 1.1 billion connected wearable devices worldwide in 2022 (Statista 2023). Pulse oximetry is a standard-of-care technique used in a range of clinical settings from intensive care to home monitoring (Ortega et al 2011, Greenhalgh et al 2021). Recently, the applications of photoplethysmography-based wearables have been expanded to include atrial fibrillation detection (Perez et al 2019), blood pressure monitoring (Vybornova et al 2021), and oxygen saturation monitoring (Spaccarotella et al 2022). Several additional potential applications of wearable photoplethysmography devices are being researched (Charlton et al 2023), including sleep staging (Kotzen et al 2022), mental health assessment (Cakmak et al 2021, Lyzwinski et al 2023), identifying obstructive sleep apnea (Behar et al 2014, 2019), and detection of peripheral arterial disease (Stansby et al 2022). Each of these applications uses PPG signal analysis to derive physiological information from the PPG.

Photoplethysmography confers several advantages over other physiological monitoring technologies, which have resulted in its widespread adoption. Measurements can be obtained quickly without the need for a trained operator, and photoplethysmography sensors are non-invasive, unobtrusive and low-cost. It is also a more compact and accessible monitoring modality than others, such as electrocardiogram (ECG) and blood pressure measurement. Furthermore, PPG measurements can be obtained without significantly disrupting daily activities, while ECG electrodes, for instance, can require careful placement and proper skin preparation to ensure accurate readings. However, a key disadvantage is that the PPG signal is highly susceptible to noise, such as in cases of poor sensor contact or motion (Li and Clifford 2012).

A key step in the use of photoplethysmography for health and fitness monitoring is the development of PPG signal analysis algorithms. Such algorithms typically extract either inter-beat-intervals (e.g. for detection of atrial fibrillation) or PPG pulse wave shape features (e.g. for estimation of blood pressure). However, unlike in other application fields, such as heart rate variability analysis, there are no standards for PPG signal analysis, and only limited open tools are available. Consequently, standardized and reproducible analysis of PPG signals is lacking. Although there are some open-source PPG toolboxes, they lack validation and are often incomplete (see table 1).

Efficient analysis of long-term continuous physiological time-series poses a challenge for many PPG toolboxes. While a number of toolboxes enable accurate peak detection, they usually do not support the detection of other fiducial points and the engineering of standard digital biomarkers (see table 1). Moreover, a number of these toolboxes are implemented in Matlab, which limits their use by the wider open-source community. Although PPGFeat toolbox can detect most of the common PPG fiducial points, it is usable using a user interface that enables the analysis of a single pulse wave only.

Despite the extensive research and applications in the field of PPG analysis, there is an urgent need to standardize approaches, terminologies, variables and definitions. Furthermore, there is no comprehensive toolbox available that covers all standard PPG biomarkers. It is important to acknowledge that certain variables may have different terminologies in the scientific literature, but our objective was to unify them to facilitate a broader understanding of PPG biomarkers. To fill this gap, we developed standardized nomenclature and toolbox. The assigned names for variables aim to provide insights into their origin, while the definitions ensure accurate interpretation and improved comprehensibility.

The primary aim of this research was to create a standardized toolbox (pyPPG) for analysis if long-term finger PPG recordings in real-time. This paper presents standardized definitions for the state-of-the-art PPG fiducial points and biomarkers implemented within the pyPPG toolbox. It provides an overview of the steps involved in raw data processing and biomarker engineering, as well as validation of the fiducial point extraction process (see figure 1). Additionally, the paper presents performance results and benchmarks them against other publicly available toolboxes.

The pyPPG toolbox provides an open-source and validated fiducial point detection and extraction of standardized, state-of-the-art digital biomarkers for the continuous PPG time series.

Two databases were used to validate the pyPPG toolbox (see table 2). The multi-ethnic study of atherosclerosis (MESA) database (Dean et al 2016, Zhang et al 2018) was used to validate the peak detector, and the PPG and blood pressure (PPG-BP) database (Liang et al 2018, Abdullah et al 2023) was used to validate the fiducial point detection algorithm. The MESA database consists of polysomnography (PSG) recordings from 2056 adults, aged 54–95 years, with subclinical cardiovascular disease, including 19 998 h of PPG recordings (Chen et al 2015, Rinkevičius et al 2023). Males accounted for 45.5% of the subjects. The database was downloaded from the National Sleep Resource Center (Zhang et al 2018). The PSG recordings in MESA were acquired at home, and included fingertip PPG signals measured at 256 Hz from the fingertip using Nonin 8000 series pulse oximeters (Nonin Medical Inc., Plymouth, US), alongside simultaneous ECG signals. The PPG-BP dataset contains 657 short (2 s) PPG recordings collected from 219 adult subjects, aged 20–89 years, with different health statuses (including healthy, hypertensive and diabetic subjects). Males accounted for 48% of the subjects. The data include fingertip PPG signals measured at 1 kHz using a SEP9AF-2 PPG sensor (SMPLUS Company, Korea). Signals were acquired using a 12 bit ADC, and the hardware applied a 0.512 Hz band-pass filter. Use of the retrospective databases available at open-access http://sleepdata.org for this research was approved by the institutional review board from the Technion-IIT Rappaport Faculty of Medicine (number 62-2019).

The pyPPG toolbox is a standardized resource for real-time analysis of long-term finger PPG recordings. The toolbox consists of five main components, as summarized in figure 1:

• 1.  
  Loading a raw PPG signal: the toolbox can accept various file formats such as .mat, .csv, .txt, or .edf. These files should contain raw PPG data along with the corresponding sampling rate.
• 2.  
  Preprocessing: the raw PPG signal is filtered to remove noise and artifacts. Subsequently, the first, second, and third derivatives (PPG’, PPG”, and PPG”’) of the PPG signal are computed and filtered. The resampling of the filtered PPG signal to 75 Hz is specifically performed for systolic peak detection.
• 3.  
  Pulse wave segmentation: the toolbox employs a peak detector to identify the systolic peaks. Based on the peak locations, the toolbox also detects the pulse onsets and offsets, which indicate the start and end of the PPG pulse waves.
• 4.  
  Fiducial points identification: for each pulse wave, the toolbox detects a set of fiducial points.
• 5.  
  Biomarker engineering: based on the fiducial points, a set of 74 PPG digital biomarkers are engineered.

The pyPPG toolbox also provides an optional PPG signal quality index based on the Matlab implementation of the work by Li and Clifford (2012). Signal quality assessment is performed based on template matching, which provides a numerical indicator for the quality of pulse waves (varying between 0 and 1). This can be used to identify distorted pulse waves. Whilst this signal quality index is provided by pyPPG, it was not used for the experiments conducted in this research.

PPG signal filtering is one of the most essential parts of preprocessing. The human heart rate ranges between 30 and 200 beats per minute (Paliakaitė et al 2020). Therefore, in PPG signal analysis, it is common to apply bandpass filtering such 0.5−8 Hz (Abdullah et al 2023), 0.5−10 Hz (Finnegan et al 2023), 0.5−15 Hz (Mejia-Mejia et al 2022), 0.5−20 Hz (Allen and Murray 2000, Liang et al 2018), or 0.5−25 Hz (Chowdhury et al 2020), to conserve the frequency content of the PPG pulse waves while filtering out lower-frequency content (e.g. baseline wander due to respiration) and higher-frequency content (e.g. muscle noise or power interference).

Whilst fiducial point detection can be simpler with lower low-pass cut-off frequencies such as 8 Hz, the drawback of using lower cut-off frequencies is that they significantly distort the pulse wave shape and reduce the accuracy with which the pulse onset and other fiducial points can be identified. Conversely, cut-off frequencies above 12 Hz can make it more complex to detect fiducial points due to the presence of extra waves in the PPG derivatives. Therefore, during the benchmarking process of other toolboxes for fiducial point detection (see section 3.2), the 0.5−12 Hz frequency band was employed for filtering purposes. Although the 0.5−12 Hz band is recommended by default for PPG analysis, user can customize the passband filter in the pyPPG toolbox. The following zero-phase filters were implemented (see figure A1):

• 1.  
  Bandpass filtering between 0.5−12 Hz: a fourth-order Chebyshev Type II filter was used for the original signal. The 12 Hz low-pass cut-off filter was used to avoid time-shifting of fiducial points (particularly pulse onset, and dicrotic notch) and to eliminate unwanted high-frequency content from the PPG derivatives. The 0.5 Hz high-pass cut-off filter was used to minimize baseline wandering whilst retaining content at low heart rates.
• 2.  
  50 ms moving average filtering (MAF): in the case of very noisy signals, some high-frequency content can remain in the band-pass filter signal. For this purpose, a 50 ms standard flat (boxcar or top-hat) MAF with a 9 Hz cut-off frequency was applied after the band-pass filtering.
• 3.  
  10 ms MAF for the PPG derivatives: to eliminate the high-frequency content in the PPG derivatives, a 10 ms standard flat (boxcar or top-hat) MAF with 45 Hz cut-off frequency was applied.

The toolbox provides default filtering parameters that are optimized for fingertip PPG signals. Users have the flexibility to customize these parameters according to their requirements, e.g. cut-off frequencies, filter order and MAF size. The applied filtering technique minimally alters the amplitude, which is also verified on the PPG-BP dataset (see figure A1).

It is common for the PPG signal to be sampled at over 100 Hz and up to 1 kHz, as, for example, in the PPG-BP dataset. However, an excessive sampling frequency may not be ideal for long-term data processing due to the computational load. The default behavior of the toolbox is to resample PPG signals at 75 Hz using the Python resample function which is based on the Fourier method. Resampling was employed to reduce the computational load of the systolic peak detector. However, for the identification of other fiducial points the original sampling frequency was used.

The toolbox identifies individual pulse waves in a PPG signal by identifying systolic peaks (sp), and then identifying the pulse onset (on) and offset (off) on either side of each systolic peak which indicate the start and end of the pulse wave, respectively.

2.4.1. Systolic peak detection

The sp is the most important fiducial point of the PPG signal (see figure 2). It is defined as the point with the highest amplitude between two consecutive pulse onsets (see figure 2). The pyPPG toolbox uses an enhanced sp detection algorithm to enable real-time analysis of long-term PPG measurements. The algorithm is an enhanced version of the Aboy beat detector (Aboy et al 2005), which performed either best (Kotzen 2022), or amongst the best (Charlton et al 2022a) in recent benchmarking studies of PPG beat detectors. We focused on improving the beat detector’s performance and reducing its computational complexity.

The original Aboy algorithm utilizes an advanced filtering technique to accurately detect systolic peaks (Aboy et al 2005). PPG recordings are segmented into 10-second windows and then filtered using three digital filters. The first filter helps to estimate the heart rate, while the second and third filters are used for peak detection. Two modifications were made to the Aboy algorithm (Aboy et al 2005). First, to enhance the speed of the previous Matlab implementation (Charlton et al 2022a), the finite impulse response (FIR) filter was replaced by a zero-phase fifth-order Chebyshev Type II infinite impulse response (IIR) filter, which applied the same cut-off frequencies as the original Aboy peak detector. Second, adaptive heart rate estimation is included to handle strong baseline wandering and rapid amplitude fluctuations (Goda et al 2023). When estimating heart rate, if the number of detected peaks is outside the expected lower and upper limits (see Goda et al 2023), then that 10 s segment of data is deemed to be low quality and the HR from the previous segment is retained. The resulting modified peak detector is denoted Aboy++.

When first presenting the Aboy++ algorithm et al (Goda et al 2023), we evaluated it on a small subset consisting of 100 recordings from the MESA dataset containing over 4.25 million reference beats. Aboy++ achieved an F1-score of 85.5% (79.78–92.57), compared to 80.99% (73.91–85.52) for the Aboy peak detector. In addition, the computational time of Aboy++ was over 57 times faster than that of Aboy. Specifically, the median peak detection time for 1 h segments was 114.24 s for Aboy, compared to 1.98 s for Aboy++.

2.4.2. Pulse onset detection

Table 4 summarises the algorithmic approaches used to detect fiducial points. The approaches were designed based on those used in PulseAnalyse. Additional approaches were created for the fiducial points which were not implemented in PulseAnalyse (for v and w points). The approaches were then refined according to the fiducial point definitions presented in table 3.

2.5.1. Dicrotic notch detection

The dicrotic notch (dn) plays an important role as a fiducial point in the analysis of PPG signals, holding immense potential for various applications such as heart disease detection (Gu et al 2008) and arterial stiffness assessment (Addison 2016). Its significance stems from its association with the duration of systole, which is known to be affected by heart disease. Additionally, appearance of the diastolic wave following the dn allows for the evaluation of arterial stiffness, with the hypothesis that the presence of the dn is influenced by the arterial stiffness. However, it should be noted that the visibility of the dn diminishes progressively with age, making it typically no longer discernible in elderly subjects (Charlton et al 2022c).

None of the existing definitions of dn are entirely satisfactory. Typically, the dn is easily recognizable when a distinct local minimum exists between the sp and the dp (see figure 2). Yet, in many cases, the dp is not clearly visible, rendering it difficult to accurately identify the dn. Dawber et al (1973) categorized different classes of dn, which are illustrated in figure 3. The fiducial point dn can be classified into four classes: Class 1, in which the dn is an incisura, Class 2, in which there is a horizontal line at the dn, Class 3, in which there is a change in gradient on the downslope, and Class 4, in which there is no clear evidence of the dn.

Another morphological approach for dn identification involves locating it at the time of zero-crossing of the PPG” between the d and e points (Chakraborty et al 2021). Yet, situations may arise where the local minimum of the dn is visible, but the occurrence of the d and e points precedes the zero-crossing point, as depicted in figure 2. Thus, debate regarding the precise location for defining the dn is ongoing.

2.5.2. Fiducial points of PPG derivatives

2.5.3. Correction of the fiducial points

We implemented standardized biomarkers, employing uniform definitions for annotation. While the PulseAnalyse toolbox provides important biomarkers, pyPPG not only includes a more extensive collection of biomarkers but also features their up-to-date implementation. More specifically, the pyPPG toolbox includes a comprehensive collection of 74 standard PPG morphological biomarkers which are calculated from the timings and amplitudes of the fiducial points (see tables 5–8). The biomarkers were categorized into four groups: (1) PPG Signal—biomarkers that are based on the location of the fiducial points of the PPG signal, (2) Signal Ratios—biomarkers that are based on ratios of the fiducial points of the PPG signal, (3) PPG Derivatives—biomarkers that are based on the location of the fiducial points of the PPG derivatives and (4) Derivatives Ratios—biomarkers that are based on ratios of the fiducial points of the PPG derivatives.

Table 5. Biomarkers derived from the PPG signal categorized by intervals, amplitudes and areas.

PPG Signal				References
1	Intervals	T pi	Pulse interval, the time between the pulse onset and pulse offset	Chowdhury et al (2020)
2		T pp	Peak-to-peak interval, the time between two consecutive systolic peaks	Kurylyak et al (2013)
3		T sys	Systolic time, the time between the pulse onset and dicrotic notch	Ahn (2017)
4		T dia	Diastolic time, the time between the dicrotic notch and pulse offset	Ahn (2017)
5		T sp	Systolic peak time, the time between the pulse onset and systolic peak	Alty et al (2003)
6		T dp	Diastolic peak time, the time between the pulse onset and diastolic peak	Chowienczyk et al (1999)
7		Δ T	Time delay, the time between the systolic peak and diastolic peak	Chowienczyk et al (1999)
8		T swx	Systolic width, the width at x% of the systolic peak amplitude between the pulse onset and systolic peak	Kurylyak et al (2013)
9		T dwx	Diastolic width, the width at x% of the systolic peak amplitude between the systolic peak and pulse offset	Kurylyak et al (2013)
10		T pwx	Pulse width, the sum of the systolic width and diastolic width at x%	Kurylyak et al (2013)
11	Amplitudes	A sp	Systolic peak amplitude, the difference in amplitude between the pulse onset and systolic peak	Chua and Heneghan (2006)
12			Dicrotic notch amplitude, the difference in amplitude between the pulse onset and dicrotic notch	Duan et al (2016)
13		A dp	Diastolic peak amplitude, the difference in amplitude between the pulse onset and diastolic peak	Duan et al (2016)
14		A off	Pulse onset amplitude, the difference in amplitude between the pulse onset and pulse offset
15	Areas	AUC pi	Area under pulse interval curve, the area under the pulse wave between pulse onset and pulse offset	Duan et al (2016)
16		AUC sys	Area under systolic curve, the area under the pulse wave between the pulse onset and dicrotic notch	Ahn (2017)
17		AUC dia	Area under diastolic curve, the area under the pulse wave between the dicrotic notch and pulse offset	Ahn (2017)

Table 7. Biomarkers derived from the PPG derivatives.

PPG derivatives				References
1	Intervals	T u	u-point time, the time between the pulse onset and u-point	ms
2		T u	v-point time, the time between the pulse onset and v-point	Suboh et al (2022)
3		T w	w-point time, the time between the pulse onset and w-point	Suboh et al (2022)
4		T a	a-point time, the time between the pulse onset and a-point	Suboh et al (2022)
5		T b	b-point time, the time between the pulse onset and b-point	Suboh et al (2022)
6		T c	c-point time, the time between the pulse onset and c-point	Suboh et al (2022)
7		T d	d-point time, the time between the pulse onset and d-point	Suboh et al (2022)
8		T e	e-point time, the time between the pulse onset and e-point	Suboh et al (2022)
9		T f	f-point time, the time between the pulse onset and f-point	Suboh et al (2022)
10		T b−c	b−c time, the time between the b-point and c-point	Charlton et al (2018)
11		T b−d	b−d time, the time between the b-point and d-point	Charlton et al (2018)
12			p1-point time, the time between the pulse onset and p1-point	Suboh et al (2022)
13			p2-point time, the time between the pulse onset and p2-point	Suboh et al (2022)
14			p1 − dia time, the time between the p1-point and diastolic peak	Peltokangas et al (2017)
15			p2 − dia time, the time between the p2-point and diastolic peak	Peltokangas et al (2017)

Table 8. Biomarkers derived from the derivatives ratios categorized into intervals, amplitudes, areas, and combinations of these.

Derivatives ratios				References
1	Intervals	Tu/Tpi	Ratio of the u-point time versus the pulse interval	Chowdhury et al (2020)
2		T v / T pi	Ratio of the v-point time versus the pulse interval	Chowdhury et al (2020)
3		T w / T pi	Ratio of the w-point time versus the pulse interval
4		T a / T pi	Ratio of the a-point time versus the pulse interval	Chowdhury et al (2020)
5		T b / T pi	Ratio of the b-point time versus the pulse interval	Chowdhury et al (2020)
6		T c / T pi	Ratio of the c-point time versus the pulse interval
7		T d / T pi	Ratio of the d-point time versus the pulse interval
8		T e / T pi	Ratio of the e-point time versus the pulse interval
9		T f / T pi	Ratio of the f-point time versus the pulse interval
10		( T u − T a )/ T pi	Ratio of the difference between the u-point-time and a-point time versus the pulse interval	Chowdhury et al (2020)
11		( T v − T b )/ T pi	Ratio of the difference between the v-point time and b-point time versus the pulse interval	Chowdhury et al (2020)
12	Amplitudes	A u /Asp	Ratio of the u-point amplitude versus the systolic peak amplitude	Alty et al (2003)
13		A v /Au	Ratio of the v-point amplitude versus the u-point amplitude
14		A w /Au	Ratio of the w-point amplitude versus the u-point amplitude
15		A b /Aa	Ratio of the b-point amplitude versus the a-point amplitude	Takazawa et al (1998)
16		A c /Aa	Ratio of the c-point amplitude versus the a-point amplitude	Takazawa et al (1998)
17		A d /Aa	Ratio of the d-point amplitude versus the a-point amplitude	Takazawa et al (1998)
18		A e /Aa	Ratio of the e-point amplitude versus the a-point amplitude	Takazawa et al (1998)
19		A f /Aa	Ratio of the f-point amplitude versus the a-point amplitude
20			Ratio of the p2-point amplitude versus the p1-point amplitude	Peltokangas et al (2017)
21		( A c − Ab )/Aa	Ratio of the difference between the b-point amplitude and c-point amplitude versus the a-point amplitude	Ahn (2017)
22		( A d − A b )/ A	Ratio of the difference between the b-point amplitude and d-point amplitude versus the a-point amplitude	Ahn (2017)
23		AGI	Aging index, (Ab -Ac -Ad -Ae )/Aa	Takazawa et al (1998)
24			Modified aging index, (Ab -Ac -Ad )/Aa	Ushiroyama et al (2005)
25			Informal aging index, (Ab -Ae )/Aa	Baek et al (2007)
26		AI	Augmentation index, (PPG(Tp2)-PPG(Tp1))/Asp	Takazawa et al (1998)
27			Reflection index of p1, Adp /(PPG(T)-PPG(Tpi (0)))	Peltokangas et al (2017)
28			Reflection index of p2, Adp /(PPG(p2)-PPG(Tpi (0)))	Peltokangas et al (2017)
29	Combined	SC	Spring constant, PPG”(Tsp )/((Asp -Au )/Asp )	Wei (2013)
30		IPAD	Inflection point area plus normalised d-point amplitude, AUCdia /AUCsys +Ad /Aa	Ahn (2017)

For a given window consisting of a set of beats, pyPPG provides the following nine general statistics for each biomarker (see appendix tables A2–A5): average (AVG), median (MED), standard deviation (SD), lower and upper quartiles (Q1, Q3), inter-quartile range (IQR), skewness (SKW, indicating a lack of symmetry in the distribution), kurtosis (KUR, indicating the pointedness of a peak in the distribution curve), and the average difference between the mean and each data value (MAD). For each biomarker, we provide these summary statistics including measures of the central tendency and dispersion.

2.7.1. Systolic peak detection

2.7.2. Fiducial point detection

2.7.3. Fiducial point benchmarking

The pyPPG peak detector was evaluated on the 2054 recordings of the MESA dataset, which included more than 91 million reference beats. The peak detection achieved a median F₁-score of 88.19% (lower—upper quartiles of 81.73%–92.71%). pyPPG demonstrated the same performance in real-time operation.

The results for the benchmarking of pyPPG against other PPG toolboxes (Pulse Analyse and PPGFeat) are presented in table 9. A total of 219 distinct pulse waves were employed for the benchmarking process. With pyPPG, the MAEs were <10 ms for all fiducial points, were less than one fifth of those for PPGFeat for all fiducial points except on, and less than those for PulseAnalyse for all except three fiducial points (sp, a, b). In comparison to the other toolboxes, pyPPG showed particular improvements in the detection of dn, p₁ and p₂. In addition, pyPPG was able to detect fiducial points (see figure A2) which were not detected by PulseAnalyse (v and w) or PPGFeat (p₁, p₂). Bland–Altman plots were generated to present the differences between the manual annotations and pyPPG fiducial points detection (see figures 4 and 5).

Note that the straight-line appearance of the Bland–Altman results for on is due to the times of the reference onsets always being equal to zero, as they were used to define the start time of each pulse wave.

The resulting systolic peak detection and fiducial points detection algorithms are packaged into an open-source Python library denoted pyPPG. In addition, a user-friendly interface is also implemented in the PhysioZoo Software. In order to ensure that pyPPG could process a large dataset without technical issues, we ran it over the full MESA database and reported standard statistics for all biomarkers (see appendix tables A2–A5).