A portable dry-electrode ECG device for rapid and accurate neonatal heart rate monitoring during resuscitation

Abstract

Each year, around 10% of infants globally will require resuscitation at birth. Pediatricians can use stethoscope, electrocardiogram (ECG) or pulse oximetry to determine heart rate (HR) which is used to guide resuscitation steps. HR must be acquired accurately and quickly. However, current HR detection modalities are either inaccurate or too slow. This work offers a novel infant heart rate detector (iHRD) using single-lead dry electrode ECG that can display HR accurately within the first 10 seconds of initial contact. A research ethics board approved validation study is conducted on 50 healthy newborns comparing iHRD’s HR with clinical HR monitors at a community hospital. 3-minute newborn single-lead ECGs and HR are recorded, and HR is annotated every 2 seconds. Statistical HR analysis is performed to ensure iHRD’s feasibility and reliability. With 2741 HR datapoints, excluding outliers, the iHRD detected HR with 94.5% accuracy with time from contact to HR display under 10 seconds. Overall, the iHRD using dry electrode single-lead ECG showed good results in providing reliable HR quickly for neonatal resuscitation efforts.

The golden minute of life is defined as the first minute after birth, a critical time that newborns must show independent breathing. However, over 10% of babies every year require medical ventilation assistance, and 1% of newborns require extensive resuscitation. It is recommended that neonatal resuscitation and/or breathing support should be performed within the first minute of birth. Due to these issues, neonatal resuscitation protocols (NRP) are developed by neonatologists, pediatric nurses, and general pediatricians, to provide a step-by-step guide to perform infant vital sign assessment and determine the appropriate ventilation and resuscitation steps¹.

Following NRP, the assessed neonatal vital signs include HR, oxygen saturation, respiratory effort, skin color, body temperature and muscle tone. The gold-standard most sensitive vital sign metric is HR which can not only be used for initial assessment but is highly relevant during vital sign re evaluation throughout the resuscitation process to determine if further intervention is required. HR relates directly to the cardio-respiratory condition of the infant. In turn, clinicians used many diagnostic tools to obtain HR. These tools included auscultation using stethoscope, Pulse Oximeters (POs), and 3-lead ECG machines.

Cardiac auscultation is defined as the listening to the heart’s sounds using an analog or digital stethoscope. Auscultation is usually identified as the go-to approach for quick infant HR measurements due to its fast use. A clinician would listen to the infant’s heart beats for a few seconds and calculate HR in BPM cognitively. For example, a pediatrician can listen to neonatal heartbeats for 6 seconds, count the number of beats in that time frame, and multiply by 10 to obtain HR in BPM. The pediatrician relays this information to the rest of the neonatal care team by tapping their fingers with the infant’s heartbeat. This approach is operator dependent which leads to inaccuracies^2–7.

The second most prominent HR modality are POs. POs are used to measure oxygen saturation and HR. POs determine HR by measuring blood flow changes in vessels using light-based sensors. Betlagi et al. recognized the importance of POs in many pediatric applications⁸. The authors determined the advantages and limitation of POs. The advantages of PO in neonatal care include its real-time capabilities, portability, non-invasiveness and cost-efficiency.⁸ However, they also recognized the limitations of the light-based signal acquisition approach in PO where skin pigmentation, light-interference and motion artifacts have an influence on signal acquisition, time to acquisiton, and ultimately HR estimations. Their work is an addition to all other studies that validate their interpretations^8–13. POs are not suitable to provide quick and reliable HR measurements in these time-sensitive scenarios. In turn, NRP guidelines recommend the use of ECG as the gold standard HR modality for HR detection in neonatal resuscitation¹.

3-lead ECG measurements are the current clinical standard for HR detection in resuscitation. Shah et al. examined the advantages of ECG in neonatal resuscitation outcomes¹⁴. Their results determined that newborns with ECG showed less aggressive interventions such as intubation and were less likely to undertake chest compressions. On the other hand, ECG compared to both PO and auscultation showed significantly better HR measurements, providing the clinicians the confidence in conducting neonatal resuscitation¹⁴. However, 3-lead ECG is not always available in delivery rooms in hospitals and can be time-consuming¹⁵. Abbey et al. showed that from time of ECG electrode placement to display of first accurate HR results in 64 seconds exceeding the golden minute limit¹⁶. Although ECG is considered the recommended approach, its costly nature and slow HR acquisition forced clinicians to still use PO and auscultation as HR modalities for neonatal resuscitation efforts as shown in Table 1.

Table 1.

State-of-the-art HR detection tools.

	Acquisition Technique	Time to HR (s)	Type of Sensor	Sensor Reusability	HR Display	Cost (USD)
3-lead ECG monitors	Cardiac electrical activity	30 to 60 s29,30	Wet electrodes	Disposable	Available	Variable (100 to 10000)
PO	PPG	60 to 120 s10,30	Infrared light	Sanitize and reuse	Available	Variable (100 to 500)
Stethoscope	Acoustic vibrations	7 to 19 s30	Disk-shaped resonator	Sanitize and reuse	N/A	Variable (50 to 300)

Many researchers and medical device companies explored the use of single-lead wireless ECG to alleviate the need for 3-lead ECG systems. These systems are built using minimum ECG hardware components which allow for ambulatory monitoring such as many ECG wearables including the Apple Watch, Alivecor, iRyhthm’s Zio Patch, Bittium Faros 360, and Bradydx’s Carnation Ambulatory Monitor^17–25. These ECG devices are suitable for adult ambulatory wireless monitoring alone and are not feasible for neonatal resuscitation efforts.

It is evident that all current HR modalities for neonatal resuscitation efforts are either too slow or inaccurate. We aimed to create a HR modality that offers quick and accurate HR for neonatal resuscitation^26,27. This study aims to propose a prototype iHRD. The iHRD uses dry electrode based single-lead ECG to estimate HR. The device also features portability, cost-effectiveness, small form factor, long battery life, and a HR display to relate HR information quickly and accurately to the neonatal care team. The iHRD is investigated under REB approved proof of concept study involving 50 healthy newborns to determine its accuracy and speed in HR acquisition in comparison to clinical HR monitors²⁶. This work explores the above-mentioned device and an in-depth analysis of the device’s development process, and validation study’s outcomes.

This study is split as follows: Section 1, Results, provides in-depth HR analyses of the iHRD in comparison to clinical HR monitors to determine accuracy. These analyses also include timing metrics to examine the speed of acquisition. Where as section 2, Discussion, provides a detailed explanation of the HR results and future directions. While section 3, Methods, provide the iHRD’s developmental process. This process includes the development of 3D printed dry electrodes, the single-lead ECG acquisition circuitry, the proposed HR detection algorithm and a brief of the proof-of-concept study.

Results

All HR data pairs versus motion-free HR data pairs

4147 HR data pairs, from the SAR-NE, dataset are used to determine iHRD’s reliability and robustness for quick accurate neonatal HR estimations. Using all HR data pairs, the iHRD’s HR and clinical HR showed weak positive correlation, as shown in Fig. 1. However, these data pairs contain both motion-artifact tainted and motion-artifact free HR pairs. Motion artifacts reduce overall accuracy due to its introduction that impacts ECG signal acquisition. On the other hand, the motion-artifact free HR data pairs amount to 2741 (66.1%) of the entire HR pairs in the dataset. Motion-free HR pairs show a stronger correlation with clinical HR, as shown in Fig. 1. This aspect reveals that the iHRD is more accurate with the absence of motion artifact noise. As shown in Table 2, the entire HR data pairs showed a high MAE of 13.1 BPM, whereas the motion-free analysis revealed a MAE of 8.5 BPM. The iHRD can minimize the presence of error, in BPM, during infant motion-free HR detection. While accuracy drops with the presence of motion artifacts in HR estimation. The entire dataset showed an average accuracy of 90.7%. While motion-free HR estimations showed an increase of accuracy to 93.5%. SD metrics also reveal that motion-free SD is less than ± 10 BPM compared to clinical HR. This SD is acceptable for clinical research settings but may not be optimal yet for clinical regulatory. Further improvements are required to improve accuracy.

Fig. 1.

(A) Correlation plot for All HR pairs with . (B) Correlation plot for Motion-free HR pairs alone with .

Table 2.

Statistical Comparison of All HR pairs versus Motion-Free HR pairs.

	All HR pairs	Motion-Free HR pairs	Statistical Significance difference ()
Pearson r Coefficient	0.64	0.88	*
Spearman Coefficient	0.74	0.87	*
NRMSE (%)	2.4%	0.61%
MAE (BPM)	13.1	8.5	* ()
Median of HR difference	7	6
SD (BPM)	20.4	9.7	* ()
Mean of HR difference	4.0	3.7
IQR	17	12
Accuracy (%)	90.7	93.5
N	4147	2741 (66.1%)

While the Bland-Altman plots, as shown in Fig. 2, validate the SD outcomes and show narrower HR differences for the motion-free HR analysis. This exemplifies that iHRD’s HR and clinical HR are closely related during no baby movement moments. All following comparative analyses will only use the motion-free HR data pairs to mitigate the influence of motion artifact noise in the analyses.

Fig. 2.

(A) BlandAltman plot for All HR pairs with Mean = 4.0. (B) BlandAltman plot for Motion-free HR pairs alone with Mean = 3.7.

iHRD HR versus Clinical HR monitors

The iHRD HRs are analyzed separately in comparison to ECG-based HR monitors which were available with NICU infants, and POs available with PPU babies, respectively. This analysis is aimed to provide an investigation into the iHRD’s clinical relevancy compared to ECG and PO separately. Both ECG and PO are identified as clinical HR benchmarks due to their use as diagnostic instruments for HR detection in the pediatric setting. Any new developed HR modalities are examined directly to either ECG or PO based HR estimations, making it a consistent research practice to determine the new technology’s performance.

For ECG-based HR monitoring, the iHRD showed strong correlation, as shown in Table 3. The trend analysis revealed iHRD’ HR following ECG-based HR estimations. Furthermore, Table 3 show that the iHRD’s HR closely aligns to clinical ECG HR monitors through a MAE of approximately ±10 BPM. However, the iHRD’s HR showed a higher SD with ECG-based versus PO-based HR modalities.

Table 3.

Statistical Comparison of ECG-HR pairs versus PO-HR pairs.

	ECG-iHRD HR (Acquired in NICU)	PO-iHRD HR (Acquired in PPU)	Statistical Significance difference ()
Pearson r Coefficient	0.83	0.88	* ()
Spearman Coefficient	0.82	0.87	* ()
NRMSE (%)	0.84%	0.56%
MAE (BPM)	10.1	8.23	* ()
Median of HR difference	3	6
SD (BPM)	13.7	8.8	* ()
Mean of HR difference	1.2	4.1
IQR	14	11
Accuracy (%)	93.1	93.6
N	375	2366

PO-based HR monitors showed better correlation with the iHRD than ECG-based monitors. Although iHRD showed good correlation outcomes, it is important to note that an unbalanced dataset is presented. ECG-based HR data pairs amount to 375 (N = 375). This size is smaller than PO’s HR data pairs of 2366, as shown in Table 3. It is difficult to clearly state that iHRD is more similar to a clinical HR monitor than the other due to the unbalanced nature of the data in this comparison.

Sex-based HR comparative analysis

Neonatal HR at birth is deemed similar between female and male sexes. This analysis aimed to examine if HR differences are revealed during iHRD HR estimation based on sex. As shown in Fig. 3, the male and female infant subgroups showed similar medians and IQRs across both clinical and iHRD HRs. These outcomes show that the iHRD can seamlessly operate with the same accuracies across infant sexes. Table 4 shows that the showed no statistically significant differences, especially in MAE, which informs the similar performance of the iHRD on both sexes. It is important to note that Pearson coefficients and SDs although showed statistical significance, these metrics are known to be sensitive to outliers. Therefore, it is important to perform further statistical tests to determine the veracity of the iHRD on both sexes and make sure of its appropriate interpretation.

Fig. 3.

(A) Box plot for female infants’ HR pairs. (B) Box plot for male infants’ HR pairs with increased outliers for both clinical and iHRD devices.

Table 4.

Statistical Comparison of female infants’ HR pairs versus male infants’ HR pairs.

	Female infants	Male infants	Statistical Significance difference (p-value< 0.05)
Pearson r Coefficient	0.86	0.88	*()
Spearman Coefficient	0.85	0.87	* ()
NRMSE (%)	0.55%	0.65%
MAE (BPM)	8.2	8.7	()
Median of HR difference	7	6
SD (BPM)	8.9	10.1	* ()
Mean of HR difference	3.9	3.6
IQR	13	11
Accuracy (%)	93.6	93.5
N	1103 (40%)	1638 (60%)

Pre-term versus early- and full-term infants

Gestation age has a significant role in neonatal physiological development. The lower the gestation age of a neonate at birth, the higher the risk of complications at birth due to physical underdevelopment. Pre-term infants also show more skin fragility, moisture, and motion. Therefore, it is imperative to investigate the iHRD’s capabilities in obtaining a robust and reliable HR depending on gestation age, more specifically pre-term (gestation age <37 weeks) versus early- and full-term infants (gestation age > 37 weeks). It is important to note that early- and full-term infants are combined as one in this analysis.

There was no statistical difference between both r correlation coefficients as shown in Table 5. This information shows that the acquired iHRD HR can follow clinical HR monitor trend across the variety of gestation ages similarly. However interestingly, the spearman coefficients, , showed statistically significant differences. Also, MAE and SD values are shown to be higher in pre-term infants showing that a larger variability is present in pre-term HR analysis when compared to early- and full-term infants.

Table 5.

Statistical HR Comparison of Pre-term and Early and Full Infants.

	Pre-term Infants	Early and Full-term Infants	Statistical Significance difference ()
Pearson r Coefficient	0.85	0.87	(p = 0.13)
Spearman Coefficient	0.82	0.86	* (p = 0.007)
NRMSE (%)	0.76%	0.57%
MAE (BPM)	10.1	8.2	* ()
Median of HR difference	7	6
SD (BPM)	12.3	9.0	* ()
Mean of HR difference	3.8	3.7
IQR	14	12
Accuracy (%)	93.1%	93.6%
N	471	2270

iHRD timing metrics

Clinical and iHRD HR analysis included timing metrics to determine the speed of the iHRD in displaying accurate HR quickly for future use in neonatal resuscitation efforts. We conducted a timing analysis to examine the iHRD’s speed. This analysis yielded that iHRD detected HR within ±10 BPM in an average of 3.9 seconds, as shown in Table 6. This timing is recorded from the moment of infant contact to HR display for the clinical team. However, the iHRD was not able to provide HR within ± 5 BPM in less than 10 seconds, a condition set forth in this study. The iHRD can provide accurate HR values similar to clinical HR monitors within ± 5 BPM in 14.2 seconds. Overall, the iHRD is presented to be accurate and reliable for displaying HR information in less than 10 seconds. Future development of the iHRD is warranted to improve its accuracy and speed of HR detection.

Table 6.

iHRD Timing and Accuracy Characteristics.

	Time to accurate HR
HR within ± 20 BPM of clinical HR	2.4 s
HR within ± 15 BPM of clinical HR	2.8 s
HR within ± 10 BPM of clinical HR	3.9 s
HR within ± 5 BPM of clinical HR	14.2 s
All HR above ± 10 BPM of clinical HR	6.1 s

Discussion

We developed a neonatal iHRD to quickly and accurately display HR for neonatal resuscitation efforts. The portable hand-held device acquires single-lead ECGs using 3D printed dry electrodes. The single-lead neonatal ECG signal undergoes signal processing and manipulation on the embedded microcontroller of the device to quickly provide HR information. We processed 2-second length of the ECG signal to estimate HR which ultimately increased the speed of the HR display. The iHRD is validated clinically on 50 healthy newborns, and a neonatal vital sign dataset, SAR-NE, is curated from this study.

A HR comparative analysis is conducted where iHRD HR is directly compared to clinical HR using statistical analyses, Bland-Altman, and correlation plots. We first examined the overall dataset, featuring 4147 HR data pairs. 66.1% (2741) of the HR data pairs do not contain any motion-artifacts. Estimated iHRD HR using motion-artifact tainted single-lead ECG signals showed lower overall correlation and higher inaccuracies as presented through the statistical analysis. The motion-tainted iHRD HR showed worse results because HR is calculated using 2 seconds of ECG signal length. This concept entails that any sudden movements by the infant during acquisition will instantly present outlier peaks in the single-lead ECG like R-peak, leading to an overestimation of HR. This reason can be clearly identified in the higher SD, MAE values, as shown in Table 2, where higher absolute errors are present during the motion versus motion-free HR analysis. On the other hand, the Bland-Altman plots for both the motion-tainted and motion-free HR outcomes showed consistent averages of 4 and 3.7 BPM, respectively. These averages represent the systemic bias present in the iHRD, which can further cause deviations from true HR. A study into the systemic bias of the iHRD is warranted and should be explored further to determine the root cause issue. For example, inherent electrical noise within the AFE circuit that was not removed can interfere with ECG signal transmission. Other noises not considered include battery noise and environmental EMI. The iHRD device does not feature a faraday shield to prevent EMI entirely. Device shielding may be warranted to lower the systemic bias shown on the device.

On the other hand, iHRD HR in comparison to ECG-based and PO-based HR yielded good correlation, SD, and MAE outcomes. iHRD HR showed similar HR resulted to ECG-based HR monitors with average error of ± 10 BPM. Whereas iHRD HR showed a better average error of ± 8.8 BPM in comparison with PO-based HR values. The main difference between ECG and PO is the type of signal acquired. ECG-based HR monitors use the electrical activity of the heart to determine HR using wet electrodes. These devices determine the R-R interval for HR detection based on QRS waveforms. On the other hand, POs use a light sensor to acquire blood flow measurements which are later used for HR estimation. Although both clinical HR modalities are different in principle, the iHRD showed approximate results to both showing its accuracy and reliability as a HR detection tool. It is important to note that this direct comparison is limited because of unbalanced data between ECG (N = 375) and PO (N = 2366). This constraint may lead to misinterpretation of the comparative outcomes. Further ECG-based HR data collection is necessary to provide a holistic interpretation with minimal bias. The differences between iHRD HR and the type of clinical HR monitor cannot be clearly identified due to this issue. However, it is evident that the proposed iHRD is suitable as a candidate HR modality when compared to both clinical HR monitors as mentioned above.

Whereas sex-based HR analysis showed the lack of differences between female and male iHRD HR accuracies. The consistency of the iHRD across sex subgroups further validates previous literature that states the lack of statistical HR differences between sexes in the first few after birth. This analysis showed the iHRD lacks sex bias, making it suitable for all sex subgroups similarly.

Gestation age differences reveal iHRD discrepancies between pre-term, early- and full-term infants. These differences can be found through statistically significant variability in spearman coefficients, as shown in 5. This information helps us reveal that physiological variability may have a role in the rate of change of HR, which is evident through this analysis framework. Pre-term infants usually show higher HRs than early- and full-term infants indicative that these changes may impact iHRD’s accuracy. It is important to note that a limitation of this analysis is the unbalanced dataset of pre-term (N = 471) versus full-term infants (N =2270). A balanced dataset is required to ensure accurate comparative analysis.

Lastly, the iHRD timing metrics revealed that the device operates as intended by providing accurate HR within 10 seconds of contact. The device achieves this goal making it suitable for further clinical studies involving real neonatal resuscitation scenarios. As shown in Table 6, achieving higher accuracy (lower MAE) may require longer acquisition times for HR display. This process is due to the adaptive threshold’s nature of continuously acclimating with more single-lead ECG signals. A main difference between clinical 3-lead ECG monitors and this iHRD is the approach to compute HR. Clinical ECG monitors rely on a minimum of 30 seconds of ECG data to calculate HR, whereas the iHRD examines 2 second ECG signals at a time. This approach speeds up HR detection but with a cost to HR accuracy. However, the provided acceptability limit of ± 10 BPM for the iHRD show that the device can operate effectively for HR detection. Also, from the previous HR analyses, we revealed that iHRD HR follows clinical HR trend. For example, as the clinical HR increases, the iHRD’s HR also increases similarly. The iHRD in its current form can reliably reveal HR trends as shown in the comparative frameworks. This aspect can help clinicians determine the success of their resuscitation when HR can be seen increasing in real-time with their efforts.

Overall, we developed an infant iHRD capable of obtaining quick and accurate HR for neonatal HR monitoring. Further development is needed to improve accurate HR estimations and comply with health regulatory requirements to increase its likeability for clinical adoption. It is important to note that the iHRD is developed to technology readiness level (TRL) 7, the iHRD have been demonstrated in an appropriate clinical environment. This level shows that the technology is validated and feasible for future testing. The iHRD’s progress to TRL 8 and 9 will require more device design reiterations, developmental testing in real conditions, and regulatory compliance before the final form of the device is achieved. Furthermore, independent evaluation is required by other research labs and regulatory institutions before clinical adoption of the iHRD. The dataset used in the validation of this technology is also available for others conducting research in neonatal monitoring and resuscitation.

Limitations and future works

Device feasibility testing is currently limited to healthy pre-term, early- and full-term newborns which do not represent the full potential of the iHRD in at birth neonatal resuscitation scenarios. Future studies are warranted to investigate the iHRD in true neonatal resuscitation.

On the other hand, healthcare professionals performing NRP will require real-time validation and feedback that displayed HR on the iHRD is accurate and reliable. This accuracy can be impacted from neonatal motion-artifact leading to unreliable HR values. Future directions will explore the development of real-time motion artifact detection offering confidence to users during device use. Also, it is important to note that the iHRD is intended to act as a decision support tool for clinicians and is not meant to be a sole solution for neonatal resuscitation.

Another limitation is device interference with chest compression. The current iHRD device requires device placement over the infant’s heart, limiting pediatrician access. The device operates by displaying HR quickly, then removed and placed after chest compressions. The future direction will explore the industrial design modification to redesign the form factor as a belt for continuous neonatal HR monitoring. The device does not interfere with any other neonatal resuscitation steps except for chest compression.

Methods

We designed and developed a novel infant single-lead ECG and HR detector that is capable of estimating HR quickly and accurately for neonatal resuscitation efforts. The design involved the creation of patent-pending 3D printed dry electrodes, a signal processing analog front end (AFE) circuitry, software-based HR detection algorithm, and Bluetooth connectivity to transmit ECG and HR information to a graphical user-interface (GUI). The device also features replaceable coin-cell batteries, a 1.3-inch OLED display to provide HR information to the user, and a low-power microcontroller to lower overall device power consumption. The device is 6 cm x 5 cm in size. Current development cost projections places the iHRD in the stethoscope price bracket as shown in Table 1.

The entire device system is designed with the clinician feedback in mind, making a small hand-held portable device with a sleek industrial design and fluid-proof casing as shown in Fig. 4. The casing contains grooves for easier handling following user feedback from pediatricians and neonatologist. The device is intended to operate solely as a support decision system in conjunction with other tools deemed appropriate for neonatal resuscitation efforts.

Fig. 4.

iHRD Design. Frontside reveals a display screen for HR. Backside reveals two 3D printed dry electrodes for single-lead ECG acquisition.

Novel 3D printed dry electrodes

The skin-electrode interface is an important consideration in the development of any ECG electrodes, especially dry electrode development. Skin naturally is hydrophobic due to the presence of dead skin cells at the top most epidermal layer. This hydrophobicity makes it difficult to detect and obtain ECG from the skin’s surface. ECG is identified as a small amplitude signal of (1-2 mV). In turn, wet commercial ECG electrodes acquire consistent ECG signals through skin preparation which involves abrasing the top layer of the skin to expose a more hydrophilic layer beneath that is optimal for ECG acquisition.

In turn, we developed 3D printed dry electrodes for neonatal HR monitoring that offset. The electrodes are created using conductive PLA (Protopasta, California) featuring carbon black composites. This material is chosen due to its availability, cost, and manufacturability. The electrodes are designed using Ultimaker Cura, and 3D printed using the printing settings; head nozzle temperature 215C, head support bed temperature 60C, fill ratio 100%, and material extrusion print speed 35 mm/s²⁸. These printing settings are chosen to maintain the PLA’s biochemical and conductive materials. While the temperature choices follow the material’s datasheet recommendations for optimal 3D printing. These 3D printed dry electrodes are successfully fabricated in under 10 minutes.

Single-lead ECG design

Following 3D printing dry electrodes for neonatal ECG acquisition, a hardware-based setup is developed to acquire single-lead ECG signals. This setup features an AD8232 AFE chip that allows for analog signal processing. A 3-48 Hz bandpass filter is implemented on the AFE. This filter is implemented to eliminate high frequency EMI noise above 48 Hz and low frequency motion artifacts below 3 Hz including respiratory signals. The AFE also features ECG signal amplification of x115 to increase its amplitude and allow for digitization. This AFE is chosen due to its fast restore features that lowers the ECG settling time to less than one second. This advantage allows us to acquire single-lead ECG signals quickly to be used for HR estimation under 10 seconds from point of contact. The acquired single-lead ECG signals are then digitized using a 12-bit ADC and transmitted using Bluetooth to a GUI for further signal processing and analysis. The 12-bit resolution is sufficient to reveal all the important PQRST information of the signal and especially the QRS waveform for HR estimation.

Proposed HR detection algorithm

2-second intervals of neonatal single-lead ECG signals are used to compute fast HR. These signals are acquired using 500 Hz sampling rate. The 2-second ECG signals undergo a high-pass filter to remove baseline wander that further eliminates noise from the iHRD hardware. This filter is implemented on the microcontroller level in C using its difference equation Eq. 1.

1

After BW filtering, the 2-second ECG signal undergoes a second order low-pass filter with a cutoff frequency of 11 Hz to keep QRS information for HR detection. The filter’s difference equation is Eq. 2 which is implemented effectively in the firmware of the microcontroller.

2

Where T is the sampling period of seconds.

Both filters are combined and chosen due to their difference equations and its flexibility in embedded implementation using a minimum amount of lines of code. These filters also minimize the use of computation resources on the embedded microcontroller.

ECG signal filtering is followed by signal manipulation to highlight the R-peak information and diminish the other ECG waveforms such as T-wave and P-waves. This manipulation is conducted using a 5-point derivative equation, Eq. 3. The QRS slopes are identified using this equation, highlighting the entire complex. After derivation, the slopes are squared point by point. The squaring technique allows us to highlight the R-peak locations on the ECGs. After signal squaring, the 2-second signals undergo a moving average filter that smooths out and magnifies the R-peaks.

3

After filtering and signal manipulation, adaptive thresholding is performed considering neonatal physiological features. Adults usually have a normal resting HR of 60-100 BPM, where for each second of sinus rhythm one or two QRS complexes are present. Whereas newborns with higher resting HR of 120 to 160 BPM, in the first 24 hours, show more than two QRS complexes in one second. This information allows us to utilize only 2 seconds of neonatal ECG sinus rhythm to accurately estimate HR in a few seconds when compared to the standard HR measurement modalities that take a minimum of 30 seconds. Using this approach, the real-time capabilities of detecting accurate HR within a few seconds of electrode contact are achievable.

iHRD proof-of-concept validation and dataset

To validate the iHRD, a proof-of-concept study is developed and implemented to test the iHRD’s capabilities in real clinical environments²⁶. This study is designed to acquire infant HR and single-lead ECG medical data using the iHRD and compare its results to currently used clinical HR monitors. This analysis is conducted to ensure that the iHRD’s HR is accurate and reliable compared to current HR modalities and can provide HR information in less than 10 seconds from initial contact.

The study is conducted on 50 healthy newborns and a dataset, SAR-NE, is formed from infants’ vital sign data²⁶. The infants were chosen randomly from the period between June 2023, and February 2025. This study was approved by Scarborough Health Network, PED-21-025, and Toronto Metropolitan University, 2023-070. The study is conducted in accordance with the relevant guidelines, regulations and the deceleration of Helsinki. The cross-sectional study is performed in the NICU, and PPU of a large community hospital in Scarborough, Canada. Guardians and Parents of recruited infants provided written informed consent. iHRD’s HR, and ECG’s or PO’s monitor HR is video recorded simultaneously. After video recordings, the infant’s data is anonymized and their HRs are annotated, following the ethics-approved protocol mentioned in²⁶. iHRD’s HR are recorded for every 2 seconds of video recordings simultaneously with clinical HRs, representing HR data pairs. Overall, 4147 HR data pairs are obtained from this study and dataset. Figure 5 shows the setup for the iHRD in the clinical study.

Fig. 5.

Snapshot from Digital Camera Point of View.

We examined the iHRD’s capabilities in comparison to clinical HR monitors used in newborn monitoring and neonatal resuscitation efforts. We created a comparative statistical framework to compare HR data pairs. This comparative analysis included computing Pearson correlation coefficients, Spearman coefficients, inter-quartile ranges (IQR), normalized root-mean square error (NRMSE), mean of HR difference, accuracy, t-test statistic, mean absolute difference (MAE), standard deviation (SD), and plotting Bland Altman and correlation plots. To further determine the differences between iHRD and clinical HR, we examined separately; sex differences, clinical HR monitor differences, and analyzed the iHRD’s HR information with and without motion-artifact presence. Lastly, we conducted timing analysis to determine the time required by the iHRD device to display an accurate HR within 10 seconds of initial device contact with the infant.

Author contributions

This study was conceptualized by AA, SK, and NM. AA and SK are involved in the heart rate detector (HRD) prototype design and production. AA and NM are involved in direct data collection. AA performed the formal analysis. All authors contributed to the writing and obtaining of the ethics-approved proposals. All authors read and approved the final manuscript.

Funding

This work has been funded by the Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant and Ontario Graduate Scholarship (OGS).

Data availability

The raw HR and ECG information is available to researchers upon request. Please email abdelrahman.abdou@torontomu.ca

Declarations

Competing interests

SK is the founding Chief Scientific Officer of Nanu Health, the company involved in commercializing the infant heart rate detector prototype discussed in the paper. He is also a co-inventor of a patent application under review for the device discussed in the paper. NM is the cofounder of and owns stock in Nanu Health.

Ethical approval

This study is approved by Toronto Metropolitan University, 2023-070, and Scarborough Health Network, PED-21-025. Written informed consent is obtained from the parents of Newborns involved in this study.