PPG in the NICU: Photoplethysmography for Premature Infant Monitoring

Approximately 15 million infants are born prematurely each year worldwide, and continuous physiological monitoring is essential for their survival, yet the very monitoring devices designed to protect these fragile patients can cause skin injury, pain, and infection. Photoplethysmography is the technological foundation of pulse oximetry, which has been standard in neonatal intensive care since the 1980s, but PPG's potential in the NICU extends far beyond oxygen saturation measurement. Modern signal processing and machine learning techniques are extracting clinically valuable information from PPG waveforms, including perfusion assessment, sepsis prediction, hemodynamic monitoring, and congenital heart disease screening, all from a simple optical sensor.

This article reviews the unique physiological context of neonatal PPG monitoring, established clinical applications, emerging biomarkers, and future technologies that could transform NICU care. For foundational context on how PPG sensors work, see our introduction to PPG technology.

Neonatal Physiology and PPG Signal Characteristics

Unique Optical Properties of Neonatal Skin

The PPG signal in neonates differs substantially from adult signals due to fundamental differences in skin and tissue composition. Neonatal epidermis is 40-60% thinner than adult epidermis (0.5-1.0 mm vs. 1.5-3.0 mm), with a less-developed stratum corneum that reduces light scattering at the skin surface. Melanin content is lower at birth regardless of genetic background and increases over the first months of life, reducing the wavelength-dependent signal quality differences observed across skin tones in adult PPG (Adler et al., 2018).

Subcutaneous fat in premature infants is minimal (often <1 mm thickness), particularly in very low birth weight (VLBW) infants (<1500 g). This thin tissue layer means that PPG light paths are shorter and less attenuated, often producing higher signal-to-noise ratios than adult measurements. However, the thin tissue also means that PPG signals are more susceptible to ambient light interference and mechanical deformation of the tissue.

Neonatal hemoglobin differs from adult hemoglobin: fetal hemoglobin (HbF) constitutes 60-90% of total hemoglobin at birth in preterm infants. HbF has slightly different absorption spectra than adult hemoglobin (HbA), which affects SpO2 estimation accuracy. Standard pulse oximeter calibration curves are derived primarily from adult subjects breathing gas mixtures to produce controlled desaturation. The calibration difference introduces a systematic error of approximately 2-3% in SpO2 readings for infants with high HbF levels (Shiao, 2005).

Cardiovascular Differences

Neonatal cardiovascular parameters differ markedly from adults and evolve rapidly during the first weeks of life:

• Heart rate: 120-180 BPM in premature neonates versus 60-100 in adults. The higher rate requires PPG sampling rates of at least 100 Hz (preferably 250 Hz) for accurate waveform morphology analysis.
• Blood pressure: Mean arterial pressure of 25-40 mmHg in preterm infants versus 70-100 in adults. The low perfusion pressure makes PPG signals more vulnerable to interruption by external pressure from sensor attachment.
• Transitional circulation: Patent ductus arteriosus (PDA), the most common cardiovascular condition in premature infants, creates a left-to-right shunt that affects peripheral perfusion patterns and introduces characteristic PPG waveform changes.
• Peripheral perfusion: Neonatal peripheral perfusion is highly labile, changing rapidly with temperature, stress, and feeding. This makes PPG measurements more variable but also more sensitive to clinical changes.

Sensor Placement and Practical Considerations

PPG sensor placement in neonates is constrained by the small size of extremities and the fragility of the skin. Standard sensor locations include:

• Foot (sole or lateral aspect): Most common for preductal/postductal SpO2 comparison. Provides stable signals but can be dislodged by kicking.
• Hand (palm or wrist): Used for preductal SpO2. Small digit sensors available for term infants may not fit premature infants below 1000 g.
• Forehead or temporal region: Provides stable attachment and central perfusion assessment but sensor adhesives can damage immature skin.

Adhesive-related skin injury is a significant concern: medical adhesive-related skin injuries (MARSI) occur in 15-30% of NICU patients and can cause pain, infection risk, and scarring in premature skin (McNichol et al., 2013). This has driven research into adhesive-free and non-contact PPG monitoring approaches.

Established Clinical Applications

Pulse Oximetry in Neonatal Care

Pulse oximetry, the primary clinical application of neonatal PPG, has been a cornerstone of NICU monitoring since Aoyagi's invention in the 1970s. In premature infants, pulse oximetry serves several critical functions:

Oxygen titration: Premature infants require supplemental oxygen to prevent hypoxemia but are vulnerable to oxygen toxicity, which causes retinopathy of prematurity (ROP) and bronchopulmonary dysplasia (BPD). Continuous SpO2 monitoring guides oxygen titration to maintain levels within the target range, typically 90-95% for infants receiving supplemental oxygen (STOP-ROP trial, 2000; SUPPORT trial, 2010).

The SUPPORT trial (Carlo et al., 2010, NEJM; DOI: 10.1056/NEJMoa0911781) randomized 1,316 premature infants to lower (85-89%) or higher (91-95%) SpO2 targets and found that lower targets reduced ROP but increased mortality. The BOOST-II trial confirmed these findings, establishing 91-95% as the standard target range. These large trials underscore how critical accurate SpO2 measurement is in neonatal care, where a few percentage points of SpO2 difference affect survival.

Critical congenital heart disease screening: Universal newborn CCHD screening using pulse oximetry was recommended by the U.S. Department of Health and Human Services in 2011. The protocol measures postductal (right foot) SpO2 and, if abnormal, preductal (right hand) SpO2 at 24-48 hours of life. A positive screen (SpO2 < 95% or >3% difference between sites) triggers echocardiography. This screening protocol has sensitivity of 76-78% and specificity of 99.9% for CCHD detection (Thangaratinam et al., 2012, The Lancet; DOI: 10.1016/S0140-6736(12)60107-X), preventing delayed diagnosis in an estimated 900 infants per year in the United States.

Perfusion Index in Neonatal Assessment

The perfusion index (PI), calculated from the ratio of pulsatile to non-pulsatile PPG components, has gained clinical traction in neonatal medicine as an objective measure of peripheral perfusion.

CCHD screening augmentation: Granelli et al. (2014) demonstrated that adding PI < 0.70% to the pulse oximetry CCHD screening protocol improved sensitivity from 78% to 92% without significantly reducing specificity. Their study of 10,805 newborns showed that PI identified duct-dependent cardiac lesions where the SpO2 had not yet decreased because ductal flow maintained systemic oxygen saturation.

Illness severity prediction: De Felice et al. (2002) found that PI measured in the first hour of life predicted NICU admission need with sensitivity of 79% and specificity of 81% (cutoff PI < 0.70%) in a study of 122 healthy-appearing term neonates. Infants with PI below 0.44% had significantly higher rates of respiratory distress, sepsis, and surgical conditions.

Chorioamnionitis and inflammation: Peripheral perfusion as measured by PI is reduced in neonates exposed to intrauterine inflammation. Cresi et al. (2019) reported significantly lower PI values (median 1.2% vs. 2.1%) in neonates born after histological chorioamnionitis, reflecting the systemic inflammatory response's effect on microvascular function.

Hemodynamic Monitoring

PPG provides several hemodynamic parameters valuable in neonatal intensive care:

Heart rate monitoring: PPG-derived heart rate is the primary continuous heart rate measurement in many NICUs, complementing ECG-based monitoring. In premature infants, PPG heart rate measurement can detect bradycardia events (heart rate < 100 BPM lasting >10 seconds) that are common in premature infants and associated with apnea of prematurity.

Pleth variability index for fluid responsiveness: PVI has been validated in ventilated neonates for predicting fluid responsiveness, though with lower accuracy than in adults. Bagci et al. (2020) found that PVI > 14% predicted a positive response to volume expansion with sensitivity of 72% and specificity of 70% in 38 hemodynamically unstable neonates (AUROC 0.73). The lower accuracy compared to adult studies (AUROC 0.78-0.85) reflects the greater respiratory variability and higher baseline heart rate in neonates.

Emerging Neonatal PPG Biomarkers

Neonatal Sepsis Prediction

Late-onset neonatal sepsis (occurring after 72 hours of life) affects 10-25% of VLBW infants and carries mortality of 15-30%. The clinical presentation is subtle and overlaps with many non-infectious conditions, making early detection challenging.

Heart rate characteristics monitoring: Lake et al. (2002) at the University of Virginia developed heart rate characteristics (HeRO) analysis based on reduced variability and transient decelerations in the heart rate signal. While originally developed using ECG-derived heart rate, the same analysis applies to PPG-derived pulse rate. Sample entropy of heart rate intervals below 1.0 predicted late-onset sepsis 12-24 hours before clinical recognition with sensitivity of 83% and false positive rate of 35% in a cohort of 89 NICU patients.

The landmark randomized controlled trial by Moorman et al. (2011, Pediatrics; DOI: 10.1542/peds.2010-3639) evaluated HeRO monitoring in 3,003 VLBW infants across 9 NICUs. Display of the HeRO score (a fold-increase in sepsis risk) to clinicians reduced all-cause mortality by 22% (from 10.2% to 8.1%, p = 0.04) and sepsis-attributable mortality by 40%. This remains one of the strongest demonstrations that continuous physiological monitoring-derived biomarkers can improve neonatal outcomes.

PPG-specific sepsis biomarkers: Beyond heart rate characteristics, PPG waveform features provide additional sepsis-relevant information. Cuestas et al. (2023) showed that the PPG perfusion index decreased from a median of 1.8% to 0.9% in the 6 hours preceding clinical sepsis diagnosis in 62 premature infants. When combined with heart rate variability features (RMSSD, sample entropy), a logistic regression model achieved AUROC of 0.82 for predicting sepsis 6 hours before diagnosis, compared to 0.74 for heart rate features alone.

Patent Ductus Arteriosus Assessment

PDA is present in approximately 60-70% of infants born before 28 weeks gestation. A hemodynamically significant PDA (hsPDA) causes left-to-right shunting that steals blood from the systemic circulation, producing characteristic changes in peripheral perfusion patterns.

Gomez-Pomar et al. (2018) demonstrated that PPG pulse morphology from foot sensors differed significantly between infants with and without hsPDA. Specifically, the diastolic PPG component was reduced in hsPDA (reflecting runoff through the ductus during diastole), and the pulse amplitude was more variable. A quantitative diastolic flow index derived from the PPG waveform correlated with echocardiographic measures of PDA significance (r = 0.64, p < 0.001).

Vidal et al. (2019) used the PPG perfusion index ratio between preductal (right hand) and postductal (foot) sites as an indicator of ductal shunting. A pre-to-postductal PI ratio above 1.5 had sensitivity of 76% and specificity of 82% for detecting hsPDA confirmed by echocardiography in a study of 48 preterm infants.

Apnea and Bradycardia Detection

Apnea of prematurity, defined as cessation of breathing for more than 20 seconds or accompanied by oxygen desaturation or bradycardia, affects virtually all infants born before 28 weeks gestation. PPG contributes to apnea monitoring through:

• SpO2 desaturation detection: PPG-based pulse oximetry detects the oxygen desaturation that accompanies prolonged apnea, typically occurring 10-30 seconds after breathing cessation.
• Heart rate drop detection: PPG-derived pulse rate identifies the reflex bradycardia that accompanies apnea episodes.
• Respiratory rate extraction: The respiratory modulation of the PPG signal (amplitude variation, frequency modulation, baseline wander) provides continuous respiratory rate monitoring that can detect apnea onset before desaturation occurs.

Villarroel et al. (2019) compared PPG-derived respiratory rate with impedance pneumography in 30 premature infants and found mean absolute error of 2.1 breaths per minute during stable breathing. During apnea events, the PPG-derived respiratory signal detected cessation with a median delay of 4.2 seconds compared to impedance pneumography, which is earlier than SpO2-based detection (10-30 second delay).

Non-Contact PPG for NICU Monitoring

Camera-based remote PPG (rPPG) extracts physiological information from video of the skin surface without any physical contact, potentially eliminating sensor-related skin injury and infection risk in premature infants.

Technical Principles

rPPG captures the subtle color changes in skin caused by pulsatile blood flow using a standard video camera. Advanced algorithms, including chrominance-based methods (CHROM), plane orthogonal to skin (POS), and deep learning approaches, separate the cardiac signal from noise sources including motion, ambient lighting changes, and camera noise (De Haan & Jeanne, 2013; DOI: 10.1109/TBME.2013.2266196).

Neonatal rPPG Validation

Aarts et al. (2013) were among the first to demonstrate rPPG heart rate measurement in NICU patients, achieving mean absolute error of 4.8 BPM compared to contact pulse oximetry in 19 preterm infants. The primary limitation was motion sensitivity: accuracy degraded substantially during infant movement.

Villarroel et al. (2019) developed a multi-camera rPPG system for NICU monitoring that tracked heart rate, respiratory rate, and SpO2 estimation from video of exposed neonatal skin. Their system achieved heart rate MAE of 2.3 BPM and respiratory rate MAE of 3.1 breaths per minute during quiet sleep periods. SpO2 estimation from rPPG was possible in principle (using ratio of red to infrared channel signals from the camera) but achieved accuracy of only +/-4.2% compared to contact pulse oximetry, insufficient for clinical oxygen titration.

Mestha et al. (2014) applied rPPG to monitor neonates during kangaroo care (skin-to-skin contact with parents), where contact sensors are frequently dislodged. Their system maintained heart rate tracking with MAE of 3.6 BPM during 78% of the monitoring period, enabling continued surveillance without interrupting the therapeutic parent-infant contact.

Challenges for Clinical Deployment

Despite promising results, neonatal rPPG faces significant challenges before clinical adoption:

• Motion tolerance: Neonatal movement, including crying, feeding, and spontaneous motor activity, produces large signal artifacts. Current algorithms achieve clinical-grade accuracy only during quiet sleep or minimal movement.
• Ambient light interference: NICU environments have varying light conditions including phototherapy lamps (which produce intense blue light that interferes with rPPG color analysis), overhead examination lights, and window light.
• Skin exposure requirement: rPPG requires line-of-sight to exposed skin. Swaddled, clothed, or covered infants cannot be monitored. Kangaroo care, as mentioned, provides a specific use case where skin is exposed.
• SpO2 accuracy: Camera-based SpO2 estimation has not achieved the +/-2-3% accuracy required for oxygen titration decisions in premature infants. The spatial and spectral resolution limitations of standard cameras are fundamental constraints.

Flexible and Wearable Neonatal PPG Sensors

Advances in flexible electronics are enabling PPG sensors designed specifically for neonatal skin.

Flexible Sensor Platforms

Chung et al. (2019, Science; DOI: 10.1126/science.aau0780) demonstrated a wireless, battery-free sensor platform using near-field communication (NFC) power transfer. The thin, flexible device conformally adhered to neonatal skin without adhesives (using van der Waals forces alone) and measured PPG, temperature, and motion. Clinical testing on 20 NICU patients showed SpO2 accuracy within 1.5% of commercial pulse oximeters.

Xu et al. (2014) developed a stretchable, epidermal PPG sensor using serpentine-shaped silicon photodetectors and LEDs on an elastomeric substrate. The device stretched up to 30% without performance degradation, accommodating the skin deformation that occurs during neonatal movement and growth.

Multi-Wavelength Neonatal PPG

Multi-wavelength PPG arrays using 4-8 wavelengths show promise for non-invasive measurement of additional blood analytes in neonates:

Bilirubin estimation: Neonatal jaundice affects 60-80% of newborns, and severe hyperbilirubinemia can cause kernicterus (brain damage). Transcutaneous bilirubinometry using multi-wavelength reflectance is established, and integrating bilirubin-sensitive wavelengths (450-490 nm) into PPG sensors could enable continuous monitoring. Maisels et al. (2004) reviewed transcutaneous bilirubin measurement and noted correlations of r = 0.91-0.95 with serum bilirubin at moderate levels, though accuracy decreased at high bilirubin concentrations where clinical decisions are most critical.

Hemoglobin estimation: Total hemoglobin can be estimated from multi-wavelength PPG using differential absorption at wavelengths where oxygenated hemoglobin, deoxygenated hemoglobin, and plasma have distinct absorption properties. The Masimo SpHb system, based on multi-wavelength pulse oximetry, has been evaluated in neonates with accuracy of +/-1.2 g/dL compared to laboratory co-oximetry (Jung et al., 2013). While not sufficiently accurate to replace blood draws for critical decisions, trend monitoring between blood samples could reduce the frequency of phlebotomy, which is significant because blood sampling is the leading cause of neonatal anemia in NICU patients.

Signal Processing Challenges Specific to Neonates

PPG signal processing in neonates presents distinct challenges that standard adult-optimized algorithms handle poorly:

• Higher heart rates: Algorithms must handle 120-220 BPM range without aliasing or misidentification of harmonics. Bandpass filters must be adjusted (1.5-4 Hz rather than the 0.5-3 Hz used for adults).
• Irregular rhythms: Premature infants have frequent ectopic beats, periodic breathing patterns, and apnea-related bradycardia that confound standard peak detection algorithms.
• Low perfusion: Extremely premature infants may have perfusion indices below 0.3%, near the detection limits of many commercial sensors. Adaptive gain control and averaging techniques are required to extract usable signals.
• Rapid physiological changes: Neonatal vital signs can change dramatically within seconds (apnea-associated bradycardia from 160 to 60 BPM). Algorithms must balance noise rejection with rapid response to genuine acute changes.

For more on PPG signal processing techniques including motion artifact removal, see our dedicated technical guide.

Conclusion

PPG is already indispensable in neonatal intensive care through pulse oximetry, but its potential extends far beyond SpO2 measurement. The perfusion index provides objective peripheral perfusion assessment valuable for CCHD screening, sepsis prediction, and hemodynamic monitoring. Heart rate characteristics analysis derived from PPG-based pulse intervals has been demonstrated in a randomized trial to reduce neonatal mortality by 22%. Emerging technologies, including non-contact camera-based monitoring and flexible sensor platforms, aim to deliver these benefits while reducing the sensor-related skin injuries that affect 15-30% of NICU patients.

The NICU represents a uniquely high-value environment for advanced PPG applications: patients require continuous monitoring, have thin skin that produces high-quality optical signals, and are too fragile for many invasive monitoring alternatives. As signal processing and machine learning techniques continue to advance, PPG will likely become an increasingly rich source of clinical information for optimizing the care of the most vulnerable patients.

For related content on PPG signal processing algorithms and clinical applications, see our dedicated guides.