Neonatal Health Screening With a Smartphone: How It Works in Africa

Every year, roughly 2.3 million newborns die within their first 28 days of life. Sub-Saharan Africa carries a disproportionate share of that burden, with a neonatal mortality rate of 27 deaths per 1,000 live births — more than ten times the rate in high-income countries. According to the UN Inter-agency Group for Child Mortality Estimation's 2024 report, the region accounts for 43% of all newborn deaths worldwide, even though it represents about 29% of global births. Most of these deaths stem from conditions that are treatable when caught early: birth asphyxia, sepsis, and complications of prematurity including jaundice. The bottleneck is not treatment. It is detection.

Neonatal health screening with a smartphone in Africa is beginning to change that equation. A phone camera, the right software, and a community health worker can now perform screening assessments that previously required clinical equipment most rural health posts simply do not have.

"The gap in neonatal survival between high-income and low-income countries is primarily a gap in screening and early detection, not in the availability of treatments. Most interventions that save newborn lives are low-cost — what is expensive is the infrastructure to identify which newborns need them." — Dr. Joy Lawn, Professor of Maternal, Reproductive and Child Health, London School of Hygiene & Tropical Medicine

Why traditional neonatal screening fails in rural Africa

The standard clinical approach to neonatal screening relies on equipment: pulse oximeters for oxygen saturation, transcutaneous bilirubinometers or serum blood draws for jaundice, electronic thermometers, and trained nursing staff to interpret the readings. In sub-Saharan Africa, this infrastructure barely exists outside referral hospitals.

A 2022 survey published in BMJ Global Health by Okomo et al. found that fewer than 12% of primary health facilities in Kenya, Nigeria, and Tanzania had functioning pulse oximeters available for neonatal use. The situation is worse at the community level. Community health workers — who are often the first and only health contact for newborns in rural areas — carry minimal equipment. A thermometer, a scale if they're lucky, and their own clinical judgment.

The result is that danger signs get missed. Neonatal jaundice, which affects an estimated 60% of term newborns and 80% of preterm newborns globally (Bhutani et al., Pediatrics, 2013), frequently goes undetected in African communities until it has progressed to acute bilirubin encephalopathy. By that point, the damage — hearing loss, cerebral palsy, death — is often irreversible. A study by Olusanya et al. in The Lancet Child & Adolescent Health (2018) estimated that extreme hyperbilirubinemia contributes to up to 114,000 neonatal deaths annually in low- and middle-income countries.

This is not a failure of CHWs. It is a failure of the tools available to them.

How smartphone cameras screen newborns

Smartphone-based neonatal screening works through two distinct technical approaches, depending on what is being measured.

Camera-based jaundice detection uses the phone's camera to photograph the newborn's skin or sclera (the white of the eye). Software algorithms analyze the color properties of the image — specifically, the ratio of blue to yellow channel values — to estimate bilirubin concentration. The approach draws on the same optical principles as transcutaneous bilirubinometry, except the sensor is a consumer camera rather than a dedicated medical device.

The BiliCam system, developed by de Greef et al. at the University of Washington, demonstrated this approach using a standard color calibration card placed next to the newborn during photography. Their validation study showed correlation coefficients above 0.85 with total serum bilirubin measurements in a diverse neonatal population. The neoSCB app, developed at University College London, takes photographs of the newborn's sclera and uses machine learning to classify jaundice risk. A 2024 validation study published in PLOS Digital Health reported sensitivity above 89% for detecting clinically significant hyperbilirubinemia.

Contactless vital signs via rPPG (remote photoplethysmography) is the second approach. The phone's camera records a short video of the newborn's face or exposed skin. Algorithms detect micro-changes in skin color caused by blood pulsing through capillaries — changes invisible to the eye but measurable by the camera sensor. From this signal, the software extracts heart rate, respiratory rate, and oxygen saturation estimates. Dewagiri et al. published a 2025 preprint (arXiv) on VideoPulse, a system specifically designed for neonatal rPPG that addresses the unique challenges of newborn physiology: faster heart rates, smaller facial regions, and more frequent movement.

Screening method	Equipment needed	Training time	Cost per screening	Works offline	Jaundice detection	Vital signs
Serum bilirubin blood draw	Lab, needles, centrifuge, reagents	Months (phlebotomy certification)	$5-15 per test	No (requires lab)	Yes (gold standard)	No
Transcutaneous bilirubinometer	Dedicated device ($3,000-8,000)	Days	$0.50-1 per test (device amortized)	Yes	Yes	No
Pulse oximeter (neonatal)	Neonatal probe + oximeter ($200-800)	Hours	$0.10-0.30 per test	Yes	No	Partial (SpO2 + HR only)
Smartphone camera (BiliCam/neoSCB)	Smartphone + color card or app	Hours	<$0.05 per test	Yes	Yes	No
Smartphone rPPG (contactless)	Smartphone only	Hours	<$0.05 per test	Yes	Emerging	Yes (HR, RR, SpO2)

The cost column matters more than it might seem. When a health ministry is deciding whether to roll out neonatal screening to 50,000 CHWs, the difference between $200 per worker in equipment costs and $0 (they already have phones) determines whether the program happens at all.

What gets screened and what gets caught

Smartphone neonatal screening is not trying to replace a neonatal intensive care unit. It is trying to answer a simpler question: does this newborn need to be referred to a facility, or can they safely stay home?

Jaundice

This is the most mature application. Neonatal jaundice is common, usually harmless, and occasionally deadly — a combination that makes screening particularly valuable. The clinical challenge is distinguishing physiological jaundice (normal, self-resolving) from pathological jaundice (dangerous, requiring phototherapy or exchange transfusion). Visual assessment by healthcare workers is unreliable, particularly for darker skin tones. Keren et al. (Pediatrics, 2009) showed that clinical visual assessment had a sensitivity of only 52% for significant hyperbilirubinemia.

Smartphone-based screening adds objectivity. The camera measures color values that the human eye cannot reliably distinguish, and the algorithm applies consistent thresholds regardless of the examiner's experience level. This does not make it a diagnostic test — it makes it a triage tool. A CHW who gets a "high risk" reading from the app refers the newborn for confirmatory testing. A CHW without the app might not recognize the problem until the baby stops feeding.

Respiratory distress

A newborn breathing too fast (tachypnea above 60 breaths per minute) or too slowly is showing signs of potential sepsis, pneumonia, or respiratory distress syndrome. Counting respiratory rate manually for a full 60 seconds on a squirming newborn is difficult and inaccurate. Smartphone-based respiratory rate counting, whether through rPPG chest movement detection or video-based algorithms that track thoracic excursion, offers continuous measurement over a longer window with less observer error.

Heart rate abnormalities

Neonatal heart rate outside the normal range (typically 120-160 bpm for term newborns) can signal cardiac defects, sepsis, or metabolic disturbance. Auscultation with a stethoscope requires training and a quiet environment — neither of which is guaranteed in a community screening context. rPPG-derived heart rate from a 30-second video provides an objective number that the CHW can compare against reference ranges.

Field realities that shape the technology

Building a neonatal screening app that works in a lab is one thing. Building one that works when a CHW is sitting on a mat in a one-room house with a kerosene lamp and a crying baby is something else.

Lighting variability. Indoor environments in rural Africa frequently lack electric lighting. Camera-based screening must function under natural light from windows, kerosene or solar lamp illumination, and mixed lighting conditions. The algorithms need to compensate for color temperature shifts that would confuse a system calibrated for clinical fluorescent lighting. BiliCam's color calibration card partially addresses this by providing a known color reference in each image, allowing the software to normalize the image before analysis.

Skin tone diversity. This is a significant technical challenge. Most rPPG and camera-based screening algorithms have been trained and validated primarily on lighter skin tones, where the optical signal-to-noise ratio is higher because melanin absorbs less of the reflected light. Performance on darker skin has historically been worse. Recent work by Addison et al. (Nature Medicine, 2023) documented pulse oximetry bias in darker-skinned patients and called for diverse training datasets. Smartphone-based systems deployed in sub-Saharan Africa must be validated across the full range of skin tones present in the populations they serve. This remains an active area of research and an honest gap in the field.

Neonatal movement. Newborns move unpredictably. They scrunch, squirm, and cry. Motion artifacts corrupt both camera-based jaundice assessment (blurred images) and rPPG signals (the algorithm cannot track skin color changes when the face keeps moving). Signal processing techniques such as motion-compensated filtering and multi-region-of-interest tracking partially mitigate this, but the practical solution is often simpler: screen when the baby is calm or sleeping, which CHWs learn to time with experience.

Connectivity and data. As covered in our earlier post on how rPPG works without internet, all screening computation happens on-device. Results sync later when connectivity is available. For neonatal screening specifically, this means the CHW gets an immediate result — "refer" or "normal" — without waiting for a server response. The referral decision happens in the home, not after returning to a coverage zone.

Where programs are running

Several organizations have moved past pilot phase into operational deployment.

In Nigeria, the Survive and Thrive Global Development Alliance (supported by USAID) has integrated smartphone-based jaundice screening into community newborn care protocols in Lagos and Kano states. Their program targets the 48-hour home visit window after birth, when jaundice risk is rising but before most families would independently seek care.

In Kenya, the Kenya Medical Research Institute (KEMRI) has partnered with academic groups to validate smartphone screening tools in county hospitals and is evaluating community-level deployment through the existing CHW infrastructure that was expanded under the 2023 Primary Health Care Act.

In Uganda, where Circadify has deployed smartphone-based vital signs screening for community health programs, the technology platform is being evaluated for neonatal applications. The on-device processing architecture that supports adult screening translates directly to neonatal use cases — the phone camera and processing pipeline are the same, with algorithmic adjustments for neonatal physiology.

Across East Africa, the Newborn Essential Solutions and Technologies (NEST360) alliance — a consortium including Rice University, Malawi's Kamuzu Central Hospital, and partners in Kenya, Nigeria, and Tanzania — has been working since 2019 to develop and deploy affordable neonatal technologies, including camera-based monitoring tools designed for low-resource settings.

Current research and evidence

The evidence base for smartphone neonatal screening is building but still early. A few studies define the current state.

Taylor et al. (Pediatrics, 2017) conducted a systematic review of mobile health interventions for neonatal care in low-income countries and found that mHealth tools improved CHW performance in newborn assessment, though they noted the need for larger trials measuring mortality outcomes rather than process indicators alone.

Rahi et al. published a 2025 review in Artificial Intelligence in Medicine evaluating AI-based neonatal jaundice detection across multiple smartphone platforms. They reported that neural network models achieved accuracy above 90% compared to serum bilirubin, though performance varied with image quality and ambient lighting conditions. The review flagged skin pigmentation diversity in training data as the most important gap limiting deployment confidence.

For contactless vital signs, the evidence in neonatal populations specifically is thinner than for adults. Villarroel et al. (Physiological Measurement, 2019) demonstrated camera-based heart rate estimation in NICU patients using overhead cameras, achieving mean absolute errors under 5 bpm. Translating this to handheld smartphone cameras in community settings adds variability that has not yet been fully characterized in published literature.

The honest summary: smartphone jaundice screening has the strongest evidence and is closest to routine deployment. Smartphone-based neonatal vital signs via rPPG is technically feasible and being actively developed, but needs more validation studies in the target populations and environments.

What comes next for neonatal smartphone screening in Africa

Two trends will likely determine how fast this field moves.

The first is integration into national CHW protocols. Individual apps and research tools exist. What is missing in most countries is formal incorporation into the standard community health worker toolkit — the official list of equipment and procedures that governments fund and train for. South Africa's recent revision of its Ward-Based Primary Health Care Outreach Team guidelines included provisions for digital health tools, and other countries are watching that process. When a health ministry adds smartphone screening to the CHW protocol, it moves from "interesting pilot" to "scaled program" overnight.

The second is algorithm performance across skin tones. The technical community knows this is the barrier. Training datasets that reflect the full melanin spectrum of sub-Saharan African newborns are being assembled by groups including the MIT Media Lab's Camera Culture group and the University of Cape Town's Division of Biomedical Engineering. Until these algorithms are validated on diverse populations and published in peer-reviewed settings, program managers will reasonably hesitate to trust them for clinical triage decisions.

There is also a regulatory question. Most smartphone screening tools currently operate as screening aids rather than diagnostic devices, which keeps them below the medical device regulatory threshold in most African jurisdictions. As the tools become more capable and programs rely on them for referral decisions, regulatory frameworks will need to catch up. The African Medicines Regulatory Harmonization initiative is beginning to address digital health tools, but guidelines specific to AI-based screening remain sparse.

Frequently asked questions

Can a smartphone really detect jaundice in a newborn?

Yes, with caveats. Smartphone camera apps can estimate bilirubin levels by analyzing skin or scleral color. Validation studies show correlation above 0.85 with lab measurements in controlled settings. The technology works as a screening tool to flag high-risk newborns for confirmatory testing — it does not replace a serum bilirubin test for diagnosis.

Does smartphone neonatal screening work on darker skin tones?

Performance varies. Most algorithms perform best on lighter skin because the optical signal is stronger with less melanin absorption. Research groups are actively working to close this gap with more diverse training datasets and skin-tone-adaptive algorithms. This is the most important technical limitation to address before widespread deployment in sub-Saharan Africa.

Do community health workers need medical training to use these tools?

The tools are designed for CHWs with standard training (typically 2-6 weeks of general community health training). The apps guide the screening process step by step and provide simple outputs: refer to facility, or continue monitoring. Field studies in Nigeria and Kenya have shown that CHWs can learn to use smartphone screening tools in a few hours of additional training.

How does this connect to reducing neonatal mortality?

Most neonatal deaths in Africa result from conditions that are treatable at health facilities — but only if the newborn reaches the facility in time. Smartphone screening at the community level catches danger signs during home visits within the first 48 hours of life, when the risk is highest and the window for intervention is narrowest. Earlier detection means earlier referral, which means better outcomes.

---

Smartphone-based neonatal screening is not a theoretical concept — it is being deployed and evaluated across multiple African countries right now. The technology works on the devices that CHWs already carry. Companies like Circadify are building the on-device processing infrastructure that makes contactless vital signs measurement possible without connectivity, without additional equipment, and without specialized training. For newborns in communities where a hospital is hours away, that capability could be the difference between a danger sign caught and one missed.