Screened once, if at all
WHO recommends eight antenatal contacts. Most programmes can fund a hemoglobin test at one or two of them, so iron deficiency that develops in the second and third trimester goes unseen.
Anemia screening for pregnancy
Anemia in pregnancy,
Anemia in pregnancy,
caught at every visit.
Nivamed estimates hemoglobin from one phone photo of the inner eyelid — so anemia gets checked at every antenatal contact, not only the visits that reach a laboratory. No needle, no tube, no return trip.
Free during pilot · No card · Unsubscribe anytime
- 30sper screening
- 110 g/LWHO pregnancy threshold
- 0consumables
Method
Palpebral conjunctiva imaging
Trimester-aware thresholds
Patient-stratified cross-validation
Reported against mean baseline
0of pregnant women worldwide are anemic (WHO)
0g/L — WHO anemia threshold in pregnancy
0antenatal contacts WHO recommends
0from photo to risk band
The gap
The most common complication of pregnancy
The most common complication of pregnancy
is also the most often missed.
Hemoglobin usually gets measured at booking, then rarely again. Anemia develops across trimesters — a single early reading says little about how a pregnancy is actually going.
The lab is the bottleneck
Venipuncture needs a phlebotomist, transport and a return visit. In rural antenatal care that loop runs longer than the clinical window it was meant to inform.
Late detection costs more
Anemia at delivery is associated with preterm birth, low birth weight and greater vulnerability to postpartum hemorrhage. Iron treatment works — but it needs weeks, not days.
How it works
Three steps, inside the visit you already run.
No dongle, no cuff, no strip. The phone in the midwife's pocket is the instrument.
Capture at intake
On-screen guides frame the palpebral conjunctiva — the inner eyelid — while the patient is being weighed or having blood pressure taken. Nivamed rejects the frame if focus, glare or exposure fall out of tolerance.
- Adds under a minute to the visit
- Auto white-balance lock
- Works offline in the field
Hold still · 2s
Model reads the colour
A convolutional backbone fine-tuned on patient-stratified folds maps tissue chromaticity to a hemoglobin value in g/L, cross-checked against a CIELAB feature model that flags disagreement.
- Per-patient cross-validation, no leakage
- Gestational age recorded with every reading
- Skin-tone stratified evaluation
a*/b* chromaticity
Act within the same visit
You get a point estimate, a confidence interval and a triage band against the WHO pregnancy threshold of 110 g/L — screen negative, borderline, or refer for CBC. Readings build a trend across the pregnancy.
- Referral threshold configurable per protocol
- Trend by gestational week, not by date
- Exportable to EMR via FHIR
Hb by gestational week
Platform
Built around the antenatal visit
A number is not a product. Nivamed ships the workflow that makes the number useful to a midwife.
Trimester-aware thresholds
The anemia cut-off moves through pregnancy. Nivamed bands each reading against the threshold for that gestational age instead of one flat number.
Calibration guardrails
Ambient light, white balance and device profile are captured with every frame. Out-of-envelope shots are rejected, not silently guessed.
Antenatal dashboard
Caseload by gestational week, referral queue, and who has missed a screening. See a patient's whole pregnancy on one trend line.
Symptom baseline
Age, trimester, fatigue, pica, dizziness and shortness of breath feed a companion model, so you always see what the image adds over the questions a midwife already asks.
Offline-first field mode
Inference runs on-device. Screenings queue locally and sync when a signal appears — designed for home visits, not hospitals.
Privacy by construction
Identifiers hashed at capture. Images never leave the device unless a clinician opts in. No laboratory value ever reaches the model at inference.
The science
We publish the numbers that could embarrass us.
Anemia screening from images has a long history of overstated accuracy. Nivamed reports against the honest baseline: what you would get by ignoring the image and guessing the cohort mean.
8.4 g/L
Mean absolute error
Baseline 11.9 g/L
0.79
AUC, anemia flag
95% CI ±0.10
5-fold
Patient-stratified CV
No patient in two folds
Reported on a development cohort of 130 pregnant patients at a single site, using the WHO pregnancy threshold of 110 g/L. This is proof-of-concept performance, not deployment evidence — multi-site and multi-device validation is in progress, and severe anemia is too rare in this cohort to model separately.
Per-region performance
MAE g/L · lower better
- Palpebral conjunctiva8.4
- Tongue9.2
- Lower lip10.1
- Nail bed11.3
- Palm crease11.7
Regions that do not beat the mean baseline are excluded from fusion.
Who it's for
One measurement, three points in maternal care
Know whether the tiredness is iron
Fatigue, breathlessness and dizziness are normal in pregnancy until they are not. Check between appointments and bring a trend to your midwife instead of a feeling.
- Trend by gestational week
- Iron and supplement reminders
- Share a PDF at your next visit
Screen at every contact, not just booking
Screen at intake while vitals are taken. Send only the borderline and low bands to venipuncture, and cut CBC volume without losing cases.
- Referral queue and thresholds
- Caseload by gestational week
- FHIR export to your EMR
Reach pregnancies the lab never sees
Equip community midwives and health workers with the phones they already own. Map anemia prevalence by district and route iron supplementation where it changes outcomes.
- Offline capture with later sync
- Prevalence dashboards by district
- Program-level export and API
How we report
Claims we refuse to make
Image-based anemia screening has a long record of impressive numbers that fail to replicate. These are the rules we hold ourselves to.
Always against baseline
Every accuracy figure is published next to what you would get by ignoring the image and guessing the cohort mean. A model that does not beat that number is not reported as working.
No patient in two folds
Cross-validation splits by patient, never by image. Splitting by image inflates results by letting the model memorise a face — the single most common flaw in this literature.
Suspicious results get audited
An AUC above 0.95 on a cohort this size means something leaked, not that the model is excellent. We chase those down before publishing, not after someone else finds them.
Pricing
Priced per screening, not per seat
Pilot partners keep pilot pricing for twelve months after general availability.
Expectant mothers
Free
For anyone tracking their own pregnancy.
- 8 screenings per month
- Trend by gestational week
- PDF export for your midwife
- On-device processing
Antenatal clinic
Contact us
Priced per screening, with volume tiers.
- Unlimited users and devices
- Referral queue and thresholds
- Caseload dashboard and audit log
- FHIR / CSV export
- Onboarding and midwife training
Maternal health program
Custom
Ministries, NGOs and multi-district programmes.
- Offline field deployment
- Prevalence mapping dashboards
- Data residency options
- Model validation on your cohort
- Dedicated support engineer
FAQ
Straight answers
Is Nivamed a replacement for antenatal blood tests?
No. Nivamed is a screening and triage tool that sits alongside routine antenatal care. It indicates whether a laboratory test is warranted and how hemoglobin is trending. Diagnosis, iron therapy decisions and anything involving severe anemia require a CBC and a clinician.
Does it work in every trimester?
Readings can be taken at any point in pregnancy. Hemoglobin falls naturally in the second trimester through plasma volume expansion, so the anemia threshold shifts with gestational age — Nivamed bands each reading against the cut-off for that week rather than one flat number.
Is it safe for the pregnancy?
Nivamed takes an ordinary photograph. There is no needle, no radiation, no contact and no light source beyond the phone screen. The measurement itself carries no physical risk. The risk to manage is a missed case, which is why borderline readings are routed to a lab rather than cleared.
How does it handle different skin tones?
Primary regions are mucosal — inner eyelid, tongue, lip — where melanin contribution is minimal. Performance is reported stratified by Fitzpatrick group, and any region where a group underperforms is dropped from fusion for that group rather than averaged away.
What ruins a reading?
Mixed lighting, coloured bulbs, heavy camera post-processing and flash reflections. The app checks white balance and exposure before accepting a frame and asks for a retake rather than returning a guess.
Where do the images go?
Inference is on-device. Images stay local unless a clinician explicitly opts into contributing them to validation, in which case identifiers are hashed at capture and no clinical laboratory values are attached.
Is it regulated or approved?
Not yet. Nivamed is currently a research and wellness tool under active clinical validation. It is not a cleared medical device and must not be used as the sole basis for any decision in antenatal care.
Screen every pregnancy that walks in,
not only the ones that reach a lab.
Pilot slots open monthly. Bring an antenatal cohort, we bring the validation protocol.
Typical onboarding: 2 weeks from signed protocol to first screening.