Protocol for a prospective accuracy study on an artificial intelligence-based ultrasound system for gestational age estimation among pregnant women in Ghana, Kenya and South Africa

This study protocol outlines a multi-country prospective cohort study in Ghana, Kenya, and South Africa designed to validate the accuracy of an AI-based handheld ultrasound system for gestational age estimation against conventional fetal biometry, aiming to support pre-eclampsia risk screening and the design of the PEARLS trial in low-resource settings.

The Gist

Imagine you are trying to bake a perfect cake, but you don't know exactly when you put the batter in the oven. If you guess the time wrong, the cake might be raw in the middle or burnt on the outside. In pregnancy, knowing the exact "baking time" (gestational age) is just as critical. It tells doctors when to start life-saving treatments, like giving aspirin to prevent a dangerous condition called pre-eclampsia.

However, in many parts of Africa, getting an accurate "timer" is hard. Usually, you need a highly trained expert (a sonographer) with a big, expensive machine to take a picture of the baby and measure it. This is like needing a master chef and a professional kitchen just to check if your cake is rising.

The Big Idea: The "Smart Camera" for Pregnancy
This paper is a plan for a study in Ghana, Kenya, and South Africa to test a new, high-tech tool. Think of it as a "Smart Camera" (an AI-powered handheld ultrasound) that can take a picture of a baby and instantly tell you how old the baby is, without needing a master chef to look at it.

The tool uses Artificial Intelligence (AI) that has been "trained" by looking at millions of pictures of babies. It's like a student who has studied every single cake recipe in the world and can now guess the baking time just by looking at the batter.

The Mission: The "Taste Test"
The researchers want to see if this new "Smart Camera" is as accurate as the "Master Chef" (the traditional ultrasound).

Here is how they are going to test it:

1. The Setup: They will invite nearly 1,000 pregnant women to visit clinics.
2. The Double Check:
   • Step A (The Gold Standard): A highly trained expert will use a traditional ultrasound to measure the baby. This is the "correct" answer, like a master chef checking the cake with a thermometer.
   • Step B (The New Tech): Immediately after, a regular nurse (who only had a short training session) will use the handheld "Smart Camera" to take a few quick, sweeping pictures of the belly. The AI will instantly shout out, "The baby is 13 weeks and 2 days old!"
3. The Comparison: The researchers will compare the two answers. Did the AI get it right? Was it off by a day? Two days? Or was it way wrong?

Why Do This?

• Speed and Scale: You can't train 10,000 expert sonographers overnight. But you can train 10,000 nurses to use a handheld device in a few days. If the AI works, it's like giving every village a "magic timer" that works instantly.
• Saving Lives: If the AI is accurate, it means doctors can start giving aspirin to women at the exact right time to prevent pre-eclampsia, saving mothers and babies from serious illness or death.
• The "Blind" Sweep: The cool part is that the nurse doesn't need to know how to measure the baby. They just follow a video guide on the screen, waving the device over the belly like a wand, and the AI does the math.

The Plan in Simple Steps:

• Who: Pregnant women in Ghana, Kenya, and South Africa.
• When: They visit twice. Once when the baby is small (first trimester) and again when the baby is bigger (second trimester).
• What: They get measured by the "Master Chef" and the "Smart Camera" at both visits.
• Goal: To prove that the "Smart Camera" is good enough to be used everywhere, replacing the need for expensive, expert-only machines in many places.

The Bottom Line
This study is like a "road test" for a new car. Before we let everyone drive it on the highway, we need to make sure it handles the bumps, the curves, and the potholes of real life just as well as the luxury cars we already have. If this AI tool passes the test, it could revolutionize prenatal care, making it possible for every pregnant woman, no matter where she lives, to know exactly how far along she is and get the care she needs.

Technical Summary

Based on the provided protocol paper, here is a detailed technical summary of the study.

1. Problem Statement

• Clinical Context: Pre-eclampsia is a leading cause of maternal mortality globally, with over 90% of deaths occurring in Low- and Middle-Income Countries (LMICs). Effective prevention relies on identifying high-risk women and administering low-dose aspirin before 20 weeks' gestation.
• The Barrier: Accurate gestational age (GA) estimation is critical for risk stratification (using tools like the Fetal Medicine Foundation algorithm) and timing interventions. However, routine access to high-quality ultrasound for GA dating is limited in sub-Saharan Africa due to a shortage of expert sonographers and expensive equipment.
• The Gap: While Artificial Intelligence (AI) and handheld ultrasound devices offer a potential solution, the specific AI algorithm (Intelligent Ultrasound ScanNav FetalCheck) has only been validated retrospectively on datasets from Australia, India, and the UK. There is no prospective evidence regarding its accuracy in real-world clinical settings within African populations or its equivalence to standard-of-care conventional ultrasound.

2. Methodology

• Study Design: A multi-country, prospective accuracy cohort study.
• Setting: 13 health facilities across Ghana, Kenya, and South Africa (ranging from primary health centers to tertiary referral hospitals).
• Population: 969 pregnant women recruited at 13 health facilities.
  • Inclusion: Pregnant women presenting for antenatal care between 11+0 and 13+6 weeks' gestation.
  • Exclusion: Known major fetal abnormalities or absent fetal heart rate.
• Intervention & Comparator:
  • Experimental (AI): A trained research midwife performs "blind" abdominal sweeps using the GE VScan Air handheld device running the Intelligent Ultrasound (IU) ScanNav FetalCheck system (v3.0). The system uses an AI algorithm to estimate GA automatically without human measurement.
  • Reference Standard (Conventional): A trained, accredited expert sonographer performs conventional ultrasound measurements according to International Society of Ultrasound in Obstetrics & Gynecology (ISUOG) standards.
    • First Trimester: Crown-Rump Length (CRL) measured three times; best measurement used with INTERGROWTH-21st charts.
    • Second Trimester: Biparietal Diameter (BPD), Head Circumference (HC), Abdominal Circumference (AC), and Femur Length (FL) measured; GA estimated via HC.
• Procedure:
  • Visit 1 (11+0 to 13+6 weeks): Participants undergo both the conventional scan (by sonographer) and the AI scan (by midwife). The operators are blinded to each other's results.
  • Visit 2 (14+0 to 27+6 weeks): Participants return for a second visit (randomly scheduled within this window) to repeat both scanning methods.
• Quality Control:
  • Sonographers must be accredited by submitting 10 first-trimester and 10 second-trimester scans meeting specific INTERGROWTH-21st quality criteria (e.g., mid-sagittal section, correct caliper placement).
  • Ongoing quality assurance involves 100% review of the first 25 scans per sonographer, followed by random 20% sampling.
• Statistical Analysis:
  • Primary analysis involves Bland-Altman plots to determine Mean Difference (bias) and 95% Limits of Agreement between AI and conventional GA.
  • Reduced Major Axis Regression (RMAR) will evaluate fixed and proportional bias.
  • Diagnostic performance (sensitivity/specificity) will be assessed for classifying pregnancies as <20 weeks.

3. Key Contributions

• First Prospective Validation in Africa: This study provides the first prospective, real-world validation of an AI-based gestational age estimation algorithm specifically within sub-Saharan African populations, addressing the lack of external validity in previous retrospective studies.
• Task-Shifting Potential: It evaluates the feasibility of using non-expert operators (midwives) with handheld devices to achieve GA estimates comparable to expert sonographers, potentially solving the workforce shortage in LMICs.
• Foundation for the PEARLS Trial: The study is a critical formative component of the larger "Preventing pre-eclampsia: Evaluating AspiRin Low-dose regimens following risk Screening" (PEARLS) trial. Validating this tool is a prerequisite for implementing large-scale pre-eclampsia screening and aspirin prevention strategies in these regions.
• Standardization: The protocol establishes rigorous accreditation and quality control standards for AI-ultrasound implementation in diverse clinical settings.

4. Results

• Status: As this is a study protocol, the paper does not contain results. It outlines the planned methodology, sample size calculations, and analysis framework.
• Projected Sample Size: The study is powered to recruit 969 women (approx. 323 per country) to achieve a precision of ±10% in the limits of agreement.
• Expected Outcomes: The study aims to determine if the AI system's Mean Absolute Error (MAE) is equivalent to conventional ultrasound (historically ~1.7–2.8 days in previous studies) and to quantify any systematic bias.

5. Significance

• Public Health Impact: If the AI system proves accurate, it could democratize access to early pregnancy dating, enabling millions of women in LMICs to receive timely pre-eclampsia screening and aspirin prophylaxis, potentially preventing thousands of maternal deaths annually.
• Scalability: The use of commercially available, portable, handheld devices reduces infrastructure costs and allows for deployment in primary care settings where specialist sonographers are unavailable.
• Global Health Equity: Successful validation would support WHO recommendations for universal antenatal ultrasound and provide a scalable model for integrating AI diagnostics into routine maternal care in resource-limited settings.
• Future Research: Positive results could pave the way for expanding AI applications to other obstetric assessments, such as fetal growth monitoring, placental health, and the detection of other pregnancy complications.