Summary
Google researchers are exploring how artificial intelligence (AI) can help pregnant women access important ultrasound information in areas where medical facilities and trained specialists are limited.
In research conducted with Northwestern Medicine and Jacaranda Health, the team tested an approach that combines simple ultrasound scans with machine learning. The method uses videos captured by healthcare workers and AI models to estimate how far along a pregnancy is and identify the baby’s position.
The research involved 2,000 mothers in total, with 1,000 participants in Nairobi, Kenya, and another 1,000 in Chicago. According to the researchers, the AI system was able to estimate gestational age and fetal presentation with accuracy comparable to that of a trained sonographer in the study.
Why Is Ultrasound Important During Pregnancy?
Ultrasound scans help healthcare professionals monitor a baby’s development and gather important information throughout pregnancy.
Doctors use these scans to estimate gestational age, check the baby’s position, assess growth, and investigate potential complications. Knowing the stage of pregnancy is particularly important when doctors need to make decisions about treatment or prepare for delivery.
For example, a baby born at 34 weeks may need different medical support from one born at 37 weeks. An accurate estimate of gestational age can help healthcare teams prepare for the baby’s needs.
However, access to ultrasound services remains limited in many parts of the world. The equipment can be expensive, and many healthcare facilities do not have enough trained sonographers to perform scans.
How Can AI Make Ultrasounds More Accessible?
Google’s research focuses on making ultrasound examinations easier to perform without requiring every healthcare worker to undergo years of specialist training.
Traditional ultrasound examinations often require a trained sonographer to move the probe carefully to capture specific images and measurements. This process takes considerable practice and experience.
The research team explored an alternative approach called a blind-sweep ultrasound. Instead of requiring operators to capture particular views manually, they move the ultrasound probe over the pregnant woman’s abdomen following a predefined pattern.
According to the researchers, healthcare workers with no previous ultrasound experience can learn this technique in approximately eight hours.
The process produces ultrasound videos that an AI model can analyze to extract useful information about the pregnancy.
How Does the AI Ultrasound System Work?
The system combines a portable ultrasound device, a simple scanning procedure, and a machine learning model.
Here is how the process works:
- Perform the scan: A trained healthcare worker moves the ultrasound probe across the abdomen using the predefined blind-sweep pattern.
- Capture ultrasound videos: The device records the images needed for the AI system to analyze.
- Check scan quality: The AI model can indicate when a sweep needs to be repeated because the captured information is insufficient.
- Estimate pregnancy stage: The model analyzes the videos to estimate gestational age.
- Identify the baby’s position: The system also assesses fetal presentation, helping determine how the baby is positioned in the womb.
The approach is designed to make useful ultrasound information available in settings where specialist imaging services are difficult to access.
Why Does On-Device AI Matter?
One important feature of the research is that the AI processing can take place directly on the device.
This means the system does not need an internet connection or Wi-Fi to generate its results. That capability could be particularly valuable in remote clinics where reliable connectivity and infrastructure are not always available.
Portable ultrasound devices can also be easier to transport than traditional machines. Some are battery-powered, which can reduce dependence on a continuous electricity supply.
Combining portable hardware with on-device AI could therefore help healthcare workers deliver certain ultrasound services in communities that have limited medical resources.
However, the availability of an offline AI system does not remove every infrastructure challenge. Healthcare facilities still need functioning equipment, appropriate training, clinical oversight, and a process for referring patients who require additional examinations.
What Did the Research Find?
The study involved 2,000 mothers across two locations: Nairobi, Kenya, and Chicago, United States. Google worked with Jacaranda Health and Northwestern Medicine on the research.
The researchers reported that their machine learning model could estimate gestational age and assess fetal presentation with accuracy comparable to a trained sonographer.
The findings suggest that AI-assisted ultrasound may help healthcare workers obtain useful pregnancy information without requiring the same level of scanning experience needed for conventional examinations.
Testing the approach across two different patient populations and healthcare settings also helped researchers investigate its potential beyond a single clinical environment.
However, these findings should not be interpreted as evidence that AI can replace sonographers or perform every type of prenatal ultrasound. The research focused on specific assessments, particularly gestational age and fetal presentation.
How Could This Help Pregnant Women in Remote Areas?
In communities with limited access to specialist healthcare, an AI-assisted ultrasound system could give frontline health workers another way to gather important information during pregnancy.
Potential benefits include:
- Improved access: More clinics may be able to offer basic ultrasound assessments.
- Less specialist training required for scanning: Healthcare workers could learn the blind-sweep technique more quickly than traditional ultrasound scanning.
- Support for clinical decisions: Estimates of gestational age and fetal presentation could help healthcare teams plan further care.
- Use in low-connectivity areas: On-device processing allows the system to operate without an internet connection.
- Portable examinations: Smaller ultrasound devices may be easier to use in community clinics and remote locations.
These benefits could be especially relevant in areas where pregnant women must travel long distances to reach facilities with ultrasound equipment and trained specialists.
AI-assisted scanning would still need to be integrated into appropriate medical workflows, with qualified professionals interpreting results and arranging further care when necessary.
Could AI Be Used for Other Medical Applications?
The researchers believe the approach could have applications beyond the two pregnancy assessments explored in this study.
Ultrasound is used in different medical situations, including evaluating injuries and examining internal organs. AI could potentially help healthcare workers capture usable images and identify information that supports clinical decisions.
However, each application would require its own research, testing, and clinical validation. A model designed to estimate gestational age cannot automatically be assumed to work reliably for unrelated medical tasks.
The wider opportunity is to combine accessible imaging devices with AI tools that help healthcare professionals gather and interpret information in settings where specialist resources are limited.
What Comes Next?
The research demonstrates the potential of AI-assisted ultrasound, but further work is important before the approach can be adopted more widely.
Researchers and healthcare providers need to establish how reliably the system performs across different populations, devices, and clinical environments. They must also determine how the results should be used in patient care and when a conventional ultrasound or specialist assessment is necessary.
Cost, equipment availability, clinical oversight, and integration into existing healthcare systems will also influence whether the technology can reach the communities that need it most.
The goal is not simply to make ultrasound equipment smaller. It is to help healthcare workers use that equipment effectively and provide pregnant women with useful medical information.
Conclusion
Google’s research shows how AI could help make certain ultrasound assessments more accessible to pregnant women, particularly in areas with limited access to trained sonographers.
By combining blind-sweep scanning with machine learning, the team explored a way for healthcare workers to collect ultrasound videos and obtain estimates of gestational age and fetal presentation without requiring an internet connection.
The findings represent a promising research direction, although AI-assisted ultrasound is not a replacement for comprehensive prenatal care or qualified medical professionals. Wider clinical testing and validation will be important to determine how the technology can safely support maternal healthcare worldwide.