A woman dies every two minutes due to complications in pregnancy or childbirth. The majority of all maternal deaths are in sub-Saharan Africa. And in Kenya, the problem is actually getting worse. 70% of the population lives in remote rural areas where they're cut off from life-saving infrastructure.

Not all mums are able to access the right care. Transport, getting to the hospital is always an issue. We don't have enough healthcare workers for the population. Big problem as I would say.

Not all maternal health clinics across Kenya are as well equipped or as well staffed as this one in Nairobi. But it's hoped the technology being trialed here could help ensure that more mothers to be get access to potentially life-saving treatment.

This is Jennifer's first glimpse of her baby. The legs.

Ultrasound scans like these play a key role in ensuring a pregnancy go smoothly. Unfortunately, there's been some bumps in the road for Jennifer.

Last month, I had a complication. I had a discharge.

She was advised to go straight to hospital to get checked out.

I was scared. I was scared. I was like, Maybe I'm losing the baby or something.

Luckily, Jennifer and her baby are both fine, but she also had problems in her previous pregnancy.

I had to go for Cicerian section.

And did your doctors tell you that it could have been prevented if you'd had an ultrasound earlier?

Yeah, they said it could have been prevented.

Where you live, is it difficult to get an ultrasound during your pregnancy?

It's quite expensive, and you have to go for a distance to get it.

Jacaranda Health, the organization that runs this clinic has partnered with tech giant Google to trial a solution that could help save lives.

It's important for pregnant women to do ultrasound throughout their pregnancy. The first importance is to check check fetal anomalies. Initially, as a nurse, we were not able to perform the ultrasound.

Until now, this ultrasound probe sends video output to a tablet or smartphone, making this system cheaper and much more portable than traditional ultrasound scanners. Artificial intelligence then interprets the image without the need for a sonographer.

The information that I'll get here will help me to save the life of a mother and also save the life of the fetus.

I've come to Google's offices in Nairobi to find out more about how their AI models are being developed and how people who aren't sonographers can be trained to use it.

If this technology is intended to be so easy that you could train anybody to do it, do you think that I could have a go?

I think you definitely could.

First, I need to measure the size of the bump and apply my ultrasound gel.

So as you smooth it, you'll start to see the imagery.

Oh, wow. Yeah.

Next, it's time to do the scan. I need to do six sweeps over the abdomen, and the app gives me instructions for every step of the process.

You're doing amazing.

Thank you. It feels like if I were more confident, this would be much quicker.

The AI models then analyze the image. They two vital pieces of information. The gestational age, that's how many weeks old the fetus is, and the fetal position or the direction the fetus is facing inside the uterus.

So it says gestational age is 23 weeks and 2 days. The fetal presentation is... I actually can't pronounce that word. Cephalic. Cephalic. What does that mean?

That just means heads down. Non-cephalic positioning could be breach or transverse, which means that the patient might need a C-section or something like that. We're not trying to replace sonographers. Humans are very important to providing care, and we just want to give them additional tools in their toolset.

Obviously, we've done this on a model that is created for training purposes. Are there challenges to doing this on real human beings who might vary from each other?

The models were trained on thousands of patients from different backgrounds, and so the model is seeing a lot of different variations.

But Google needs to collect an even greater diversity of patient data to ensure the system can work for as many people as possible.

Ai is only ever going to be as good as the data that it's trained on. It's always very difficult to get hold of very good data. This is something that could take a couple of years to do. It may well show that actually this is fine to use an in X country, but actually it doesn't work in Y country. So that's where the testing comes in. And if you were to deploy it wider, it's to make sure that you're not creating that bias. You've got to make sure that this is 100% foolproof. There is no margin for error.

Around 100 scans have been collected in clinical settings in Kenya so far. A lot more than that will be needed before the system is ready for routine use. Although the research is still in its early stages, Jakeranda Health hopes the trial marks the beginning of better days for maternal health in Kenya.

10% of mums in rural Kenya have access to ultrasound services. If this is available, it will definitely be a game changer..