AAIC 2026 | The development of population-specific brain age prediction models in African populations

Udunna Anazodo

Good afternoon. I’m Dr Udunna Anazodo. I’m at AAIC 2026 with Tolulope Olusuyi. We are both working at Medical Artificial Intelligence Laboratory, Lagos, Nigeria. And I’m based at the Montreal Neurological Institute in McGill in Canada. And today we’re going to be talking about a project that we both collaborated on. It’s a brain age model that Tolu worked really hard to develop for looking at brain aging using T1-weighted MRI scans in African populations. And so we’re just going to discuss that a little bit today as a way to highlight some of the promising upcoming research coming out of Africa, which is, I think, I would say, led quite a bit from our lab in Lagos, Nigeria. So I’m going to start off with some questions, and Tolu, we’re just going to have a nice conversation about your project. Okay, so can you first of all just give us a quick overview of your project and what motivated it? Let’s just kind of start with that.

Tolulope Olusuyi

Okay, just like you’ve mentioned, brain age in African population. There are existing brain age models out there at the moment, but they’re mostly trained on data sets that are not from our population. Data set from Europe, from North America, basically from a different population that are not African, so the motivation was to have a brain age model that is specific to our population because we believe those other models when maybe when they are being tested or evaluated on our data sets and the performance will not be what it should be, so yeah, there is a need for a population-specific brain age model for Africans.

Udunna Anazodo

So let’s just take a step back. Can you describe what brain age is? And why do we have to create this model? What are we looking at when we’re trying to use these models to look at maybe as a biomarker, as some people have suggested? So what is brain age and what are brain age models?

Tolulope Olusuyi

Okay, so there is the chronological age of an individual, like someone is 30 years old, and there is how old the brain might present to be, so the brain of the 30-year-old could appear older or lesser than 30. So being able to predict or tell how old the brain is could tell us if that person is at risk of neurological conditions, so the need for a brain age model is to be able to predict the age of the brain, so we’re able to say oh, this guy is at risk of certain neurological disorder. What was the other question again? Okay. I think, yeah, that’s the brain age model.

Udunna Anazodo

So this is similar to what people have done. So there are biological age models that have been developed for looking at, say, other parts of the body. So you’re sort of describing what you’ve done, sort of trying to look at is somebody’s brain at risk of neurological disorder, even if they appear chronologically or their birth ages. Okay, so in terms of brain age model, you mentioned that there are existing models, yeah, but you’ve looked at those existing models and your understanding is that these models may not predict brain age well in African populations, why do you believe that that’s the case?

Tolulope Olusuyi

Okay, so, so basically, usually machine learning models or deep learning models, they kind of, it’s what you feed in as in training data that you sort of expect as results. So when these models are not representative of certain data sets from a certain population or a group of data from somewhere, then they are not expected to predict well on that data set, and this is also something that we’ve seen, we tested a few models and we saw that the performance on our data set was lower, very, very lower than what was reported in the original research work.

Udunna Anazodo

So what is the risk, so if we, if you didn’t do this research, so if we decided or if other researchers decided to use these established models, these established models are trained with very large data sets, one of the models was trained with almost 20,000 brain MRIs, and our model was trained with about 200 brain MRIs, so what is the risk of saying, OK, we don’t have 20,000 brain scans, can we just take the model and apply it to our population? What is the risk of doing that?

Tolulope Olusuyi

Okay, just like I mentioned, we’ll check out some of these models. So when we make use of these models that are not representative of our data set, and the model predicts that, oh, this 30-year old person is 50, it could be wrong because it could either overestimate the person’s brain age or underestimate it, so we are at risk of misdiagnosis or let’s say someone who is at risk, oh, you’re fine, you’re good, you can go, so yeah, they were risking misdiagnosis or misclassification, I guess, yeah. Also the whole essence of the brain age prediction is early detection of this neurological disorder, so if we’re not able to early detect, then we are risking missing a diagnosis, yeah.

Udunna Anazodo

So I’ve seen a few brain age posters, a few brain age talks at the conference, have you seen some?

Tolulope Olusuyi

Yeah, yeah, I saw one yesterday from a team in Uganda, and I think there was one that is not particularly brain age, but from some guys in the UK here.

Udunna Anazodo

Yeah, and the team in Uganda, what was their project on?

Tolulope Olusuyi

Also, brain age, but not particularly predicting or building a model, but looking at it from, I’m not so sure at this very moment, but not brain age model. Okay, I took some pictures to check out later.

Udunna Anazodo

Okay, so it looks like there are not that many brain age models that are coming out of places like Africa or low-income countries. Do you know one of the reasons why there’s this big disparity in our understanding of brain aging using models like brain age modeling or even using, because some brain age modeling are done not only on MRI or T1-weighted images, some of them are done using other bio-based, biological-based markers? So what is the disparity from your few years of doing this research, or few months, I should say, of doing this research? What is your understanding of the disparity and what is driving that disparity?

Tolulope Olusuyi

I think one thing I would say would be the lack of data set, and I think one thing that could have led to lack of the data set or the size of the data that is available would be how much it cost to actually capture those images or the fact that people don’t usually show up at the hospitals or diagnosis center until they start presenting symptoms, so to get research-grade quality data sets, to actually do some of this predictive modeling, it’s quite expensive, and yeah, usually people don’t just show up at the diagnostic center until they are presented.

Udunna Anazodo

So the two key challenges you mentioned is, one, the availability of the imaging data, so if the scanners are not there, not there, if the patients are not coming in, or if these individuals are not coming in as early on when they are not symptomatic, then we don’t have good data sets to now build predictive models that could predict before a person shows up with the symptoms. Okay, and then so what, what is your understanding of what we can do to start to close that gap, so we can create really good imaging data sets that are coming from Africa, from Lagos, for example, or from other centers? You mentioned the work you saw in Uganda.

Tolulope Olusuyi

Yeah, so I think one thing would be that we need more data, we need more representative data sets, and one way to achieve that would be to get more data, one way to get more data would be, maybe grants for people to, I don’t know, it costs a lot.

Udunna Anazodo

Yeah, so grants to just look at healthy people, right, yeah, because that’s those type of grants are far fewer and in between in Africa, and oftentimes when there are research grants, so first of all, health systems, the lab, the Medical Artificial Intelligence lab in Lagos, we’re fortunate enough to have the lab in the radiology facility in a private radiology clinic, Crest Radiology, and the Crest Radiology has five radiology sites, and we happen to be in one of the sites where the MRI scanner, 1.5 T Siemens M scanner, is, but that’s not how a lot of spaces and places are set up in Africa, and you mentioned funding, and one issue with funding is when funding comes, first of all, there’s not a lot of funding in many African countries that are reserved for health research, and when it is reserved for health research, they tend to think that the funding should go towards a typical health research needs, which is infectious disease, maternal health, they don’t see Alzheimer’s disease as an African problem, they still think it’s a problem for the West to solve, and so when there are limited health research funding, it doesn’t typically go to imaging. Imaging, like you said, is very expensive, and when there are now external funding coming from, from most of the West, those fundings are typically also earmarked for looking at infectious disease, a typical what they think are African health need problems. And when we do have the small little money that trickling in to look at dementia or a non-communicable disease, they’re typically earmarked for patients, and so there’s not very few funding to look at healthy individuals, specific data set, all of these brain age models are built on normative, on normal appearing brain, because the whole point is to predict brain aging in a healthy population with a need of eventually applying that to a population that have a condition like Alzheimer’s disease. So yes, you’re right, without funding being one key, then the data sets are not there. Another, I think, important thing to consider is people like you, you know, having people like you trained to do this type of research, having the environment where you have the computer, the computing resources, you have this environment, we’re very fortunate to have computers donated to us by the Global Health Labs, for example, and some of these computers, we’re putting it to good use, being able to build and test these models, so if we don’t have the computational environment, if we don’t have the neuroscientists like yourself that are well-trained to do some of these brain-age models, we’re going to continue seeing these disparities, and like you said, we do understand that we cannot copy-paste models, even if the models are big, well-built models, we have to start to build our own, which is what you’ve done and done very well.

I think one other thing that you are working on that we may want to inform the audience is your brain, Africa Brain Atlas, so do you want to describe what the Africa Brain Atlas is, quickly, and how it’s tied to the brain age project, and why we need to create this brain imaging template that people tend to use to do image analysis?

Tolulope Olusuyi

All right, so I think when patients or individuals’ images are first being acquired, there should be a reference that we’ll take a look at, compare that image to, does this look like what it should look like typically? So the brain atlas or the African brain atlas is going to be the first reference template for what an African brain should look like. So at the instance, we are able to take a look at the individual’s brain, sort of register it or compare it to the brain atlas, and you’re able to tell, oh, there’s something not quite right with this, and then with them, we can then make use of the brain age model and try to predict what the brain age is, so we could sort of do some correction, some predictions also, or elimination, okay, after we’ve been able to recognize that this brain doesn’t appear, ah, it should appear, and then we’re able to then predict the brain age, they were able to maybe quite bring down our suspicions, then we’re able to say we suspect that this person might be presenting with this particular neurological disorder. So yeah, the brain atlas that we’re working on, it’s going to be a reference for what the African brain is supposed to appear, could appear.

Udunna Anazodo

So yeah, so the African brain template it’s a, almost like a space where we can register everybody’s, so your head, my head, everybody’s head here, not the same shape and size, typically when we do brain image analysis, we try to register everybody’s head into common space, so then we can be able to see this type of campus and this person’s head, and is the same, the location is the same location, XYZ coordinate, so it’s almost like a reference map or a geolocation map, if you want to say that, and that allows us to compare everybody’s brain one-to-one or patient-to-controls, because we have this common atlas where we sort of registered and put everybody’s brain into. Now, the issue with doing that is that a lot of these atlases are created with North American or European data sets, and we’ve learned from our Chinese colleagues, our Indian colleagues, that, you know, we have to think really carefully how we sort of force everybody’s head to fit within this space, because there are obviously maybe other reasons why our population brains are very specific, and when we start to force them into this template, we might remove some of this population-specific differences, and so we’ve seen that from the Chinese brain template, the Indian brain template, the Korean brain template, you’re right, they need to create an African brain template, and maybe even more population-specific brain templates, and so we, you’ve started leading this work to create an African brain template using normative data sets as well, from Nigeria, as well, and with the intention of increasing the size of that data with normative data coming from other African countries.

So I think just to wrap up here, what do you see going forward as the future of neuroscience and neuroimaging, very specifically within the context of dementia. We’ve talked about creating this predictive, probably early biomarker using brain age model, and using something that’s very, let’s say more easier to do, as anatomical scans, where do you see, from your early career, as you’re starting off, where do you see the future of neuroimaging as a very useful tool in the toolkit of dementia diagnosis, dementia mechanism, looking at mechanism, where do you see the future of that going, it’s coming from somebody like you that’s studying your career in, in Nigeria.

Tolulope Olusuyi

Okay, just like we’ve been saying from the start of this, I think that, so with more predictive models like we’re trying to work on, we’re able to predict early, yeah, so the early detection or early prediction of Alzheimer’s diseases or dementia, we’re able to sort of get people to get treated on time, start receiving treatment, and I think that could bring about decrease in maybe death from these conditions, or decrease, and most of the times when people start presenting with symptoms, they don’t even know that is what that’s the condition that they have, until typically people just assume dementia or Alzheimer’s, you have to not be yourself, forget things all the time, and even when maybe simpler symptoms are showing up, they are not aware that’s the condition that they have, but with a predictive model like this that can be easily deployed in low-resource settings like ours, then we are able to make early detection.

Udunna Anazodo

Okay, so you think the future of dementia imaging in Africa should be more towards low-cost solutions that are easily deployed, deployed on a community basis level, like you mentioned, the patients may come in sometimes not aware of the symptoms, mean, and if there’s not any tools to start to figure out what that symptom is presenting, or even if it’s just a bit, yes or no, at a very community level, so the future should be more focused on biomarkers that are more deployable and community setting, do you think these biomarkers can also include stuff like EEG, wearable devices, where we can also apply predictive models like what you’re proposing?

Tolulope Olusuyi

Yes, I think so, so I think beyond MRI scans, then we could also look into, I mean, I know what is currently going on on that as well, we can look into imaging that are easier collected or maybe cheaper, in the way, like the EEG, very good devices, yeah.

Udunna Anazodo

Yeah, because you’re building an EEG model, yes, you’re building an EEG model, right? You’re building an EEG Africa-based model as well, so these are more deployable, so the future is looking at some of these more deployable, very predictive models that can help at the community level. Yeah, fantastic work, I think we can thank our host, VJNeurology, for this opportunity to share work in Africa, Tolu, how can they find you?

Tolulope Olusuyi

Okay, on LinkedIn, the name is Tolulope Olusuyi, and they could check out our lab as well, MAI Lab, that’s Medical Artificial Intelligence in Lagos, Nigeria.

Udunna Anazodo

Thank you very much for this opportunity, thank you very much, see you in the next one.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.