Translating Imaging-based Brain Age Models for Predicting Accelerated Brain Ageing in African Populations: A Pilot Study on a Nigerian Cohort.

The human brain is not a static organ; it changes shape and structure as we grow older. Just as our skin develops wrinkles or our bones lose density over time, the brain undergoes a natural, gradual transformation. Scientists have learned to measure these changes using magnetic resonance imaging, or MRI, to create a concept called "brain age." This is a number that estimates how old a person's brain looks based on its physical structure, rather than how many years they have actually lived. If a brain appears older than the person's actual age, it can signal that the organ is under stress or at risk for diseases like dementia. For years, researchers have used this idea to try to spot health problems early. However, the tools used to calculate this brain age have a significant flaw: they were almost entirely built using data from people in North America, Europe, and Australia. These models assume that all human brains age in the same way, regardless of where a person lives or what their background is. This assumption ignores the fact that genetics, environment, and even the specific machines used to take the pictures can influence how a brain looks.

A team of researchers set out to test whether these Western-made tools work for people in Africa, specifically in Nigeria. They drew from the Africa Neuroimaging Archive (AfNiA), a multi-institutional database containing over 11,000 clinical brain scans from across the continent. From this vast archive, they carefully selected 309 scans from Nigerian adults who were neurologically healthy, meaning they had no known brain diseases or abnormalities. These specific scans came from a variety of sources, including public and private hospitals in Nigeria, and were taken using five different types of MRI machines. Many of these images were not the crisp, high-resolution 3D pictures typically used in research; instead, they were often faster, lower-quality 2D slices taken to move patients through the clinic quickly. To make these images usable for their study, the team used a computer program to convert the 2D slices into 3D images, effectively filling in the gaps to create a complete picture of the brain. They then ran these images through a rigorous quality check, removing any scans that were too blurry or distorted to be useful, ensuring that their final dataset was as reliable as possible given the real-world constraints.

With their clean dataset ready, the team trained a new computer model to learn the relationship between brain structure and age specifically for this Nigerian population. They taught the model to look at the size and shape of different brain regions and predict how old the person was. When they tested this new model, it made predictions that were off by an average of about 9.8 years before they applied a standard statistical adjustment. After making a simple correction to account for a common tendency in these models to overestimate the age of younger people and underestimate the age of older people, the error dropped significantly to just 6.4 years. This level of accuracy is comparable to the best models trained on large, high-quality datasets from the Global North, but it was achieved using data that was far more representative of the local population.

To see how well the old, imported models would perform, the researchers tested four different established brain age tools on the same Nigerian data. These tools had been trained on tens of thousands of scans from people in the UK, the US, and Europe. The results were stark. The imported models made much larger errors, with some being off by more than 16 years on average. One of the most advanced deep learning models, which had been praised for its ability to generalize to new groups, failed completely on a small portion of the scans, unable to process the images at all. Another model performed so poorly that it actually predicted the wrong direction of aging for the group. The researchers found that these models were essentially trying to force the Nigerian brains to fit a pattern they had never seen, leading to confusion and inaccurate results. The study showed that even the most sophisticated tools cannot simply be transferred across continents without losing their reliability.

The study also highlighted the practical challenges of bringing these technologies to resource-limited settings. While the new model they built was accurate, the process of preparing the images took a tremendous amount of time. Converting and analyzing the scans for all 309 participants required nearly 40 days of continuous computer processing. This bottleneck, caused by the need to break down the brain images into specific parts before the model could read them, is a major hurdle for using these tools in busy hospitals where speed is essential. The researchers noted that while newer, faster methods exist, they often require expensive, powerful computers that are not available in many parts of Africa. This creates a difficult trade-off: the most accurate models are too slow and expensive to use widely, while the faster models are often less accurate or fail entirely on local data.

Ultimately, this work demonstrates that the path to better brain health in Africa requires a different approach. It is not enough to take the tools developed in wealthy nations and apply them to African populations. The researchers showed that by using local data to build local models, they could achieve results that were both more accurate and more relevant to the people being studied. They concluded that the future of global brain health depends on creating a diverse ecosystem of tools, each validated for the specific populations they are designed to help. As they continue to gather more data from across the continent, they hope to refine these models further, eventually creating a system that can reliably identify individuals at risk of accelerated brain aging, allowing for earlier intervention and better care for millions of people who have long been left out of the conversation.