In a busy public hospital in Lagos, a doctor reviews a scan that could determine the next step in a patient’s care.
The image is clear. The machine worked. But the decision cannot be made yet because no radiologist is available to interpret the results.
So the patient waits. Sometimes for hours. Sometimes for days.
In cases like stroke, internal bleeding, or suspected cancer, that delay is not just inconvenient. It changes outcomes. It turns manageable conditions into critical ones. It quietly decides who gets treated early and who arrives too late.
This is not an isolated breakdown. It is a structural weakness built into Nigeria’s healthcare system.
Nigeria has fewer than 300 practising radiologists serving a population of over 200 million people, according to the Radiological Society of Nigeria. But the issue is not just the number. The specialists who are available are often overwhelmed, working through long queues of scans under pressure. When that workload becomes too much, delays are inevitable.
And when diagnosis slows down, everything downstream suffers.
Take stroke care as an example. The first few hours after a stroke are critical. Imaging is needed quickly to determine whether the patient is experiencing a blockage or bleeding in the brain. A delay in interpreting that scan can mean the difference between recovery and permanent disability.
The same pattern appears in cancer diagnosis. A tumour that could have been detected early is instead discovered at a later stage, when treatment is more invasive and less effective. In both cases, the issue is not the absence of technology. It is the gap between having the scan and acting on it.
This is where artificial intelligence becomes necessary.
AI-powered imaging systems can analyse scans in seconds, flag abnormalities, and bring urgent cases to the front. They do not replace radiologists. They support them and reduce pressure on a system that is already stretched thin.
Unlike human specialists, AI systems do not get tired or slow down after long hours. They can process large volumes of scans consistently, without the mental fatigue that comes with heavy workloads. That consistency matters, especially where every delay has consequences.
The impact of early detection is well established. The World Health Organisation has shown that early diagnosis improves survival rates for diseases like cancer and tuberculosis. The challenge in Nigeria has never been knowing this. It has been acting on it fast enough.
In a recent study on lumbar spine MRI analysis, I developed a convolutional neural network to detect serious disc bulges from axial MRI images. The dataset comprised scans from over 500 patients with lower back pain, covering key spinal segments such as L3-L4, L4-L5, and L5-S1.
These were real clinical images, with variations in anatomy, image quality, and disease presentation. The model was trained to distinguish between serious and non-serious disc bulges, achieving an accuracy, precision, recall, and F1-score of 89 per cent.
But accuracy alone is not enough.
One of the biggest barriers to adopting AI in healthcare is trust. Many systems operate as black boxes, producing results without clear reasoning. To address this, the study incorporated Local Interpretable Model-Agnostic Explanations (LIME), allowing the model to highlight the specific regions of the MRI that influenced its decision.
This makes AI usable in practice. A clinician can see not just the prediction, but why it was made.
More importantly, this approach is not limited to a single dataset or country.
The same model can be retrained using Nigerian imaging data, adapting to local patient demographics and disease patterns. With the right collaboration between hospitals and developers, systems like this can be deployed without having to start from scratch.
That is what makes this practical.
If systems like this can be built and adapted, then the question is no longer whether the technology works. The real question is why it is not yet part of everyday clinical practice.
The answer is not complexity. It is about priority.
The federal government should fund pilot AI imaging programs in selected public hospitals within the next 12 months, focusing on high-volume centres where delays are most critical.
Teaching hospitals should work with local developers and researchers to build and validate systems using Nigerian data.
At the same time, regulatory bodies must establish clear standards for safety, data privacy, and accountability. Adoption without oversight creates risks. But hesitation creates greater ones.
None of this requires perfect conditions. It requires direction.
Because at this point, delayed diagnosis is no longer just a system inefficiency.
It is a decision.
A decision to accept slower care.
A decision to accept avoidable complications.
A decision to accept outcomes that could have been different.
And that is a decision Nigeria can no longer afford to keep making.
