How Agxio built an AI to detect skin cancer

AI and machine learning are powerful tools with applications across every facet of our lives. And one place where they are making big waves is in the medical services. But why is AI so good at medical research and how did Agxio build an AI to detect skin cancer? We’ll explain.

Why is AI good at medical research?

AI is very good at recognising patterns and it’s much faster than people. Proper AI tools (not just fancy automation) can pinpoint signs and symptoms more accurately than humans also.

It’s said that a GP is right 52% of the time and a specialist is right 80% of the time. Well, specially-trained AI and ML agents who are focused on solving just one problem are like hundreds of human specialists – all working on one thing. AI isn’t just mimicking human thinking; it’s multiplying it to create breakthrough solutions to global challenges. AI can sift through medical data faster and more consistently than any human and that’s why it’s so good at spotting early warning signs of disease, creating treatments and speeding up diagnoses. But there are still challenges to overcome and one of those is image noise.

However, our recent work creating an AI model to detect skin cancers is a success story that shows that progress is possible.

Building our AI model to detect 20 types of skin cancer

In our recent news post, we announced that “Agxio recently worked with a private dermatology clinic and technology provider to a series of hospital trusts to develop a ground-breaking AI prediction engine for the top 20 skin cancer conditions which account for c95+% of all cases.” What this story doesn’t tell you is how we got there. Our team worked on this Explainable Artificial Intelligence or XAI project for 6 months and it failed over and over. The image noise of freckles and other benign skin marks in these medical images was a real problem to overcome. 

The tree model

To tackle the challenge, we had to create a whole new model for Apollo – a tree of models that worked to take the data in an image and translate it into numbers so it’s easier for the AI to understand and categorise. Then, we delivered that information transparently – explaining to the human specialists why the AI arrived at a specific diagnosis. For example, using heatmaps to highlight cancerous regions in histopathological slides for the skin cancer specialists to review instead of just marking a specimen ‘cancerous’.

Histopathological slides are a crucial tool for dermatopathologists and dermatologists in diagnosing and managing skin cancers, but the sheer volume of samples waiting for categorisation often means long wait times and delayed diagnosis. An AI like ours can prioritise and segment inbound patient photos, slides and samples to help medical specialists manage their workloads more efficiently.

Other applications for this approach to dealing with image noise are in:

• Autonomous cars
• Astronomy
• Military
• Crop scouting
• Livestock management
• Novel drug discovery

By keeping the specialist involved and working to augment human expertise; not replace it, Agxio built an AI to detect skin cancer that’s just as effective as a clinician and can help cut down on waiting time for diagnosis. If you’d like to investigate how our tree model or other innovative AI systems can support your operation as well, please get in touch.