Artificial intelligence can help predict illnesses before they happen – but our healthcare systems need to catch up.

It was a problem that had baffled biologists for half a century. But AI solved it in days. Last year, London-based AI lab DeepMind was able to predict how a protein folds into a unique three-dimensional shape.  

It was a stunning breakthrough.

Not just because of the speed of the achievement, but also its significance. Many diseases are linked to the way proteins function. And determining their 3D structure can help with the development of new drugs and treatments.

In recent years, I’ve been amazed by the speed of AI progress in healthcare. For example, it’s been used to analyse retinal images to predict Alzheimer’s disease, and spot breast cancer as accurately (or better) than a radiologist.

These advances could replace invasive and time-consuming tests, help predict disease risk, and potentially prevent people from getting ill in the first place. But if we want to bring all of this promise to mainstream medicine, the systems underpinning modern healthcare need a major overhaul.

A vision of an integrated healthcare system

Many healthcare providers around the world operate on siloed systems. Patient data is highly fragmented, making it difficult to access and share. So when we visit a doctor’s surgery, their ability to use data as a diagnostic tool is relatively limited.

This same infrastructure also limits AI-based diagnostics. Not just because data is siloed, but because of its relative paucity. To be truly effective, any AI tool we use needs access to a large and varied data set – ideally population-sized.

What could this look like? We don’t need to imagine.

Israel’s hugely successful vaccine programme has been built on a digitised healthcare system that’s been in place for decades. Patient records sit on a centralised database, which has allowed Israel to build up a vast pool of data that can help, in part, to develop new treatments.

This offers a blueprint for how AI can be utilised more effectively.

With millions of health records catalogued and tagged, an algorithm can be trained to spot patterns within the data that correlate with certain diseases. A patient’s medical history, blood tests, medications, activity levels, ethnicity (and in the future, maybe their genome) would all help to create a risk profile.

During a routine check-up, a doctor could then use this profile to aid decision-making. Perhaps deciding to fast-track someone for a precautionary screening. Or suggesting lifestyle changes based on their propensity for a certain disease.

AI-based healthcare: beyond the hospital

I’m especially excited by the prospect of connecting data and AI tools across an entire healthcare system.

During a routine eye exam, an optician might notice an abnormality in your blood vessels, then use an AI model to check if it warrants further investigation. This data could then be shared with your GP and hospital, accompanied with an automated appointment request.

To achieve this vision, legacy systems need to be migrated to the cloud, with a centralised system that tags and processes a growing library of diagnostic data. Data could then be easily accessed, shared, and swiftly acted upon.

Ongoing modernisation would be essential to connect this system with new AI applications and patient-facing apps. This would enable automated appointments and allow doctors to easily contact us. Imagine if this system also connected with the wearables on our wrists, allowing it to gain even deeper real-time health insights.  

For me, the ultimate goal is to democratise this form of healthcare – so diagnostics can be run even in the poorest and most remote villages. In theory, all you’d need is a testing tool connected to the internet that can transmit images or test data, and receive results back for rapid diagnosis.

I appreciate that sounds ambitious, but the technology is already in place to make this happen.

Concerns over privacy

Ultimately, the kind of healthcare system I’ve described requires huge amounts of trust. People must be willing to share large amounts of their medical details, and allow (to some degree) an algorithm to decide how they’re treated.

Faster treatment, disease prevention, and better patient outcomes are undoubtedly positive things. But are they enough to overcome potential privacy concerns? We also need to factor in the regulatory aspects of data, which would be especially tricky for cross-border diagnosis.

I think AI has huge potential within healthcare. But what about you? Leave a comment to let me know your thoughts on this topic, including any additional barriers or unexplored opportunities.