Drawing and Predicting Lines: The Role of Artificial Intelligence in Assisting Medical Professionals

Traditionally, radiologists analyse MRI scans, ultrasounds or X-rays themselves to see what is wrong with a patient. And before a treatment, the radiologist or doctor also draws various lines on a scan. If, for example, a tumour is to be treated with radiation therapy, the beam needs to be aimed precisely to ensure the surrounding organs receive as little radiation as possible. Lines are therefore drawn around the tumour and the organs on the scans.

This kind of manual labour takes a lot of time. And the outcome can be a bit different each day because people vary in their work. Automated methods have long been used to support radiologists and physicians, but AI has been changing the field dramatically for a few years now. AI is fast and always does the same. And it can now help draw those lines. By training algorithms, Staring, who is Professor of Machine Learning for Medical Imaging, is trying to get AI to work as well as radiologists, or even better.

Manageable workload

Healthcare needs this automation, says Staring, to keep the workload manageable. ‘Staff are absent because of the pressure, and that’s not healthy.’ AI can help reduce that pressure.

How does AI already help with image analysis? In his inaugural lecture, Staring gives three examples: AI can determine the size of certain tumours very precisely; it can discover breast cancer slightly more often than people can (with fewer false positives); and it can map changes over time in scans. The latter would take a person weeks for a single patient.

‘We want to use AI to predict how the tumour will look six months from now.’

Doubting algorithm

Staring and his team are working on a tool that can predict whether a tumour behind the ear (acoustic neuroma) will grow in the future. If the tumour is stable, it is not treated. It sits against the brain, which is a risky area to operate on, and radiation therapy can cause side effects too. ‘But if you know the tumour is going to grow, you will intervene’, says Staring. ‘We want to use AI to predict how the tumour will look six months from now.’

As the future is difficult to predict, the algorithm also indicates how reliable the prediction is. ‘An algorithm can also have its doubts’, says Staring. ‘By gaining an idea of the extent of that doubt, a doctor knows what he can and cannot rely on. That can help him make a decision.’

Hard work and frequent failure

It is not easy to get such a tool working, but that, says Staring, is the beauty of his job. ‘You have to work hard and you often fail. But you try for a while and the computer finally does what you want. That’s what I really like. And then it’s also something people need. It’s great that by developing technology I can have an impact on society.’