Within healthcare, it has been known for several years that lung cancer can be detected at an earlier stage using a CT scan. What many people do not yet know is that, in many cases, radiologists can also detect early signs of COPD and cardiovascular disease using the same scans. Researchers in the B3CARE project are investigating early risk indicators of these diseases on radiological images. In particular, they aim to use artificial intelligence to speed up the assessment of these images.
In the project, researchers bring together images of CT and MRI scans from large-scale studies such as ImaLife (Imaging in Lifelines). With this data, they then train computers to recognise early signs of Big-3 diseases. This is done using biomarkers. A biomarker is a measurable property that allows you to identify (the risk of) a disease and predict its course. For example, a biomarker for lung cancer is an unevenness in the lung: a lung nodule. This unevenness may be visible on a low-dose CT scan. An early sign of risk of cardiovascular disease is the presence and severity of calcification of coronary arteries.
CT scan of the heart. In green, calcifications are indicated by artificial intelligence software.
'Early detection and prevention of Big-3 diseases is crucial,' says Rozemarijn Vliegenthart, project leader of B3CARE. Vliegenthart is a radiologist at UMCG and cohort data expert at UMCG's Data Science Center in Health (DASH). 'Together, these diseases are expected to cause the most deaths by 2050. Unfortunately, most people are only diagnosed when the disease is already at an advanced stage and cure is no longer possible. This is particularly the case with lung cancer; at the time of diagnosis, the disease has already metastasised in most patients and only 15% survive 5 years. If the Big-3 diseases are detected at an earlier stage, timely treatment can be provided and worse disease symptoms can be prevented in many patients'.
Photo: Rozemarijn Vliegenthart
In the B3CARE project, researchers are preparing a combined screening for all three diseases. Analyses by the project team show that adding COPD and cardiovascular disease to lung cancer screening greatly improves cost-effectiveness. This means that more quality life-years are gained, at lower costs. But screening will also produce many thousands of scans every year. Assessing these is a huge job for radiologists. ‘To support them in this, we are applying artificial intelligence-based software,' says Vliegenthart. ‘A large group of Lifelines participants had a scan of heart and lungs. On these, we used manual software to determine the presence and size of lung nodules. Artificial intelligence-based software will probably be able to do this automatically in the future, without a radiologist having to find and measure it himself. We are now testing new, automatic software and comparing it with manual measurements. In this, we are working closely with academic and industrial partners who have developed automatic software'.
But are Big-3 diseases actually always detectable with such Big-3 screening? Not according to Vliegenthart: 'If Big-3 screening is introduced, it will always be a trade-off between advantages and disadvantages. A CT scan involves a low dose of potentially harmful X-rays and, moreover, early signs of Big-3 diseases are rare at a younger age. It is therefore to be expected that mainly middle-aged or slightly older people will be considered for screening, likely in combination with risk factors like smoking. Big-3 diseases that occur at younger ages are then missed. Repeated screening for lung cancer is also very important, every year or every other year. Only then, the earlier stages of lung cancer can still be detected in time.’
Software based on artificial intelligence is expected to help measure relevant biomarkers in the future. Vliegenthart: 'The computer can then use that information to suggest a test result, after which the radiologist makes the final evaluation. Based on this, the screening participant will then receive personalised advice, for example on lifestyle, preventive medication, referral and repeat screening.'
Similar software is also under development for screening for other diseases, including breast cancer. Given the large number of radiological examinations to be reviewed, artificial intelligence can support filtering out normal examinations or, put the suspicious ones at the front of the queue.
Want more information about the B3CARE project? Visit the website B3care.nl or contact Rozemarijn Vliegenthart.
