Leiden team wins challenge for faster MRI scan through artificial intelligence
Researchers from Leiden, in cooperation with Philips, have won a challenge in which international research groups dedicate themselves to accelerating MRI scans with the help of artificial intelligence (AI). They developed an algorithm with which it is possible to use eight times less data than normal and still reconstruct an MRI image of a knee that is almost as good as one using the usual amount of data.
In the fastMRI challenge, organised by the Facebook AI research lab and New York University, artificial intelligence specialists were challenged to apply their knowledge to making MRI scans faster and more efficient. The 34 teams taking part were supplied with a raw data set of a few hundred MRI scans of knees. They also received a number of incomplete data sets. Their task was to develop a self-learning algorithm with which the complete MRI image could still be ‘reverse engineered’ from the incomplete data sets. Including less data during the MRI scan means the scan time is shorter.
Close cooperation is the key to success
The Philips & LUMC team was judged the best in the most difficult part of the competition: reconstructing a nearly intact image from just one-eighth of the data. The group of researchers worked full-time for months on the self-learning algorithm. AI expert Marius Staring and MRI specialist Thijs van Osch (both from the Radiology department) are the team leaders from LUMC and they are proud of the achievement. Staring: ‘The key to success was the close cooperation between AI experts, MRI technicians and clinicians from both LUMC and Philips. There was very intensive cooperation with Philips, which enabled us to test all sorts of ideas very quickly. I am proud of the LUMC researchers Sahar Yousefi and Mohamed Elmahdy who, together with colleagues at Philips, have delivered a top performance.’
Benefits for patient and practitioner
According to the researchers, artificial intelligence shows great promise for the world of imaging. ‘An MRI scan can currently easily take from a quarter of an hour to half an hour. If we can reduce that to a few minutes, it will benefit both the patient and the practitioner,’ explains Van Osch. ‘We look forward to continuing the collaboration with Philips, in order to do further research, but also to pass on the benefits quickly to the patient.’
Text: Berit Sinterniklaas