Home >> News >> Machine learning could offer faster, more precise results for cardiac MRI scans

Machine learning could offer faster, more precise results for cardiac MRI scans

Analysis of heart structure and function using MRI can be performed significantly faster with similar precision to experts across a wide range of diseases when using automated machine learning, new research shows.

Analysing heart function on cardiac MRI scans takes approximately 13 minutes for humans. Utilizing machine learning, a scan can be analysed with comparable precision in approximately 4 seconds, according to a paper co-authored by UCL academics and published in Circulation: Cardiovascular Imaging, an American Heart Association journal.

Healthcare professionals regularly use cardiac MRI to make measurements of heart structure and function. These measurements guide clinical decisions including timing of cardiac surgery and implantation of defibrillators.

Improving the performance of these measures is therefore likely to improve patient management and potentially outcomes.

In the UK, it is estimated that close to 150,000 cardiac MRI scans are required each year. Based on this number of scans and the amount of time saved from analysing scans, researchers believe that utilizing artificial intelligence (AI) to read scans could potentially lead to 54 clinician-days per year being saved at each UK health centre.

Researchers trained a neural network to read the cardiac MRI scans and results of close to 600 patients. When tested for precision on 110 separate patients from multiple centres, researchers found that there was no significant difference between the fully automated AI and an expert, or trainee.

Dr Charlotte Manisty, Senior Lecturer at UCL and a Consultant Cardiologist at the Barts Heart Centre and UCLH, said: “Cardiovascular MRI offers unparalleled image quality for assessing cardiac structure and function, however current manual analysis remains crude. Automated machine learning techniques offer the potential to change this and radically improve efficiency, although demonstrating superiority over human observers has yet to be shown.

“Our dataset of patients with a range of cardiac diseases scanned twice enabled us to demonstrate that the greatest sources of measurement error arise from human factors, and that automated techniques are at least as good as humans, with the potential soon to be ‘super-human’ – transforming clinical and research measurement precision.”