Artificial intelligence interprets heart scans to assess mortality risks

1 min read

Artificial intelligence could one day help doctors to predict which of their patients are at greatest risk of dying of a heart condition, allowing them to be treated more effectively.

Researchers at the MRC London Institute of Medical Sciences (LMS) have for the first time used machine learning to interpret heart scans, to predict how long patients will live.

The research, published in the journal Radiology, found that the AI software could predict survival at one year with up to 80 per cent accuracy, according to Declan O’Regan, who led the project.

“We studied patients with pulmonary hypertension, which is a devastating disease that causes heart failure, where pressure builds up in the blood vessels of the lungs and then feeds back into the heart,” said O’Regan. “It can cause about one third of people to die within five years of diagnosis,” he said.

The key to treating the disease is to identify which patients are at greatest risk of developing heart failure, so that they can be targeted with the most intensive treatment, he said. However, predictions made today are often inaccurate.

The machine learning software automatically analyses moving images of a patient’s heart captured during an MRI scan. It then uses advanced image processing to build a virtual 3D heart, which replicates the way over 30,000 points in the heart contract during each beat.

The researchers fed the system historic data from over 256 patients with pulmonary hypertension. By linking these data with its models, it learned which attributes of a heart, its shape and structure, put an individual at a given risk of heart failure.

“It can find the really earliest signs of heart failure that are difficult for humans to spot,” said O’Regan.

The researchers now plan to test the software on data from a different hospital to that from Hammersmith Hospital, where it was developed, in order to verify its accuracy.

They also plan to use more advanced AI, according to O’Regan.

“We want to use techniques like deep learning, to try to improve on the techniques that we have, and to build a more detailed picture of the whole heart,” he said.

The researchers ultimately hope to develop software that could also be used to make predictions about which type of treatment will work best in each patient.