A virtual 3D heart, developed by University of Auckland bioengineers, could provide a major boost to the treatment of atrial fibrillation – the most common heart rhythm problem.
“This is a very important step toward much improved clinical diagnosis, patient stratification and clinical guidance during ablation treatment for patients with atrial fibrillation,” senior author and research team leader Dr. Jichao Zhao, with the Auckland Bioengineering Institute at the university, noted in a statement.
The key is the team's novel machine learning algorithm, called AtriaNet.
“Existing approaches for analyzing atrial structures in 3D, especially from late gadolinium-enhanced (LGE)-MRIs, rely heavily on manual segmentation methods, which are extremely labor-intensive and prone to errors,” the team of international researchers noted in their paper, appearing in imaging journal IEEE Transactions on Medical Imaging.
This makes “a robust and automated method” for analyzing atrial structures in 3D – like AtriaNet – of high interest.
AtriaNet is a convolutional neural network. It pulls its data from 154 3D LGE-MRIs of patients with atrial fibrillation to automatically segment the left atrial (LA) epicardium and endocardium.
This atrial segmentation is a crucial task in the medical management of AF patients, noted the researchers.
“This advanced machine learning algorithm is way faster and more accurate than any other approaches,” Zhao noted, adding that, “it will help doctors to pinpoint precise locations of diseased tissue in the upper chambers (atria) of the heart by creating a 3D virtual heart within a minute.”
The software can run on a regular personal computer and create a 3D virtual heart for a new patient, with an accuracy of 94 percent.
“Current clinical treatment is unsatisfactory, mainly due to a lack of understanding of the human atrial tissue which directly sustains AF,” added Zhao, observing that atrial fibrillation can lead to serious consequences including tripling the risk of heart failure and bumping up stroke risk by a factor of five.
