Deep Learning for Cardiac MR Image Quality Assessment - CIBM

A peer-reviewed research article on Deep Learning, reveals that an Artificial intelligence (AI) algorithm can achieve similar results to a human expert in terms of assessing image quality on 3D whole-heart cardiac magnetic resonance (MR) imaging.

The article, entitled “Deep Learning to Automate Reference-Free Image Quality Assessment of Whole Heart MR Images”, was published May 27, 2020 in Radiology: Artificial Intelligence. The abstract reports that an artificial intelligence–based algorithm can mimic expert visual image quality assessment and allows for fast and automated image quality grading of three-dimensional whole-heart MR images.

CrossSectionColorCoded — Cross-section of the heart where the colors indicate how much each part of the anatomy contribute to make up the quality grade of the dataset. It shows that the network picks up small vessels and small structures and is not just basing the grade on noise level or sharpness of the whole volume.

The scientific article was picked up by AuntMinnie.com, a comprehensive community Internet site for radiologists and related professionals in the medical imaging industry. The article, “AI can handle quality assessment of 3D cardiac MRI” was posted on 2 June, 2020.

Professor Matthias Stuber, Section Head of CIBM MRI CHUV-UNIL and senior author of the study said,

“The fact the media has picked up on the article is a testimony of our collaborative, translational and interdisciplinary work. I think it nicely exemplifies what the CIBM is about, and how such collaborative research can thrive in a hospital setting. In fact, this concerns data that we had begun acquiring some seven years ago, with a technique that we developed 10 years ago, and now we were able to harvest these data and perform some AI research.”

The findings were published through a collaboration between researchers and clinicians at the CHUV in Radiology and Cardiology, with contributions of the CIBM Signal Processing EPFL UNIGE Section (Dimitri Van De Ville), and a long standing Siemens Healthcare collaboration (Davide Piccini & Tobias Kober).

Citation: RSNA, Radiology: Artificial Intelligence, “Deep Learning to Automate Reference-Free Image Quality Assessment of Whole-Heart MR Images, Davide Piccini*, Robin Demesmaeker*, John Heerfordt, Jérôme Yerly, Lorenzo Di Sopra, Pier Giorgio Masci, Juerg Schwitter, Dimitri Van De Ville, Jonas Richiardi, Tobias Kober, Matthias Stuber.

* D.P. and R.D. contributed equally to this work.