Abstract #4784

Deep Learning for Automated Medical Image Quality Assessment: Proof of Concept in Whole-Heart Magnetic Resonance Imaging

Robin Demesmaeker^1,2, John Heerfordt^1,3, Tobias Kober^1,3,4, Jérôme Yerly^3,5, Pier Giorgio Masci⁶, Dimitri Van De Ville^2,7, Matthias Stuber^3,5, and Davide Piccini^1,3,4

¹Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, ²Institute of Bioengineering/Center for Neuroprosthetics, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, ³Department of Radiology, University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, ⁴LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, ⁵Center for Biomedical Imaging (CIBM), Lausanne, Switzerland, ⁶Division of Cardiology and Cardiac MR Center, University Hospital of Lausanne (CHUV), Lausanne, Switzerland, ⁷Department of Radiology and Medical Informatics, University Hospital of Geneva (HUG), Geneva, Switzerland

We aim at developing a fully automated algorithm which quantitatively gauges the quality of medical images using deep learning to mimic human perception. An automated image quality assessment algorithm based on a deep convolutional neural regression network is designed, optimized, trained, validated and tested on a clinical database of 3D whole-heart cardiac MRI scans. The algorithm was successfully trained and validated, yielding a regression performance in the range of the intra- and inter-observer agreement. These results show the relevance of deep learning concepts to image quality analysis, in particular to volumetric cardiac MR imaging.

This abstract and the presentation materials are available to members only; a login is required.

Join Here