A Lightweight ConvNet for 4D Multi-Structure Segmentation of Cardiac Cine-MRI

K.G. van Leeuwen, J.W. Benjamins, I. Everts, T. Hendriks, W. Nijhof, F. van der Heijden, P. van der Harst

Abstract: Cardiac magnetic resonance imaging (MRI) is the gold standard for heart function assessment. We present a method for automatic segmentation of the left ventricular cavity (LVC), right ventricular cavity (RVC), and left ventricle myocardium (LVM) with a simplified, lightweight variation of the U-net trained in the cloud allowing for high throughput analysis. Dice indices on the test set were 0.95 for the LVC and 0.85 for both the LVM as RVC. A dashboard was created for visualization of the segmentation inferred cardiac function parameters. We demonstrate that this simple model trained on only a subset of the data yields satisfying segmentation results that can clinically aid cardiologists in diagnosis and treatment of heart failure patients as well as facilitate big data research on cardiac function.

Keywords: segmentation, U-net, cardiac MRI, cine-MRI, convolutional neural network, medical image analysis, cloud services

Author Affiliation: Siemens Healthineers, University of Twente, University Medical Center Groningen, GoDataDriven