Meeting Banner
Abstract #3560

Comparison of 3D convolutional neural networks and loss functions for ventilated lung segmentation using multi-nuclear hyperpolarized gas MRI

Joshua R Astley1,2, Alberto M Biancardi1, Paul J Hughes1, Laurie J Smith1, Helen Marshall1, Guilhem J Collier1, James Eaden1, Nicholas D Weatherley1, Jim M Wild1, and Bilal A Tahir1,2
1POLARIS, University of Sheffield, Sheffield, United Kingdom, 2Oncology and Metabolism, University of Sheffield, Sheffield, United Kingdom

Deep learning has shown great promise for numerous medical image segmentation tasks, including delineation of ventilated lung volumes from hyperpolarized gas MRI. We previously demonstrated the utility of a VNet convolutional neural network (CNN), trained on a combination of 3He and 129Xe scans, in producing accurate segmentations that outperform conventional methods. In this work, we compared the performance of several 3D CNNs and loss functions for segmentation of ventilated lungs on a significantly larger and more diverse multi-nuclear hyperpolarised gas MRI dataset using several training strategies. We observe that the UNet CNN provides the best performing model for our dataset.

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

Join Here