2019/02/28 by Carlo Biffi, Juan J. Cerrolaza, Biffi, Carlo +11 · 2 citations
Medicine · #Radiomics and Machine Learning in Medical Imaging #Cardiac Valve Diseases and Treatments #Advanced MRI Techniques and Applications
paper · pdf · doi:10.48550/arxiv.1902.11000
Accurate segmentation of heart structures imaged by cardiac MR is key for the\nquantitative analysis of pathology. High-resolution 3D MR sequences enable\nwhole-heart structural imaging but are time-consuming, expensive to acquire and\nthey often require long breath holds that are not suitable for patients.\nConsequently, multiplanar breath-hold 2D cine sequences are standard practice\nbut are disadvantaged by lack of whole-heart coverage and low through-plane\nresolution. To address this, we propose a conditional variational autoencoder\narchitecture able to learn a generative model of 3D high-resolution left\nventricular (LV) segmentations which is conditioned on three 2D LV\nsegmentations of one short-axis and two long-axis images. By only employing\nthese three 2D segmentations, our model can efficiently reconstruct the 3D\nhigh-resolution LV segmentation of a subject. When evaluated on 400 unseen\nhealthy volunteers, our model yielded an average Dice score of 87.92 \± 0.15\nand outperformed competing architectures.\n