Anatomically-guided deep learning for left ventricle geometry generation with uncertainty quantification based on short-axis MR images

Recent advances in deep learning have greatly improved the ability to generate analysis models from medical images. In particular, great attention is focused on quickly generating models of the left ventricle from cardiac magnetic resonance imaging (cMRI) to improve the diagnosis and prognosis of millions of patients. However, even state-of-the art frameworks present challenges, such as discontinuities of the cardiac tissue and excessive jaggedness along the myocardial walls. These geometrical features are often anatomically incorrect and may lead to unrealistic results once the geometrical models are employed in computational analyses. In this work, we propose an anatomically-guided deep learning model to overcome these limitations while preserving the advantages of state-of-the-art frameworks, such as computational efficiency, robustness, and generalization capabilities. Our novel anatomically-guided neural networks are formed by a UNet followed by a B-spline head, which acts as a regularization layer during training. The B-spline head aggregates the prediction into a single connected region, removes any undesired tissue islands, and produces a smooth continuous contour. In addition, the introduction of the B-spline head contributes to achieve a robust uncertainty quantification of the left ventricle inner and outer walls. Our results show that the proposed model generates anatomically consistent geometries while achieving an agreement with the ground truth images comparable to state-of-the-art frameworks and simultaneously improving the geometry uncertainty quantification in comparison to classic UNet models. The examples presented here, as well as source codes, are all open-source under the GitHub repository https://github.com/CBL-UCF/unet_ag.