Keywords: Deep learning, CT ventilation, functional lung imaging, image synthesis
TL;DR: We show that a synergy between conventional CT-ventilation modelling and deep learning can improve the performance of functional lung image synthesis.
Abstract: Hyperpolarized gas MRI can visualize and quantify regional lung ventilation with exquisite detail but requires highly specialized equipment and exogenous contrast. Alternative, non-contrast techniques, including CT-based models of ventilation have shown moderate spatial correlations with hyperpolarized gas MRI. Here, we propose a hybrid framework that integrates CT-ventilation modelling and deep learning approaches. The hybrid model/DL framework generated synthetic ventilation images which accurately replicated gross ventilation defects in hyperpolarized gas MRI scans, significantly outperforming other model- and DL-only approaches. Our results show that a synergy between conventional CT-ventilation modelling and DL can improve the performance of functional lung image synthesis.
Paper Type: both
Primary Subject Area: Image Synthesis
Secondary Subject Area: Application: Radiology
Paper Status: original work, not submitted yet
Source Code Url: If the paper is accepted, a github link to the source code will be provided in the final version of the paper.
Data Set Url: The dataset used is part of the larger AAPM-endorsed CTVIE19 (http://aapmchallenges.cloudapp.net/competitions/35) grand challenge dataset. We plan to release the dataset subsequently once the grand challenge paper is published later this year. The data is tied to formal transfer agreements that dictate that it can only be publicly released after this paper's publication.
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.