Generating high-resolution synthetic CT from lung MRI with ultrashort echo-times: initial evaluation in cystic fibrosis

Arthur Longuefosse (1) , Julien Raoult , Ilyes Benlala (2) , Baudouin Denis de Senneville (3, 4) , Thomas Benkert , Julie Macey (2) , Stéphanie Bui (2) , Patrick Berger (2) , Gilbert Ferretti , Jean-Yves Gaubert , Renan Liberge , Antoine Hutt , Baptiste Morel , François Laurent (2) , Fabien Baldacci (1) , Gaël Dournes (2)

1 LaBRI - Laboratoire Bordelais de Recherche en Informatique
2 CRCTB - Centre de recherche Cardio-Thoracique de Bordeaux [Bordeaux]
3 IMB - Institut de Mathématiques de Bordeaux
4 MONC - Modélisation Mathématique pour l'Oncologie

Background: Lung MRI with ultrashort echo times (UTE-MRI) enables high-resolution and radiation-free morphological imaging, however, its imaging quality is still lower than CT. Purpose: To assess the image quality and clinical applicability of synthetic CTs generated from lung UTE-MRI using Generative Adversarial Networks (GAN). Methods and Materials: This retrospective study included patients with Cystic Fibrosis (CF) who underwent both UTE-MRI and CT on the same day at six institutions between January 2018 and December 2022. The 2D GAN algorithm was trained using paired MR-to-CT slices and tested, along with an external dataset. Imaging quality was assessed quantitatively by measuring apparent contrast-to-noise-ratio, apparent signal-to-noise-ratio, and overall noise and qualitatively using visual scores for features including artifacts. Two readers evaluated CF-related structural abnormalities and used them to determine clinical Bhalla scores. Results: The training, test, and external datasets were comprised of 82 CF patients (mean age, 21 ± 11 [SD]; 42 males), 28 patients (mean age, 18 ± 11; 16 males), and 46 patients (mean age, 20 ± 11; 24 males), respectively. In the test dataset, the contrast-to-noise ratio of synthetic CTs (median, 303 [interquartile range: 221, 382]) was higher than that of UTE-MRI (9.3 [6.6, 35]; P < .001). The median signal-to-noise ratio was similar between the synthetic and corresponding CT (88 [84, 92] vs 88 [86, 91]; P = 0.96). Synthetic CT had a lower noise level than CT (median score, 22 [18, 30] vs 42 [32, 50]; P < .001) and the lowest level of artifacts (median score, 0 [0, 0]; P ≤ .001). The concordance between Bhalla scores for synthetic and real CTs was almost perfect (ICC, ≥ 0.92). Conclusion: Synthetic CTs showed almost perfect concordance with CTs for the depiction of Cystic Fibrosis-related pulmonary alterations and had better image quality than lung MRI with ultrashort echo-times.

Mots clés: CF; cystic fibrosis UTE-MRI; ultrashort echo time MRI PFT; pulmonary function test GAN; generalized adversarial networks ICC; intraclass correlation coefficient

Domaines: Sciences du Vivant [q-bio]; Informatique [cs]

Arthur Longuefosse, Julien Raoult, Ilyes Benlala, Baudouin Denis de Senneville, Thomas Benkert, et al.. Generating high-resolution synthetic CT from lung MRI with ultrashort echo-times: initial evaluation in cystic fibrosis. Radiology, 2023, 308 (1), ⟨10.1148/radiol.230052⟩. ⟨hal-04268631⟩