Conjugate gradient and deep learning reconstructions: reduced time without affecting image quality and nodule detection.
Authors
Affiliations (8)
Affiliations (8)
- Department of Diagnostic Radiology, Fujita Health University School of Medicine, Toyoake, Japan. [email protected].
- Joint Research Laboratory of Advanced Medical Imaging and Artificial Intelligence, Fujita Health University School of Medicine, Toyoake, Japan. [email protected].
- Department of Diagnostic Radiology, Fujita Health University School of Medicine, Toyoake, Japan.
- Joint Research Laboratory of Advanced Medical Imaging and Artificial Intelligence, Fujita Health University School of Medicine, Toyoake, Japan.
- Canon Medical Systems Corporation, Otawara, Japan.
- Department of Radiology, Fujita Health University Hospital, Toyoake, Japan.
- Department of Radiology, Fujita Health University Bantane Hospital, Nagoya, Japan.
- Department of Radiology, Fujita Health University Okazaki Medical Center, Okazaki, Japan.
Abstract
The purpose of this study was to determine the utility of conjugate gradient reconstruction (CG Recon) and deep learning reconstruction (DLR) for reducing scan time while maintaining the image quality and nodule detection capability on lung MRI with ultrashort TE (UTE-MRI) as compared with grid reconstruction (Grid Recon). In the in vitro and in vivo studies, the NEMA phantom and 35 patients with pulmonary nodules were scanned by UTE-MRI with original (TE<sub>original</sub>), 1/2 (UTE<sub>1/2</sub>), and 1/4 (UTE<sub>1/4</sub>) spoke numbers obtained by both methods and reconstructed with and without DLR. In this study, the standard protocol was UTE<sub>original</sub> obtained by Grid Recon without DLR. Then, signal-to-noise ratios (SNR) of the phantom, lung and lesion were assessed. In the in vivo study, overall image quality and nodule detection capability were visually assessed on each UTE-MRI. Quantitative and qualitative indices were then compared between the standard protocol and others. Finally, a receiver operating characteristic (ROC) analysis was performed to compare the standard and other protocols. In in vitro and in vivo studies, all SNRs were significantly different between the standard protocol and each UTE-MRI with CG Recon and DLR (p < 0.05). Overall image quality of the standard protocol differed significantly from that of all UTE<sub>1/4</sub>s (p < 0.05). The area under the curve of each UTE<sub>Original</sub> obtained by CG Recon was significantly larger than that of the standard protocol (p < 0.05). CG Recon and DLR can reduce scan time while maintaining image quality and nodule detection capabilities on lung UTE-MRI. Question To determine the utility of conjugate gradient reconstruction (CG Recon) to reduce scan time without nodule detection capability on MRI with ultrashort TE (UTE-MRI). Findings Nodule detection capability was not significantly decreased by CG Recon with or without deep learning reconstruction when reducing scan time from the standard UTE-MRI protocol. Clinical relevance Conjugate gradient reconstruction (CG Recon) and deep learning reconstruction (DLR) have the potential to reduce scan time while maintaining image quality and nodule detection capability in lung MR imaging with ultrashort TE.