Model-based deep learning reconstruction using folded image training strategy (FITS) for abdominal 3D T1-weighted images

概要

Purpose: To evaluate the feasibility of folded image training strategy (FITS) and the clinical image quality of images reconstructed using the proposed improved model-based deep learning network trained with FITS (FITS-iMoDL) on abdominal magnetic resonance imaging.

Methods: This retrospective study included 122 clinical cases of 3D T1-weighted imaging on upper abdominal examinations. The proposed network was trained using a training set (n=40) and tested with the other cases (test set, n=77). In the experimental analyses, peak signal-to-noise ratio (PSNR) and structure similarity index (SSIM) of images reconstructed with FITS-iMoDL were compared with those with the following reconstruction methods; conventional model-based deep learning (conv-MoDL), MoDL trained with FITS (FITS-MoDL), total variation regularized compressed sensing (CS), parallel imaging (CG-SENSE). In the clinical analysis, signal-to-noise ratio (SNR) and image contrast were measured on the reference, FITS-iMoDL, and CS images. Three independent radiologists evaluated the image quality of FITS-iMoDL and CS images using a 5-point scale in terms of liver edge, depiction of hepatic vessels, depiction of pancreas, lesion conspicuity, noise, aliasing and motion artifact, blurring, and overall quality. The mean opinion score (MOS) was calculated for each case.

Results: The PSNR of FITS-iMoDL was significantly higher than that of FITS-MoDL, conv-MoDL, CS, and CG-SENSE (p<0.001). The SSIM of FITS-iMoDL was significantly higher than those of the others (p<0.001), except for FITS-MoDL with an acceleration factor of 2 (p=0.056). In the clinical analysis, the SNR of FITS-iMoDL was significantly higher than that of the reference and CS (p<0.0001). Image contrast was equivalent within an equivalence margin of 10% among these three image sets (p<0.0001). MOS was significantly improved in FITS-iMoDL (p<0.001), except for hepatic vessels (p=0.42).

Conclusion: Proposed method, FITS-iMoDL, was useful for the reconstruction of 3D T1-weighted imaging. FITS-iMoDL would yield higher quality of images in clinical abdominal MRI.