New publication in NMR in Biomedicine

Symbolic picture for the article. The link opens the image in a large view.
5. February 2025
We are excited to share that our latest research, “Deep Learning‐Based Accelerated MR Cholangiopancreatography Without Fully‐Sampled Data”, has been published in NMR in Biomedicine! (http://DOI: 10.1002/nbm.70002)
This study demonstrates how deep learning(DL)-based reconstructions can significantly accelerate MR cholangiopancreatography (MRCP) without full-sampled ground truth. We explored both supervised and self-supervised deep learning approaches. DL reconstructions could reduce acquisition times while preserving high image quality.
This research was a collaborative project between Friedrich-Alexander-Universität Erlangen-Nürnberg and Siemens Healthineers AG.
Figure 1. Results of retrospective undersampling at (a) 3T and (b) 0.55T: We use GRAPPA with two-fold acceleration as the ground truth for comparison with CG-SENSE, CS, SSV, and SV reconstructions with six-fold acceleration. Each column corresponds to a reconstruction method and shows three different presentation forms: Maximum Intensity Projection (MIP) (top), a cropped view of the hepatobiliary duct of the MIP (middle), and a representative single slice (bottom). In (a), the blue circles indicate aliasing artifacts, and the orange circles display the representative sharpness of the common bile duct. In (b), the turquoise circles show details of the hepatobiliary ducts. PSNR in dB and SSIM in % of the six-fold reconstructions with respect to GRAPPA R=2 are shown in the top-right corner of the MIPs. The red and green arrows in (a) indicate the signal intensity source for the line profile representation for Figure 2.
Figure 1. Line profiles from the retrospective study. The line profiles are taken from the same position of the corresponding reconstructions and one representative position is presented as the red arrow (a) and the green arrow (b) in Figure 1(a). Pearson product-moment correlation coefficients (PPMCC) against GRAPPA R=2 are shown in the legend for the corresponding reconstructions to quantify the similarity to the reference profile. SV presents the highest PPMCC among others, meaning that the line profile of SV is the most similar to GRAPPA R=2. The blue arrows in (a) indicate aliasing artifacts of CG-SENSE and CG. The pink circle in (b) demonstrates that DL reconstructions preserve the signal intensity of the gallbladder.

Category: Allgemein