4017

Deep-learning-based phase correction during reconstruction of high-resolution, multi-shot reduced-FOV pancreatic DWI.
Ryan L Brunsing1, Alexandra H Besser2, Arnaud Guidon3, Xinzeng Wang3, and Patricia Lan3
1Radiology, Stanford, Stanford, CA, United States, 2Stanford, Stanford, CA, United States, 3GE Healthcare, Waukesha, WI, United States

Synopsis

Keywords: Pancreas, Diffusion/other diffusion imaging techniques, Pancreas, Deep-learning, Phase Correction, DWI, rFOV, multishot, SNR

Motivation: Diffusion weighted imaging (DWI) is valuable in pancreatic imaging but suffers from artifacts and low SNR. The combination of reduced FOV imaging with a multishot data sampling strategy (rFOV-msDWI) improves artifacts from susceptibility and allows higher achievable resolution but still suffers from low SNR.

Goal(s): Here we report early findings from an ongoing pilot study

Approach: Evaluate a DL-based phase correction algorithm for improved SNR in patients undergoing rFOV-msDWI of the pancreas.

Results: DL-based phase correction subjectively improves image quality.

Impact: DL-based phase correction may improve rFOV-msDWI of the pancreas. Further evaluation is warranted.

Background: Diffusion weighted imaging (DWI) is valuable in pancreatic imaging as it can help differentiate benign from malignant lesions, detect pancreatic ductal adenocarcinoma (PDAC), and assess PDAC treatment response [1]–[4], however conventional DWI methods (single-shot echo-planar imaging for example) are prone to artifact. The combination of reduced FOV imaging with a multishot data sampling strategy (rFOV-msDWI) improves artifacts from susceptibility and allows higher achievable resolution [5] but still suffers from low SNR. Delineation of small lesions or accurate delineation of the tumor margin is challenging by any current imaging modality but is critical for early detection of cancer and determination of surgical candidacy, when it can be treated with curative intent [6]. With sufficient resolution DWI may be able to solve this problem, but further tools to address the already low SNR are needed to realize DWIs full potential in the pancreas.

Goal: To determine if DL-phase correction during image reconstruction can improve image quality in rFOV-msDWI of the pancreas.

Approach: This single-center prospective pilot study was performed with Institutional Review Board approval and an informed consent waiver. The initial three patients undergoing pancreatic DWI with rFOV-msDWI (Parameters in Table 1) were included in this preliminary assessment. Raw data was collected and processed through DL Phase Correction (DLPC), which aims to generate a high-quality phase (high resolution and SNR) for phase correction using a deep-learning based network, which was trained from a database of over 10,000 images with various SNR levels and background phases [7]. This correction step was combined with a commercially available DL-based denoising tool [8] to further improve in-plane resolution and SNR of the DL phase corrected image. Two sets of images from the same set of raw MR data were produced: original image and DL Phase Corrected (DLPC) images. High b-value datasets and ADC maps were reviewed by a radiologist with subspecialty training in abdominal MRI and 10 years of experience and a descriptive summary presented. Additional DWI sequences included a full FOV respiratory triggered msDWI in all cases (2-shots, slice thickness 5mm, in-plane resolution 2.2 x 2.2 mm, TE~55ms, TR~8500ms).

Results: In all three patients, DLPC subjectively improved perceived SNR and boundary sharpness on the high b-value images from rFOV-msDWI, both in the pancreatic tissue (Case 1: Figure 1) and on a cystic pancreatic lesion (Case 2: Figure 2). Similar results were noted on the ADC maps, which showed superior perceived resolution and improved SNR relative to the no-DL corrected images. In one patient with a 12 cm pancreatic cystic lesion (Case 3: Figure 3), high-resolution rFOV-msDWI with DLPC resolved both thin internal linear structures (solid arrow) and tiny 2-3mm nodular structures (dashed and dotted arrows) better than msDWI or rFOV DWI alone.

Conclusion: DL-based phase correction may improve rFOV-msDWI of the pancreas. Further evaluation is warranted.

Acknowledgements

No acknowledgement found.

References

[1] M. Zhu et al., “Accuracy of quantitative diffusion-weighted imaging for differentiating benign and malignant pancreatic lesions: a systematic review and meta-analysis,” Eur Radiol, vol. 31, no. 10, pp. 7746–7759, Oct. 2021, doi: 10.1007/s00330-021-07880-3.

[2] S. Ichikawa, M.-L. Kromrey, U. Motosugi, and H. Onishi, “Optimal target b-value on computed diffusion-weighted magnetic resonance imaging for visualization of pancreatic ductal adenocarcinoma and focal autoimmune pancreatitis,” Abdom Radiol, vol. 46, no. 2, pp. 636–646, Feb. 2021, doi: 10.1007/s00261-020-02695-0.

[3] H. Fn et al., “Computed high-b-value high-resolution DWI improves solid lesion detection in IPMN of the pancreas,” European radiology, May 2023, doi: 10.1007/s00330-023-09661-6.

[4] F. Donati, C. Casini, R. Cervelli, R. Morganti, and P. Boraschi, “Diffusion-weighted MRI of solid pancreatic lesions: Comparison between reduced field-of-view and large field-of-view sequences,” European Journal of Radiology, vol. 143, p. 109936, Oct. 2021, doi: 10.1016/j.ejrad.2021.109936.

[5] Y. Bai et al., “MRI: Evaluating the Application of FOCUS-MUSE Diffusion-Weighted Imaging in the Pancreas in Comparison With FOCUS, MUSE, and Single-Shot DWIs,” Journal of Magnetic Resonance Imaging, vol. n/a, no. n/a, doi: 10.1002/jmri.28382.

[6] H. Rhee and M.-S. Park, “The Role of Imaging in Current Treatment Strategies for Pancreatic Adenocarcinoma,” Korean J Radiol, vol. 22, no. 1, pp. 23–40, Jan. 2021, doi: 10.3348/kjr.2019.0862.

[7] Wang X, Litwiller D, Guidon A, Lan PS, Sprenger T., “Robust complex signal averaging for diffusion weighted imaging.,” presented at the ISMRM Annual Meeting Proceedings., 2023.

[8] R. M. Lebel, “Performance characterization of a novel deep learning-based MR image reconstruction pipeline,” arXiv:2008.06559 [cs, eess], Aug. 2020, Accessed: Nov. 04, 2021. [Online]. Available: http://arxiv.org/abs/2008.06559

[9] N. Chen, A. Guidon, H.-C. Chang, and A. W. Song, “A robust multi-shot scan strategy for high-resolution diffusion weighted MRI enabled by multiplexed sensitivity-encoding (MUSE),” Neuroimage, vol. 72, pp. 41–47, May 2013, doi: 10.1016/j.neuroimage.2013.01.038.

Figures

Figure 1: Table of MRI parameters for the reduced field-of-view multi-shot DWI used in this study (rFOV-msDWI). All scans were performed on Premier 3T MRI scanners (GE Healthcare, Waukesha, WI) using 30-channel phased-array flexible coils.

Figure 2: Case 1: 68-year-old female with dilated bile duct on ultrasound. DLPC dramatically improves the perceived SNR and sharpness of the margin of the pancreas (outlined with dashed line on reference image).

Figure 3: Case 2: 61-year-old male with suspected pancreatic cyst. DLPC dramatically improves the perceived SNR and sharpness of the margin of the 1.3 cm pancreatic body cyst (outlined with dashed line on reference image with arrow).

Figure 4: Case 3: 48-year-old female. High-resolution delayed post-contrast axial 3D T1w SPGR (1.1mm isotropic resolution) and axial T2w single-shot FSE images show a 12cm pancreatic cystic lesion concerning for malignancy. Solid arrow shows a thin linear structure within the cyst. Dashed and dotted arrows show 2-3mm nodular structures along some of the internal septations.

Proc. Intl. Soc. Mag. Reson. Med. 32 (2024)
4017
DOI: https://doi.org/10.58530/2024/4017