LAVA HyperSense and deep-learning reconstruction for near-isotropic (3D) enhanced magnetic resonance enterography in patients with Crohn's disease: utility in noise reduction and image quality improvement

Abstract

Purpose: This study aimed to compare near-isotropic contrast-enhanced T1-weighted (CE-T1W) magnetic resonance enterography (MRE) images reconstructed with vendor-supplied deep-learning reconstruction (DLR) with those reconstructed conventionally in terms of image quality.

Methods: A total of 35 patients who underwent MRE for Crohn's disease between August 2021 and February 2022 were included in this retrospective study. The enteric phase CE-T1W MRE images of each patient were reconstructed with conventional reconstruction and no image filter (original), with conventional reconstruction and image filter (filtered), and with a prototype version of AIR^TM Recon DL 3D (DLR), which were then reformatted into the axial plane to generate six image sets per patient. Two radiologists independently assessed the images for overall image quality, contrast, sharpness, presence of motion artifacts, blurring, and synthetic appearance for qualitative analysis, and the signal-to-noise ratio (SNR) was measured for quantitative analysis.

Results: The mean scores of the DLR image set with respect to overall image quality, contrast, sharpness, motion artifacts, and blurring in the coronal and axial images were significantly superior to those of both the filtered and original images (P < 0.001). However, the DLR images showed a significantly more synthetic appearance than the other two images (P < 0.05). There was no statistically significant difference in all scores between the original and filtered images (P > 0.05). In the quantitative analysis, the SNR was significantly increased in the order of original, filtered, and DLR images (P < 0.001).

Conclusion: Using DLR for near-isotropic CE-T1W MRE improved the image quality and increased the SNR.

Keywords: Crohn’s disease; MR enterography; deep learning; image quality; noise reduction.

Figure 1

Dynamic enteric phase contrast-enhanced T1-weighted images of a 30-year-old man with active Crohn’s disease. Three sets of coronal images were obtained, each with (a) conventional reconstruction and no image filter (original); (b) with conventional reconstruction and image filter (filtered); and (c) with deep-learning reconstruction (DLR) at the noise-reduction level of 75% (DLR). The DLR image (c) shows increased sharpness of bowel walls and mesenteric vessels, enabling better visualization of active inflammation of bowel segments and adjacent comb signs (arrows). A reduction of noise with a slight synthetic appearance is also noted in the DLR image (c) compared with the other two images (a, b).

Figure 2

Dynamic enteric phase contrast-enhanced T1-weighted images of an 18-year-old man with active Crohn’s disease. Three sets of images were obtained and reformatted in the axial plane with a 1.4-mm slice thickness: (a) with conventional reconstruction and no image filter (original); (b) with conventional reconstruction and image filter (filtered); and (c) with deep-learning reconstruction (DLR) at the noise-reduction level of 75% (DLR). The DLR image (c) better visualizes active inflammation in the ascending colon (arrows) with lower noise, a sharper margin of the bowel wall and vascularity, and better contrast of bowel wall stratification compared with the other two images (a, b). In the qualitative analysis, the DLR image was given a higher score (4 or 5) by two readers regarding overall image quality, contrast, sharpness, motion artifacts, and blurring than the other two images. The consecutive images of the same patient are also presented in Supplementary Video 1, available in the online supplement.

Figure 3

Dynamic enteric phase contrast-enhanced T1-weighted images of a 28-year-old woman with active Crohn’s disease. Three sets of images were obtained, each with (a) conventional reconstruction and no image filter (original); (b) with conventional reconstruction and image filter (filtered); and (c) with deep-learning reconstruction (DLR) at the noise-reduction level of 75% (DLR). Axial images with a 1.4-mm slice thickness were reformatted from coronal images, respectively (d-f). The DLR images in both the coronal and axial planes (c,f) show lower noise, a sharper margin of the bowel wall and vascularity, and better contrast of bowel wall stratification than the other two images (a, b, d, e). In the coronal image sets, a small sinus tract is detected in the proximal ileum (arrows), which is better visualized in the DLR image (c). The corresponding penetrating lesion in the proximal ileum is also well visualized in the sets of axial images (arrows) and is most clearly visible in DLR (f), which helps increase diagnostic confidence.

Supplementary Figure 1

The phantom experiments to evaluate the effects of different levels of noise reduction on deep-learning reconstruction (DLR). Phantom magnetic resonance images were processed using the prototype DLR with tunable noise-reduction factors of 0%, 25%, 50%, and 75%. As the noise-reduction levels increased, the signal-to-noise ratio calculated by placing a region of interest in each phantom image improved by 61.9, 66.6, 77.5, and 91.8.

Supplementary Figure 2

The coronal images of dynamic enteric phase contrast-enhanced T1-weighted magnetic resonance enterography processed using deep-learning reconstruction (DLR) with different noise-reduction factors of (a) 0%, (b) 25%, (c) 50%, and (d) 75%. During preliminary reading sessions, the optimal denoising level was determined based on the consensus of the expert readers. The denoising level of 75% was chosen as it was found to yield the highest signal-to-noise ratio and the best image sharpness among the evaluated noise-reduction factors. Despite some synthetic appearance, it was concluded that the overall synthetic appearance produced by DLR was acceptable.

Supplementary Figure 3

Example of three representative locations selected to measure the signal-to-noise ratio (SNR) in each coronal (a-c) and axial (d-f) plane. For the coronal plane, images showing both external iliac and femoral arteries at the level where multiple small bowel loops are visible (a), aortic bifurcation (b), and kidneys at the level where the ascending and descending colon are visible (c) were chosen to calculate the SNR in the anterior, middle, and posterior portions of the abdominal cavity, respectively. For the axial plane, images at the level of (d) the superior mesenteric artery, (e) aortic bifurcation, and (f) the pelvic cavity showing superior gluteal veins were chosen to calculate the SNR in the upper, middle, and lower portions of the abdominal cavity, respectively.

Supplementary Figure 4

Flow chart of noise estimation in a coronal image at the level of the kidneys. For the noise estimation, the modified (HH) sub-band that contains coefficients corresponding only to noise was obtained by removing the HH sub-band coefficients corresponding to edges using discrete wavelet transform and edge map.

Supplementary Figure 5

Dynamic enteric phase contrast-enhanced T1-weighted images of a 41-year-old man with active Crohn’s disease. Three sets of coronal images were obtained with a 1.4-mm slice thickness: (a) with conventional reconstruction and no image filter (original); (b) with conventional reconstruction and image filter (filtered); and (c) with deep-learning reconstruction (DLR) at the noise-reduction level of 75% (DLR). The portion of the small bowel mesentery and bowel loops is magnified to better illustrate the structures in each image. The DLR image (c) demonstrates reduced noise, whereas the other two images (a, b) still have noise that is well visualized in the background mesenteric fat. Note that better contrast and sharpness are seen for mesenteric vessels in DLR (c). However, DLR (c) typically demonstrates a synthetic appearance, which refers to a “plastic” or “cartoon-like” appearance.

Supplementary Figure 6

Dynamic enteric phase contrast-enhanced T1-weighted images of a 27-year-old woman with active Crohn’s disease. The coronal images (a, b) were obtained using deep-learning reconstruction at the noise-reduction level of 75% and then reformatted to axial images (c, d). In the coronal images, an inflammatory mass in the ileocecal area (arrowheads) is noted with two fistulous tracts connected to the bowel loops (arrows). In the axial images, one of the fistulous tracts (arrows in a, c) reveals a connection between the inflammatory mass (Im) and the cecum (Ce), whereas the other tract (arrows in b and d) is a bidirectional fistula communicating between the cecum (Ce), terminal ileum (Ti), and inflammatory mass (Im). Two fistulas are demonstrated as more caudal to the ileocecal valve (not shown). Note that the relationship of the bidirectional fistulous tract between the bowel loops is clearly visible in the axial images, which help clarify anatomic detail when reviewed along with coronal images. The consecutive images of the same patient are also presented in Supplementary Video 2, available in the online supplement.