Impact of deep-learning-based reconstruction on MR enterography

by Yedaun Lee, MD, Inje University Heaundae Paik Hospital, Busan, Republic of Korea

Yedaun Lee, MD

Inje University Heaundae Paik Hospital

Busan, Republic of Korea

MR enterography is a common exam for diagnosing and evaluating progression of Crohn’s disease. At Inje University Heaundae Paik Hospital, approximately 100 Crohn’s disease patients undergo this exam each year to evaluate severity of active inflammation or monitor treatment response.

To detect findings of active inflammation of Crohn’s disease (bowel wall enhancement, stratification, perienteric infiltration, small abscess, fistula tract) and tracing bowel segment, the MR enterography exam should provide fine anatomic details and good image quality.

Due to bowel peristalsis and limited breath-holding duration, shorter scan times can lead to better image quality. Also, shorter scan times are good for patient compliance.

In our institution, MR enterography is performed after oral administration of 1000 mL of polyethylene glycol 3350. Scans are acquired on a SIGNA™ Architect 3.0T system with two 30-channel AIR™ Anterior Array Coils.

To avoid bowel peristalsis, 2.5 mg of cimetropium bromide (Alpit, Hana Pharmaceutical Co., Korea) is administered intravenously at three different intervals during the imaging exam.

Our standard MR enterography protocol includes: coronal T2-weighted fast spin echo sequences with and without fat saturation; coronal T2-like steady- state gradient-echo sequences with fat saturation; coronal free-breathing diffusion-weighted imaging with b-values of 0 and 900 sec/mm²; and coronal T1-weighted spoiled gradient- echo sequences with fat saturation, including unenhanced, enteric and portal phases after intravenous administration of gadoterate meglumine.

For additional enhanced axial T1-weighted images, we typically obtained these on the delayed phase after acquiring the dynamic coronal images. These delayed phases may be helpful for differentiating fibrosis from active inflammation. However, these late-delayed axial images were not helpful for evaluating active inflammation because bowel wall enhancement and mesenteric vascularity are best visualized in the earlier phases of a dynamic study.

In November 2020, we upgraded our software to SIGNA™Works AIR™ IQ Edition, which includes AIR™ Recon DL, a deep-learning-based reconstruction solution. Now, with AIR™ Recon DL, we can obtain high-resolution, near- isotropic coronal T1-weighted images with good image quality‡. This allows for high-quality, axial-reformatted images of simultaneous dynamic phases of the coronal images, which is helpful for depicting anatomic details of bowel segments or extraluminal complications such as fistulas or abscesses.

The typical imaging parameters of the fat-suppressed LAVA acquisition currently in use at our institution are: voxel size = 1.3 x 1.5 x 1.6 mm, flip angle = 10, NEX = 1, and bandwidth = 83.33 kHz. The scan time was approximately 17 seconds, varying slightly to accommodate the size of the patient.

Conclusion

Our results showed that compared to filtered and conventional images, AIR™ Recon DL significantly increased SNR, improved overall image quality and improved sharpness of the bowel wall and mesenteric vessels. The shorter scan times further decreased the radiologist’s perception of motion artifact, enabling better visualization of active inflammation of the bowel segment and depiction of the comb sign. Associated with Crohn’s disease, the comb sign is the engorgement of the mesenteric vessels with vascular dilatation, tortuosity with spacing of the vasa recta, and prominence of surrounding mesenteric fat resembling a comb.

Use of AIR™ Recon DL enables high-resolution, near-isotropic contrast-enhanced T1-weighted MR enterography with sufficient image quality. It allows for reformatting of high- quality axial images to be obtained from the same dynamic phase as the coronal images, which is helpful for depicting anatomic details of bowel segments or extraluminal complications such as fistulas or abscesses.

Figure 1. Dynamic enteric phase contrast-enhanced T1w images of a 26–year-old patient with Crohn’s disease. Images were obtained using LAVA HyperSense, 1.3 x 1.5 x 1.6 mm, 16 sec. Three sets of images were obtained: (A) with conventional reconstruction and no image filter, (B) with conventional reconstruction and image filter and (C) with AIR™ Recon DL at the high noise reduction level. (D-F) Axial images with 1.4 mm slice thickness were reformatted from coronal images. (C, F) Images with AIR™ Recon DL in both coronal and axial planes show a reduction in background noise and sharper margin of the bowel wall than (A, B, D, E) the other images with conventional reconstruction.Dynamic enteric phase contrast-enhanced T1w images of a 26–year-old patient with Crohn’s disease. Images were obtained using LAVA HyperSense, 1.3 x 1.5 x 1.6 mm, 16 sec. Three sets of images were obtained: (A) with conventional reconstruction and no image filter, (B) with conventional reconstruction and image filter and (C) with AIR™ Recon DL at the high noise reduction level. (D-F) Axial images with 1.4 mm slice thickness were reformatted from coronal images. (C, F) Images with AIR™ Recon DL in both coronal and axial planes show a reduction in background noise and sharper margin of the bowel wall than (A, B, D, E) the other images with conventional reconstruction.

Figure 2. Dynamic enteric phase contrast-enhanced T1w images of a 19–year-old patient with Crohn’s disease. Images were obtained using LAVA HyperSense, 1.3 x 1.5 x 1.6 mm, 16 sec. Three sets of images were obtained: (A) with conventional reconstruction and no image filter, (B) with conventional reconstruction and image filter and (C) with AIR™ Recon DL at the high noise reduction level. (D-F) Axial images with 1.4 mm slice thickness were reformatted from coronal images. (C, F) The images with AIR™ Recon DL show increased sharpness of the bowel wall, enabling better visualization of active inflammation of bowel segments and ulcer (arrows).

Figure 3. Dynamic enteric phase contrast-enhanced T1w images of a 39–year-old patient with active Crohn’s disease. Three sets of images were obtained: (A) with conventional reconstruction and no image filter, (B) with conventional reconstruction and image filter and (C) with AIR™ Recon DL at the high noise reduction level. (D-F) Axial images with 1.4 mm slice thickness were reformatted from coronal images, respectively. (C, F) Images with AIR™ Recon DL in both coronal and axial planes show a reduction in background noise and sharper margin of bowel wall and vascularity than the other images with conventional reconstruction. In coronal image sets, active inflammation is shown in the mid to distal ileum and a small abscess is detected in the distal ileum (arrows), which is better visualized in (C) the image with AIR™ Recon DL. The corresponding penetrating lesion in the distal ileum is also well-visualized in the axial images (arrows), most clearly visible in (F) the image with AIR™ Recon DL, which is helpful for increasing diagnostic confidence.