Impact of Deep Learning Denoising Algorithm on Diffusion Tensor Imaging of the Growth Plate on Different Spatial Resolutions.

To assess the impact of a deep learning (DL) denoising reconstruction algorithm applied to identical patient scans acquired with two different voxel dimensions, representing distinct spatial resolutions, this IRB-approved prospective study was conducted at a tertiary pediatric center in compliance with the Health Insurance Portability and Accountability Act. A General Electric Signa Premier unit (GE Medical Systems, Milwaukee, WI) was employed to acquire two DTI (diffusion tensor imaging) sequences of the left knee on each child at 3T: an in-plane 2.0 × 2.0 mm2 with section thickness of 3.0 mm and a 2 mm3 isovolumetric voxel; neither had an intersection gap. For image acquisition, a multi-band DTI with a fat-suppressed single-shot spin-echo echo-planar sequence (20 non-collinear directions; b-values of 0 and 600 s/mm2) was utilized. The MR vendor-provided a commercially available DL model which was applied with 75% noise reduction settings to the same subject DTI sequences at different spatial resolutions. We compared DTI tract metrics from both DL-reconstructed scans and non-denoised scans for the femur and tibia at each spatial resolution. Differences were evaluated using Wilcoxon-signed ranked test and Bland-Altman plots. When comparing DL versus non-denoised diffusion metrics in femur and tibia using the 2 mm × 2 mm × 3 mm voxel dimension, there were no significant differences between tract count (p = 0.1, p = 0.14) tract volume (p = 0.1, p = 0.29) or tibial tract length (p = 0.16); femur tract length exhibited a significant difference (p < 0.01). All diffusion metrics (tract count, volume, length, and fractional anisotropy (FA)) derived from the DL-reconstructed scans, were significantly different from the non-denoised scan DTI metrics in both the femur and tibial physes using the 2 mm3 voxel size (p < 0.001). DL reconstruction resulted in a significant decrease in femorotibial FA for both voxel dimensions (p < 0.01). Leveraging denoising algorithms could address the drawbacks of lower signal-to-noise ratios (SNRs) associated with smaller voxel volumes and capitalize on their better spatial resolutions, allowing for more accurate quantification of diffusion metrics.

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.

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 AIRTM 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.

Thin-Slice Pituitary MRI with Deep Learning-based Reconstruction: Diagnostic Performance in a Postoperative Setting.

Background Achieving high-spatial-resolution pituitary MRI is challenging because of the trade-off between image noise and spatial resolution. Deep learning-based MRI reconstruction enables image denoising with sharp edges and reduced artifacts, which improves the image quality of thin-slice MRI. Purpose To assess the diagnostic performance of 1-mm slice thickness MRI with deep learning-based reconstruction (DLR) (hereafter, 1-mm MRI+DLR) compared with 3-mm slice thickness MRI (hereafter, 3-mm MRI) for identifying residual tumor and cavernous sinus invasion in the evaluation of postoperative pituitary adenoma. Materials and Methods This single-institution retrospective study included 65 patients (mean age ± standard deviation, 54 years ± 10; 26 women) who underwent a combined imaging protocol including 3-mm MRI and 1-mm MRI+DLR for postoperative evaluation of pituitary adenoma between August and October 2019. Reference standards for correct diagnosis were established by using all available imaging resources, clinical histories, laboratory findings, surgical records, and pathology reports. The diagnostic performances of 3-mm MRI, 1-mm slice thickness MRI without DLR (hereafter, 1-mm MRI), and 1-mm MRI+DLR for identifying residual tumor and cavernous sinus invasion were evaluated by two readers and compared between the protocols. Results The performance of 1-mm MRI+DLR in the identification of residual tumor was comparable to that of 3-mm MRI (area under the receiver operating characteristic curve [AUC], 0.89-0.92 vs 0.85-0.89, respectively; P ≥ .09). In the identification of cavernous sinus invasion, the diagnostic performance of 1-mm MRI+DLR was higher than that of 3-mm MRI (AUC, 0.95-0.98 vs 0.83-0.87, respectively; P ≤ .02). Conventional 1-mm MRI (AUC, 0.82-0.83) showed comparable diagnostic performance to 3-mm MRI (AUC, 0.83-0.87) (P ≥ .38). With 1-mm MRI+DLR, residual tumor was diagnosed in 20 patients and cavernous sinus invasion was diagnosed in 14 patients, in whom these diagnoses were not made with 3-mm MRI. Conclusion In the postoperative evaluation of pituitary adenoma, 1-mm slice thickness MRI with deep learning-based reconstruction showed higher diagnostic performance than 3-mm slice thickness MRI in the identification of cavernous sinus invasion and comparable diagnostic performance to 3-mm slice thickness MRI in the identification of residual tumor. © RSNA, 2020 Online supplemental material is available for this article.

Comparing accuracy and reproducibility of sequential and Hadamard-encoded multidelay pseudocontinuous arterial spin labeling for measuring cerebral blood flow and arterial transit time in healthy subjects: A simulation and in vivo study.

PURPOSE: To compare performance of sequential and Hadamard-encoded pseudocontinuous arterial spin labeling (PCASL). MATERIALS AND METHODS: Monte Carlo simulations and in vivo experiments were performed in 10 healthy subjects. Field strength and sequence: 5-delay sequential (5-del. Seq.), 7-delay Hadamard-encoded (7-del. Had.), and a single-delay (1-del.) PCASL, without and with vascular crushing at 3.0T. The errors and variations of cerebral blood flow (CBF) and arterial transit time (ATT) from simulations and the CBF and ATT estimates and variations in gray matter (GM) with different ATT ranges were compared. Pairwise t-tests with Bonferroni correction were used. RESULTS: The simulations and in vivo experiments showed that 1-del. PCASL underestimated GM CBF due to insufficient postlabeling delay (PLD) (37.2 ± 8.1 vs. 47.3 ± 8.5 and 47.3 ± 9.0 ml/100g/min, P ≤ 6.5 × 10-6 ), while 5-del. Seq. and 7-del. Had. yielded comparable GM CBF (P ≥ 0.49). 5-del. Seq. was more reproducible for CBF (P = 4.7 × 10-4 ), while 7-del. Had. was more reproducible for ATT (P = 0.033). 5-del. Seq. was more prone to intravascular artifacts and yielded lower GM ATTs compared to 7-del. Had. without crushing (1.13 ± 0.18 vs. 1.23 ± 0.13 seconds, P = 2.3 × 10-3 ), but they gave comparable ATTs with crushing (P = 0.12). ATTs measured with crushing were longer than those without crushing (P ≤ 6.7 × 10-4 ), but CBF was not affected (P ≥ 0.16). CONCLUSION: The theoretical signal-to-noise ratio (SNR) gain through Hadamard encoding was confirmed experimentally. For 1-del., a PLD of 1.8 seconds is recommended for healthy subjects. With current parameters, 5-del. Seq. was more reproducible for CBF, and 7-del. Had. for ATT. Vascular crushing may help reduce variations in multidelay experiments without compromising tissue CBF or ATT measurements. LEVEL OF EVIDENCE: 1 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2018;47:1119-1132.