Abstract
Objectives: – To evaluate the feasibility of generating high-resolution CT-like bone images from shoulder 3D-VIBE CAIPIRINHA Dixon MRI using vendor-independent deep-learning reconstruction. Methods: – This IRB-approved retrospective study included patients who underwent shoulder MRI and CT between January 2023 and 2024. MRI was performed on a 3.0-T system with a dedicated coil. Patients with study dates >6 months apart or poor image quality were excluded. A 3D multiecho VIBE CAIPIRINHA-Dixon sequence was postprocessed with a DL-based tool. In-phase and opposed-phase images before (Ori-In, Ori-Op) and after DL processing (DL-In, DL-Op) were evaluated by 2 musculoskeletal radiologists for pathology, anatomic visibility, artifacts, and glenoid dimensions, using a Likert scale. CT served as the gold standard. Relative contrast-to-noise (rCNR) and signal-to-noise ratios (rSNR) were assessed by radiologists and a computer scientist. Statistical tests included Friedman, Wilcoxon Rank, Kruskal-Wallis, and intraclass correlation coefficients. Results: – Ninety-nine patients (52 males, 47 females; age range: 17-87) were included. Cortical visibility score of humerus was significantly higher in DL-Op [median value: 2.75, interquartile range (IQR): 1.5 to 3.5 than Ori: 2 (1.5 to 2.44), P<0.001]. Trabecular score was higher on DL-In and Op (2.63 and 2.75) than Ori (2, P<0.001). DL-In images had the minimal artifacts [2 (1.5 to 2.38)] with the best rSNR (humerus: 110.82; glenoid: 92.14) and rCNR (humerus: 39.45; glenoid: 21.40). Ori-In underestimated longitudinal diameter compared with CT (P=0.046); transverse diameters were similar across sequences. The median DL-in and DL-op scores of rotator cuff calcification were 2.38 (IQR: 2.5 to 3.06) and 2.25 (2.06 to 2.94), significantly higher compared with the 1.25 (1 to 2.75) of fat-saturated T2WI. Inter-reader agreement ranged from 0.62 to 0.82 (P<0.001). Conclusions: – DL-3D-VIBE CAIPIRINHA Dixon MRI can generate high-quality CT-like images of shoulder joint, providing a potential for a radiation-free bone imaging.
| Original language | English |
|---|---|
| Pages (from-to) | 454-463 |
| Number of pages | 10 |
| Journal | Journal of Head Trauma Rehabilitation |
| DOIs | |
| State | Accepted/In press - 2026 |
Bibliographical note
Publisher Copyright:Copyright © 2025 Wolters Kluwer Health, Inc. All rights reserved.
Keywords
- bone MRI
- deep learning
- MRI, magnetic resonance imaging; CT, computed tomography; GRAPPA, GeneRalized Autocalibrating Partially Parallel Acquisitions; CAIPIRINHA or CAIPI, controlled aliasing in parallel imaging resulting in high acceleration; CS, compressed sensing; DL, deep learning; rCNR, relative contrast-to-noise ratio; rSNR, relative signal-to-noise ratio; Ori-In, original in-phase; Ori-Op, original opposed-phase; DL-In, swift-reconstructed in-phase, DL-Op, DL-reconstructed opposed-phase; HU, Hounsfield unit.
- pseudo-CT
- shoulder imaging
Fingerprint
Dive into the research topics of 'Enhancing Bone MRI With Vendor-Independent Deep Learning: A Comparative Study of CT and 3D VIBE CAIPI-Dixon Sequences for Shoulder Assessment'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver