Deep learning reconstructed T2-weighted Dixon imaging of the spine: Impact on acquisition time and image quality

Purpose:<bold> </bold>To assess the image quality and impact on acquisition time of a novel deep learning based T2 Dixon sequence (T2(DL)) of the spine. Methods:<bold> </bold>This prospective, single center study included n = 44 consecutive patients with a clinical indication for lumbar MRI at our university radiology department between September 2022 and March 2023. MRI examinations were performed on 1.5-T and 3-T scanners (MAGNETOM Aera and Vida; Siemens Healthineers, Erlangen, Germany) using dedicated spine coils. The MR study protocol consisted of our standard clinical protocol, including a T2 weighted standard Dixon sequence (T2(std)) and an additional T2(DL) acquisition. The latter used a conventional sampling pattern with a higher parallel acceleration factor. The individual contrasts acquired for Dixon water-fat separation were then reconstructed using a dedicated research application. After reconstruction of the contrast images from k-space data, a conventional water-fat separation was performed to provide derived water images. Two readers with 6 and 4 years of experience in interpreting MSK imaging, respectively, analyzed the images in a randomized fashion. Regarding overall image quality, banding artifacts, artifacts, sharpness, noise, and diagnostic confidence were analyzed using a 5-point Likert scale (from 1 = non-diagnostic to 5 = excellent image quality). Statistical analyses included the Wilcoxon signed-rank test and weighted Cohen's kappa statistics. Results: Forty-four patients (mean age 53 years (+/- 18), male sex: 39 %) were prospectively included. Thirty-one examinations were performed on 1.5 T and 13 examinations on 3 T scanners. A sequence was successfully acquired in all patients. The total acquisition time of T2(DL) was 93 s at 1.5-T and 86 s at 3-T, compared to 235 s, and 257 s, respectively for T2(std) (reduction of acquisition time: 60.4 % at 1.5-T, and 66.5 % at 3-T; p < 0.01). Overall image quality was rated equal for both sequences (median T2(DL): 5[3 -5], and median T2(std): 5 [2 -5]; p = 0.57). T2(DL) showed significantly reduced noise levels compared to T2(std) (5 [4 -5] versus 4 [3 -4]; p < 0.001). In addition, sharpness was rated to be significantly higher in T2(DL) (5 [4 -5] versus 4 [3 -5]; p < 0.001). Although T2(DL) displayed significantly more banding artifacts (5 [2 -5] versus 5 [4 -5]; p < 0.001), no significant impact on readers diagnostic confidence between sequences was noted (T2(std): 5 [2 -5], and T2(DL): 5 [3 -5]; p = 0.61). Substantial inter-reader and intrareader agreement was observed for T2(DL) overall image quality (kappa: 0.77, and kappa: 0.8, respectively). Conclusion: T2(DL) is feasible, yields an image quality comparable to the reference standard while substantially reducing the acquisition time.