Evaluation of Iterative Denoising 3-Dimensional T2-Weighted Turbo Spin Echo for the Diagnosis of Deep Infiltrating Endometriosis

Abstract

ObjectivesThe primary end point of this study was to evaluate the image quality and reliability of a highly accelerated 3-dimensional T2 turbo spin echo (3D-T2-TSE) sequence with prototype iterative denoising (ID) reconstruction compared with conventional 2D T2 sequences for the diagnosis of deep infiltrating endometriosis (DIE). The secondary end point was to demonstrate the 3D-T2-TSE sequence image quality improvement using ID reconstruction.Material and MethodsPatients were prospectively enrolled to our institution for pelvis magnetic resonance imaging because of a suspicion of endometriosis over a 4-month period. Both conventional 2D-T2 (sagittal, axial, coronal T2 oblique to the cervix) and 3D-T2-TSE sequences were performed with a scan time of 7 minutes 43 seconds and 4 minutes 58 seconds, respectively. Reconstructions with prototype ID (3D-T2-denoised) and without prototype ID (3D-T2) were generated inline at the end of the acquisition. Two radiologists independently evaluated the image quality of 3D-T2, 3D-T2-denoised, and 2D-T2 sequences. Diagnosis confidence of DIE was evaluated for both 3D-T2-denoised and 2D-T2 sequences. Intraobserver and interobserver agreements were calculated using Cohen κ coefficient.ResultsNinety female patients were included. Both readers found that the ID algorithm significantly improved the image quality and decreased the artifacts of 3D-T2-denoised compared with 3D-T2 sequences (P < 0.001). A significant image quality improvement was found by 1 radiologist for 3D-T2-denoised compared with 2D-T2 sequences (P = 0.002), whereas the other reader evidenced no significant difference. The interobserver agreement of 3D-T2-denoised and 2D-T2 sequences was 0.84 (0.73–0.95) and 0.78 (0.65–0.9), respectively, for the diagnosis of DIE. Intraobserver agreement for readers 1 and 2 was 0.86 (0.79–1) and 0.83 (0.76–1), respectively. For all localization of DIE, interobserver and intraobserver agreements were either almost perfect or substantial for both 3D-T2-denoised and 2D-T2 sequences.ConclusionsThree-dimensional T2-denoised imaging is a promising tool to replace conventional 2D-T2 sequences, offering a significant scan time reduction without compromising image quality or diagnosis information for the assessment of DIE.

Analysis of a Deep Learning-Based Superresolution Algorithm Tailored to Partial Fourier Gradient Echo Sequences of the Abdomen at 1.5 T

“…4,12 Previous works demonstrated the usefulness of denoising methods in abdominopelvic MRI for improvement of image quality and compensation of SNR loss. [13][14][15][16] Furthermore, it was most recently shown that deep learning (DL) techniques exhibit much potential for improvement of MR image quality and drastic reduction of acquisition time in abdominopelvic MRI. [17][18][19] The most recent developments of DLbased superresolution algorithms including partial Fourier acquisition allow to retrospectively omit portions of the acquired data mimicking a more aggressive partial Fourier setting including theoretical acquisition time reduction of the reconstructed images.…”

Quantitative Assessment of Iteratively Denoised 3D SPACE with Inner-Volume Excitation and Simultaneous Multi-Slice BLADE for Optimizing Female Pelvis Magnetic Resonance Imaging at 1.5 T

“…Another study comparing the image quality and reliability of 3D SPACE with ID and conventional 2D T2w for the diagnosis of deep infiltrating endometriosis determined that the ID algorithm significantly improved the image quality compared with 3D T2w (p < 0.001). One reader found a significant improvement in image quality compared with 2D T2w (p < 0.01), while the other reader found no difference (11).…”

“…The combination of SPACE acquisition with CAIPIRI-NHA acceleration (27,28), reconstructed using a prototype ID algorithm, demonstrated remarkable results in several body and musculoskeletal applications, including the uterus (11,19,29). 3D SPACE with ID was shown to be beneficial by generating a solitary set of volumetric T2w images with the opportunity for multiplanar reconstructions, thus eradicating the risk of false slice orientation (11). Moreover, the acquisition time of 3D SPACE with ID was shorter than that of separately acquired multiple stBLADE orientations.…”

“…Alternatively, isotropic 3D sequences can be used, in which a volume dataset is acquired in a single acquisition similar to computed tomography imaging, allowing retrospective reconstruction of the four endometrial axes and additional oblique planes. In this context, a recent study revealed excellent intraobserver agreements for both denoised 3D Sampling Perfection with Application-optimized Contrasts using different flip angle Evolutions (SPACE) and conventional 2D T2w sequences for the assessment of deep infiltrating endometriosis, while 3D T2w significantly reduced the scan time (11).…”

A Faster Prostate MRI: Comparing a Novel Denoised, Single‐Average T2 Sequence to the Conventional Multiaverage T2 Sequence Regarding Lesion Detection and PI‐RADS Score Assessment

“…However, T 2 ‐weighted imaging is a time‐consuming component of the prostate MRI examination, typically involving repeated acquisitions with signal averaging in order to obtain a sufficient signal‐to‐noise ratio for interpretation. Denoising approaches 7,8 can improve the subjective perception of image quality when combined with shorter acquisitions, and in some scenarios have been shown to allow for equivalent performance of diagnostic tasks compared to longer acquisitions without denoising 9–14 . However, any postprocessing method targeted to improving measures of image quality must be investigated to determine whether they alter the diagnostic assessment of images, either positively or negatively.…”