2741
Feasibility of High-Resolution SSFSE MR imaging using Deep Learning Reconstruction in Assessment of Ovarian Volume and Follicle Count1Department of Radiology, Renmin Hospital of Wuhan University, Wuhan, China, 2Reproductive Medicine Center, Renmin Hospital of Wuhan University, Wuhan, China, 3MR Research China, GE Healthcare, Beijing, China
Synopsis
Keywords: Image Reconstruction, Artifacts, Single-shot fast spin-echo; PROPELLER; Follicle number per ovary; Ovarian volume
Motivation: Transvaginal ultrasonography (TVUS) often underestimates follicle count (FC) compared to MRI. The repeatability of FC and ovarian volume (OV) assessment was still affected by motion artifact on conventional T2-weighted fast spin echo images.
Goal(s): To propose a more reliable MRI technique in assessing the FC and OV.
Approach: High-resolution single-shot fast spin echo (SSFSE) sequence was used to accelerate the acquisition speed, and AIRTM Recon DL was used to compensate for noise in this study.
Results: Contributing to the improved time resolution and reduced noise, SSFSE-DL demonstrated better repeatability in FC and OV assessment compared to the widely used motion-robust PROPELLER technique.
Impact: High-resolution SSFSE sequence with DL reconstruction can be a reliable imaging method in the assessment of ovarian morphology. It has a potential in determining the threshold value of FC for PCOM identification in future studies.
Introduction
Transvaginal ultrasonography (TVUS) is currently the most preferred method to determine polycystic ovary morphology (PCOM) depending on its economy and efficiency.1, 2 TVUS-assessed follicle count (FC) is underestimated in comparison with MRI;3, 4 however, motion artifact affects the visualization of ovarian morphology on the conventional fast spin-echo (FSE) T2-weighted images.5 There are two main approaches to reduce motion artifact, including the utility of radially-filling k-space pattern such as PROPELLER or BLADE that was widely used in uterine and ovaries6, 7 and fast-speed acquisition such as SSFSE.8 SSFSE is a more motion-robust sequence compared to PROPELLER but high-resolution ovary MRI is not a preferred sequence due to inherent blurring and relatively low SNR.9 Recently, deep learning (DL) reconstruction has effectively reduced noise in different body part of MRI.10, 11 Therefore, the aim of this study was to investigate the application value of the combination of SSFSE sequence and DL reconstruction in the ovary MRI by comparing the image quality and repeatability of follicle number per ovary (FNPO) and OV assessment with those of SSFSE and PROPELLER images.Methods
Patients: 16 participants with clinical confirmed (n = 16) and suspected (n = 6) polycystic ovary syndrome (PCOS) including 4 adolescent girls and 18 adult women (mean age = 23.7 ± 5.0 years, range between 15 and 32 years) were involved in this study. A total of 44 ovaries were evaluated. Image acquisition: All participants underwent ovary MRI on a 3.0T MR scanner (SIGNA Architect; GE Medical Systems, Milwaukee, WI, USA) with a 30-channel phased-array coil. Both PROPELLER and SSFSE T2-weighted sequences in three planes were acquired with matched slices. SSFSE sequences were generated using conventional reconstruction (abbreviated as SSFSE) and AIRTM Recon DL (GE Healthcare, abbreviated as SSFSE-DL). Image analysis: Subjective image evaluation including the blurring artifacts, subjective noise, conspicuity of follicles and conspicuity of ovarian border were conducted by two radiologists. The assessments of FNPO and OV were performed twice by observer 1 and once by observer 2. Statistical Analysis: The qualitative results were compared using the Wilcoxon signed-rank tests and the inter-observer agreements were compared using the Cohen’s weighted kappa. The repeatability for FNPO and OV assessment were estimated by calculating the intraclass correlation coefficients (ICCs) and Bland-Altman plots. A paired t test was used to compare the absolute values of intra-observer and inter-observer differences.Results
SSFSE-DL outperformed SSFSE and PROPELLER at the aspect of blurring artifacts, subjective noise, and the conspicuity of follicles and the ovarian border (P < 0.05), except for the subjective noise and conspicuity of the ovarian border as assessed by observer 1 (P > 0.05) (Table 1 and Fig. 1). SSFSE-DL also demonstrated the best intra-observer and inter-observer agreements for both FNPO and OV assessment (Table 2) and the narrowest 95% LOAs (Fig. 2). Furthermore, the absolute values of intra-observer and inter-observer differences for FNPO and OV assessment on SSFSE-DL were the lowest among the three methods (P < 0.05). Notably, the absolute values of intra-observer and inter-observer differences for FNPO and OV assessments on SSFSE-DL showed no statistical disparities (P > 0.05), but the absolute values of intra-observer differences were significantly lower than those of inter-observer differences on SSFSE and PROPELLER (P < 0.05) (Table 3).Discussion
In our study, we utilized SSFSE sequences to expedite the acquisition speed, enabling each slice to be captured in less than one second. SSFSE facilitated the mitigation of respiratory motion and bowel motility, resulting in a pronounced reduction of motion-related blurring, in contrast to the PROPELLER method. Furthermore, the implementation of the AIRTM Recon DL with SSFSE-DL images notably reduced subjective noise compared to the SSFSE images. In contrast to the standard inverse Fourier transform reconstruction, AIRTM Recon DL utilizes a convolution neural network (CNN) model with the Rectified Linear Unit (ReLu) comprising over 100 thousand distinct internal cores to recognize 440-million parameters and retain essential information.12 Therefore, SSFSE-DL images exhibited enhanced conspicuity of follicles and ovarian border. Furthermore, different from the prevalent usage of the ellipsoid formula (length × width × height × 0.523), our approach involved a semi-automatic OV measurement technique aimed at deriving an accurate three-dimensional volume of the ovaries. The traditional formula relies on the assumption of an ellipsoidal shape of the ovaries, and its utilization may potentially introduce measurement errors when compared to actual OV. Consequently, SSFSE-DL images yielded superior consistency in both FNPO and OV assessment.Conclusion
High-resolution SSFSE-DL significantly improve the repeatability of FNPO and OV assessment, further elevating the reliability of PCOM determination.Acknowledgements
Funding: This work was supported by the Key Laboratory Project of Hubei Province (No. 2021KFY005).References
1. Dewailly D, Lujan ME, Carmina E, et al. Definition and significance of polycystic ovarian morphology: a task force report from the Androgen Excess and Polycystic Ovary Syndrome Society. Hum Reprod Update. 2014;20:334-352.
2. Teede HJ, Misso ML, Costello MF, et al. Recommendations from the international evidence-based guideline for the assessment and management of polycystic ovary syndrome. Fertil Steril. 2018;110:364-379.
3. Brown M, Park AS, Shayya RF, et al. Ovarian imaging by magnetic resonance in adolescent girls with polycystic ovary syndrome and age-matched controls. J Magn Reson Imaging. 2013;38:689-693.
4. Lujan ME, Jarrett BY, Brooks ED, et al. Updated ultrasound criteria for polycystic ovary syndrome: reliable thresholds for elevated follicle population and ovarian volume. Hum Reprod. 2013;28:1361-1368.
5. Wang SJ, Zhang MM, Duan N, et al. Using transvaginal ultrasonography and MRI to evaluate ovarian volume and follicle count of infertile women: a comparative study. Clin Radiol. 2022;77:621-627.
6. Tamhane AA, Arfanakis K. Motion correction in periodically‐rotated overlapping parallel lines with enhanced reconstruction (PROPELLER) and turboprop MRI. Magn Reson Med. 2009;62:174-182.
7. Lane BF, Vandermeer FQ, Oz RC, et al. Comparison of Sagittal T2-Weighted BLADE and Fast Spin-Echo MRI of the Female Pelvis for Motion Artifact and Lesion Detection. Am J Roentgenol. 2011;197:W307-W313.
8. Tsuboyama T, Onishi H, Nakamoto A, et al. Impact of Deep Learning Reconstruction Combined With a Sharpening Filter on Single-Shot Fast Spin-Echo T2-Weighted Magnetic Resonance Imaging of the Uterus. Invest Radiol. 2022;57:379-386.
9. Tsuboyama T, Takei O, Okada A, et al. Comparison of HASTE with multiple signal averaging versus conventional turbo spin echo sequence: a new option for T2-weighted MRI of the female pelvis. Eur Radiol. 2020;30:3245-3253.
10. Chaudhari AS, Fang Z, Kogan F, et al. Super‐resolution musculoskeletal MRI using deep learning. Magn Reson Med. 2018;80:2139-2154.
11. Lin DJ, Johnson PM, Knoll F, et al. Artificial Intelligence for MR Image Reconstruction: An Overview for Clinicians. J Magn Reson Imaging. 2021;53:1015-1028.
12. RM L. Performance characterization of a novel deep learning-based MR image reconstruction pipeline. 2020.
Figures
