Sage Journals: Discover world-class research

Abstract

French

Purpose:

In pelvic MRI, Turbo Spin Echo (TSE) pulse sequences are used for T2-weighted imaging. However, its lengthy acquisition time increases the potential for artifacts. Deep learning (DL) reconstruction achieves reduced scan times without the degradation in image quality associated with other accelerated techniques. Unfortunately, a comprehensive assessment of DL-reconstruction in pelvic MRI has not been performed. The objective of this prospective study was to compare the performance of DL-TSE and conventional TSE pulse sequences in a broad spectrum of pelvic MRI indications.

Methods:

Fifty-five subjects (33 females and 22 males) were scanned at 3 T using DL-TSE and conventional TSE sequences in axial and/or oblique acquisition planes. Two radiologists independently assessed image quality in 6 categories: edge definition, vessel margin sharpness, T2 Contrast Dynamic Range, artifacts, overall image quality, and lesion features. The contrast ratio was calculated for quantitative assessment. A two-tailed sign test was used for assessment.

Results:

The 2 readers found DL-TSE to deliver equal or superior image quality than conventional TSE in most cases. There were only 3 instances out of 24 where conventional TSE was scored as providing better image quality. Readers agreed on DL-TSE superiority/inferiority/equivalence in 67% of categories in the axial plane and 75% in the oblique plane. DL-TSE also demonstrated a better contrast ratio in 75% of cases. DL-TSE reduced scan time by approximately 50%.

Conclusion:

DL-accelerated TSE sequences generally provide equal or better image quality in pelvic MRI than standard TSE with significantly reduced acquisition times.

Keywords

pelvic diagnostic imaging artificial intelligence magnetic resonance imaging radiology methods T2-weighted pulse sequence

Introduction

MRI is used widely to assess anatomy, pathology, and treatment response in pelvic diseases.^1-15 A critical component of pelvic MRI exams is the acquisition of T2-weighted images.^{2,4,7,8,16-19} These provide high tissue contrast that enables the delineation of anatomy and pathology in a broad range of pelvic diseases.^2-4,18,20,21 T2-weighted images are typically acquired with a Turbo Spin Echo (TSE) or Fast Spin Echo (FSE) pulse sequence. However, a major drawback of diagnostic-quality TSE scans is the long acquisition time. This leads to longer overall patient examination times, as well as an increased likelihood of artifacts.^20,21

The long acquisition time of TSE has motivated the use of acceleration techniques such as parallel imaging (PI) and compressed sensing (CS).^22-28 While these techniques decrease scan time, they can result in reduced signal-to-noise ratio (SNR) or additional artifacts.^22,29 As an alternative, deep learning (DL) reconstruction has been used to achieve highly reduced scan times, but without the degradation in image quality associated with other accelerated techniques.^{14,16,17,24,30-32} Previously, there have been some promising studies demonstrating the potential of the DL-accelerated TSE sequence for evaluation of targeted pelvic malignancy protocols (eg, prostate and rectal cancer). However, a comprehensive comparison to standard TSE pulse sequences for broad applications across general pelvic anatomy has not been performed.^{14,16,17,24,30-34} Therefore, the objective of this study was to compare the performance of a DL-accelerated TSE pulse sequence to a standard TSE pulse sequence for a broad spectrum of pelvic MRI indications. Demonstration of the superiority of DL-TSE over standard TSE sequences can support replacement of the latter in wide ranging pelvic MRI applications with great potential for enhancing clinical efficiency and improved utilization of MRI resources.

Materials and Methods

Study Population

Study approval was obtained from the institutional research ethics board, and written informed consent was secured from all patients. This prospective study enrolled 55 patients undergoing pelvic MRI scans from January to September 2023. The patient cohort was comprised of 33 females and 22 males, with a mean age of 54 years (range of 21-81). Patients were referred for a pelvic MRI for clinical indications that included gastrointestinal pathology (29 patients), gynecological conditions (21 patients), pelvic sarcoma (5 patients), and a genitourinary condition (1 patient). Patients referred for prostate MRI were not included in this study, since it would have required a different TSE protocol. Table 1 summarizes the age, gender, and clinical protocol for the study cohort.

Table 1.

Patient Demographics Data.

Patient characteristics	Values
Number of patients	55
Age
Mean and range	54 (28-81)
Gender
Male	22
Female	33
Protocol indicated
Gastrointestinal	27
Rectum	25
Anal	2
Genitourinary	2
Bladder	1
Testes	1
Gynecological	21
Endometrium	11
Cervix	4
Ovarian	6
Pelvic/sarcoma	5

Imaging Protocol

MRI exams were performed on a 3 Tesla system (MAGNETOM Verio, Siemens Healthineers, Erlangen, Germany), employing a 32-channel body coil with the compatibility of parallel imaging. All patients were scanned with both a standard 2D TSE sequence as well as an investigational DL-accelerated 2D TSE pulse sequence (Siemens Healthineers, Erlangen, Germany). Both scans were performed during the same imaging session. Depending on the clinical indication, scans were performed in either an axial plane (45 patients), an oblique plane (25 patients), or both planes (15 patients). Representative scan parameters for the standard and DL-accelerated TSE sequences are listed in Table 2. However, the exact FOV, number of slices, and scan angles deviated slightly from these values depending on the patient and clinical indication. The conventional TSE parameters reflect current clinical practice at our institution. The DL-TSE parameters were selected to assess the increased performance capabilities of the pulse sequence specifically with regards to resolution and shorter scan time versus the conventional TSE sequence.

Table 2.

Sample Imaging Protocol (Subject to Minor Changes Based on Variations Due to Patient Body Habitus and Clinical Indication).

Parameters	Axial T2 2D TSE		Oblique (mainly axial) T2 2D TSE
Parameters	Standard	DL-based WIP	Standard	DL-based WIP
FOV (mm²)	200 × 200	200 × 200	200 × 200	200 × 200
Voxel size (mm³)	0.6 × 0.6 × 4	0.3 × 0.3 × 4	0.6 × 0.6 × 3	0.3 × 0.3 × 3
Slice gap (mm)	1	1	0	0
Number of slices	46	46	24	24
Averages	2	1	2	1
TR (ms)	5450	6840	5110	5110
TE (ms)	91	104	91	104
Flip angle (°)	140	160	120	160
Bandwidth (Hz/pixel)	391	200	248	200
Parallel imaging acceleration	2	3	2	3
Turbo factor	20	25	18	25
Echo spacing (ms)	11.4	11.6	11.4	11.6
Scan time (s)	360	152	375	112

Note. FOV = field-of-view.

Deep Learning Reconstruction

The DL-based image reconstruction employs a variational neural network that receives under-sampled k-space data as well as pre-calculated coil sensitivity maps as input and generates the image through a fixed number of iterations that alternate between a parallel imaging-based data consistency and a neural network-based image enhancement.³⁵ The parameters of the neural networks as well as step sizes for the data consistency are considered as trainable and determined through end-to-end supervised training of the unrolled architecture. Trainings pairs were derived from fully sampled acquisitions providing target images and retrospectively under-sampled k-space data as input. About 25 000 fully sampled slices from healthy volunteers acquired on 1.5 and 3 T scanners (MAGNETOM scanners, Siemens Healthineers, Erlangen, Germany) in different body regions using a broad range of image contrasts were used as training data. Implemented in PyTorch, the model was trained on an Nvidia Tesla V100-SXM2 GPU cluster and exported for prospective use in the scanner reconstruction pipeline. The reconstruction was provided by Siemens Healthineers as a research application and has been previously evaluated for different clinical applications.^30,32

Image Quality Analyses

All images were anonymized and reviewed by 2 abdominal radiologists (with 2 and 24 years of MRI experience) using the institutional PACS system (Coral PACS, Version: 2.5.0.0). Both radiologists were blinded to all sequence information, and pulse sequence order was randomized for presentation to the radiologists. Each radiologist independently scored image quality on a 5-point scale in the following categories (including subcategories), except for the T2 Contrast Dynamic Range, which used a 3-point scale:

(1) Edge definition (rectum, bladder, prostate/uterus)

(2) Vessel margin sharpness

(3) T2 Contrast Dynamic Range (T2CDR)

(4) Artifacts (noise, T2 blur, motion, rectal gas)

(5) Overall image quality

(6) Lesion features (edge definition and conspicuity)

The definitions of the above categories are self-explanatory, except for T2CDR. This was defined as the ability of the sequence to provide qualitative (visual) contrast differences in T2 signal intensity between pelvic soft tissue structural elements, such as that of smooth muscle (eg, bowel wall, uterus, etc.) against pelvic fat, skeletal muscle, and or fluid signal intensity. High T2 contrast differences between these pelvic structures are critical for characterization of pelvic pathology as well as cancer detection and staging. Since this variable involved an overall visual qualitative assessment by the readers, a 3-point scale was deemed adequate for comparison of the 2 TSE sequences.

Including subcategories, there were a total of 12 different scores for each patient. Table 3 lists the scoring scales used for each of these 12 categories. As an additional quantitative measure of image quality, the contrast ratio (CR) was assessed for each pulse sequence. This was defined as the difference in signal between ischiorectal fat pads and the gluteus medius muscle, divided by the signal in the ischiorectal fat pads:

C o n t r a s t R a t i o (C R) = \frac{{S i g n a l}_{i s c h i o r e c t a l} - {S i g n a l}_{G l u t e u s}}{{S i g n a l}_{i s c h i o r e c t a l}}

(1)

Table 3.

Image Evaluation Categories and Scoring Scales.

Metric	Score
Metric	1	2	3	4	5
Edge definition (rectum)	Non-diagnostic	Ill defined	Moderately defined	Well defined	Excellently defined
Edge definition (bladder)
Edge definition (uterus/prostate)
Vessel margin sharpness	Non-diagnostic	Unsharp	Suboptimally sharp	Optimally sharp	Excellently sharp
T2 contrast dynamic range	Suboptimal	Adequate	Excellent	N/A	N/A
Artifacts (noise)	Non-diagnostic	Substantial impact on diagnosis	Moderate impact on diagnosis	Little impact on image diagnosis	No artifact
Artifacts (T2 blur)
Artifacts (motion)
Artifacts (rectal gas)
Overall image quality	Non-diagnostic	Substantial deficits in image quality	Acceptable image quality	Good image quality	Excellent image quality
Lesion feature (edge definition)	Non-diagnostic	Ill defined	Moderately defined	Well defined	Excellently defined
Lesion feature (conspicuity)	Lesion unidentifiable	Poor differentiation between lesion and normal anatomy	Lesion with suboptimal defined edges	Well seen lesion with adequately defined edges	Well seen lesion with very well defined edges

Signal was calculated as the mean value over a circular region-of-interest (ROI) in each of the regions. Measurements were taken on both left and right sides.

Acquisition Time

The acquisition times for the DL-TSE and standard TSE sequences in axial and oblique planes (when performed) were recorded in all patients for direct examination time comparisons.

Statistical Analyses

The null hypothesis tested in this study was:

H0: There is no difference in image quality (ie, categories listed in Table 3) or CR (equation (1)) between DL- and conventional-TSE pulse sequences

This hypothesis was tested with a two-tailed Sign Test using Matlab (MathWorks, R2013b). Statistical significance was set at the 95% confidence level. All tests were performed separately for axial and oblique orientations. For the image quality tests, a Bonferroni correction was applied to account for the multiple test categories under investigation. For each image quality category (Table 3), scoring was performed separately for each of the 2 readers. Reader agreement was defined as the fraction of image quality categories where both readers agreed on the superiority, inferiority, or equivalence of DL-TSE compared to conventional TSE.

Results

Image Results

Figures 1 and 2 illustrate some example DL- and conventional TSE images. Figure 1 shows the comparison of axial T2-weighted images acquired with the standard-TSE against the DL-TSE. Overall, the axial DL-TSE renders sharper anatomical edge definition than the axial standard-TSE without introducing artifacts. Lesion-wise, the axial DL-TSE provides sharper delineation of the uterine mass than the axial standard-TSE, although both sequences were deemed diagnostic (Figure 1A and B). In another patient, Figure 1C and D, the axial DL-TSE shows significantly reduced motion artifacts and superior delineation of the low rectal cancer within its posterior wall. Figure 2 compares the oblique T2-weighted images between the standard-TSE and DL-TSE. Like Figure 1, more definitive tissue delineations are observed generally for the oblique DL-TSE compared to the oblique standard-TSE – for example of the uterine junctional zone (Figure 2A and B). As shown in another patient in Figure 2C and D, the oblique DL-TSE provides clearly superior diagnostic details of a rectal tumor than the oblique standard-TSE, with significantly reduced motion blur and precise rectal tumor delineation. This is especially critical for optimal rectal cancer staging and management.

Figure 1.

Comparison of axial T2-weighted images acquired with standard TSE versus DL-TSE. The images include: (A) standard TSE and (B) DL-TSE of a patient being evaluated for a uterine mass (indicated by the arrows), and (C) standard TSE and (D) DL-T-TSE of a patient with rectal cancer (indicated by the arrows).

Figure 2.

Comparison of oblique T2-weighted images acquired with standard TSE versus DL-TSE. The images include: (A) standard TSE and (B) DL-TSE demonstrating comparative delineation of uterine junctional zone and zonal anatomy (indicated by the arrow), and (C) standard TSE and (D) DL-TSE in a patient with rectal cancer (indicated by the arrow).

Qualitative Evaluation

Tables 4 and 5 list the results for the image quality comparison in axial and oblique scan planes respectively. Overall, in 10 out of 12 categories in the axial plane, and 11 out of 12 categories in the oblique orientation, both readers independently found DL-TSE to provide image quality with either no significant difference or better image quality than conventional TSE. Going into more detail for the individual categories, both readers found significantly better image quality for DL-TSE in 6 out of 12 categories in both scan planes: edge definition (rectum, bladder, and uterus/prostate), vessel margin sharpness, overall image quality, and edge definition), while neither reader found a significant difference in T2CDR and rectal gas artifacts in either of the scan planes. For the remaining 4 categories: artifacts (noise, T2 blur, motion) and lesion conspicuity, there was disagreement between the readers in one or both orientations. There were 2 cases within these latter categories (T2 blur and motion artifacts in the axial orientation) where reader #2 found conventional TSE was better than DL-TSE. However, reader #1 disagreed and found the obverse. There was one case (noise in the oblique orientation) where reader #1 found conventional TSE to be better than DL-TSE, while reader #2 found no difference. Overall, the 2 readers agreed on superiority/inferiority/equivalence of DL-TSE compared to conventional TSE in 8/12 (=67%) of the categories in the axial plane, and 9/12 (=75%) of the categories in the oblique plane.

Table 4.

Image Quality Results of the Two-Tailed Sign Test for the Standard TSE Sequence and the DL-TSE Sequence in the Axial Orientation.

Table 5.

Image Quality Results of the Two-Tailed Sign Test for the Standard TSE Sequence and the DL-TSE Sequence in the Oblique Orientation.

Note. Gray highlights (underline) indicate statistical significance for DL-TSE score greater (less) than TSE at the 95% confidence level. Asterisks indicate categories with reader disagreement on the superiority/inferiority/equivalence of DL-TSE compared to conventional TSE.

Quantitative Evaluation

Table 6 summarize the CR analyses. The DL-TSE CR is significantly higher than conventional TSE in all cases except for the left sided oblique CR. In the latter case, conventional TSE CR was higher. Although all results were statistically significant, the differences in CR only amounted to a few percent in absolute terms.

Table 6.

Contrast Ratio (CR) Results of the Two-Tailed Sign Rank Test for the Standard TSE Sequence and the DL-Based TSE Sequence.

Note. Gray highlights (underline) indicate statistical significance for DL-TSE CR greater (less) than TSE at the 95% confidence level.

Fat means Ischiorectal Fat Pad.

Muscle means Gluteus Medius Muscle.

Acceleration Improvement

DLR-accelerated sequences enabled a significant reduction in scan acquisition times. On average, the axial DL-TSE accelerated the scan to 152.0 ± 10.4 seconds from 360.0 ± 17.7 seconds rendered by the axial standard-TSE. Similarly, the oblique DL-TSE achieved a scan time reduction to 112.0 ± 0.9, compared to 375.0 ± 22.5 seconds of the oblique standard-TSE.

Discussion

This prospective study compared a DL-accelerated T2 2D TSE sequence to a standard T2 2D TSE sequence in the context of pelvic MRI. In 10 out of 12 categories in the axial plane, and 11 out of 12 categories in the oblique orientation, DL-TSE provided image quality with either no significant difference or improved image quality compared to conventional TSE. In half of the image quality assessment categories, both readers scored the DL-TSE sequence as significantly better than the conventional TSE sequence. In 2 of the categories, T2CDR and rectal gas artifacts, both readers did not detect any significant difference in image quality. In the case of T2CDR, no significant difference is expected, since it should depend primarily on the TE value, rather than the reconstruction technique. Based on the reader scores, rectal gas artifacts were absent in this patient data set for both pulse sequences. For the remaining categories (artifacts and conspicuity), there was a discrepancy between the 2 readers on the superiority/inferiority/equivalence of DL-TSE compared to conventional TSE. Discrepancies also arose between axial and oblique orientations in these categories. A potential explanation for the observed discordance could be that these categories represent more subtle aspects of image quality and or differences in readers’ experience. On the other hand, given that DL-TSE has approximately half the scan time compared to the conventional TSE, it was somewhat surprising that DL-TSE was not clearly superior to conventional TSE in the motion artifact category. This may be due to the source of the motion artifacts. Artifacts arising from motion that occurs sporadically (eg, gross patient movement) are expected to be reduced with the use of a shorter scan time; those arising from rhythmic physiological motion (eg, peristalsis or breathing) would likely remain (unless scan times were much shorter – on the order of a few seconds). Unfortunately, the nature of the motion artifacts was not captured in the scoring system used in this study (Table 3).

The quantitative results of this study indicated that the CR of DL-TSE was superior to those of conventional TSE images in 3 out of 4 cases and worse in one. However, although statistically significant, the difference in CR values was relatively small (a few percent) in all cases. This small difference is likely an additional reason why neither reader noted a significant difference in the T2 contrast dynamic range image quality category. It is important to emphasize that the equivalent or better image quality provided by DL-TSE was obtained in about half the scan time of conventional TSE. Thus, it can be concluded that DL-TSE provides a viable clinical alternative to TSE for routine pelvic MRI examinations.

There are several recent studies which are in accordance with the present results –specifically, that DL-accelerated sequences provide equal or better image quality to their standard equivalents.^16,17,30-34 It should be emphasized, though, that most previous studies focused on a narrow range of pelvic anatomy – primarily the prostate or female pelvis. The present study more generally included all consenting patients scheduled for non-prostatic pelvic MRI. Additionally, this study evaluated a broader range of image quality categories than these previous studies. For example, none of the previously mentioned studies has explicitly evaluated motion artifacts, T2CDR, or rectal gas artifacts.

Limitations

One limitation of this study was the number of patients. This study included consenting patients for any type of non-prostate pelvic MRI exam. Therefore, although the total study enrollment was 55, the numbers of patients scanned for any one specific clinical indication was lower (Table 1). As a result, no sub-analysis of the results by clinical indication was performed. Related to this limitation, some anatomy is not well represented in our patient group (eg, anus, bladder, and testes). However, since we were non-selective in patient recruitment, the mix of indications is representative of our clinical practice. Also note that since this study was designed as an assessment of image quality only, there was no direct assessment of diagnostic performance.

A second limitation was the modest reader agreement (~70%) on the superiority/inferiority/equivalence of DL-TSE compared to conventional TSE. For example, in some cases, one reader would score DL-TSE as significantly better than conventional TSE, while the other reader would find no significant difference. However, despite discordant reader agreement in ~30% of the cases, there were only 3 instances out of 24 (=12 categories × 2 readers) where either reader scored the conventional TSE superior to DL-TSE. Even in these cases, the other reader did not agree with the result. In all the rest of the cases, DL-TSE was scored equal to or better than conventional TSE. A related limitation is that only 2 readers were used to assess image quality in this study. Additional reader(s) may have helped resolve some of the cases of reader disagreement. However, due to the relatively few cases where conventional TSE was scored better than DL-TSE by either reviewer, it is unlikely that additional reviewers would have changed the overall conclusions of this study.

One factor that potentially confounded the comparison between conventional and DL-TSE was the fact that the scan parameters were not exactly the same (see Table 2). In particular, note that the DL-TSE sequence had a finer resolution and shorter scan time. Indeed, these parameter differences may well have been the reason for some of the better image quality scores achieved by DL-TSE in Tables 4 and 5. The reason that identical parameters were not used was because this wouldn’t have provided a realistic clinical assessment of the improved capabilities offered by the DL-TSE pulse sequence – the entire motivation for using this sequence in the first place. Conversely, improving the resolution of the conventional TSE sequence to match that of DL-TSE would have resulted in clinically unsuitable long scan times. Therefore, the parameters chosen for this study represent a fair assessment of each pulse sequence as it would be implemented clinically.

Finally, although both the DL-accelerated and standard TSE sequences provided convincing depictions of pelvis pathology, the correlation between the image results and pathological results, were outside the scope of this study.

Conclusions

This study demonstrated that DL-TSE overall provides equivalent or improved image quality than conventional TSE for most pelvic MRI examinations. Moreover, it achieves this in approximately half the acquisition time, while providing better CR in most cases. This could be used to enhance clinical efficiency, optimize imaging resource allocation in pelvic MRI, enhancing patient experience while maintaining or improving image quality. DL-TSE is therefore a viable option for replacement of conventional TSE in pelvic MRI.

Footnotes

Acknowledgements

The authors would like to thank Dr. Gerald Moran, Research Collaborations Manager at Siemens Healthcare Limited, Oakville, ON, Canada.

ORCID iD

Marshall S. Sussman

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Lumeng Cui and Dominik Nickel report financial support was provided by Siemens Healthineers AG. Other authors report no specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Lumeng Cui and Dominik Nickel are employees of Siemens.

References

Expert Panel on GYN and OB Imaging, Reinhold

Ueno

, et al. ACR Appropriateness Criteria^® pretreatment evaluation and follow-up of endometrial cancer. J Am Coll Radiol. 2020;17(11S):S472-S486.

Gollub

Arya

Beets-Tan

, et al. Use of magnetic resonance imaging in rectal cancer patients: Society of Abdominal Radiology (SAR) rectal cancer disease-focused panel (DFP) recommendations 2017. Abdom Radiol (NY). 2018;43(11):2893-2902.

Nougaret

Horta

Sala

, et al. Endometrial cancer MRI staging: updated guidelines of the European Society of Urogenital Radiology. Eur Radiol. 2019;29(2):792-805.

Beets-Tan

RGH

Lambregts

DMJ

Maas

, et al. Magnetic resonance imaging for clinical management of rectal cancer: updated recommendations from the 2016 European Society of Gastrointestinal and Abdominal Radiology (ESGAR) consensus meeting. Eur Radiol. 2018;28(4):1465-1475.

Ang

De Robles

Kang

Winn

. Accuracy of pelvic magnetic resonance imaging in local staging for rectal cancer: a single local health district, real world experience. ANZ J Surg. 2021;91(1-2):111-116.

Pálsdóttir

Fridsten

Blomqvist

, et al. Interobserver agreement of transvaginal ultrasound and magnetic resonance imaging in local staging of cervical cancer. Ultrasound Obstet Gynecol. 2021;58(5):773-779.

Yoshizako

Yoshida

Asou

Nakamura

Kitagaki

. Comparison between turbo spin-echo and echo planar diffusion-weighted imaging of the female pelvis with 3T MRI. Acta Radiol Open. 2021;10(2):2058460121994737.

Sahin

Panico

Ursprung

, et al. Non-contrast MRI can accurately characterize adnexal masses: a retrospective study. Eur Radiol. 2021;31(9):6962-6973.

Siddiqui

Bari

. Accuracy of MRI pelvis in the diagnosis of ovarian endometrioma: using histopathology as gold standard. Cureus. 2021;13(12):e20650.

10.

Saida

Mori

Hoshiai

, et al. Diagnosing ovarian cancer on MRI: a preliminary study comparing deep learning and radiologist assessments. Cancers. 2022;14(4):987.

11.

Zheng

Miao

Liu

, et al. Multiparametric MRI-based radiomics model to predict pelvic lymph node invasion for patients with prostate cancer. Eur Radiol. 2022;32(8):5688-5699.

12.

Mao

Chen

Gao

Xiong

Lin

. A deep learning-based automatic staging method for early endometrial cancer on MRI images. Front Physiol. 2022;13:974245.

13.

Peacock

Manisundaram

Dibrito

, et al. Magnetic resonance imaging directed surgical decision making for lateral pelvic lymph node dissection in rectal cancer after total neoadjuvant therapy (TNT). Ann Surg. 2022;276(4):654-664.

14.

Kim

Lee

C-M

Jang

Kim

Lim

Kim

. Deep learning-based imaging reconstruction for MRI after neoadjuvant chemoradiotherapy for rectal cancer: effects on image quality and assessment of treatment response. Abdom Radiol (NY). 2023;48(1):201-210.

15.

Millischer

Santulli

Da Costa

, et al. Adolescent endometriosis: prevalence increases with age on magnetic resonance imaging scan. Fertil Steril. 2023;119(4):626-633.

16.

Lee

Hwang

Park

, et al. Utility of accelerated T2-weighted turbo spin-echo imaging with deep learning reconstruction in female pelvic MRI: a multi-reader study. Eur Radiol. 2023;33(11):7697-7706.

17.

Kim

Choi

Lee

Han

Mostapha

Nickel

. Deep learning-accelerated T2-weighted imaging versus conventional T2-weighted imaging in the female pelvic cavity: image quality and diagnostic performance. Acta Radiol. 2024;65(5):499-505.

18.

Fernandes

Gollub

Brown

. The importance of MRI for rectal cancer evaluation. Surg Oncol. 2022;43:101739.

19.

Herrmann

Benkert

Brendlin

, et al. Shortening acquisition time and improving image quality for pelvic MRI using deep learning reconstruction for diffusion-weighted imaging at 1.5 T. Acad Radiol. 2024;31(3):921-928.

20.

Maheshwari

Nougaret

Stein

, et al. Update on MRI in evaluation and treatment of endometrial cancer. Radiographics. 2022;42(7):2112-2130.

21.

Ringe

Wang

Deng

, et al. Abbreviated MRI protocols in the abdomen and pelvis. J Magn Reson Imaging. 2024;59(1):58-69.

22.

Deshmane

Gulani

Griswold

Seiberlich

. Parallel MR imaging. J Magn Reson Imaging. 2012;36(1):55-72.

23.

Ding

. Compare image quality of T2-weighted imaging with different phase acceleration factors. Medicine (Baltimore). 2023;102(10):e33234.

24.

Ueda

Ohno

Yamamoto

, et al. Compressed sensing and deep learning reconstruction for women’s pelvic MRI denoising: utility for improving image quality and examination time in routine clinical practice. Eur J Radiol. 2021;134:109430.

25.

Choi

Lee

Jung

Han

. High-resolution 3D T2-weighted SPACE sequence with compressed sensing for the prostate gland: diagnostic performance in comparison with conventional T2-weighted images. Abdom Radiol (NY). 2023;48(3):1090-1099.

26.

Pruessmann

Weiger

Scheidegger

Boesiger

. SENSE: sensitivity encoding for fast MRI. Magn Reson Med. 1999;42(5):952-962.

27.

Griswold

Jakob

Heidemann

, et al. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med. 2002;47(6):1202-1210.

28.

Lustig

Donoho

Pauly

. Sparse MRI: the application of compressed sensing for rapid MR imaging. Magn Reson Med. 2007;58(6):1182-1195.

29.

Jaspan

Fleysher

Lipton

. Compressed sensing MRI: a review of the clinical literature. Br J Radiol. 2015;88(1056):20150487.

30.

Gassenmaier

Afat

Nickel

, et al. Accelerated T2-Weighted TSE Imaging of the Prostate Using Deep Learning Image Reconstruction: A Prospective Comparison with Standard T2-Weighted TSE Imaging. Cancers. 2021;13(14):3593.

31.

Johnson

Tong

Donthireddy

, et al. Deep learning reconstruction enables highly accelerated biparametric MR imaging of the prostate. J Magn Reson Imaging. 2022;56(1):184-195.

32.

Ren

Liu

, et al. Comparison of a deep learning-accelerated T2-weighted turbo spin echo sequence and its conventional counterpart for female pelvic MRI: reduced acquisition times and improved image quality. Insights Imaging. 2022;13(1):193.

33.

Tsuboyama

Onishi

Nakamoto

, 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(6):379-386.

34.

Yang

Zou

Liu

W (Vivian)

, et al. High-resolution single-shot fast spin-echo MR imaging with deep learning reconstruction algorithm can improve repeatability and reproducibility of follicle counting. J Clin Med. 2023;12(9):3234.

35.

Hammernik

Klatzer

Kobler

, et al. Learning a variational network for reconstruction of accelerated MRI data. Magn Reson Med. 2018;79(6):3055-3071.

Deep Learning Reconstruction for T2 Weighted Turbo-Spin-Echo Imaging of the Pelvis: Prospective Comparison With Standard T2-Weighted TSE Imaging With Respect to Image Quality,Lesion Depiction,and Acquisition Time