Skip to main content
European Radiology Experimental
Download PDF
  • Original article
  • Open access
  • Published:

Efficacy of compressed sensing and deep learning reconstruction for adult female pelvic MRI at 1.5 T

European Radiology Experimental volume 8, Article number: 103 (2024) Cite this article

Abstract

Background

We aimed to determine the capabilities of compressed sensing (CS) and deep learning reconstruction (DLR) with those of conventional parallel imaging (PI) for improving image quality while reducing examination time on female pelvic 1.5-T magnetic resonance imaging (MRI).

Methods

Fifty-two consecutive female patients with various pelvic diseases underwent MRI with T1- and T2-weighted sequences using CS and PI. All CS data was reconstructed with and without DLR. Signal-to-noise ratio (SNR) of muscle and contrast-to-noise ratio (CNR) between fat tissue and iliac muscle on T1-weighted images (T1WI) and between myometrium and straight muscle on T2-weighted images (T2WI) were determined through region-of-interest measurements. Overall image quality (OIQ) and diagnostic confidence level (DCL) were evaluated on 5-point scales. SNRs and CNRs were compared using Tukey’s test, and qualitative indexes using the Wilcoxon signed-rank test.

Results

SNRs of T1WI and T2WI obtained using CS with DLR were higher than those using CS without DLR or conventional PI (p < 0.010). CNRs of T1WI and T2WI obtained using CS with DLR were higher than those using CS without DLR or conventional PI (p < 0.003). OIQ of T1WI and T2WI obtained using CS with DLR were higher than that using CS without DLR or conventional PI (p < 0.001). DCL of T2WI obtained using CS with DLR was higher than that using conventional PI or CS without DLR (p < 0.001).

Conclusion

CS with DLR provided better image quality and shorter examination time than those obtainable with PI for female pelvic 1.5-T MRI.

Relevance statement

CS with DLR can be considered effective for attaining better image quality and shorter examination time for female pelvic MRI at 1.5 T compared with those obtainable with PI.

Key Points

  • Patients underwent MRI with T1- and T2-weighted sequences using CS and PI.

  • All CS data was reconstructed with and without DLR.

  • CS with DLR allowed for examination times significantly shorter than those of PI and provided significantly higher signal- and CNRs, as well as OIQ.

Graphical Abstract

Background

Magnetic resonance imaging (MRI) is widely accepted as an effective modality for evaluating female pelvic diseases, with T2-weighted sequences yielding excellent soft tissue contrast of the pelvic organ [1].

According to Organization for Economic Co-operation and Development (OECD) data, MRI examinations have been increasingly used in many OECD countries, with the average number of MRI examinations in OECD countries in 2020 having increased roughly 1.6 times over that observed in 2010. Although the average number of MRI units in OECD countries in 2020 has increased about 1.5 times over that in 2010, waiting times for MRI examinations have tended to be longer because the duration of MRI examinations has become longer [2]. Thus, the resultant increase in examination duration and waiting times is one of the major issues for MRI [3].

Currently, many academic institutions use 3-T rather than 1.5-T field strength to perform female pelvic MRI due to the higher signal-to-noise ratio (SNR) of 3-T systems, although no clinical advantages of 3-T systems have been demonstrated [4,5,6,7,8,9,10,11,12,13,14,15]. In some institutions and guidelines, the use of the 1.5-T systems is preferred over 3-T systems in pregnant patients [16]. Moreover, 1.5-T systems are considered as valuable as 3-T systems on whole-body imaging, as well as diffusion-weighted image (DWI) alone and routine MR imaging with DWI, although 3-T have been suggested as more useful than 1.5-T systems for prostate MRI without endorectal coils [6, 17, 18].

Since 2019, the efficacy of compressed sensing (CS) and deep learning reconstruction (DLR) has been tested and compared with that of routinely applied parallel imaging (PI) using 3-T systems not only for female pelvic studies, but also for other body areas or clinical purposes [15, 19,20,21,22,23,24,25,26,27,28,29,30,31,32]. Although T2-weighted sequences are considered the most important ones for routine pelvic MRI, other sequences such as those T1-weighted, are also used. To date, no major studies have assessed the efficacy of CS and DLR for female pelvic MRI in comparison with that of PI at 1.5 T. The purpose of this study was to determine the efficacy of CS and DLR for shorter examination time and better image quality compared with those obtainable with PI using female pelvic 1.5-T MRI.

Methods

Protocol, support, and funding

This retrospective study was approved by the Institutional Review Board of Fujita Health University Hospital (HM20-075, date of approval 21st July 2020) and technically and financially supported by Canon Medical Systems Corporation (Otawara, Japan). The IRB waived the need for written informed consent from each of the subjects in this study. Four of the authors are employees of Canon Medical Systems Corporation (K.Y., N.Y., M.I., and M.Y.) but did not have control over any of the data used in this study.

Subjects

Between January and March 2021, 53 consecutive patients with female pelvic diseases (mean age 45 years, range 22–85 years) underwent 1.5-T pelvic MRI using CS and DLR, as well as conventional PI. The exclusion criteria were: (1) no availability of T1-weighted images (T1WI) and T2-weighted images (T2WI); (2) previous surgery for female pelvic diseases; and (3) contraindications to MRI (such as pacemakers, ferromagnetic implants, etc.). One patient was excluded due to criterion 1. Finally, 52 consecutive patients with female pelvic diseases (mean age 44 years, range 22–85 years) were included. The patient selection flowchart is shown in Fig. 1.

Fig. 1
figure 1

Patient flowchart showing inclusions and exclusions

Full size image

MRI protocol

All examinations were performed with a 1.5-T system (Vantage Orian; Canon Medical Systems Corporation, Otawara, Japan) using a multiple phased-array surface coil: 16 channels for Atlas SPEEDER Body and 32 channels for Atlas SPEEDER Spine (Canon Medical Systems). To determine the utility of CS and DLR, each subject was studied with T1- and T2-weighted fast spin-echo sequences using CS and conventional PI. CS data was then reconstructed with and without DLR. In this study, CS (Compressed SPEEDER, Canon Medical Systems) was used with a reduction factor of 2 (i.e., k-space 50% undersampling) for both T1- and T2-weighted sequences.

For T1-weighted sequences, the scan parameters were: repetition time 515 ms; echo time 15 ms; excitation/refocus flip angles 90°/180°; plane: axial; acquisition matrix 224 × 256; reconstruction matrix 512 × 512; slice thickness 4 m; interslice gap 0.4 mm; field of view 250 × 250 mm2. For T2-weighed sequences, they were 5,096 ms, 93.5 ms, 90°/160°, sagittal, 352 × 352, 704 × 704, 4 mm, 0.4 mm, 250 × 250 mm2, respectively. The number of excitations (NEX) for both T1- and T2-weighted sequences with CS was 1, while it was 2 for those with PI. All CS data was reconstructed with and without DLR.

Details of DLR have been presented in previous papers [19,20,21,22, 24,25,26,27, 29,30,31,32]. In brief, the DLR method provided by Canon Medical Systems (Advanced intelligent Clear IQ Engine-AiCE) and used in this study is a convolutional neural network. In the study presented here, times for each data acquisition and reconstruction procedure were combined and recorded as “examination time” for each sequence for each patient.

Image analysis

All quantitative and qualitative assessments of image quality were performed on a commercially available workstation (Vitrea version 7.4, Canon Medical Systems).

Quantitative analysis

A board-certified female pelvic radiologist (T.U.) with 12 years of experience performed region of interest (ROI)-based measurements. Circular ROIs of 5–10 mm in diameter were placed over the myometrium, iliac muscle, multifidus muscle fat tissue of the gluteal region, and straight muscle of the abdomen on T1WI and T2WI for each patient. As for quantitative image indexes, SNR and contrast-to-noise ratio (CNR) were calculated by means of previously published formulas [19, 33,34,35,36]. For SNR, they were the following:

$${{{\rm{SNR}}}}_{{{\rm{T}}}1{{\rm{WI\; axial}}}}={{{\rm{SI}}}}_{{{\rm{Iliac\; muscle}}}}/{{{\rm{SD}}}}_{{{\rm{Iliac\; muscle}}}}$$
$${{{\rm{SNR}}}}_{{{\rm{T}}}2{{\rm{WI\; sagittal}}}}={{{\rm{SI}}}}_{{{\rm{Multifidus\; muscle}}}}/{{{\rm{SD}}}}_{{{\rm{Multifidus\; muscle}}}}$$

where SI is signal intensity and SD is the standard deviation.

For T1-weighted axial imaging, SIIliac muscle is the averaged SI of the iliac muscle within the ROI placed on either side, and SDIliac muscle is the average SD of the iliac muscle within the ROI on either side. For T2-weighted sagittal imaging, SIMultifidus muscle is the average SI of the multifidus muscle within the ROI placed on the same slice, and SDMultifidus muscle is the average SD of the multifidus muscle within the ROI on the same slice.

For CNR, the formulas were the following:

$${{{\rm{CNR}}}}_{{{\rm{T}}}1{{{\rm{WI}}}\; {{\rm{axial}}}}}=({{{\rm{SI}}}}_{{{{\rm{Fat}}}\; {{\rm{tissue}}}\; {{\rm{of}}}\; {{\rm{gluteal}}}\; {{\rm{region}}}}}{{{-}}}{{{\rm{SI}}}}_{{{{\rm{Iliac}}}\; {{\rm{muscle}}}}})/{{{\rm{SD}}}}_{{{{\rm{Iliac}}}\; {{\rm{muscle}}}}}$$
$${{{\rm{CNR}}}}_{{{\rm{T}}}2{{{\rm{WI}}}\; {{\rm{sagittal}}}}}= \,({{{\rm{SI}}}}_{{{\rm{Myometrium}}}}{{{-}}}{{{\rm{SI}}}}_{{{{\rm{Straight}}}\; {{\rm{muscle}}}\; {{\rm{of}}}\; {{\rm{abdomen}}}}}) \\ /{{{\rm{SD}}}}_{{{{\rm{Straight}}}\; {{\rm{muscle}}}\; {{\rm{of}}}\; {{\rm{abdomen}}}}}$$

For T1-weighted axial imaging, SIIliac muscle and SDIliac muscle are, respectively, the SI and SD of the iliac muscle within the ROI placed on the same slice as the previously determined SIFat tissue of the gluteal region. For T2-weighted sagittal imaging, the SIStraight muscle of the abdomen and SDStraight muscle of the abdomen are, respectively, the SI and SD of the straight muscle of the abdomen within the ROI placed on the same slice as the previously determined SIMyometrium.

Qualitative analysis

For qualitative assessment of image quality, the same female pelvic radiologist (T.U.) who performed quantitative assessment and a board-certified radiologist with 28 years experience (Y.O.) independently visually evaluated overall image quality (OIQ), artifacts, and diagnostic confidence level (DCL) by using a 5-point scoring system. OIQ was rated as: 1, poor; 2, fair; 3, moderate; 4, good; and 5, excellent. Artifacts were scored as: 1, no visualization of artifacts; 2, visualization of few artifacts; 3, visualization of a few artifacts; 4, visualization of several artifacts; and 5, visualization of marked number of artifacts. DCL was rated as with: 1, ≤ 20% confidence (very unsure); 2, 21–40% confidence; 3, 41–60% confidence; 4, 61–80% confidence; and 5, 81–100% confidence (highly confident). All final visual scores for each patient were determined by the consensus of the two readers.

Statistical analysis

For determining the efficacy of CS for reducing examination time while maintaining or improving image quality by DLR, mean examination times for T1WI and T2WI were compared for CS with and without DLR and for PI by means of Tukey’s honestly significant difference—HSD test. Examination time was determined in this study as the combined acquisition and reconstruction times. To compare quantitative image quality indexes for the three methods, SNRs and CNRs of T1WI and T2WI were compared for CS with and without DLR and for PI by means of Tukey’s test.

Interobserver agreement for each qualitative index was evaluated by using Cohen’s linearly weighted κ statistics and χ2 test. Interobserver agreements were considered as poor for κ < 0.21, fair for κ = 0.21–0.40, moderate for κ = 0.41–0.60, substantial for κ = 0.61–0.80, and excellent for κ = 0.81–1.00. Next, each of the qualitative indexes for the three methods was also compared by means of the Wilcoxon signed—rank test.

A p-value less than 0.05 was considered significant for all statistical analyses. All statistical analyses were performed with commercially available statistical software (JMP 14.2, SAS Institute Japan, Tokyo, Japan).

Results

All the patients had clinically or pathologically diagnosed uterine and ovarian diseases consisting of 57 benign uterine or ovarian lesions, eight malignant uterine or ovarian lesions, and three others (one pelvic abscess, two functional ovarian cysts). Details of patient characteristics are shown in Table 1. A representative case is shown in Figs. 2 and 3.

Table 1 Patient characteristics (histopathology of pelvic disease) of the 52 consecutive female patients included in this study (mean age 44 years, range 22–85 years)
Full size table
Fig. 2
figure 2

A 50-year-old female patient with uterine myoma on T2WI obtained with PI (a), CS with DLR (b), and CS without DLR (c). Examination times were 285 s, 107 s, and 103 s, respectively. CS with DLR (b) reduced image noise and showed uterine myoma, as well as normal structure more clearly than PI (a) or CS without DLR (c)

Full size image
Fig. 3
figure 3

A 32-year-old female patient with ovarian mature teratoma on T1WI obtained with PI (a), CS with DLR (b), and CS without DLR (c). All images clearly display the ovarian mature teratoma. Examination times were 243 s, 62 s, and 58 s, respectively. CS with DLR (b) reduced image noise and showed the ovarian mature teratoma more clearly than does conventional PI (a) or CS without DLR (c)

Full size image

Table 2 shows a comparison of the results for mean examination time and quantitative image quality for the three methods on T1WI and T2WI. Comparison of mean examination times showed those for CS with and without DLR on T1WI and T2WI were significantly shorter than those for PI (p < 0.001).

Table 2 Comparison of examination time and quantitative image quality for the three methods for T1WI and T2WI
Full size table

Comparison of SNRs for T1WI and T2WI for the three methods demonstrated that SNRs of T1WI and T2WI obtained by using CS with DLR were significantly higher than without the use of DLR (p < 0.001). Moreover, SNRs obtained by using CS with DLR were significantly higher than those using PI for both T1WI (p < 0.001) and T2WI (p = 0.01).

Comparisons of CNRs among the three methods showed that CNRs obtained by using CS with DLR were significantly higher than those obtained without DLR for both T1WI (p < 0.001) and T2WI (p < 0.001), as they were for results obtained with PI (p < 0.001 and p = 0.003, respectively).

Tables 3 and 4 show the results for interobserver agreement and qualitative image quality index for T1WI and T2WI. Interobserver agreements for OIQs, artifacts, and DCLs resulting from the use of all methods on T1WI and T2WI were rated as ‘substantial’ or ‘excellent’ (p < 0.001).

Table 3 OIQ, artifacts, and DCL for the three methods on T1WI
Full size table
Table 4 OIQ, artifacts, and DCL for the three methods on T2WI
Full size table

Comparison among the three methods of each of the OIQs for T1WI and T2WI showed that OIQs of T1WI and T2WI obtained by using CS with DLR were significantly higher than those obtained by using PI or CS without DLR (p < 0.001). Furthermore, a comparison among the three methods of artifacts for T1WI and T2WI showed that for T1WI and T2WI obtained by using CS with DLR, there were significantly fewer than by using PI or CS without DLR (p < 0.001). Finally, a comparison of DCLs among the three methods showed that DCLs of T2WI obtained using CS with DLR were significantly higher than those using PI or CS without DLR (p < 0.001).

Discussion

Our findings demonstrate CS and DLR, when compared with PI, can improve image quality, including CNR, while shortening examination time for not only T2WI, but also T1WI for female pelvic MRI using a 1.5-T system. To the best of our knowledge, this study is the first to demonstrate the clinical potentials of CS and DLR for female pelvic MRI in comparison with conventional PI or CS alone using a 1.5-T MR system.

Comparisons of mean examination times for T1WI and T2WI showed those using CS were significantly shorter than for PI, regardless of whether each of the MR images was obtained by using CS with or without DLR. These findings are compatible with those previously published for using a 3-T system [19,20,21, 29].

Comparisons of quantitative image quality showed SNRs and CNRs of T1WI and T2WI obtained by using CS with DLR were significantly higher than those of others. These findings suggest DLR reduces image noise and improves SNRs and CNRs when T1WI and T2WI of patients with female pelvic diseases are obtained by using CS. Moreover, with the use of CS and a reduction in NEX, DLR is potentially superior to PI for improving image noise, SNRs, and CNRs on T1WI and T2WI obtained by CS, as compared with PI, whose NEX was double. In view of the results of this, as well as previous studies using a 3-T MR system, CS should thus be better used in combination with DLR for routine clinical practice rather than PI for female pelvic MR imaging using not only 1.5-T, but also 3-T systems [20, 21, 27, 30,31,32].

Interobserver agreements for all qualitative indexes on T1WI and T2WI obtained with the three methods were rated as ‘substantial’ or ‘excellent’. Our evaluation for each index can thus be considered reproducible [37].

Comparisons of all qualitative image qualities show that OIQ and artifacts on T1WIs obtained by using CS with DLR were superior to those on other T1WIs, although there were no significant differences in DCL evaluation among the three methods. Moreover, each of the qualitative indexes for T2WI obtained by means of CS with DLR was significantly higher than that of other T2WIs. These results are compatible with the quantitative image quality analysis results in this study, which could have been anticipated in view of previously published results when using 3-T systems [19,20,21,22, 24, 29]. In comparison with currently used conventional PI for female pelvic MRI, DLR may well be more effective for better quantitative and qualitative image quality on T1WI and T2WI, while CS should be used in combination with DLR for shorter examination times when using a 1.5-T MR system in routine clinical practice.

There were several limitations to this study. First, the number of subjects was limited, and their underlying pathological and clinical conditions varied; in addition, the number of malignant lesions examined in this study was limited. Second, we used a different NEX for CS for comparison with conventional PI. The NEX for conventional PI was 2 (NEX 1 could not be used for conventional PI due to SNR limitations of the MRI system and sequences, i.e., sensitivity and noise performance of acquisition and receive coil). Therefore, this difference could have affected our results for comparison mean examination times. Third, we tested CS and DLR to obtain T1WI and T2WI on a 1.5-T system but did not use another field strength. Moreover, diagnostic performances by the three methods were not compared. Fourth, this study only evaluated unenhanced T1WI and T2WI and did not assess other important sequences such as diffusion-weighted imaging or dynamic contrast-enhanced imaging. Fifth, the reproducibility of the provided results was not investigated for other MRI systems and CS and DLR techniques from other vendors. Sixth, in this study, we calculated SNR as the mean SI within ROI divided by SD within ROI according to past literature [19, 33,34,35,36]. However, there are different equations for assessing SNR in other papers. Therefore, our results had some biases based on differences in SNR equations as compared with other methods. Further investigations are warranted to determine the real significance of these techniques for routine female pelvic MRIs.

In conclusion, CS and DLR are more effective than PI for shortening examination time and improving the image quality of female pelvic MRI on a 1.5-T system. The addition of DLR renders CS even more effective for improving image quality and improving not only SNR, but also CNR on T1WI and T2WI in this setting.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Abbreviations

CNR:

Contrast-to-noise ratio

CS:

Compressed sensing

DCL:

Diagnostic confidence level

DLR:

Deep learning reconstruction

DWI:

Diffusion-weighted image

NEX:

Number of excitations

OIQ:

Overall image quality

PI:

Parallel imaging

ROI:

Region of interest

SD:

Standard deviation

SI:

Signal intensity

SNR:

Signal-to-noise ratio

T1WI:

T1-weighted images

T2WI:

T2-weighted images

References

  1. Proscia N, Jaffe TA, Neville AM, Wang CL, Dale BM, Merkle EM (2010) MRI of the pelvis in women: 3D versus 2D T2-weighted technique. AJR Am J Roentgenol 195:254–259. https://doi.org/10.2214/ajr.09.3226

    Article  PubMed  Google Scholar 

  2. OECD (2023) Magnetic resonance imaging (MRI) units (indicator) (2023) Available via https://data.oecd.org/healthcare/magnetic-resonance-imaging-mri-exams.htm. Accessed 13 Aug 2023

  3. Touska P, Connor SEJ (2019) Recent advances in MRI of the head and neck, skull base and cranial nerves: new and evolving sequences, analyses and clinical applications. Br J Radiol 92:20190513. https://doi.org/10.1259/bjr.20190513

    Article  PubMed  PubMed Central  Google Scholar 

  4. Kataoka M, Kido A, Koyama T et al (2007) MRI of the female pelvis at 3 T compared to 1.5 T: evaluation on high-resolution T2-weighted and HASTE images. J Magn Reson Imaging 25:527–534. https://doi.org/10.1002/jmri.20842

    Article  PubMed  Google Scholar 

  5. Takahashi M, Uematsu H, Hatabu H (2003) MR imaging at high magnetic fields. Eur J Radiol 46:45–52. https://doi.org/10.1016/s0720-048x(02)00331-5

    Article  PubMed  Google Scholar 

  6. Morakkabati-Spitz N, Bastian PJ, Gieseke J et al (2008) MR imaging of the prostate at 3.0 T with external phased array coil—preliminary results. Eur J Med Res 13:287–291

    CAS  PubMed  Google Scholar 

  7. Merkle EM, Dale BM, Paulson EK (2006) Abdominal MR imaging at 3 T. Magn Reson Imaging Clin N Am 14:17–26. https://doi.org/10.1016/j.mric.2005.12.001

    Article  PubMed  Google Scholar 

  8. Schick F (2005) Whole-body MRI at high field: technical limits and clinical potential. Eur Radiol 15:946–959. https://doi.org/10.1007/s00330-005-2678-0

    Article  PubMed  Google Scholar 

  9. Hussain SM, van den Bos IC, Oliveto JM, Martin DR (2006) MR imaging of the female pelvis at 3T. Magn Reson Imaging Clin N Am 14:537–544. https://doi.org/10.1016/j.mric.2007.01.008

    Article  PubMed  Google Scholar 

  10. Ramalho M, Altun E, Herédia V, Zapparoli M, Semelka R (2007) Liver MR imaging: 1.5 T versus 3 T. Magn Reson Imaging Clin N Am 15:321–347. https://doi.org/10.1016/j.mric.2007.06.003

    Article  PubMed  Google Scholar 

  11. Chang KJ, Kamel IR, Macura KJ, Bluemke DA (2008) 3.0-T MR imaging of the abdomen: comparison with 1.5-T. Radiographics 28:1983–1998. https://doi.org/10.1148/rg.287075154

    Article  PubMed  Google Scholar 

  12. Bernstein MA, Huston 3rd J, Ward HA (2006) Imaging artifacts at 3.0 T. J Magn Reson Imaging 24:735–746. https://doi.org/10.1002/jmri.20698

    Article  PubMed  Google Scholar 

  13. Hata H, Inoue Y, Nakajima A, Komi S, Miyatake H (2017) Influence of the magnetic field strength on image contrast in Gd-EOB-DTPA-enhanced MR imaging: comparison between 1.5 T and 3.0 T. Magn Reson Med Sci 16:109–114. https://doi.org/10.2463/mrms.mp.2015-0158

    Article  PubMed  Google Scholar 

  14. Kataoka M, Isoda H, Maetani Y et al (2007) MR imaging of the female pelvis at 3 Tesla: evaluation of image homogeneity using different dielectric pads. J Magn Reson Imaging 26:1572–1577. https://doi.org/10.1002/jmri.21173

    Article  PubMed  Google Scholar 

  15. Akai H, Yasaka K, Sugawara H et al (2023) Commercially available deep-learning-reconstruction of MR Imaging of the knee at 1.5 T has higher image quality than conventionally-reconstructed imaging at 3 T: a normal volunteer study. Magn Reson Med Sci 22:353–360. https://doi.org/10.2463/mrms.mp.2022-0020

    Article  PubMed  Google Scholar 

  16. Jabehdar Maralani P, Kapadia A, Liu G et al (2022) Canadian association of radiologists recommendations for the safe use of MRI during pregnancy. Can Assoc Radiol 73:56–67. https://doi.org/10.1177/08465371211015657

    Article  Google Scholar 

  17. Ohno Y, Takeshi Y, Takenaka D, Koyama H, Aoyagi K, Yui M (2020) Comparison of diagnostic accuracy for TNM stage among whole-body MRI and coregistered PET/MRI using 1.5-T and 3-T MRI systems and integrated PET/CT for non-small cell lung cancer. AJR Am J Roentgenol 215:1191–1198. https://doi.org/10.2214/AJR.19.22565

    Article  PubMed  Google Scholar 

  18. Park HJ, Lee SY, Rho MH et al (2016) Single-shot echo-planar diffusion-weighted MR imaging at 3 T and 1.5 T for differentiation of benign vertebral fracture edema and tumor infiltration. Korean J Radiol 17:590–597. https://doi.org/10.3348/kjr.2016.17.5.590

    Article  PubMed  PubMed Central  Google Scholar 

  19. Matsuyama T, Ohno Y, Yamamoto K et al (2022) Comparison of utility of deep learning reconstruction on 3D MRCPs obtained with three different k-space data acquisitions in patients with IPMN. Eur Radiol 32:6658–6667. https://doi.org/10.1007/s00330-022-08877-2

    Article  PubMed  Google Scholar 

  20. Obama Y, Ohno Y, Yamamoto K et al (2022) MR imaging for shoulder diseases: effect of compressed sensing and deep learning reconstruction on examination time and imaging quality compared with that of parallel imaging. Magn Reson Imaging 94:56–63. https://doi.org/10.1016/j.mri.2022.08.004

    Article  PubMed  Google Scholar 

  21. Ueda T, Ohno Y, Yamamoto K et al (2021) 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 134:109430. https://doi.org/10.1016/j.ejrad.2020.109430

    Article  PubMed  Google Scholar 

  22. Kidoh M, Shinoda K, Kitajima M et al (2019) Deep learning based noise reduction for brain MR imaging: tests on phantoms and healthy volunteers. Magn Reson Med Sci 19:195–206. https://doi.org/10.2463/mrms.mp.2019-0018

    Article  PubMed  PubMed Central  Google Scholar 

  23. Kiryu S, Akai H, Yasaka K et al (2023) Clinical impact of deep learning reconstruction in MRI. Radiographics 43:e220133. https://doi.org/10.1148/rg.220133

    Article  PubMed  Google Scholar 

  24. Kashiwagi N, Tanaka H, Yamashita Y et al (2021) Applicability of deep learning-based reconstruction trained by brain and knee 3-T MRI to lumbar 1.5-T MRI. Acta Radiol Open 10:20584601211023939. https://doi.org/10.1177/20584601211023939

    Article  PubMed  PubMed Central  Google Scholar 

  25. Oshima S, Fushimi Y, Miyake KK et al (2023) Denoising approach with deep learning-based reconstruction for neuromelanin-sensitive MRI: image quality and diagnostic performance. Jpn J Radiol 41:1216–1225. https://doi.org/10.1007/s11604-023-01452-9

  26. Fujiwara M, Kashiwagi N, Matsuo C et al (2023) Ultrafast lumbar spine MRI protocol using deep learning-based reconstruction: diagnostic equivalence to a conventional protocol. Skeletal Radiol 52:233–241. https://doi.org/10.1007/s00256-022-04192-5

    Article  PubMed  Google Scholar 

  27. Kashiwagi N, Sakai M, Tsukabe A et al (2022) Ultrafast cervcial spine MRI protocol using deep learning-based reconstruction: diagnostic equivalence to a conventional protocol. Eur J Radiol 156:110531. https://doi.org/10.1016/j.ejrad.2022.110531

    Article  PubMed  Google Scholar 

  28. Zerunian M, Pucciarelli F, Caruso D et al (2023) Fast high-quality MRI protocol of the lumbar spine with deep learning-based algorithm: an image quality and scanning time comparison with standard protocol. Skeletal Radiol 53:151–159. https://doi.org/10.1007/s00256-023-04390-9

    Article  PubMed  PubMed Central  Google Scholar 

  29. Ueda T, Ohno Y, Yamamoto K et al (2022) Deep learning reconstruction of diffusion-weighted MRI improves image quality for prostatic imaging. Radiology 303:373–381. https://doi.org/10.1148/radiol.204097

    Article  PubMed  Google Scholar 

  30. Uetani H, Nakaura T, Kitajima M et al (2022) Hybrid deep-learning-based denoising method for compressed sensing in pituitary MRI: comparison with the conventional wavelet-based denoising method. Eur Radiol 32:4527–4536. https://doi.org/10.1007/s00330-022-08552-6

    Article  PubMed  Google Scholar 

  31. Shiraishi K, Nakaura T, Uetani H et al (2023) Combination use of compressed sensing and deep learning for shoulder magnetic resonance imaging with various sequences. J Comput Assist Tomogr 47:277–283. https://doi.org/10.1097/rct.0000000000001418

    Article  PubMed  Google Scholar 

  32. Akai H, Yasaka K, Sugawara H et al (2023) Acceleration of knee magnetic resonance imaging using a combination of compressed sensing and commercially available deep learning reconstruction: a preliminary study. BMC Med Imaging 23:5. https://doi.org/10.1186/s12880-023-00962-2

    Article  PubMed  PubMed Central  Google Scholar 

  33. Chen Z, Sun B, Duan Q et al (2019) Three-dimensional breath-hold MRCP using SPACE pulse sequence at 3 T: comparison with conventional navigator-triggered technique. AJR Am J Roentgenol 213:1247–1252. https://doi.org/10.2214/AJR.19.21399

    Article  PubMed  Google Scholar 

  34. Yokoyama K, Nakaura T, Iyama Y et al (2016) Usefulness of 3D hybrid profile order technique with 3-T magnetic resonance cholangiography: comparison of image quality and acquisition time. J Magn Reson Imaging 44:1346–1353. https://doi.org/10.1002/jmri.25289

    Article  PubMed  Google Scholar 

  35. Zerunian M, Pucciarelli F, Caruso D et al (2024) Fast high-quality MRI protocol of the lumbar spine with deep learning-based algorithm: an image quality and scanning time comparison with standard protocol. Skeletal Radiol 53:151–159. https://doi.org/10.1007/s00256-023-04390-9

    Article  PubMed  Google Scholar 

  36. Uetani H, Nakaura T, Kitajima M et al (2021) A preliminary study of deep learning-based reconstruction specialized for denoising in high-frequency domain: usefulness in high-resolution three-dimensional magnetic resonance cisternography of the cerebellopontine angle. Neuroradiology 63:63–71. https://doi.org/10.1007/s00234-020-02513-w

    Article  PubMed  Google Scholar 

  37. Mandrekar JN (2011) Measures of interrater agreement. J Thorac Oncol 6:6–7. https://doi.org/10.1097/JTO.0b013e318200f983

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

This study was technically and financially supported by Canon Medical Systems Corporation. Large language models (LLMs) were not used in this study.

Funding

This study was technically and financially supported by Canon Medical Systems Corporation. Among the authors, Drs. Ohno and Nagata got research grants from Canon Medical Systems Corporation.

Author information

Authors and Affiliations

  1. Department of Diagnostic Radiology, Fujita Health University School of Medicine, Toyoake, Japan

    Takahiro Ueda, Masahiko Nomura, Yoshiyuki Ozawa & Yoshiharu Ohno

  2. Canon Medical Systems Corporation, Otawara, Japan

    Kaori Yamamoto & Natsuka Yazawa

  3. Department of Radiology, Fujita Health University Bantane Hospital, Nagoya, Japan

    Ikki Tozawa

  4. Joint Research Laboratory of Advanced Medical Imaging, Fujita Health University School of Medicine, Toyoake, Japan

    Masato Ikedo, Masao Yui, Hiroyuki Nagata & Yoshiharu Ohno

Authors
  1. Takahiro Ueda
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Kaori Yamamoto
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Natsuka Yazawa
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Ikki Tozawa
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Masato Ikedo
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Masao Yui
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Hiroyuki Nagata
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Masahiko Nomura
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Yoshiyuki Ozawa
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Yoshiharu Ohno
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

TU: conceptualization, data curation, formal analysis, investigation, methodology, project administration, resources, visualization, writing—original draft, and writing—review and editing. KY: software and writing—review and editing. NY: software and writing—review and editing. IT: data curation. MI: software and writing—review and editing. MY: software and writing—review and editing. HN: data curation, funding acquisition, investigation, methodology, resources, visualization, and writing—review and editing. MN: data curation, methodology, supervision, validation, and visualization. Y Ozawa: data curation, methodology, project administration, resources, software, supervision, validation, visualization, and writing—review and editing. Y Ohno: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, software, supervision, validation, visualization, writing—original draft, and writing—review and editing.

Corresponding author

Correspondence to Takahiro Ueda.

Ethics declarations

Guarantor

This study was approved by the Institutional Review Board (IRB) of Fujita Health University Hospital. The IRB waived the need for written informed consent from each of the subjects in this study because of the retrospective study.

Consent for publication

Not applicable.

Competing interests

This study was technically and financially supported by Canon Medical Systems Corporation. Among the authors, Y Ohno and MN received research grants from Canon Medical Systems Corporation. KY, NY, MI, and MY are employees of Canon Medical System Corporation, but did not have control over any of the data and information submitted for publication or which data and information were to be included in this study. The remaining authors report no conflicts of interest.

Additional information

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Ueda, T., Yamamoto, K., Yazawa, N. et al. Efficacy of compressed sensing and deep learning reconstruction for adult female pelvic MRI at 1.5 T. Eur Radiol Exp 8, 103 (2024). https://doi.org/10.1186/s41747-024-00506-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s41747-024-00506-5

Keywords