Brain magnetic resonance imaging (MRI) at 5 T offers unprecedented spatial resolution but is often limited by long scan times. Acceleration techniques, such as compressed sensing (CS) and artificial intelligence-assisted compressed sensing (ACS), have the potential to speed up the acquisition process while maintaining image quality. This study aims to evaluate and compare the performance of CS and ACS (with various acceleration factors) in brain MRI imaging at 5 T.

Methods

In this study, we enrolled 12 healthy volunteers and compared ACS-accelerated 5 T brain MRI with conventional methods of CS. The ACS acceleration factors for the brain protocol, consisting of 3D T1-weighted sequences and 2D T2-weighted sequences, were optimized in a pilot study on healthy volunteers (acceleration factor, 2.06–3.41× in T2-weighted imaging and 3.52–8.49× in T1-weighted imaging). We evaluated the images acquired from patients using various acceleration methods on the basis of acquisition times, the signal-to-noise ratio (SNR), the contrast-to-noise ratio, subjective image quality, and diagnostic agreement.

Results

Our findings revealed that ACS acceleration significantly reduced the acquisition times for T1- and T2-weighted sequences by up to 43% and 53%, respectively, compared with traditional CS at 5 T. Importantly, this acceleration was achieved while maintaining excellent image quality, demonstrated by higher or comparable SNR and contrast-to-noise ratio values.

Conclusions

The optimal ACS acceleration factors for 5 T brain MRI were determined to be 2.73× for 2D T2-weighted sequences and 6.5× for 3D T1-weighted sequences. ACS not only facilitates rapid imaging but also ensures comparable image quality and diagnostic performance, highlighting its potential to revolutionize high-field MRI scanning.

Abbreviations

ACS: artificial intelligence-assisted compressed sensing
AI: artificial intelligence
CNN: convolutional neural network
CNR: contrast-to-noise ratio
CS: compressed sensing
FOV: field of view
ICC: intraclass correlation coefficient
MRI: magnetic resonance imaging
PI: parallel imaging
ROI: region of interest
SAR: specific absorption rate
SNR: signal to-noise ratio
T: Tesla
T1WI: T1-weighted imaging
T2WI: T2-weighted imaging

1 INTRODUCTION

Magnetic resonance imaging (MRI) of the brain plays a crucial role in the diagnosis and treatment of various neurological disorders. However, a significant challenge encountered in clinical brain scanning is the lengthy acquisition time, which can often lead to patient discomfort, motion artifacts, and reduced image clarity. Long scan durations not only compromise patient comfort but also increase the probability of motion interference, particularly in patients with neurological impairments and pediatric patients. As a result, there is a pressing need to develop advanced accelerated scan protocols that can shorten the acquisition time while maintaining or even improving image quality [1, 2].

Against this backdrop, our study opts for a 5 Tesla (T) MRI scanner. This choice is driven by its distinct advantages over both 3 and 7 T scanners. Compared with a 3.0 T scanner, a 5 T scanner offers superior signal intensity and contrast, resulting in higher-resolution imaging that reveals finer anatomical details [3, 4]. Moreover, although a 7 T scanner boasts even greater resolution [5, 6], it faces technical challenges such as B1 field inhomogeneity and elevated specific absorption rate (SAR) distributions, posing safety concerns and limiting its clinical use for whole-body imaging [7-9]. In contrast, the recently introduced 5 T clinical MRI scanner addresses these issues, providing excellent image homogeneity, contrast uniformity, and reduced SAR risks. Therefore, employing a 5 T scanner ensures a balance between achieving the high-resolution imaging essential for precise analysis and maintaining patient safety, making it the most suitable option for exploring brain ACS under various acceleration factors.

Compressed sensing (CS) technology has emerged as a promising approach to accelerate MRI acquisition. By exploiting the sparsity of images in certain transform domains, CS allows for the reconstruction of high-quality images from undersampled k-space data [10, 11]. However, the application of CS at high acceleration factors is often limited by a reduction in the signal-to-noise ratio (SNR) and an increase in image artifacts. These limitations pose challenges to achieving significant acceleration without compromising image quality [12].

To address these challenges, artificial intelligence (AI)-assisted compressed sensing (ACS) has been introduced as an innovative acceleration solution [13-15]. ACS combines the principles of CS with the power of deep learning algorithms, specifically convolutional neural networks (CNNs). By pre-training CNNs on large datasets of fully sampled MR images, ACS is able to learn the underlying image structure information. This pre-trained knowledge is then used during the reconstruction of undersampled images, enabling high acceleration factors while also maintaining high image quality.

The technology of CS has been extensively studied in the context of brain structural imaging [16, 17], neuroimaging [18, 19], and brain diseases [20], yet ACS remains relatively untested in head research [21]. Specifically, the imaging quality of 5 T MR imaging of the brain using ACS has not been validated. In this study, we aim to evaluate the performance of ACS in brain MRI at 5 T, comparing it with conventional CS and analyzing the effects of different ACS acceleration factors for T1- and T2-weighted sequences on image quality metrics such as the SNR and contrast-to-noise ratio (CNR), as well as performing subjective image assessments. By doing so, we seek to determine the optimal acceleration factors for ACS that balance scan time reduction and image fidelity. The results of this study have the potential to provide valuable insights aiding the clinical application of ACS for brain MRI, ultimately contributing to improved patient comfort, scanner use, and diagnostic accuracy.

2 MATERIALS AND METHODS

2.1 Participants

The study cohort comprised 12 healthy volunteers (7 males and 5 females; mean age, 42.6 ± 7.6 years; age range: 27–56 years) who were recruited specifically for this research. All volunteers underwent a thorough screening process to ensure they met the inclusion criteria, which were as follows: (1) no history of significant medical conditions or neurological disorders; (2) no contraindications for MRI scanning; and (3) willingness to participate in the study and adhere to the required protocols. Exclusion criteria included pregnancy, claustrophobia, and any metal implants that could interfere with MRI scanning. The study was conducted in accordance with the ethical guidelines of our institution, and written informed consent was obtained from all volunteers prior to their participation.

2.2 MR image acquisition

All participants underwent scanning on a 5 T MR scanner (uMRJupiter; United Imaging Healthcare, Shanghai, China). Participants were positioned supine, using a 48-channel cranial coil head to ensure optimal signal reception. Their heads were stabilized throughout the scanning process to minimize movement artifacts. The scanning procedure encompassed the entire brain and employed both 2D T2-weighted fast spin echo sequences and 3D T1-weighted gradient echo sequences with the acceleration techniques of CS and ACS. Detailed parameters for each sequence are listed in Tables 1 and 2.

TABLE 1. Imaging parameters of T2WI sequences.

Parameters	T2-CS	T2-ACS1	T2-ACS2	T2-ACS4	T2-ACS6
FOV (mm × mm)	200 × 230	200 × 230	200 × 230	200 × 230	200 × 230
Slice thickness (mm)	3.0	3.0	3.0	3.0	3.0
Acquisition matrix	570 × 656	570 × 656	570 × 656	570 × 656	570 × 656
Repetition time (ms)	11,729	11,729	11,729	11,729	11,729
Echo time (ms)	111.36	100.76	100.76	100.76	100.76
Refocusing flip angle (°)	110	110	110	110	110
Bandwidth (Hz/Pixel)	260	260	260	260	260
Echo train length	21	23	23	23	23
Acceleration factor	2.06	2.10	2.27	2.73	3.41
Acquisition time (min:s)	7:01	6:15	5:40	4:41	3:43

Abbreviations: ACS, artificial intelligence-assisted compressed sensing; CS, compressed sensing; FOV, field of view; T2WI, T2-weighted imaging.

TABLE 2. Imaging parameters of T1WI sequences.

Parameters	T1-CS	T1-ACS1	T1-ACS2	T1-ACS4	T1-ACS6
FOV (mm × mm × mm)	240 × 256 × 210	240 × 256 × 210	240 × 256 × 210	240 × 256 × 210	240 × 256 × 210
Slice thickness (mm)	0.8	0.8	0.8	0.8	0.8
Acquisition matrix	300 × 320	300 × 320	300 × 320	300 × 320	300 × 320
Repetition time (ms)	9.2	9.2	9.2	9.2	9.2
Echo time (ms)	3.2	3.2	3.2	3.2	3.2
Refocusing flip angle (°)	9	9	9	9	9
Bandwidth (Hz/Pixel)	160	160	160	160	160
Echo train length	165	165	165	165	165
Acceleration factor	3.62	3.52	4.66	6.5	8.49
Acquisition time (min:s)	4:16	4:24	3:21	2:23	1:50

Abbreviations: ACS, artificial intelligence-assisted compressed sensing; CS, compressed sensing; FOV, field of view; T1WI, T1-weighted imaging.

2.3 ACS image reconstruction

ACS reconstruction is an advanced method that combines the benefits of CS, parallel imaging (PI), partial Fourier, and AI for accelerated MRI. The core of ACS lies in its integration of these techniques within an iterative reconstruction framework. Specifically, the method exploits the sparsity constraint inherent in CS to reconstruct images from highly undersampled k-space data, thereby resulting in accelerated image acquisition. PI techniques further accelerate the scan by exploiting the spatial sensitivity of multiple receiver coils. The partial Fourier method also contributes to scan time reduction by acquiring only half of the k-space data along the phase-encoding direction.

To address the uncertainty associated with traditional AI-based methods, ACS incorporates the output of a trained AI module, typically a CNN, as an additional regularization term within the CS reconstruction framework (Equation (1)). This approach constrains the AI predictions within the mathematical model, ensuring consistency with the acquired undersampled k-space data. Consequently, ACS reconstruction not only benefits from the noise reduction and artifact suppression capabilities of AI but also maintains the physical constraints inherent in CS and PI. The combined power of these techniques enables ACS to produce high-quality images with significantly reduced scan times, making it a promising solution for clinical MRI applications.

{\mathrm{a}\mathrm{r}\mathrm{g}\mathrm{m}\mathrm{i}\mathrm{n}}_{x}\Vert Ex-y{{\Vert }^{2}}_{2}+{\lambda }_{1}\Vert \psi (x){\Vert }_{1}+{\lambda }_{2}\Vert \phi ({x}_{\mathrm{A}\mathrm{I}},x){\Vert }_{w}+{\lambda }_{3}\Vert \mathrm{P}\mathrm{I}{\Vert }_{m}+{\lambda }_{4}\Vert \mathrm{P}\mathrm{F}{\Vert }_{n}

()

where x is the image to be reconstructed, E denotes the production of Fourier encoding with a binary k-space sampling mask, y represents the acquired k-space data, ψ denotes the sparse transform, x_AI is the reconstructed image of the trained AI module, and λ_i is the constraint for each term.

2.4 Quantitative analysis

In our study, transversal T1-weighted imaging (T1WI) and sagittal T2-weighted imaging (T2WI) images were chosen for quantitative assessments. The average and standard deviation signal intensity of the following sites were calculated with RadiAnt DICOM Viewer software (version 2020.1; Medixant [https://www.radiantviewer.com]). For each obtained cranial MR image, we delineated the region of interest (ROI) in six clinically relevant areas: the gray matter and white matter in the frontal lobe of the cerebrum, the gray matter and white matter in the cerebellum, the putamen (shell nucleus of the basal ganglia), and the corpus callosum. For each structure, we selected three different regions to delineate the ROI for signal acquisition, and we used the corresponding perpendicular (“top”) regions for noise data acquisition. For transverse T1WI and sagittal T2WI sequences, the circular ROIs were approximately 12.2–15.9 mm². The SNR and CNR were calculated using the Equations (2) and (3) [22]:

\mathrm{S}\mathrm{N}\mathrm{R}=\frac{{\mathrm{S}\mathrm{I}}_{\text{tissue}}}{{\mathrm{S}\mathrm{D}}_{\text{background}}}

()

\mathrm{C}\mathrm{N}\mathrm{R}=\frac{{\mathrm{S}\mathrm{I}}_{\text{tissue}1}-{\mathrm{S}\mathrm{I}}_{\text{tissue}2}}{\sqrt{{\mathrm{S}\mathrm{D}}_{\text{tissue}1}^{2}+{\mathrm{S}\mathrm{D}}_{\text{tissue}2}^{2}}}

()

2.5 Quality assessment

The image quality assessment in this study was conducted independently by two radiologists, both possessing 3 years of clinical experience, who remained blinded to the nature of the images they were evaluating. The overall quality of the individual images was assessed using a modified five-point ordinal Likert scale considering sharpness, distortion, and artifacts as the key metrics for evaluating the overall image quality. The scoring criteria were as follows: (1) nondiagnostic, indicating images deemed not clinically useful; (2) poor, signifying noticeable motion-related artifacts or image distortion, yet still partially diagnostic; (3) adequate, showing moderate motion-related artifacts or image distortion, but of sufficient quality for ventricular contour tracing; (4) good, exhibiting only mild motion-related artifacts or image distortion; and (5) excellent, denoting minimal to virtually no motion-related artifacts or image distortion. This systematic approach ensured a rigorous and standardized evaluation of image quality.

2.6 Statistical analysis

The Shapiro–Wilk test was initially conducted to assess the normality of the data distribution. Subsequently, the Friedman test with Bonferroni correction was used to ascertain whether significant variations existed among CS and ACS images with different acceleration factors, specifically evaluating the SNR and CNR.

Additionally, intraclass correlation coefficients (ICCs) were computed to assess interobserver reliability. Agreement was classified on the basis of the ICCs, as follows: 0.8–1.0, excellent agreement; 0.6–0.8, substantial agreement; 0.4–0.6, moderate agreement; 0.2–0.4, fair agreement; and 0.0–0.2, poor agreement. Statistical significance was determined by a p < 0.05. All statistical analyses were executed using SPSS version 26.0 (IBM Corp., Armonk, NY, USA).

3 RESULTS

In our study, we compared the performance of traditional CS with varying acceleration factors of ACS in 3D T1- and 2D T2-weighted MRI sequences. For the CS technique, acceleration factors of 3.62 and 2.06 were employed, whereas the ACS method used acceleration factors ranging from 3.52× to 8.49× for the 3D T1 sequence and from 2.10× to 3.41× for the 2D T2 sequence. As the ACS acceleration factor increased, we observed a significant reduction in scanning time. Specifically, for the 3D T1 sequence, the scanning duration gradually decreased from 4 min 24 s to 1 min 50 s. Similarly, for the 2D T2 sequence, the scanning time was reduced from 6 min 15 s to 3 min 43 s. In terms of image quality, we assessed the SNR and CNR, as shown in Figures 1-3. Notably, in the 2D T2 sequence, ACS demonstrated consistently superior performance compared with traditional CS across different acceleration factors (Figure 1). However, in the 3D T1 sequence, ACS and traditional CS exhibited comparable performance (Figure 2). The overall subjective image quality of T1- and T2-weighted images with different acceleration methods or factors was evaluated, and the results are displayed in Table 3. The interobserver agreement was strong for all techniques, with ICCs exceeding 0.8.

Details are in the caption following the image — **FIGURE 1**
Open in figure viewer PowerPoint

T2WI SNR of the (a) prefrontal white matter, (b) cerebellar white matter, (c) corpus callosum, (d) prefrontal gray matter, (e) cerebellar gray matter, and (f) putamen (shell nucleus of the basal ganglia). Statistically significant differences (p < 0.05) are marked with asterisks. ACS, artificial intelligence-assisted compressed sensing; CS, compressed sensing; SNR, signal to-noise ratio; T2WI, T2-weighted imaging.

TABLE 3. Results of the subjective evaluations of two readers.

T2WI		1	2	3	4	5	ICC	T1WI		1	2	3	4	5	ICC
		Image quality score								Image quality score
CS	Reader 1	1	1	2	4	2	0.877	CS	Reader 1	0	1	2	3	4	0.864
	Reader 2	0	1	3	2	4			Reader 2	0	0	1	4	5
ACS2	Reader 1	0	0	1	3	6	0.892	ACS1	Reader 1	0	0	1	3	6	0.897
	Reader 2	0	0	0	5	5			Reader 2	0	0	1	4	5
ACS4	Reader 1	0	0	1	5	4	0.885	ACS2	Reader 1	0	0	2	2	6	0.875
	Reader 2	0	0	1	4	5			Reader 2	0	0	0	4	6
ACS5	Reader 1	0	0	0	4	6	0.869	ACS4	Reader 1	0	0	1	3	6	0.839
	Reader 2	0	0	1	3	6			Reader 2	0	0	2	3	5
ACS6	Reader 1	0	1	0	4	5	0.842	ACS6	Reader 1	0	1	1	4	4	0.896
	Reader 2	0	0	2	4	4			Reader 2	0	0	1	4	5

Abbreviations: ACS, artificial intelligence-assisted compressed sensing; CS, compressed sensing; ICC, intraclass correlation coefficient; T1WI, T1-weighted imaging; T2WI, T2-weighted imaging.

4 DISCUSSION

This study comprehensively investigated the performance of traditional CS and advanced compressed sensing (ACS) techniques at various acceleration factors in both 2D T2 and 3D T1 sequences.

The application of 5 T ultra-high-field MRI provides the necessary hardware support for whole-body high-resolution imaging, enabling the visualization of finer structures within the human body. This enhanced resolution offers a stronger foundation for clinical diagnosis. However, it also leads to increased scanning durations and an elevated risk of exceeding the SAR limits. Prolonged scan times can affect patient comfort, increase the likelihood of motion artifacts, and potentially compromise image quality. Therefore, accelerated scanning at 5 T becomes imperative.

CS is a signal-processing technique that allows for the efficient recovery of sparse or compressible signals from a small number of non-adaptive linear measurements. In the context of MRI, CS has been widely used to accelerate image acquisition by reducing the number of k-space samples required, thereby shortening scan times [23]. This technology exploits the inherent sparsity of MRI images in certain transform domains, such as wavelet or gradient domains, to reconstruct high-quality images from undersampled data [24]. ACS, on the other hand, takes CS a step further by integrating CS, PI, partial Fourier transformation, and deep learning-based reconstruction. ACS leverages deep learning techniques to enhance the quality of image reconstruction, especially in challenging scenarios where traditional CS might struggle [25]. For instance, in the case of brain MRI, ACS can use prior knowledge about the anatomy and structure of the brain to improve image quality and reduce artifacts [21].

In term of practical application of ACS, the selection of the acceleration factor plays a pivotal role in balancing scanning duration and image quality. An appropriate acceleration factor will ensure that the tissue structures remain clear, facilitating accurate clinical diagnosis, while also considering the patient's tolerance with respect to scanning time to minimize motion artifacts. Our findings suggest that for the 2D T2 sequence, an acceleration factor of 2.73× is optimal. This factor results in a scanning time reduction of 2 min 20 s (33.25% reduction) compared with traditional CS. For the 3D T1 sequence, the optimal acceleration factor was found to be 6.5×, leading to a scanning time decrease of 2 min 18 s (44.14% reduction) compared with standard CS.

In addition to its rapid acquisition time, our results indicate that the ACS-accelerated 2D T2 sequence offers substantial benefits in terms of achieving higher SNR and CNR. However, significantly improved T2WI image quality was seen with ACS compared with CS (Figure 4). Nevertheless, as the ACS acceleration factor increases, the image quality does not consistently align with the observed trends in SNR and CNR. Specifically, for acceleration factors ranging from 1 to 4, the image benefits from artifact reduction (as indicated by the green arrow) and improved sharpness (as indicated by the yellow arrow) with an increase in the acceleration factor (Figure 4). However, when the acceleration factor is set to 6, image artifacts become more pronounced, and sharpness decreases (Figure 4). These observations are also consistent with the conclusions drawn from the image quality assessment.

Furthermore, our findings reveal that the ACS-accelerated 3D T1 sequence not only decreases scan time as acceleration factors increase but also maintains an SNR and CNR comparable to CS. Additionally, on the basis of the images and image quality scores presented in Figure 5 and Table 3, it is evident that when the acceleration factor is 4, the sharpness of the image surpasses that of images with other acceleration factors (as indicated by the yellow arrow). Moreover, the distortion and artifact levels are minimal (as indicated by the blue arrow).

In the comparative experiment of different acceleration factors for brain ACS, there were several limitations that require attention and further exploration. First, the sample size of this experiment was limited to only 12 participants, which may not provide sufficient statistical power for comprehensive conclusions to be drawn. Future studies should aim to collect larger samples to enhance the validity and reliability of the findings. Second, all participants in this study were healthy individuals. This restricts the generalizability of the results to patients with actual brain pathologies. To fully assess the efficacy of ACS acceleration techniques in clinical settings, it is essential to include a more diverse population, especially those with known brain diseases. Observing how ACS performs at different acceleration factors in identifying and discriminating lesion areas would provide valuable insights into its diagnostic accuracy. Moreover, although the current study focused on the technical aspects of ACS acceleration, future research should also consider integrating pathological diagnoses to validate the diagnostic accuracy of the imaging techniques. This would provide a more comprehensive understanding of the relationship between imaging results and actual pathological conditions. Lastly, the role of spatial resolution in diagnostic accuracy was not fully explored in this study. Future research could investigate how ACS methods can be optimized to achieve higher resolution while maintaining acceptable acquisition times. Determining the optimal balance between resolution and scan duration, especially for patients who may have limited tolerance because of pain or discomfort, remains an important area for further study.

5 CONCLUSIONS

Overall, our findings suggest that ACS offers significant advantages in terms of reducing scanning time while maintaining or even improving image quality, particularly in 2D T2 sequences. These results pave the way for further optimization of MRI protocols, balancing scanning efficiency and image quality.

AUTHOR CONTRIBUTIONS

Jiaqi Wang: Data curation (equal); writing—review and editing (equal). Liqiang Zhou: Software (equal); writing—original draft (equal).

ACKNOWLEDGMENTS

The authors thank the research technologists at United Imaging Healthcare, including Guobin Li, Jianmin Yuan, and Ran Tang, who supported this work.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

ETHICS STATEMENT

Not applicable.

INFORMED CONSENT

All participants provided written informed consent at the time of entering this study.

Open Research

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.

REFERENCES

1Bratke G, Rau R, Weiss K, Kabbasch C, Sircar K, Morelli JN, et al. Accelerated MRI of the lumbar spine using compressed sensing: quality and efficiency. J Magn Reson Imag. 2019; 49(7): e164–e175. https://doi.org/10.1002/jmri.26526
10.1002/jmri.26526
PubMed Web of Science® Google Scholar
2Morita K, Nakaura T, Maruyama N, Iyama Y, Oda S, Utsunomiya D, et al. Hybrid of compressed sensing and parallel imaging applied to three-dimensional isotropic T₂-weighted turbo spin-echo MR imaging of the lumbar spine. Magn Reson Med Sci. 2020; 19(1): 48–55. https://doi.org/10.2463/mrms.mp.2018-0132
10.2463/mrms.mp.2018-0132
PubMed Web of Science® Google Scholar
3Krautmacher C, Willinek WA, Tschampa HJ, Born M, Träber F, Gieseke J, et al. Brain tumors: full- and half-dose contrast-enhanced MR imaging at 3.0 T compared with 1.5 T: initial experience. Radiology. 2005; 237(3): 1014–1019. https://doi.org/10.1148/radiol.2373041672
10.1148/radiol.2373041672
PubMed Google Scholar
4Vaughan JT, Snyder CJ, DelaBarre LJ, Bolan PJ, Tian J, Bolinger L, et al. Whole-body imaging at 7 T: preliminary results. Magn Reson Med. 2009; 61(1): 244–248. https://doi.org/10.1002/mrm.21751
10.1002/mrm.21751
PubMed Web of Science® Google Scholar
5Bangerter NK, Taylor MD, Tarbox GJ, Palmer AJ, Park DJ. Quantitative techniques for musculoskeletal MRI at 7 Tesla. Quant Imag Med Surg. 2016; 6(6): 715–730. https://doi.org/10.21037/qims.2016.12.12
10.21037/qims.2016.12.12
PubMed Google Scholar
6Pohmann R, Speck O, Scheffler K. Signal-to-noise ratio and MR tissue parameters in human brain imaging at 3, 7, and 9.4 Tesla using current receive coil arrays. Magn Reson Med. 2016; 75(2): 801–809. https://doi.org/10.1002/mrm.25677
10.1002/mrm.25677
PubMed Web of Science® Google Scholar
7Winter L, Özerdem C, Hoffmann W, Santoro D, Müller A, Waiczies H, et al. Design and evaluation of a hybrid radiofrequency applicator for magnetic resonance imaging and RF induced hyperthermia: electromagnetic field simulations up to 14.0 tesla and proof-of-concept at 7.0 Tesla. PLoS One. 2013; 8(4):e61661. https://doi.org/10.1371/journal.pone.0061661
10.1371/journal.pone.0061661
CAS PubMed Web of Science® Google Scholar
8Kraff O, Theysohn JM, Maderwald S, Saylor C, Ladd SC, Ladd ME, et al. MRI of the knee at 7.0 Tesla. Rofo. 2007; 179(12): 1231–1235. https://doi.org/10.1055/s-2007-963607
10.1055/s-2007-963607
CAS PubMed Web of Science® Google Scholar
9Karamat MI, Darvish-Molla S, Santos-Diaz A. Opportunities and challenges of 7 Tesla magnetic resonance imaging: a review. Crit Rev Biomed Eng. 2016; 44(1–2): 73–89. https://doi.org/10.1615/critrevbiomedeng.2016016365
10.1615/CritRevBiomedEng.2016016365
PubMed Google Scholar
10Lustig M, Donoho D, Pauly JM. Sparse MRI: the application of compressed sensing for rapid MR imaging. Magn Reson Med. 2007; 58(6): 1182–1195. https://doi.org/10.1002/mrm.21391
10.1002/mrm.21391
CAS PubMed Web of Science® Google Scholar
11Jaspan ON, Fleysher R, Lipton ML. Compressed sensing MRI: a review of the clinical literature. Br J Radiol. 2015; 88(1056):20150487. https://doi.org/10.1259/bjr.20150487
10.1259/bjr.20150487
PubMed Web of Science® Google Scholar
12Ravishankar S, Bresler Y. MR image reconstruction from highly undersampled k-space data by dictionary learning. IEEE Trans Med Imag. 2011; 30(5): 1028–1041. https://doi.org/10.1109/TMI.2010.2090538
10.1109/TMI.2010.2090538
PubMed Web of Science® Google Scholar
13Eulig E, Maier J, Knaup M, Bennett NR, Hörndler K, Wang AS, et al. Deep learning-based reconstruction of interventional tools and devices from four X-ray projections for tomographic interventional guidance. Med Phys. 2021; 48(10): 5837–5850. https://doi.org/10.1002/mp.15160
10.1002/mp.15160
PubMed Web of Science® Google Scholar
14Freedman JN, Gurney-Champion OJ, Nill S, Shiarli AM, Bainbridge HE, Mandeville HC, et al. Rapid 4D-MRI reconstruction using a deep radial convolutional neural network: Dracula. Radiother Oncol. 2021; 159: 209–217. https://doi.org/10.1016/j.radonc.2021.03.034
10.1016/j.radonc.2021.03.034
PubMed Web of Science® Google Scholar
15Hamilton JI, Currey D, Rajagopalan S, Seiberlich N. Deep learning reconstruction for cardiac magnetic resonance fingerprinting T₁ and T₂ mapping. Magn Reson Med. 2021; 85(4): 2127–2135. https://doi.org/10.1002/mrm.28568
10.1002/mrm.28568
PubMed Web of Science® Google Scholar
16Vranic JE, Cross NM, Wang Y, Hippe DS, de Weerdt E, Mossa-Basha M. Compressed sensing-sensitivity encoding (CS-SENSE) accelerated brain imaging: reduced scan time without reduced image quality. AJNR Am J Neuroradiol. 2019; 40(1): 92–98. https://doi.org/10.3174/ajnr.A5905
10.3174/ajnr.A5905
CAS PubMed Web of Science® Google Scholar
17Dieckmeyer M, Roy AG, Senapati J, Wachinger C, Grundl L, Döpfert J, et al. Effect of MRI acquisition acceleration via compressed sensing and parallel imaging on brain volumetry. MAGMA. 2021; 34(4): 487–497. https://doi.org/10.1007/s10334-020-00906-9
10.1007/s10334-020-00906-9
PubMed Google Scholar
18Cho SJ, Choi YJ, Chung SR, Lee JH, Baek JH. High-resolution MRI using compressed sensing-sensitivity encoding (CS-SENSE) for patients with suspected neurovascular compression syndrome: comparison with the conventional SENSE parallel acquisition technique. Clin Radiol. 2019; 74(10): 817.e9–817.e14. https://doi.org/10.1016/j.crad.2019.06.023
10.1016/j.crad.2019.06.023
CAS PubMed Web of Science® Google Scholar
19Molnar U, Nikolov J, Nikolić O, Boban N, Subašić V, Till V. Diagnostic quality assessment of compressed SENSE accelerated magnetic resonance images in standard neuroimaging protocol: choosing the right acceleration. Phys Med. 2021; 88: 158–166. https://doi.org/10.1016/j.ejmp.2021.07.003
10.1016/j.ejmp.2021.07.003
PubMed Google Scholar
20Hou Y, Liu Q, Chen J, Wu B, Zeng F, Yang Z, et al. Application value of T2 fluid-attenuated inversion recovery sequence based on deep learning in static lacunar infarction. Acta Radiol. 2023; 64(4): 1650–1658. https://doi.org/10.1177/02841851221134114
10.1177/02841851221134114
PubMed Google Scholar
21Liu K, Xi B, Sun H, Wang J, Chen C, Wen X, et al. The clinical feasibility of artificial intelligence-assisted compressed sensing single-shot fluid-attenuated inversion recovery (ACS-SS-FLAIR) for evaluation of uncooperative patients with brain diseases: comparison with the conventional T2-FLAIR with parallel imaging. Acta Radiol. 2023; 64(5): 1943–1949. https://doi.org/10.1177/02841851221139125
10.1177/02841851221139125
PubMed Google Scholar
22Lee SH, Lee YH, Suh JS. Accelerating knee MR imaging: compressed sensing in isotropic three-dimensional fast spin-echo sequence. Magn Reson Imag. 2018; 46: 90–97. https://doi.org/10.1016/j.mri.2017.10.018
10.1016/j.mri.2017.10.018
PubMed Web of Science® Google Scholar
23Lusting M, Donoho DL, Santos JM, Pauly JM. Compressed sensing MRI. IEEE Signal Process Mag. 2008; 25(2): 72–82. https://doi.org/10.1109/MSP.2007.914728
10.1109/MSP.2007.914728
Web of Science® Google Scholar
24Sheng R, Zheng L, Jin K, Sun W, Liao S, Zeng M, et al. Single-breath-hold T2WI liver MRI with deep learning-based reconstruction: a clinical feasibility study in comparison to conventional multi-breath-hold T2WI liver MRI. Magn Reson Imag. 2021; 81: 75–81. https://doi.org/10.1016/j.mri.2021.06.014
10.1016/j.mri.2021.06.014
CAS PubMed Google Scholar
25Lin DJ, Johnson PM, Knoll F, Lui YW. Artificial intelligence for MR image reconstruction: an overview for clinicians. J Magn Reson Imag. 2021; 53(4): 1015–1028. https://doi.org/10.1002/jmri.27078
10.1002/jmri.27078
PubMed Web of Science® Google Scholar

Citing Literature

Volume2, Issue6

THEMED SECTION: Innovation and Application of Artificial Intelligence in Radiology

December 2024

Pages 584-593

An investigation into the applicability of rapid artificial intelligence-assisted compressed sensing in brain magnetic resonance imaging performed at 5 Tesla field strength