Deep Learning-Based Reconstruction for Three-Dimensional Volumetric Brain MRI: A Qualitative and Quantitative Assessment

Deep Learning-Based Reconstruction for Three-Dimensional Volumetric Brain MRI: A Qualitative and Quantitative Assessment | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Deep Learning-Based Reconstruction for Three-Dimensional Volumetric Brain MRI: A Qualitative and Quantitative Assessment Yeseul Kang, Sang-Young Kim, Jun Hwee Kim, Nak-Hoon Son, Chae Jung Park This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5345149/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 27 Mar, 2025 Read the published version in BMC Medical Imaging → Version 1 posted 4 You are reading this latest preprint version Abstract Background: To evaluate the performance of a deep learning reconstruction (DLR) based on Adaptive-Compressed sensing (CS)-Network for brain MRI and validate it in a clinical setting Methods: Ten healthy volunteers and 22 consecutive patients were prospectively enrolled. Volunteers underwent 3D brain MRI including T1 without CS factor (9:16min, reference standard); with CS factor of 2 without DLR (CS2, 4:6min); with CS factor of 2 with DLR (DLR-CS2); with CS factor of 4 without DLR (CS4, 2:6min); and with CS factor of 4 with DLR (DLR-CS4). The patients’ MRI included the CS2 and DLR-CS4. The volumes of lateral ventricles, hippocampus, choroid plexus, and white matter hypointensity were calculated and compared among the sequences. Three radiologists independently assessed anatomical conspicuity, overall image quality, artifacts, signal-to-noise ratio (SNR), and sharpness using a 5-point scale for each sequence. Results: Applying acceleration factors of 2 and 4 reduced the scan time to 65.4% and 33.5%, respectively, of that of the reference standard. Volumes of all the measured subregions showed no significant differences among different sequences in all participants. In qualitative analysis, the interrater agreement was excellent (κ=0.844–0.926). In volunteers, quality of DLR-CS4 were comparable to those of CS2 for all metrics except for the overall image quality and SNR despite a 51.2% scan time reduction. In patients, DLR-CS4 showed quality comparable to that of CS2 for all metrics. Conclusions: DLR allowedthe scan time reduction by at least half without sacrificing image quality and volumetric quantification accuracy, supporting its reliability and efficiency. Magnetic resonance imaging deep learning compressed sensing qualitative assessment quantitative assessment Figures Figure 1 Figure 2 Figure 3 Introduction Three-dimensional (3D) high-resolution structural MRI is fundamental to evaluate various neurological diseases. 3D T1-weighted imaging is a basic MRI sequence used for the evaluation of structural anatomy and volumetric analysis. To obtain acceptable image quality in a limited scan time, many accelerating techniques have been developed, including parallel imaging, compressed sensing, and partial Fourier [ 1 , 2 ]. However, these techniques frequently and inevitably introduce noise and degrade spatial resolution. Recently, deep learning (DL) has been suggested for the reconstruction of accelerated MR scans as well as for image denoising, gaining increased attention in recent years [ 3 – 6 ]. MRI scanners equipped with DL reconstruction (DLR) for denoising are already available for clinical use. One of the recently developed DLR methods called Adaptive-Compressed sensing (CS)-Network utilizes a novel convolutional neural networks (CNN) to integrate and enhance conventional CS-SENSE algorithms (i.e., CS + SENSE) [ 7 ]. As the traditional CS-based reconstruction results in compromised diagnostic value due to difficulties in choosing the optimal sparsity transforms [ 8 ], DL has replaced wavelet-based sparsity transform with multi-scale sparsifying transform in this novel method. With this strategy, the performance of denoising can be improved through automatically optimized the sparsity constraint during the iterative reconstruction. Recent clinical studies using either the prototype or commercial versions of the DLR based on Adaptive-CS-Network have demonstrated improved image quality in the imaging studies of the prostate [ 9 ], musculoskeletal area [ 10 , 11 ], breasts [ 12 ], heart [ 13 ], and MR cholangiopancreatography [ 14 ] and proved its feasibility; however, it has not been validated yet in the field of neuroradiology. We, therefore, aimed to evaluate the performance of a novel DLR tool in adult brain 3D T1-weighted images (T1) with quantitative and qualitative assessments and to validate its feasibility and clinical usefulness in a prospective setting. Materials and methods Participants The Institutional Review Board of our tertiary institution approved this prospective study (No. 9-2024-0091). Informed consents were obtained from all participants. This study adhered both the Declaration of Helsinki and the Health Insurance Portability and Accountability Act (HIPAA). In this prospective study, 10 healthy volunteers (median age: 38 [interquartile range, IQR, 31–39] years; male:female, 4:6) and 22 consecutive patients (median age: 76 [IQR 74–79] years; male:female, 9:13) undergoing clinical brain MRI at our tertiary institution were recruited. The patients were imaged for a variety of clinical indications: subjective cognitive decline (n = 12), mild cognitive impairment (n = 1), Alzheimer’s disease (n = 2), Parkinson’s disease (n = 1), infarction (n = 2), and other conditions, including visual disturbance, trauma, carotid artery stenosis, and hydrocephalus. MRI acquisition All MRI examinations were performed using a 3T MR scanner (Ingenia Elition; Philips Healthcare, Amsterdam, Netherlands). Different imaging protocols were used for healthy volunteers and patients. Detailed MRI parameters are presented in Supplementary Material. For the healthy volunteers, five different 3D T1-weighted images (T1) were obtained as follows: 1) 3D T1 without CS factor (reference standard, 9 min 16 s); 2) 3D T1 with CS factor of 2 without DLR (CS2, 4 min 6 s); 3) 3D T1 with CS factor of 2 with DLR (DLR-CS2, 4 min 6 s); 4) 3D T1 with CS factor of 4 without DLR (CS4, 2 min 6 s); and 5) 3D T1 with CS factor of 4 with DLR (DLR-CS4, 2 min 6 s). Twenty-two patients underwent brain MRI, including imaging with two 3D T1 sequences, as follows: 1) CS2 (4 min 6 s) and 2) DLR-CS4 (2 min 6 s). CS2 is a standard brain MRI sequence used at our institution. Therefore, patients additionally underwent a faster 3D T1 scan by increasing the CS factor to 4, which was then reconstructed with DLR (DLR-CS4) to evaluate the feasibility of DLR in the patient population. DLR The DLR used in this study builds on parallel imaging SENSE combined with CS, CS-SENSE in which the prior information such as coil sensitivity profile and low-resolution background information is used for data reliability during the iterative reconstruction. The applied DL, Adaptive-CS-Network was first introduced for the FastMRI challenge to reconstruct under-sampled knee MRI data [7] and it is now extended to all image contrasts and application domains. Briefly, the Adaptive-CS-Net is an advancement of the deep learning–based iterative shrinkage-thresholding algorithm (ISTA) network proposed by Zhang and Ghanem [15]. The power of Adaptive-CS-Net comes from the use of multi-scale and multi-slice computations in iterative learning-based reconstruction scheme in which MR images are efficiently denoised by multiple reconstruction blocks. Similar to CS-SENSE reconstruction, the network is fed by raw k-space data, coil sensitivity and coarse background information, but the sparsity constraining step of the iterative reconstruction is DL-based. In each individual block of the network, data consistency checking with the incoming raw k-space data is performed. The Adaptive-CS-Net used in this study was initially trained with a large dataset from various anatomical regions and acceleration factors. The algorithm was refined to run on standard reconstruction hardware, in contrast with a previously reported network [7]. Reconstruction was performed on the scanner and required approximately 80 s for the CS DL reconstructions, compared with approximately 40 s for the standard CS reconstructions. Qualitative analysis To qualitatively compare image quality among different sequences, two neuroradiologists (C.J.P. and J.H.K., with 4 and 5 years of experience in neuroradiology, respectively) and one junior resident evaluated the following four image quality parameters on a 5-point Likert scale (1 = unacceptable; 2 = poor; 3 = acceptable; 4 = good; 5 = excellent): anatomical conspicuity, overall image quality, artifacts, signal-to-noise ratio (SNR), and sharpness. The anatomic conspicuity was assessed by evaluating the delineation of existing pathologies contrast to the adjacent normal brain structures or delineation of subregions of deep gray matter in cases with no abnormalities on 3D T1. Gray-white matter differentiation was used to evaluate sharpness. The reviewers evaluated all image series in the Picture Archiving and Communication Systems database. No restrictions were applied to window-level setting adjustments regarding the time or ability to scroll through the images. The reviewers were blinded to patient information, including disease and scan parameters, to identify the type of sequence. Quantitative analysis For quantitative analysis, we obtained the volumes of lateral ventricle (LV), hippocampus, choroid plexus (CP), white matter hypointensity (WMH), and total intracranial volume (ICV) from each sequence using FastSurfer software. FastSurfer is a fast and extensively validated DL pipeline for fully automated processing of structural human brain MRIs (Henschel et al., NeuroImage 2020; 219:117012). This enabled whole-brain segmentation into 95 classes within 50 s per participant. The details and validity thereof can be found elsewhere (https://deep-mi.org/research/fastsurfer/). The regional volume was expressed as the ratio of the regional volume to the total ICV (ratio to the total ICV × 10 3 ). Statistical analysis Based on the results of the Shapiro–Wilk test, a comparative test was conducted using parametric and non-parametric methods. Qualitative analysis results were compared between the different sequences using the Kruskal–Wallis test and Mann–Whitney U test for healthy volunteers and patients, respectively. The interrater agreement between the three radiologists was evaluated on a total of 94 images from all study participants (n = 32) using the weighted kappa (κ). A κ value ≤ 0.20 indicated slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–0.99, almost perfect agreement [16]. Quantitative volumetric results were compared using an analysis of variance and independent two-sample t-tests for healthy volunteers and patients, respectively. A Bland–Altman analysis was performed to assess the quantitative volumetric biomarker equivalence of the datasets. Statistical analyses were performed using SAS software (version 9.4, SAS Institute). Results with a two-tailed P value < 0.05 were considered statistically significant. The Dwass–Steel–Critchlow–Fligner correction method was used to prevent an increase in type 1 errors in multiple testing using the post hoc method for the Kruskal–Wallis test. Results By applying CS factors of 2 and 4 to the reference standard, the scan time was reduced to 65.4% and 33.5%, respectively. Comparing CS2 and CS4, CS4 enables faster scanning, with a scan time which is 51.2% of that of CS2. Qualitative results The weighted κ values that assessed interrater agreement showed almost perfect agreement in anatomic conspicuity, overall image quality, artifacts, SNR, and sharpness (ranging from 0.844 to 0.926, Table 1). In healthy volunteers, applying a CS factor of 2 to the reference standard significantly degraded the quality of all metrics (P-values < 0.05) except SNR (P value = 0.241). When DLR was applied to CS2, it significantly improved the quality of all metrics to the extent of the reference standard (median score of 5 for all metrics, [IQR, 4–5], Table 2). CS4 exhibited the worst performance among the five sequences, with median scores of 3 for all metrics, which were significantly lower than those of CS2 (all P-values <.001, Table 2). CS4 images frequently showed blurriness between the cortical gray matter, subcortical white matter, and grainy artifacts in the white matter (Figure 1). When DLR was applied to CS4 (DLR-CS4), the quality of all metrics was significantly improved, compared with CS4 (all P-values <.001), with decreased artifacts (median 4, [IQR 4–5]), improved anatomic conspicuity (median 4, [IQR 4–4]), overall image quality (median 4, [IQR 4–4]), sharpness (median 4, [IQR 4–4]), and increased SNR (median 4.5 [IQR 4-5]). In addition, when comparing DLR-CS4 with CS2, DLR-CS4 achieved comparable quality in anatomic conspicuity (DLR-CS4 vs. CS2, median 4 [IQR, 4–4] vs. median 4 [IQR, 4–4]), artifacts (median 4 [IQR, 4–5] vs. median 5 [IQR, 4–5]), and sharpness (median 4 [IQR, 4–4] vs. median 4 [IQR, 4–4]). However, the overall image quality was significantly better with CS2 (median 5 [IQR 5–5]) than with DLR-CS4 (median 4 IQR [4–4]). SNR was also significantly higher in CS2 (median 5 [IQR 5–5]) than DLR-CS4 (median 4.5 IQR [4–5]). (P = 0.021, Table 2 and Fig 1). The detailed P values from comparisons of qualitative analysis results among sequences are presented in the Supplementary Table S1. In the patient group, DLR-CS4 achieved comparable quality for all metrics, compared with CS2, despite a scan time reduction by approximately half, with all median scores ranging from 4 to 5 (Table 3, Fig 2). Anatomic conspicuity between pathologies, such as old infarctions or ischemic lesions and normal adjacent white matter was compensated for in DLR-CS4 by applying DLR, which yielded comparable scores of anatomic conspicuity between CS2 and DLR-CS4. Diffuse and grainy artifacts, which were frequently observed in CS4, were also decreased by applying DLR to CS4 at a level similar to that of CS2. SNR decrease which was observed in CS4 was also increased by applying DLR (Fig 2). Quantitative results The volumes of LV, hippocampus, CP, and WMH were not significantly different among the different 3D T1 sequences in both healthy volunteers and patients (P values = 0.972–0.999) (Table 4). The Bland–Altman plot graph analysis further demonstrated a strong agreement between the quantitative values from CS2 and DLR-CS4 (Fig 3). Discussion We investigated whether compressed sensing-based DLR is a feasible tool for brain imaging by applying different acceleration factors with and without DLR and comparing different sequences for both healthy volunteers and patients. The major findings were as follows: (1) DL-reconstructed CS2 images achieved comparable and excellent quality for all metrics compared with the reference standard, with an approximately 60% reduced scan time; (2) DL-reconstructed CS4 images achieved a quality comparable to CS2 images in approximately half the scan time in both healthy volunteers and patients; and (3) volumetric analysis results did not differ significantly when either different accelerations or DLR were applied. These findings suggest that DLR enables acceptable quality in a significantly reduced scan time and provides accurate volumetric analysis results, which makes it a feasible tool for integration into clinical settings for brain imaging. Recently, various commercially available DLR tools have been developed and have attracted interest because they produce high-quality images with improved SNR in reduced scan times [ 17 ]. Studies have investigated the reliability of the DLR tools for general clinical use in various imaging fields, and several DL-based reconstructions applied to brain MRI showed promising results, by improving the image quality or aiding diagnosis in brain imaging. The vendor-agnostic DICOM-based DL-enhancement software has proven its value because it provides acceptable image quality for 3D brain MRI sequences with a significant scan time reduction in both healthy controls and patients [ 18 , 19 ]. One of the vendor-driven DLRs which uses CNN to reconstruct the image directly from k- space data has been applied to the sella MRI in the postoperative setting after pituitary adenoma surgery and compared with a 3-mm sequence, which revealed that the 1-mm sequence with DLR showed higher diagnostic performance in identifying cavernous sinus invasion and comparable diagnostic performance in identifying residual tumors [ 20 ]. This software has also been validated for use in pediatric brain MRI; it can reduce noise and truncation artifacts, and improve lesion conspicuity and overall image quality in T2-weighted images [ 21 ]. Another vendor-driven DLR tool based on Adaptive-CS-Network been recently developed and its clinical utility has been proved in recent clinical MRI studies of the head and neck [ 22 ], prostate[ 9 ], musculoskeletal area [ 10 , 11 ], breasts [ 12 ], heart [ 13 ], and MR cholangiopancreatography [ 14 ]; however, little is known about its usefulness in the field of brain imaging. Because the validation of Adaptive-CS-Network DLR is still a prerequisite for its clinical use in brain imaging, we applied it to 3D T1 images in brain MRI, with different acceleration factors. For healthy volunteers, 3D T1 images with different CS factors (0, reference standard; 2, and 4) were obtained to evaluate the extent to which DLR can improve the image quality in an accelerated scan, with different CS factors. For patients, an accelerated 3D T1 with a CS factor of 4 was additionally obtained for comparison with a standard 3D T1 with a CS factor of 2. We found that, despite a significant scan time reduction by increasing the CS factor, image quality was maintained by applying DLR. Furthermore, the accuracy of the volumetric analysis results using 3D T1 was maintained when DLR was applied. Considering the high clinical demands for brain MRI acquisition in a hectic schedule, DLR can be suggested as a feasible and efficient tool for integration into brain MRI to reduce the scan time without sacrificing image quality. Because acquiring 3D T1 images without the CS factor (reference standard) is not practical due to its long scan time (9 min 16 s), CS2 was the standard 3D T1 sequence in our institution. It is clinically important to determine whether we can further reduce the scan time by approximately half by increasing the CS factor to 4 while simultaneously maintaining image quality by applying DLR. We observed that the qualitative analysis results of images acquired using CS2 and DLR-CS4 were comparable between healthy volunteers and patients. For healthy volunteer scans, the overall image quality and SNR of CS2 was significantly better than that of DLR-CS4 and all other quality metrics were comparable between the two sequences. Considering that all the quality metrics of CS4 were significantly worse than those of CS2, DLR had a significant impact on the improvement in the qualitative analysis results. The values of denoising and improved image quality were validated in our patients’ scans, and all quality metrics showed no significant differences between CS2 and DLR-CS4. Therefore, DLR can be safely applied to the standard 3D T1 sequence while reducing the scan time by approximately half, by increasing the CS factor. The DLR tool used in this study achieved improved image quality through iterative image processing using a DL-based image-sparsifying approach for denoising and artifact [ 7 ]. In our study, we observed that the artifacts were substantially reduced by applying DLR. DLR-CS4 frequently showed diffuse, grainy artifacts across the images and throughout the white matter, resulting in a median score of 3 in the qualitative assessment. In general, the introduction of noise and artifacts is inevitable when scan time is reduced by increasing the CS factor. These artifacts decreased when DLR was applied, thereby increasing the score to a level similar to that of CS2. For patient scans, artifacts were present in both 3D T1 with CS2 and DLR-CS4; however, they did not hinder brain pathology or interrupt the reading of MRIs. Therefore, we suggest that DLR can successfully remove the artifacts. Several reports suggest that DLR denoising itself can cause specific artifacts or may enhance artifacts due to the high signal-to-noise ratio of DLR [ 17 ]. However, we did not notice any peculiar artifacts caused by or related to DLR in our study. Further studies focusing on the evaluation of the prominent artifacts caused by DLR denoising should be performed. Quantitative volumetric MR analytical tools are widely used to evaluate patients with variable clinical diagnoses, including neurodegenerative diseases. Accurate results are important for clinical and research purposes. FreeSurfer, the most popular software for volumetric analysis, provides segmentation masks and calculates the volumes for detailed brain subregions. Therefore, it is necessary to validate whether DL-enhanced 3D T1 images provide accurate volumetric analysis results using FreeSurfer, as DLR will become more widely used in the future. In our study, we observed no significant differences in the volumes of the brain subregions obtained from different sequences (different acceleration factors with and without DLR). In particular, we chose the hippocampus, WMH, and CP for volumetric analysis together with the lateral ventricles and ICV, as these are frequently investigated in various neurological diseases, including neurodegenerative diseases [ 23 – 25 ]. Therefore, it can be concluded that DLR does not affect the accuracy of volumetric analysis results from 3D T1-weighted images, and that the volumes calculated from DL-enhanced T1 images can be safely used for both clinical and research purposes. This study has some limitations. First, this was a prospective single-center study with a relatively small number of patients. Our results need to be validated in future studies with larger sample sizes and multiple institutions. Second, we only focused on 3D high-resolution T1-weighted images, which are the fundamental sequences for the evaluation of brain structural anatomy, midline structures, and volumetric analysis. Because 3D T2-weighted and fluid-attenuated inversion recovery images are essential for the evaluation of brain pathology, the application of DLR to these sequences should be evaluated in the future. Third, we did not calculate the values of SNR or contrast-to-noise ratio for the specific lesion in our study. Instead, we assessed image SNR and sharpness by radiologists in the qualitative analysis, and observed DLR increased SNR and maintained sharpness. Fourth, whether applying CS-based DLR in a same scan time can improve diagnostic performance in patients with a specific disease was not evaluated in this study. Rather, we focused on the efficiency that DLR might provide, whether it can maintain image quality in a significant reduced scan time, therefore, can be a good solution for tight MRI schedules. In future studies, this CS-based DLR can be applied to patients with a specific disease and can be validated whether it can improve diagnostic performance or help neuroradiologists’ decisions. In conclusion, the novel DLR tool based on Adaptive-CS-Network in combination with an accelerated MRI protocol allows significant scan time reduction while maintaining image quality and high volumetric quantification accuracy compared with the standard sequence. Our study supports the reliability and efficiency of DL-based reconstruction, which can be safely incorporated into clinical practice. Abbreviations CNN Convolutional neural network CP Choroid plexus CS Compressed sensing DL Deep learning DLR Deep learning reconstruction ICV Intracranial volume ISTA Iterative shrinkage-thresholding algorithm LV Lateral ventricle WMH White matter hypointensity Declarations Ethics approval and consent to participate The Institutional Review Board of our tertiary institution approved this prospective study (No. 9-2024-0091). Informed consents were obtained from all participants. This study adhered both the Declaration of Helsinki and the Health Insurance Portability and Accountability Act (HIPAA). Consent for publication Not applicable Competing interests The authors declare no competing interests. Funding The authors did not receive support from any organization for the submitted work. Author Contribution All authors read and approved the final manuscript. YK: Data curation, Formal analysis, Investigation, Methodology, Visualization, and Writing – original draft. SK: Conceptualization, Formal analysis, Investigation, Methodology, Project administration, and Software. JHK: Data curation, Supervision, Validation, Visualization, and Writing – review and editing. NS: Formal analysis, Investigation, Methodology, Resources, Software, and Writing – review and editing. CJP: Conceptualization, Investigation, Project administration, Resources, Validation, and Writing – review and editing. Acknowledgements Not applicable. Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. References Liu F, Duan Y, Peterson BS, Kangarlu A: Compressed sensing MRI combined with SENSE in partial k-space . Phys Med Biol 2012, 57 (21):N391-403. Tsao J, Kozerke S: MRI temporal acceleration techniques . J Magn Reson Imaging 2012, 36 (3):543-560. Ke Z, Cheng J, Ying L, Zheng H, Zhu Y, Liang D: An unsupervised deep learning method for multi-coil cine MRI . Phys Med Biol 2020, 65 (23):235041. Kidoh M, Shinoda K, Kitajima M, Isogawa K, Nambu M, Uetani H, Morita K, Nakaura T, Tateishi M, Yamashita Y et al : Deep Learning Based Noise Reduction for Brain MR Imaging: Tests on Phantoms and Healthy Volunteers . Magn Reson Med Sci 2020, 19 (3):195-206. Kim M, Kim HS, Kim HJ, Park JE, Park SY, Kim YH, Kim SJ, Lee J, Lebel MR: Thin-Slice Pituitary MRI with Deep Learning-based Reconstruction: Diagnostic Performance in a Postoperative Setting . Radiology 2021, 298 (1):114-122. Yasaka K, Akai H, Sugawara H, Tajima T, Akahane M, Yoshioka N, Kabasawa H, Miyo R, Ohtomo K, Abe O et al : Impact of deep learning reconstruction on intracranial 1.5 T magnetic resonance angiography . Jpn J Radiol 2022, 40 (5):476-483. Pezzotti N, Yousefi S, Elmahdy MS, Van Gemert JHF, Schuelke C, Doneva M, Nielsen T, Kastryulin S, Lelieveldt BP, Van Osch MJ: An adaptive intelligence algorithm for undersampled knee MRI reconstruction . IEEE Access 2020, 8 :204825-204838. Fuin N, Bustin A, Küstner T, Oksuz I, Clough J, King AP, Schnabel JA, Botnar RM, Prieto C: A multi-scale variational neural network for accelerating motion-compensated whole-heart 3D coronary MR angiography . Magn Reson Imaging 2020, 70 :155-167. Bischoff LM, Peeters JM, Weinhold L, Krausewitz P, Ellinger J, Katemann C, Isaak A, Weber OM, Kuetting D, Attenberger U et al : Deep Learning Super-Resolution Reconstruction for Fast and Motion-Robust T2-weighted Prostate MRI . Radiology 2023, 308 (3):e230427. Foreman SC, Neumann J, Han J, Harrasser N, Weiss K, Peeters JM, Karampinos DC, Makowski MR, Gersing AS, Woertler K: Deep learning-based acceleration of Compressed Sense MR imaging of the ankle . Eur Radiol 2022, 32 (12):8376-8385. Feuerriegel GC, Weiss K, Kronthaler S, Leonhardt Y, Neumann J, Wurm M, Lenhart NS, Makowski MR, Schwaiger BJ, Woertler K et al : Evaluation of a deep learning-based reconstruction method for denoising and image enhancement of shoulder MRI in patients with shoulder pain . European Radiology 2023, 33 (7):4875-4884. Yang F, Pan X, Zhu K, Xiao Y, Yue X, Peng P, Zhang X, Huang J, Chen J, Yuan Y et al : Accelerated 3D high-resolution T2-weighted breast MRI with deep learning constrained compressed sensing, comparison with conventional T2-weighted sequence on 3.0 T . European Journal of Radiology 2022, 156 :110562. Wu X, Tang L, Li W, He S, Yue X, Peng P, Wu T, Zhang X, Wu Z, He Y et al : Feasibility of accelerated non-contrast-enhanced whole-heart bSSFP coronary MR angiography by deep learning–constrained compressed sensing . European Radiology 2023, 33 (11):8180-8190. Zhang Y, Peng W, Xiao Y, Ming Y, Ma K, Hu S, Zeng W, Zeng L, Liang Z, Zhang X et al : Rapid 3D breath-hold MR cholangiopancreatography using deep learning–constrained compressed sensing reconstruction . European Radiology 2023, 33 (4):2500-2509. Zhang J, Ghanem B: ISTA-Net: Interpretable Optimization-Inspired Deep Network for Image Compressive Sensing . In: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition: 18-23 June 2018 2018 ; 2018: 1828-1837. Fleiss JL: Measuring nominal scale agreement among many raters . Psychological bulletin 1971, 76 (5):378. Kiryu S, Akai H, Yasaka K, Tajima T, Kunimatsu A, Yoshioka N, Akahane M, Abe O, Ohtomo K: Clinical Impact of Deep Learning Reconstruction in MRI . RadioGraphics 2023, 43 (6):e220133. Bash S, Wang L, Airriess C, Zaharchuk G, Gong E, Shankaranarayanan A, Tanenbaum LN: Deep Learning Enables 60% Accelerated Volumetric Brain MRI While Preserving Quantitative Performance: A Prospective, Multicenter, Multireader Trial . AJNR Am J Neuroradiol 2021, 42 (12):2130-2137. Rudie JD, Gleason T, Barkovich MJ, Wilson DM, Shankaranarayanan A, Zhang T, Wang L, Gong E, Zaharchuk G, Villanueva-Meyer JE: Clinical Assessment of Deep Learning-based Super-Resolution for 3D Volumetric Brain MRI . Radiol Artif Intell 2022, 4 (2):e210059. Kim M, Kim HS, Kim HJ, Park JE, Park SY, Kim Y-H, Kim SJ, Lee J, Lebel MR: Thin-Slice Pituitary MRI with Deep Learning–based Reconstruction: Diagnostic Performance in a Postoperative Setting . Radiology 2021, 298 (1):114-122. Kim S-H, Choi YH, Lee JS, Lee SB, Cho YJ, Lee SH, Shin S-M, Cheon J-E: Deep learning reconstruction in pediatric brain MRI: comparison of image quality with conventional T2-weighted MRI . Neuroradiology 2023, 65 (1):207-214. Fujima N, Nakagawa J, Kameda H, Ikebe Y, Harada T, Shimizu Y, Tsushima N, Kano S, Homma A, Kwon J et al : Improvement of image quality in diffusion-weighted imaging with model-based deep learning reconstruction for evaluations of the head and neck . Magnetic Resonance Materials in Physics, Biology and Medicine 2023. Brewer JB, Magda S, Airriess C, Smith ME: Fully-automated quantification of regional brain volumes for improved detection of focal atrophy in Alzheimer disease . AJNR Am J Neuroradiol 2009, 30 (3):578-580. Jeong SH, Park CJ, Jeong HJ, Sunwoo MK, Ahn SS, Lee SK, Lee PH, Kim YJ, Sohn YH, Chung SJ: Association of choroid plexus volume with motor symptoms and dopaminergic degeneration in Parkinson's disease . J Neurol Neurosurg Psychiatry 2023, 94 (12):1047-1055. Shen Q, Loewenstein DA, Potter E, Zhao W, Appel J, Greig MT, Raj A, Acevedo A, Schofield E, Barker W et al : Volumetric and visual rating of magnetic resonance imaging scans in the diagnosis of amnestic mild cognitive impairment and Alzheimer's disease . Alzheimers Dement 2011, 7 (4):e101-108. Tables Table 1. Interrater agreement Total 94 images from both volunteers + patients (n=32) Rater A Rater B Kappa 95% confidence interval P value Anatomic conspicuity Rater 1 Rater 2 0.888 (0.806, 0.971) <.001 Rater 1 Rater 3 0.904 (0.828, 0.981) <.001 Rater 2 Rater 3 0.920 (0.850, 0.999) <.001 Overall image quality Rater 1 Rater 2 0.926 (0.863, 0.989) <.001 Rater 1 Rater 3 0.924 (0.858, 0.990) <.001 Rater 2 Rater 3 0.880 (0.801 0.959) <.001 Artifacts Rater 1 Rater 2 0.923 (0.856, 0.990) <.001 Rater 1 Rater 3 0.865 (0.779, 0.951) <.001 Rater 2 Rater 3 0.910 (0.840, 0.980) <.001 Signal-to-noise ratio Rater 1 Rater 2 0.922 (0.845, 0.995) <.0001 Rater 1 Rater 3 0.883 (0.794, 0.973) <.0001 Rater 2 Rater 3 0.851 (0.752, 0.951) <.0001 Sharpness Rater 1 Rater 2 0.890 (0.811, 0.969) <.001 Rater 1 Rater 3 0.844 (0.752, 0.936) <.001 Rater 2 Rater 3 0.892 (0.813, 0.971) <.001 Table 2. Multirater quantitative assessments in volunteers Rating criterion Reference standard CS2 DLR-CS2 CS4 DLR-CS4 P value Median IQR Median IQR Median IQR Median IQR Median IQR CS2 vs. CS4 CS2 vs. DLR-CS4 Anatomic conspicuity 5 (5,5) 4 (4,4) 5 (5,5) 3 (3,3) 4 (4,4) <.001 0.241 Overall image quality 5 (5,5) 5 (5,5) 5 (5,5) 3 (3,3) 4 (4,4) <.001 <.001 Artifacts 5 (5,5) 5 (4,5) 5 (5,5) 3 (3,3) 4 (4,5) <.001 0.123 Signal-to-noise ratio 5 (5,5) 5 (5,5) 5 (5,5) 3 (3,4) 4.5 (4,5) <.001 0.021 Sharpness 5 (5,5) 4 (4,4) 5 (5,5) 3 (3,3) 4 (4,4) <.001 0.887 CS2, compressed sensing factor 2 without deep learning reconstruction; DLR-CS2, compressed sensing factor 2 with deep learning reconstruction; CS4, compressed sensing factor 4 without deep learning reconstruction; DLR-CS4, compressed sensing factor 4 with deep learning reconstruction; IQR, interquartile range Table 3. Multirater quantitative assessments in patients Rating criterion CS2 DLR-CS4 P value Median IQR Median IQR Anatomic conspicuity 4.5 (4, 5) 4 (4, 5) 0.864 Overall image quality 4 (4, 5) 4 (4, 5) 0.188 Artifacts 4 (4 ,5) 4 (4, 4) 0.226 Signal-to-noise ratio 5 (4, 5) 5 (4, 5) 0.320 Sharpness 4 (4, 5) 4 (4, 5) 0.057 CS2, compressed sensing factor 2 without deep learning reconstruction; DLR-CS4, compressed sensing factor 4 with deep learning reconstruction; IQR, interquartile range Table 4. Quantitative assessments in both volunteers and patients Regional volumes Reference standard CS2 DLR-CS2 CS4 DLR-CS4 P value Mean SD Mean SD Mean SD Mean SD Mean SD Volunteers LV volume, ratio of ICV × 10 3 0.011 0.004 0.011 0.004 0.011 0.004 0.011 0.004 0.011 0.004 >.999 * Hippocampal volume, ratio of ICV × 10 3 0.008 0.002 0.008 0.002 0.008 0.002 0.008 0.002 0.008 0.002 0.999 * CP volume, ratio of ICV × 10 3 0.001 0.000 0.001 0.000 0.001 0.000 0.001 0.000 0.001 0.000 0.999 * WMH volume, ratio of ICV × 10 3 0.001 0.000 0.001 0.000 0.001 0.000 0.001 0.000 0.001 0.000 0.960 * Patients LV volume, ratio of ICV × 10 3 0.028 0.012 0.028 0.012 0.938 † Hippocampal volume, ratio of ICV × 10 3 0.015 0.006 0.015 0.006 0.943 † CP volume, ratio of ICV × 10 3 0.001 0.000 0.001 0.000 0.884 † WMH volume, ratio of ICV × 10 3 0.004 0.005 0.004 0.005 0.899 † LV, lateral ventricle; CP, choroid plexus; WMH, white matter hypointensity; ICV, intracranial volume; CS2, compressed sensing factor 2 without deep learning reconstruction; DLR-CS2, compressed sensing factor 2 with deep learning reconstruction; CS4, compressed sensing factor 4 without deep learning reconstruction; DLR-CS4, compressed sensing factor 4 with deep learning reconstruction; SD, standard deviation Calculated from * ANOVA and † independent two sample t-test Additional Declarations No competing interests reported. Supplementary Files Additionalfile1.docx Cite Share Download PDF Status: Published Journal Publication published 27 Mar, 2025 Read the published version in BMC Medical Imaging → Version 1 posted Editorial decision: Revision requested 06 Nov, 2024 Editor assigned by journal 06 Nov, 2024 Submission checks completed at journal 31 Oct, 2024 First submitted to journal 28 Oct, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5345149","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":374685182,"identity":"a41afba0-080c-4070-8e44-24a0a080facf","order_by":0,"name":"Yeseul Kang","email":"","orcid":"","institution":"Yonsei University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yeseul","middleName":"","lastName":"Kang","suffix":""},{"id":374685187,"identity":"060f0237-5114-4da7-a580-52b0161a15d6","order_by":1,"name":"Sang-Young Kim","email":"","orcid":"","institution":"Philips Healthcare","correspondingAuthor":false,"prefix":"","firstName":"Sang-Young","middleName":"","lastName":"Kim","suffix":""},{"id":374685188,"identity":"ab335f98-8dfa-4660-91de-11a37807b238","order_by":2,"name":"Jun Hwee Kim","email":"","orcid":"","institution":"Yonsei University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jun","middleName":"Hwee","lastName":"Kim","suffix":""},{"id":374685192,"identity":"062c5fd7-0e84-4bc0-99e6-3be1938de6d3","order_by":3,"name":"Nak-Hoon Son","email":"","orcid":"","institution":"Keimyung University","correspondingAuthor":false,"prefix":"","firstName":"Nak-Hoon","middleName":"","lastName":"Son","suffix":""},{"id":374685194,"identity":"4c18ae5f-614d-4380-817a-7d5b837c38af","order_by":4,"name":"Chae Jung Park","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3klEQVRIiWNgGAWjYDACCQbGA2AGD/MBBsYG4rQwQLWwJZCshceAOC38s5sPHPxRYZdn3nPmmzTvjjoG/vYDBCy5cyzhMM+Z5GKZs73bpHnPHGaQOJOAX4uBRI7BYcY25sQZ/LxALW1AG24QcJiBRP6Hgz//1QO18DwDaqljkCesJYfhAG/D4cQZvD1sQC3MDAaEtEjcSDM4zHPseLEEzzFjy7lth3kMCfmFf0byw4c/aqrzJHiSH95421YnJ3f8AAFroABkMgswjhh4iFMP1cL8gWjlo2AUjIJRMKIAAAA2RI+nWaVMAAAAAElFTkSuQmCC","orcid":"","institution":"Yonsei University College of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Chae","middleName":"Jung","lastName":"Park","suffix":""}],"badges":[],"createdAt":"2024-10-28 08:08:03","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5345149/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5345149/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12880-025-01647-8","type":"published","date":"2025-03-27T15:56:55+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":70296327,"identity":"5f5c889c-d815-4d09-b249-4ab66f95d81f","added_by":"auto","created_at":"2024-12-02 00:37:01","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2852411,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative figures from a healthy volunteer (upper row: coverage of the whole brain; lower low: magnification of cortical/subcortical areas). Due to denoising of deep learning reconstruction (DLR), the signal-to-noise ratio significantly increased in CS2 with DLR (DLR-CS2) compared with CS2 without DLR and in CS4 with DLR (DLR-CS4) compared with CS4 without DLR. DLR-CS2 provided images with quality comparable to that of the reference standard. CS4 without DLR showed the worst results for qualitative analysis among the five sequences, with notable artifacts in the white matter and reduced sharpness in the border between the gray and white matter. Applying DLR to CS4 (DLR-CS4) enhanced the image quality by decreasing artifacts in the white matter and increasing sharpness in the border between gray and white matter, making its quality comparable to that of CS2. It is notable that DLR-CS4 and CS2 achieved similar qualitative analysis results, whereas the scan time of DLR-CS4 was reduced by approximately half (2 min 6 s vs. 4 min 6 s)\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-5345149/v1/e9850dd909937960b5434602.png"},{"id":70296328,"identity":"d9397cbd-61e1-4963-86ea-1fa1553bac13","added_by":"auto","created_at":"2024-12-02 00:37:01","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":8299075,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative figures from three patients (left: 3D T1 with CS factor of 2 without DLR [CS2]; right: 3D T1 with CS factor of 4 with DLR [DLR-CS4]). (a) Patient 1: All the metrics were comparable as per the qualitative analysis results, including artifacts diffusely scattered in the white matters and gray-white matter differentiation. (b) Patient 2: The delineation of old infarctions in the corona radiata did not differ significantly between the two sequences. (c) Patient 3: Diffuse, grainy artifacts throughout the white matter were significantly worse in DLR-CS4 compared with CS2; meanwhile the other metrics did not differ significantly\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-5345149/v1/9f6737cb95e712fd69fe0c28.png"},{"id":70296326,"identity":"26e323fd-93e5-4db8-b05c-5d359618756e","added_by":"auto","created_at":"2024-12-02 00:37:01","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":629448,"visible":true,"origin":"","legend":"\u003cp\u003eBland–Altman results for CS2 versus DLR-CS4. The plots demonstrate a linear distribution without significant scatter, indicating consistent concordance between CS2 and DLR-CS4 in the quantitative assessment of volumes of lateral ventricle (LV), hippocampus, choroid plexus (CP), and white matter hypointensity (WMH)\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-5345149/v1/43eaa6b3bb48da2555050b10.png"},{"id":79604891,"identity":"e93f6222-c4b1-4044-9dd1-dafe4b250ac8","added_by":"auto","created_at":"2025-03-31 16:08:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":12639720,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5345149/v1/b25a36de-47f2-48f1-a10b-e2e82ffdc6cd.pdf"},{"id":70296325,"identity":"e8dea9b1-ab91-48aa-aa4f-57b8916924f0","added_by":"auto","created_at":"2024-12-02 00:37:01","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":21226,"visible":true,"origin":"","legend":"","description":"","filename":"Additionalfile1.docx","url":"https://assets-eu.researchsquare.com/files/rs-5345149/v1/8ea9d789327f73eb37f81c80.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Deep Learning-Based Reconstruction for Three-Dimensional Volumetric Brain MRI: A Qualitative and Quantitative Assessment","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThree-dimensional (3D) high-resolution structural MRI is fundamental to evaluate various neurological diseases. 3D T1-weighted imaging is a basic MRI sequence used for the evaluation of structural anatomy and volumetric analysis. To obtain acceptable image quality in a limited scan time, many accelerating techniques have been developed, including parallel imaging, compressed sensing, and partial Fourier [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. However, these techniques frequently and inevitably introduce noise and degrade spatial resolution.\u003c/p\u003e \u003cp\u003eRecently, deep learning (DL) has been suggested for the reconstruction of accelerated MR scans as well as for image denoising, gaining increased attention in recent years [\u003cspan additionalcitationids=\"CR4 CR5\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. MRI scanners equipped with DL reconstruction (DLR) for denoising are already available for clinical use. One of the recently developed DLR methods called Adaptive-Compressed sensing (CS)-Network utilizes a novel convolutional neural networks (CNN) to integrate and enhance conventional CS-SENSE algorithms (i.e., CS\u0026thinsp;+\u0026thinsp;SENSE) [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. As the traditional CS-based reconstruction results in compromised diagnostic value due to difficulties in choosing the optimal sparsity transforms [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], DL has replaced wavelet-based sparsity transform with multi-scale sparsifying transform in this novel method. With this strategy, the performance of denoising can be improved through automatically optimized the sparsity constraint during the iterative reconstruction.\u003c/p\u003e \u003cp\u003eRecent clinical studies using either the prototype or commercial versions of the DLR based on Adaptive-CS-Network have demonstrated improved image quality in the imaging studies of the prostate [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], musculoskeletal area [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], breasts [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], heart [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], and MR cholangiopancreatography [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] and proved its feasibility; however, it has not been validated yet in the field of neuroradiology. We, therefore, aimed to evaluate the performance of a novel DLR tool in adult brain 3D T1-weighted images (T1) with quantitative and qualitative assessments and to validate its feasibility and clinical usefulness in a prospective setting.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003e\u003cstrong\u003eParticipants\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Institutional Review Board of our tertiary institution approved this prospective study (No. 9-2024-0091). Informed consents were obtained from all participants. This study adhered both the Declaration of Helsinki and the Health Insurance Portability and Accountability Act (HIPAA). In this prospective study, 10 healthy volunteers (median age: 38 [interquartile range, IQR, 31\u0026ndash;39] years; male:female, 4:6) and 22 consecutive patients (median age: 76 [IQR 74\u0026ndash;79] years; male:female, 9:13) undergoing clinical brain MRI at our tertiary institution were recruited. The patients were imaged for a variety of clinical indications: subjective cognitive decline (n = 12), mild cognitive impairment (n = 1), Alzheimer\u0026rsquo;s disease (n = 2), Parkinson\u0026rsquo;s disease (n = 1), infarction (n = 2), and other conditions, including visual disturbance, trauma, carotid artery stenosis, and hydrocephalus.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMRI acquisition\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll MRI examinations were performed using a 3T MR scanner (Ingenia Elition; Philips Healthcare, Amsterdam, Netherlands). Different imaging protocols were used for healthy volunteers and patients. Detailed MRI parameters are presented in Supplementary Material. For the healthy volunteers, five different 3D T1-weighted images (T1) were obtained as follows: 1) 3D T1 without CS factor (reference standard, 9 min 16 s); 2) 3D T1 with CS factor of 2 without DLR (CS2, 4 min 6 s); 3) 3D T1 with CS factor of 2 with DLR (DLR-CS2, 4 min 6 s); 4) 3D T1 with CS factor of 4 without DLR (CS4, 2 min 6 s); and 5) 3D T1 with CS factor of 4 with DLR (DLR-CS4, 2 min 6 s). Twenty-two patients underwent brain MRI, including imaging with two 3D T1 sequences, as follows: 1) CS2 (4 min 6 s) and 2) DLR-CS4 (2 min 6 s). CS2 is a standard brain MRI sequence used at our institution. Therefore, patients additionally underwent a faster 3D T1 scan by increasing the CS factor to 4, which was then reconstructed with DLR (DLR-CS4) to evaluate the feasibility of DLR in the patient population.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDLR\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe DLR used in this study builds on parallel imaging SENSE combined with CS, CS-SENSE in which the prior information such as coil sensitivity profile and low-resolution background information is used for data reliability during the iterative reconstruction. The applied DL, Adaptive-CS-Network was first introduced for the FastMRI challenge to reconstruct under-sampled knee MRI data [7] and it is now extended to all image contrasts and application domains. Briefly, the Adaptive-CS-Net is an advancement of the deep learning\u0026ndash;based iterative shrinkage-thresholding algorithm (ISTA) network proposed by Zhang and Ghanem [15]. The power of Adaptive-CS-Net comes from the use of multi-scale and multi-slice computations in iterative learning-based reconstruction scheme in which MR images are efficiently denoised by multiple reconstruction blocks. Similar to CS-SENSE reconstruction, the network is fed by raw k-space data, coil sensitivity and coarse background information, but the sparsity constraining step of the iterative reconstruction is DL-based. In each individual block of the network, data consistency checking with the incoming raw k-space data is performed. The Adaptive-CS-Net used in this study was initially trained with a large dataset from various anatomical regions and acceleration factors. The algorithm was refined to run on standard reconstruction hardware, in contrast with a previously reported network [7]. Reconstruction was performed on the scanner and required approximately 80 s for the CS DL reconstructions, compared with approximately 40 s for the standard CS reconstructions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQualitative analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo qualitatively compare image quality among different sequences, two neuroradiologists (C.J.P. and J.H.K., with 4 and 5 years of experience in neuroradiology, respectively) and one junior resident evaluated the following four image quality parameters on a 5-point Likert scale (1 = unacceptable; 2 = poor; 3 = acceptable; 4 = good; 5 = excellent): anatomical conspicuity, overall image quality, artifacts, signal-to-noise ratio (SNR), and sharpness. The anatomic conspicuity was assessed by evaluating the delineation of existing pathologies contrast to the adjacent normal brain structures or delineation of subregions of deep gray matter in cases with no abnormalities on 3D T1. Gray-white matter differentiation was used to evaluate sharpness. The reviewers evaluated all image series in the Picture Archiving and Communication Systems database. No restrictions were applied to window-level setting adjustments regarding the time or ability to scroll through the images. The reviewers were blinded to patient information, including disease and scan parameters, to identify the type of sequence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQuantitative analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor quantitative analysis, we obtained the volumes of lateral ventricle (LV), hippocampus, choroid plexus (CP), white matter hypointensity (WMH), and total intracranial volume (ICV) from each sequence using FastSurfer software. FastSurfer is a fast and extensively validated DL pipeline for fully automated processing of structural human brain MRIs (Henschel et al., NeuroImage 2020; 219:117012). This enabled whole-brain segmentation into 95 classes within 50 s per participant. The details and validity thereof can be found elsewhere (https://deep-mi.org/research/fastsurfer/). The regional volume was expressed as the ratio of the regional volume to the total ICV (ratio to the total ICV \u0026times; 10\u003csup\u003e3\u003c/sup\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBased on the results of the Shapiro\u0026ndash;Wilk test, a comparative test was conducted using parametric and non-parametric methods. Qualitative analysis results were compared between the different sequences using the Kruskal\u0026ndash;Wallis test and Mann\u0026ndash;Whitney U test for healthy volunteers and patients, respectively. The interrater agreement between the three radiologists was evaluated on a total of 94 images from all study participants (n = 32) using the weighted kappa (\u0026kappa;). A \u0026kappa; value \u0026le; 0.20 indicated slight agreement; 0.21\u0026ndash;0.40, fair agreement; 0.41\u0026ndash;0.60, moderate agreement; 0.61\u0026ndash;0.80, substantial agreement; and 0.81\u0026ndash;0.99, almost perfect agreement [16]. Quantitative volumetric results were compared using an analysis of variance and independent two-sample t-tests for healthy volunteers and patients, respectively. A Bland\u0026ndash;Altman analysis was performed to assess the quantitative volumetric biomarker equivalence of the datasets. Statistical analyses were performed using SAS software (version 9.4, SAS Institute). Results with a two-tailed P value \u0026lt; 0.05 were considered statistically significant. The Dwass\u0026ndash;Steel\u0026ndash;Critchlow\u0026ndash;Fligner correction method was used to prevent an increase in type 1 errors in multiple testing using the post hoc method for the Kruskal\u0026ndash;Wallis test.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eBy applying CS factors of 2 and 4 to the reference standard, the scan time was reduced to 65.4% and 33.5%, respectively. Comparing CS2 and CS4, CS4 enables faster scanning, with a scan time which is 51.2% of that of CS2.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQualitative results\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe weighted\u0026nbsp;\u0026kappa; values that assessed interrater agreement showed almost perfect agreement in anatomic conspicuity, overall image quality, artifacts, SNR, and sharpness (ranging from 0.844 to 0.926, Table 1).\u0026nbsp;In healthy volunteers, applying\u0026nbsp;a CS factor of 2 to the reference standard significantly degraded the quality\u0026nbsp;of all metrics (P-values \u0026lt; 0.05) except SNR (P value = 0.241). When DLR was applied to CS2, it significantly improved the quality of all metrics to the extent of\u0026nbsp;the reference standard (median score of 5 for all metrics, [IQR, 4\u0026ndash;5], Table 2). CS4 exhibited the worst performance\u0026nbsp;among\u0026nbsp;the five sequences, with median scores of 3\u0026nbsp;for all metrics, which were significantly lower than those of CS2 (all P-values \u0026lt;.001, Table 2). CS4 images frequently showed blurriness between\u0026nbsp;the cortical gray matter, subcortical white matter, and grainy artifacts in the white matter (Figure 1). When DLR was applied to CS4 (DLR-CS4), the\u0026nbsp;quality of all metrics was significantly improved, compared with CS4 (all P-values \u0026lt;.001), with decreased artifacts (median 4, [IQR 4\u0026ndash;5]), improved anatomic conspicuity (median 4, [IQR 4\u0026ndash;4]), overall image quality (median 4, [IQR 4\u0026ndash;4]), sharpness (median 4, [IQR 4\u0026ndash;4]), and increased SNR (median 4.5 [IQR 4-5]). In addition, when comparing DLR-CS4 with CS2, DLR-CS4 achieved comparable quality in anatomic conspicuity (DLR-CS4 vs. CS2, median 4 [IQR, 4\u0026ndash;4] vs. median 4 [IQR, 4\u0026ndash;4]), artifacts (median 4 [IQR, 4\u0026ndash;5] vs. median 5 [IQR, 4\u0026ndash;5]), and sharpness (median 4 [IQR, 4\u0026ndash;4] vs. median 4 [IQR, 4\u0026ndash;4]). However,\u0026nbsp;the overall image quality was significantly better with CS2 (median 5 [IQR 5\u0026ndash;5])\u0026nbsp;than with DLR-CS4 (median 4 IQR [4\u0026ndash;4]). SNR was also significantly higher in CS2 (median 5 [IQR\u0026nbsp;5\u0026ndash;5]) than DLR-CS4 (median 4.5 IQR\u0026nbsp;[4\u0026ndash;5]). (P = 0.021, Table 2 and Fig 1). The detailed P values from comparisons of qualitative analysis results among sequences are presented in the Supplementary Table S1.\u003c/p\u003e\n\u003cp\u003eIn the patient group, DLR-CS4 achieved comparable quality for all metrics, compared with CS2, despite a scan time reduction by approximately half, with all median scores ranging from 4 to 5 (Table 3, Fig 2). Anatomic conspicuity between pathologies, such as old infarctions or ischemic lesions and normal adjacent white matter was compensated for in DLR-CS4 by applying DLR, which yielded comparable scores of anatomic conspicuity between CS2 and DLR-CS4. Diffuse and grainy artifacts, which were frequently observed in CS4, were also decreased by applying DLR to CS4 at a level similar to that of CS2. SNR decrease which was observed in CS4 was also increased by applying DLR (Fig 2).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQuantitative results\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe volumes of LV, hippocampus, CP, and WMH were not significantly different among the different 3D T1 sequences in both healthy volunteers and patients (P values = 0.972\u0026ndash;0.999) (Table 4). The Bland\u0026ndash;Altman plot graph analysis further demonstrated a strong agreement between the quantitative values from CS2 and DLR-CS4 (Fig 3).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe investigated whether compressed sensing-based DLR is a feasible tool for brain imaging by applying different acceleration factors with and without DLR and comparing different sequences for both healthy volunteers and patients. The major findings were as follows: (1) DL-reconstructed CS2 images achieved comparable and excellent quality for all metrics compared with the reference standard, with an approximately 60% reduced scan time; (2) DL-reconstructed CS4 images achieved a quality comparable to CS2 images in approximately half the scan time in both healthy volunteers and patients; and (3) volumetric analysis results did not differ significantly when either different accelerations or DLR were applied. These findings suggest that DLR enables acceptable quality in a significantly reduced scan time and provides accurate volumetric analysis results, which makes it a feasible tool for integration into clinical settings for brain imaging.\u003c/p\u003e \u003cp\u003eRecently, various commercially available DLR tools have been developed and have attracted interest because they produce high-quality images with improved SNR in reduced scan times [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Studies have investigated the reliability of the DLR tools for general clinical use in various imaging fields, and several DL-based reconstructions applied to brain MRI showed promising results, by improving the image quality or aiding diagnosis in brain imaging. The vendor-agnostic DICOM-based DL-enhancement software has proven its value because it provides acceptable image quality for 3D brain MRI sequences with a significant scan time reduction in both healthy controls and patients [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. One of the vendor-driven DLRs which uses CNN to reconstruct the image directly from \u003cem\u003ek-\u003c/em\u003espace data has been applied to the sella MRI in the postoperative setting after pituitary adenoma surgery and compared with a 3-mm sequence, which revealed that the 1-mm sequence with DLR showed higher diagnostic performance in identifying cavernous sinus invasion and comparable diagnostic performance in identifying residual tumors [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. This software has also been validated for use in pediatric brain MRI; it can reduce noise and truncation artifacts, and improve lesion conspicuity and overall image quality in T2-weighted images [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Another vendor-driven DLR tool based on Adaptive-CS-Network been recently developed and its clinical utility has been proved in recent clinical MRI studies of the head and neck [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], prostate[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], musculoskeletal area [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], breasts [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], heart [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], and MR cholangiopancreatography [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]; however, little is known about its usefulness in the field of brain imaging. Because the validation of Adaptive-CS-Network DLR is still a prerequisite for its clinical use in brain imaging, we applied it to 3D T1 images in brain MRI, with different acceleration factors. For healthy volunteers, 3D T1 images with different CS factors (0, reference standard; 2, and 4) were obtained to evaluate the extent to which DLR can improve the image quality in an accelerated scan, with different CS factors. For patients, an accelerated 3D T1 with a CS factor of 4 was additionally obtained for comparison with a standard 3D T1 with a CS factor of 2. We found that, despite a significant scan time reduction by increasing the CS factor, image quality was maintained by applying DLR. Furthermore, the accuracy of the volumetric analysis results using 3D T1 was maintained when DLR was applied. Considering the high clinical demands for brain MRI acquisition in a hectic schedule, DLR can be suggested as a feasible and efficient tool for integration into brain MRI to reduce the scan time without sacrificing image quality.\u003c/p\u003e \u003cp\u003eBecause acquiring 3D T1 images without the CS factor (reference standard) is not practical due to its long scan time (9 min 16 s), CS2 was the standard 3D T1 sequence in our institution. It is clinically important to determine whether we can further reduce the scan time by approximately half by increasing the CS factor to 4 while simultaneously maintaining image quality by applying DLR. We observed that the qualitative analysis results of images acquired using CS2 and DLR-CS4 were comparable between healthy volunteers and patients. For healthy volunteer scans, the overall image quality and SNR of CS2 was significantly better than that of DLR-CS4 and all other quality metrics were comparable between the two sequences. Considering that all the quality metrics of CS4 were significantly worse than those of CS2, DLR had a significant impact on the improvement in the qualitative analysis results. The values of denoising and improved image quality were validated in our patients\u0026rsquo; scans, and all quality metrics showed no significant differences between CS2 and DLR-CS4. Therefore, DLR can be safely applied to the standard 3D T1 sequence while reducing the scan time by approximately half, by increasing the CS factor.\u003c/p\u003e \u003cp\u003eThe DLR tool used in this study achieved improved image quality through iterative image processing using a DL-based image-sparsifying approach for denoising and artifact [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. In our study, we observed that the artifacts were substantially reduced by applying DLR. DLR-CS4 frequently showed diffuse, grainy artifacts across the images and throughout the white matter, resulting in a median score of 3 in the qualitative assessment. In general, the introduction of noise and artifacts is inevitable when scan time is reduced by increasing the CS factor. These artifacts decreased when DLR was applied, thereby increasing the score to a level similar to that of CS2. For patient scans, artifacts were present in both 3D T1 with CS2 and DLR-CS4; however, they did not hinder brain pathology or interrupt the reading of MRIs. Therefore, we suggest that DLR can successfully remove the artifacts. Several reports suggest that DLR denoising itself can cause specific artifacts or may enhance artifacts due to the high signal-to-noise ratio of DLR [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. However, we did not notice any peculiar artifacts caused by or related to DLR in our study. Further studies focusing on the evaluation of the prominent artifacts caused by DLR denoising should be performed.\u003c/p\u003e \u003cp\u003eQuantitative volumetric MR analytical tools are widely used to evaluate patients with variable clinical diagnoses, including neurodegenerative diseases. Accurate results are important for clinical and research purposes. FreeSurfer, the most popular software for volumetric analysis, provides segmentation masks and calculates the volumes for detailed brain subregions. Therefore, it is necessary to validate whether DL-enhanced 3D T1 images provide accurate volumetric analysis results using FreeSurfer, as DLR will become more widely used in the future. In our study, we observed no significant differences in the volumes of the brain subregions obtained from different sequences (different acceleration factors with and without DLR). In particular, we chose the hippocampus, WMH, and CP for volumetric analysis together with the lateral ventricles and ICV, as these are frequently investigated in various neurological diseases, including neurodegenerative diseases [\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Therefore, it can be concluded that DLR does not affect the accuracy of volumetric analysis results from 3D T1-weighted images, and that the volumes calculated from DL-enhanced T1 images can be safely used for both clinical and research purposes.\u003c/p\u003e \u003cp\u003eThis study has some limitations. First, this was a prospective single-center study with a relatively small number of patients. Our results need to be validated in future studies with larger sample sizes and multiple institutions. Second, we only focused on 3D high-resolution T1-weighted images, which are the fundamental sequences for the evaluation of brain structural anatomy, midline structures, and volumetric analysis. Because 3D T2-weighted and fluid-attenuated inversion recovery images are essential for the evaluation of brain pathology, the application of DLR to these sequences should be evaluated in the future. Third, we did not calculate the values of SNR or contrast-to-noise ratio for the specific lesion in our study. Instead, we assessed image SNR and sharpness by radiologists in the qualitative analysis, and observed DLR increased SNR and maintained sharpness. Fourth, whether applying CS-based DLR in a same scan time can improve diagnostic performance in patients with a specific disease was not evaluated in this study. Rather, we focused on the efficiency that DLR might provide, whether it can maintain image quality in a significant reduced scan time, therefore, can be a good solution for tight MRI schedules. In future studies, this CS-based DLR can be applied to patients with a specific disease and can be validated whether it can improve diagnostic performance or help neuroradiologists\u0026rsquo; decisions.\u003c/p\u003e \u003cp\u003eIn conclusion, the novel DLR tool based on Adaptive-CS-Network in combination with an accelerated MRI protocol allows significant scan time reduction while maintaining image quality and high volumetric quantification accuracy compared with the standard sequence. Our study supports the reliability and efficiency of DL-based reconstruction, which can be safely incorporated into clinical practice.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eCNN Convolutional neural network\u003c/p\u003e\n\u003cp\u003eCP Choroid plexus\u003c/p\u003e\n\u003cp\u003eCS Compressed sensing\u003c/p\u003e\n\u003cp\u003eDL Deep learning\u003c/p\u003e\n\u003cp\u003eDLR Deep learning reconstruction\u003c/p\u003e\n\u003cp\u003eICV Intracranial volume\u003c/p\u003e\n\u003cp\u003eISTA Iterative shrinkage-thresholding algorithm\u003c/p\u003e\n\u003cp\u003eLV Lateral ventricle\u003c/p\u003e\n\u003cp\u003eWMH White matter hypointensity\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e\n\u003cp\u003eThe Institutional Review Board of our tertiary institution approved this prospective study (No. 9-2024-0091). Informed consents were obtained from all participants. This study adhered both the Declaration of Helsinki and the Health Insurance Portability and Accountability Act (HIPAA).\u003c/p\u003e\n\u003ch2\u003eConsent for publication\u003c/h2\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003ch2\u003eCompeting interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThe authors did not receive support from any organization for the submitted work.\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eAll authors read and approved the final manuscript. YK: Data curation, Formal analysis, Investigation, Methodology, Visualization, and Writing \u0026ndash; original draft. SK: Conceptualization, Formal analysis, Investigation, Methodology, Project administration, and Software. JHK: Data curation, Supervision, Validation, Visualization, and Writing \u0026ndash; review and editing. NS: Formal analysis, Investigation, Methodology, Resources, Software, and Writing \u0026ndash; review and editing. CJP: Conceptualization, Investigation, Project administration, Resources, Validation, and Writing \u0026ndash; review and editing.\u003c/p\u003e\n\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003ch2\u003eAvailability of data and materials\u003c/h2\u003e\n\u003cp\u003eThe datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eLiu F, Duan Y, Peterson BS, Kangarlu A: \u003cstrong\u003eCompressed sensing MRI combined with SENSE in partial k-space\u003c/strong\u003e. \u003cem\u003ePhys Med Biol \u003c/em\u003e2012, \u003cstrong\u003e57\u003c/strong\u003e(21):N391-403.\u003c/li\u003e\n\u003cli\u003eTsao J, Kozerke S: \u003cstrong\u003eMRI temporal acceleration techniques\u003c/strong\u003e. \u003cem\u003eJ Magn Reson Imaging \u003c/em\u003e2012, \u003cstrong\u003e36\u003c/strong\u003e(3):543-560.\u003c/li\u003e\n\u003cli\u003eKe Z, Cheng J, Ying L, Zheng H, Zhu Y, Liang D: \u003cstrong\u003eAn unsupervised deep learning method for multi-coil cine MRI\u003c/strong\u003e. \u003cem\u003ePhys Med Biol \u003c/em\u003e2020, \u003cstrong\u003e65\u003c/strong\u003e(23):235041.\u003c/li\u003e\n\u003cli\u003eKidoh M, Shinoda K, Kitajima M, Isogawa K, Nambu M, Uetani H, Morita K, Nakaura T, Tateishi M, Yamashita Y\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eDeep Learning Based Noise Reduction for Brain MR Imaging: Tests on Phantoms and Healthy Volunteers\u003c/strong\u003e. \u003cem\u003eMagn Reson Med Sci \u003c/em\u003e2020, \u003cstrong\u003e19\u003c/strong\u003e(3):195-206.\u003c/li\u003e\n\u003cli\u003eKim M, Kim HS, Kim HJ, Park JE, Park SY, Kim YH, Kim SJ, Lee J, Lebel MR: \u003cstrong\u003eThin-Slice Pituitary MRI with Deep Learning-based Reconstruction: Diagnostic Performance in a Postoperative Setting\u003c/strong\u003e. \u003cem\u003eRadiology \u003c/em\u003e2021, \u003cstrong\u003e298\u003c/strong\u003e(1):114-122.\u003c/li\u003e\n\u003cli\u003eYasaka K, Akai H, Sugawara H, Tajima T, Akahane M, Yoshioka N, Kabasawa H, Miyo R, Ohtomo K, Abe O\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eImpact of deep learning reconstruction on intracranial 1.5 T magnetic resonance angiography\u003c/strong\u003e. \u003cem\u003eJpn J Radiol \u003c/em\u003e2022, \u003cstrong\u003e40\u003c/strong\u003e(5):476-483.\u003c/li\u003e\n\u003cli\u003ePezzotti N, Yousefi S, Elmahdy MS, Van Gemert JHF, Schuelke C, Doneva M, Nielsen T, Kastryulin S, Lelieveldt BP, Van Osch MJ: \u003cstrong\u003eAn adaptive intelligence algorithm for undersampled knee MRI reconstruction\u003c/strong\u003e. \u003cem\u003eIEEE Access \u003c/em\u003e2020, \u003cstrong\u003e8\u003c/strong\u003e:204825-204838.\u003c/li\u003e\n\u003cli\u003eFuin N, Bustin A, K\u0026uuml;stner T, Oksuz I, Clough J, King AP, Schnabel JA, Botnar RM, Prieto C: \u003cstrong\u003eA multi-scale variational neural network for accelerating motion-compensated whole-heart 3D coronary MR angiography\u003c/strong\u003e. \u003cem\u003eMagn Reson Imaging \u003c/em\u003e2020, \u003cstrong\u003e70\u003c/strong\u003e:155-167.\u003c/li\u003e\n\u003cli\u003eBischoff LM, Peeters JM, Weinhold L, Krausewitz P, Ellinger J, Katemann C, Isaak A, Weber OM, Kuetting D, Attenberger U\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eDeep Learning Super-Resolution Reconstruction for Fast and Motion-Robust T2-weighted Prostate MRI\u003c/strong\u003e. \u003cem\u003eRadiology \u003c/em\u003e2023, \u003cstrong\u003e308\u003c/strong\u003e(3):e230427.\u003c/li\u003e\n\u003cli\u003eForeman SC, Neumann J, Han J, Harrasser N, Weiss K, Peeters JM, Karampinos DC, Makowski MR, Gersing AS, Woertler K: \u003cstrong\u003eDeep learning-based acceleration of Compressed Sense MR imaging of the ankle\u003c/strong\u003e. \u003cem\u003eEur Radiol \u003c/em\u003e2022, \u003cstrong\u003e32\u003c/strong\u003e(12):8376-8385.\u003c/li\u003e\n\u003cli\u003eFeuerriegel GC, Weiss K, Kronthaler S, Leonhardt Y, Neumann J, Wurm M, Lenhart NS, Makowski MR, Schwaiger BJ, Woertler K\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eEvaluation of a deep learning-based reconstruction method for denoising and image enhancement of shoulder MRI in patients with shoulder pain\u003c/strong\u003e. \u003cem\u003eEuropean Radiology \u003c/em\u003e2023, \u003cstrong\u003e33\u003c/strong\u003e(7):4875-4884.\u003c/li\u003e\n\u003cli\u003eYang F, Pan X, Zhu K, Xiao Y, Yue X, Peng P, Zhang X, Huang J, Chen J, Yuan Y\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eAccelerated 3D high-resolution T2-weighted breast MRI with deep learning constrained compressed sensing, comparison with conventional T2-weighted sequence on 3.0 T\u003c/strong\u003e. \u003cem\u003eEuropean Journal of Radiology \u003c/em\u003e2022, \u003cstrong\u003e156\u003c/strong\u003e:110562.\u003c/li\u003e\n\u003cli\u003eWu X, Tang L, Li W, He S, Yue X, Peng P, Wu T, Zhang X, Wu Z, He Y\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eFeasibility of accelerated non-contrast-enhanced whole-heart bSSFP coronary MR angiography by deep learning\u0026ndash;constrained compressed sensing\u003c/strong\u003e. \u003cem\u003eEuropean Radiology \u003c/em\u003e2023, \u003cstrong\u003e33\u003c/strong\u003e(11):8180-8190.\u003c/li\u003e\n\u003cli\u003eZhang Y, Peng W, Xiao Y, Ming Y, Ma K, Hu S, Zeng W, Zeng L, Liang Z, Zhang X\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eRapid 3D breath-hold MR cholangiopancreatography using deep learning\u0026ndash;constrained compressed sensing reconstruction\u003c/strong\u003e. \u003cem\u003eEuropean Radiology \u003c/em\u003e2023, \u003cstrong\u003e33\u003c/strong\u003e(4):2500-2509.\u003c/li\u003e\n\u003cli\u003eZhang J, Ghanem B: \u003cstrong\u003eISTA-Net: Interpretable Optimization-Inspired Deep Network for Image Compressive Sensing\u003c/strong\u003e. In: \u003cem\u003e2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition: 18-23 June 2018 2018\u003c/em\u003e; 2018: 1828-1837.\u003c/li\u003e\n\u003cli\u003eFleiss JL: \u003cstrong\u003eMeasuring nominal scale agreement among many raters\u003c/strong\u003e. \u003cem\u003ePsychological bulletin \u003c/em\u003e1971, \u003cstrong\u003e76\u003c/strong\u003e(5):378.\u003c/li\u003e\n\u003cli\u003eKiryu S, Akai H, Yasaka K, Tajima T, Kunimatsu A, Yoshioka N, Akahane M, Abe O, Ohtomo K: \u003cstrong\u003eClinical Impact of Deep Learning Reconstruction in MRI\u003c/strong\u003e. \u003cem\u003eRadioGraphics \u003c/em\u003e2023, \u003cstrong\u003e43\u003c/strong\u003e(6):e220133.\u003c/li\u003e\n\u003cli\u003eBash S, Wang L, Airriess C, Zaharchuk G, Gong E, Shankaranarayanan A, Tanenbaum LN: \u003cstrong\u003eDeep Learning Enables 60% Accelerated Volumetric Brain MRI While Preserving Quantitative Performance: A Prospective, Multicenter, Multireader Trial\u003c/strong\u003e. \u003cem\u003eAJNR Am J Neuroradiol \u003c/em\u003e2021, \u003cstrong\u003e42\u003c/strong\u003e(12):2130-2137.\u003c/li\u003e\n\u003cli\u003eRudie JD, Gleason T, Barkovich MJ, Wilson DM, Shankaranarayanan A, Zhang T, Wang L, Gong E, Zaharchuk G, Villanueva-Meyer JE: \u003cstrong\u003eClinical Assessment of Deep Learning-based Super-Resolution for 3D Volumetric Brain MRI\u003c/strong\u003e. \u003cem\u003eRadiol Artif Intell \u003c/em\u003e2022, \u003cstrong\u003e4\u003c/strong\u003e(2):e210059.\u003c/li\u003e\n\u003cli\u003eKim M, Kim HS, Kim HJ, Park JE, Park SY, Kim Y-H, Kim SJ, Lee J, Lebel MR: \u003cstrong\u003eThin-Slice Pituitary MRI with Deep Learning\u0026ndash;based Reconstruction: Diagnostic Performance in a Postoperative Setting\u003c/strong\u003e. \u003cem\u003eRadiology \u003c/em\u003e2021, \u003cstrong\u003e298\u003c/strong\u003e(1):114-122.\u003c/li\u003e\n\u003cli\u003eKim S-H, Choi YH, Lee JS, Lee SB, Cho YJ, Lee SH, Shin S-M, Cheon J-E: \u003cstrong\u003eDeep learning reconstruction in pediatric brain MRI: comparison of image quality with conventional T2-weighted MRI\u003c/strong\u003e. \u003cem\u003eNeuroradiology \u003c/em\u003e2023, \u003cstrong\u003e65\u003c/strong\u003e(1):207-214.\u003c/li\u003e\n\u003cli\u003eFujima N, Nakagawa J, Kameda H, Ikebe Y, Harada T, Shimizu Y, Tsushima N, Kano S, Homma A, Kwon J\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eImprovement of image quality in diffusion-weighted imaging with model-based deep learning reconstruction for evaluations of the head and neck\u003c/strong\u003e. \u003cem\u003eMagnetic Resonance Materials in Physics, Biology and Medicine \u003c/em\u003e2023.\u003c/li\u003e\n\u003cli\u003eBrewer JB, Magda S, Airriess C, Smith ME: \u003cstrong\u003eFully-automated quantification of regional brain volumes for improved detection of focal atrophy in Alzheimer disease\u003c/strong\u003e. \u003cem\u003eAJNR Am J Neuroradiol \u003c/em\u003e2009, \u003cstrong\u003e30\u003c/strong\u003e(3):578-580.\u003c/li\u003e\n\u003cli\u003eJeong SH, Park CJ, Jeong HJ, Sunwoo MK, Ahn SS, Lee SK, Lee PH, Kim YJ, Sohn YH, Chung SJ: \u003cstrong\u003eAssociation of choroid plexus volume with motor symptoms and dopaminergic degeneration in Parkinson\u0026apos;s disease\u003c/strong\u003e. \u003cem\u003eJ Neurol Neurosurg Psychiatry \u003c/em\u003e2023, \u003cstrong\u003e94\u003c/strong\u003e(12):1047-1055.\u003c/li\u003e\n\u003cli\u003eShen Q, Loewenstein DA, Potter E, Zhao W, Appel J, Greig MT, Raj A, Acevedo A, Schofield E, Barker W\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eVolumetric and visual rating of magnetic resonance imaging scans in the diagnosis of amnestic mild cognitive impairment and Alzheimer\u0026apos;s disease\u003c/strong\u003e. \u003cem\u003eAlzheimers Dement \u003c/em\u003e2011, \u003cstrong\u003e7\u003c/strong\u003e(4):e101-108.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1. Interrater agreement\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"633\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal 94 images from both volunteers + patients (n=32)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRater A\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRater B\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eKappa\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e95% confidence interval\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003eAnatomic conspicuity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.888\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.806, 0.971)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.904\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.828, 0.981)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.920\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.850, 0.999)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003eOverall image quality\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.926\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.863, 0.989)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.924\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.858, 0.990)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.880\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.801 0.959)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003eArtifacts\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.923\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.856, 0.990)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.865\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.779, 0.951)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.910\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.840, 0.980)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003eSignal-to-noise ratio\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.922\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.845, 0.995)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.0001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.883\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.794, 0.973)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.0001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.851\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.752, 0.951)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.0001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003eSharpness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.890\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.811, 0.969)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.844\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.752, 0.936)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31.2303%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003eRater 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e0.892\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.8202%;\"\u003e\n \u003cp\u003e(0.813, 0.971)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11.9874%;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e\u003cbr\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2. Multirater quantitative assessments in volunteers\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"955\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRating criterion\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 151px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eReference standard\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 121px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCS2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 106px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDLR-CS2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCS4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDLR-CS4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eIQR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003eIQR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003eIQR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003eIQR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003eIQR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003eCS2 vs. CS4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eCS2 vs. DLR-CS4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAnatomic conspicuity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e(4,4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e(3,3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e(4,4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e0.241\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOverall image quality\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e(3,3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e(4,4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eArtifacts\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e(4,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e(3,3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e(4,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e0.123\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSignal-to-noise ratio\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e(3,4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e4.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e(4,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e0.021\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSharpness\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e(4,4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 47px;\"\u003e\n \u003cp\u003e(5,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e(3,3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e(4,4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 76px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e0.887\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eCS2, compressed sensing factor 2 without deep learning reconstruction; DLR-CS2, compressed sensing factor 2 with deep learning reconstruction; CS4, compressed sensing factor 4 without deep learning reconstruction; DLR-CS4, compressed sensing factor 4 with deep learning reconstruction; IQR, interquartile range\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3. Multirater quantitative assessments in patients\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003eRating criterion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCS2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDLR-CS4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 87px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 87px;\"\u003e\n \u003cp\u003eIQR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 87px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 87px;\"\u003e\n \u003cp\u003eIQR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAnatomic conspicuity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e4.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e0.864\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOverall image quality\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e0.188\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eArtifacts\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4 ,5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e0.226\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSignal-to-noise ratio\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e0.320\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSharpness\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e0.057\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eCS2, compressed sensing factor 2 without deep learning reconstruction; DLR-CS4, compressed sensing factor 4 with deep learning reconstruction; IQR, interquartile range\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4. Quantitative assessments in both volunteers and patients\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"936\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 265px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRegional volumes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 114px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eReference standard\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCS2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDLR-CS2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCS4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDLR-CS4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVolunteers\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003eLV volume, ratio of ICV \u0026times; 10\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026gt;.999\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003eHippocampal volume,\u0026nbsp;ratio of ICV\u0026nbsp;\u0026times;\u0026nbsp;10\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e0.999\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003eCP volume, ratio of ICV\u0026nbsp;\u0026times;\u0026nbsp;10\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e0.999\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003eWMH volume, ratio of ICV\u0026nbsp;\u0026times;\u0026nbsp;10\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e0.960\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePatients\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003eLV volume, ratio of ICV \u0026times; 10\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.028\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.012\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.028\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.012\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.938\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003eHippocampal volume,\u0026nbsp;ratio of ICV\u0026nbsp;\u0026times;\u0026nbsp;10\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.015\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.015\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.943\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003eCP volume, ratio of ICV\u0026nbsp;\u0026times;\u0026nbsp;10\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.884\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 265px;\"\u003e\n \u003cp\u003eWMH volume, ratio of ICV\u0026nbsp;\u0026times;\u0026nbsp;10\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.005\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e0.005\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.899\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eLV, lateral ventricle; CP, choroid plexus; WMH, white matter hypointensity; ICV, intracranial volume; CS2, compressed sensing factor 2 without deep learning reconstruction; DLR-CS2, compressed sensing factor 2 with deep learning reconstruction; CS4, compressed sensing factor 4 without deep learning reconstruction; DLR-CS4, compressed sensing factor 4 with deep learning reconstruction; SD, standard deviation\u003c/p\u003e\n\u003cp\u003eCalculated from \u003csup\u003e*\u003c/sup\u003eANOVA and\u003csup\u003e\u0026nbsp;\u003c/sup\u003e\u003csup\u003e\u0026dagger;\u003c/sup\u003eindependent two sample t-test\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Magnetic resonance imaging, deep learning, compressed sensing, qualitative assessment, quantitative assessment","lastPublishedDoi":"10.21203/rs.3.rs-5345149/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5345149/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003eTo evaluate the performance of a deep learning reconstruction (DLR) based on Adaptive-Compressed sensing (CS)-Network for brain MRI and validate it in a clinical setting\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e Ten healthy volunteers and 22 consecutive patients were prospectively enrolled. Volunteers underwent 3D brain MRI including T1 without CS factor (9:16min, reference standard); with CS factor of 2 without DLR (CS2, 4:6min); with CS factor of 2 with DLR (DLR-CS2); with CS factor of 4 without DLR (CS4, 2:6min); and with CS factor of 4 with DLR (DLR-CS4). The patients’ MRI included the CS2 and DLR-CS4. The volumes of lateral ventricles, hippocampus, choroid plexus, and white matter hypointensity were calculated and compared among the sequences. Three radiologists independently assessed anatomical conspicuity, overall image quality, artifacts, signal-to-noise ratio (SNR), and sharpness using a 5-point scale for each sequence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e Applying acceleration factors of 2 and 4 reduced the scan time to 65.4% and 33.5%, respectively, of that of the reference standard. Volumes of all the measured subregions showed no significant differences among different sequences in all participants. In qualitative analysis, the interrater agreement was excellent (κ=0.844–0.926). In volunteers, quality of DLR-CS4 were comparable to those of CS2 for all metrics except for the overall image quality and SNR despite a 51.2% scan time reduction. In patients, DLR-CS4 showed quality comparable to that of CS2 for all metrics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003eDLR allowedthe scan time reduction by at least half without sacrificing image quality and volumetric quantification accuracy, supporting its reliability and efficiency.\u003c/p\u003e","manuscriptTitle":"Deep Learning-Based Reconstruction for Three-Dimensional Volumetric Brain MRI: A Qualitative and Quantitative Assessment","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-02 00:36:56","doi":"10.21203/rs.3.rs-5345149/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-11-06T06:48:04+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-11-06T06:28:14+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-11-01T02:33:48+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2024-10-28T07:52:09+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"10e687ae-158d-4996-8b4f-11c6aa674700","owner":[],"postedDate":"December 2nd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-03-31T16:01:39+00:00","versionOfRecord":{"articleIdentity":"rs-5345149","link":"https://doi.org/10.1186/s12880-025-01647-8","journal":{"identity":"bmc-medical-imaging","isVorOnly":false,"title":"BMC Medical Imaging"},"publishedOn":"2025-03-27 15:56:55","publishedOnDateReadable":"March 27th, 2025"},"versionCreatedAt":"2024-12-02 00:36:56","video":"","vorDoi":"10.1186/s12880-025-01647-8","vorDoiUrl":"https://doi.org/10.1186/s12880-025-01647-8","workflowStages":[]},"version":"v1","identity":"rs-5345149","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5345149","identity":"rs-5345149","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}