Share：
Share this content in WeChat
X
[Chinese] [PDF] 38 1
• Technical Article •
Evaluation of image quality and diagnostic value of T2-FLAIR in intracranial space occupying lesions based on different levels of deep learning reconstruction
MIAO Zhiming  TAN Yong  ZHANG Jing  YANG Jing  ZHOU Xiaoyu  HU Yixin  TANG Yu  LIN Meng  ZHANG Jiuquan 

DOI:10.12015/issn.1674-8034.2026.02.019.


[Abstract] Objective Based on 1.5 T MRI, this study aims to compare the image quality and clinical value of conventional T2-fluid attenuated inversion recovery (T2-FLAIR) reconstruction and fast T2-FLAIR deep learning (DL) reconstruction in patients with intracranial space occupying lesions, and explore the optimal DL reconstruction parameters.Materials and Methods A total of 104 patients with intracranial space occupying lesions were prospectively enrolled, and routine T2-FLAIR and fast T2-FLAIR (parallel imaging, PI acceleration factor 2) were collected separately. Conventional T2-FLAIR adopts traditional reconstruction, denoted as NDL; Fast T2-FLAIR selects DL reconstruction levels 2, 3, and 4, denoted as PI-DL2, PI-DL3, and PI-DL4. The four groups of images were evaluated quantitatively and qualitatively by two doctors in blind state, and the size and quantity of lesions were recorded. Quantitative evaluation includes signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR). Qualitative evaluation includes image sharpness, noise, gray white contrast, artifacts, lesion display, diagnostic confidence, and overall image quality.Results The conventional T2-FLAIR scan time was 2 minutes and 8 seconds, while the fast T2-FLAIR scan time was 1 minute and 20 seconds, resulting in a time reduction of approximately 37.5%. Quantitative analysis showed that compared with NDL, the SNR of each DL reconstruction group (level 2, 3, 4) was improved, and increased with the increase of DL level (P < 0.05). The CNR of PI-DL4 was significantly higher than the other three groups (P < 0.05), while there was no statistically significant difference in the CNR of PI-DL2 in the corpus callosum, brainstem, and cerebellar regions compared to the NDL group (P > 0.05). In terms of qualitative evaluation, the consistency of the two diagnostic physicians' evaluations is good. PI-DL4 performed the best in terms of image sharpness, noise control, and overall image quality (P < 0.05). There was no statistically significant difference in gray white matter contrast, lesion display, and diagnostic confidence between PI-DL4 and PI-DL3 (P > 0.05). There was no statistically significant difference between PI-DL2 and NDL in various qualitative evaluation indicators (P > 0.05). In the detection of lesions, the detection rate of DL group was higher than that of NDL group, and there was no statistically significant difference in size measurement (P > 0.05).Conclusions In 1.5 T MRI, combining DL reconstruction algorithm with PI acceleration technology can significantly improve the image quality and lesion display ability of T2-FLAIR sequence, and effectively shorten the scanning time. Because DL level 4 may reduce the contrast of some lesions, DL level 3 is recommended as the best reconstruction parameter of T2-FLAIR sequence for intracranial space occupying lesions.
[Keywords] T2-fluid attenuated inversion recovery；deep learning reconstruction；image quality；intracranial space occupying lesions；magnetic resonance imaging

MIAO Zhiming1   TAN Yong1   ZHANG Jing1   YANG Jing1   ZHOU Xiaoyu2   HU Yixin1   TANG Yu1   LIN Meng1   ZHANG Jiuquan1, 2*  

1 Department of Radiology, Chongqing University Cancer Hospital, Chongqing 400030, China

2 School of Medicine, Chongqing University, Chongqing 400030, China

Corresponding author: ZHANG J Q, E-mail: zhangjq_radiol@foxmail.com

Conflicts of interest   None.

Received  2025-09-04
Accepted  2026-01-26
DOI: 10.12015/issn.1674-8034.2026.02.019
DOI:10.12015/issn.1674-8034.2026.02.019.

[1]
ZHANG Y H, YU R Q, YU B, et al. Application value of diffusion-weighted imaging based on deep learning reconstruction algorithm in cranial MRI examination[J]. Chin J Magn Reson Imaging, 2025, 16(7): 65-71. DOI: 10.12015/issn.1674-8034.2025.07.010.
[2]
MA W F, GE X, DANG P, et al. Application of Sy-MRI combined with DWI in predicting MGMT methylation in glioma[J]. Chin J Magn Reson Imaging, 2023, 14(7): 18-24, 48. DOI: 10.12015/issn.1674-8034.2023.07.004.
[3]
HAMEL C, AVARD B, DEA N, et al. Canadian association of radiologists central nervous system diagnostic imaging referral guideline[J]. Can Assoc Radiol J, 2025, 76(3): 435-443. DOI: 10.1177/08465371241311247.
[4]
CAMPION A, IV M. Brain tumor imaging: review of conventional and advanced techniques[J]. Semin Neurol, 2023, 43(6): 867-888. DOI: 10.1055/s-0043-1776765.
[5]
TREVISI G, MANGIOLA A. Current knowledge about the peritumoral microenvironment in glioblastoma[J/OL]. Cancers, 2023, 15(22): 5460 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/38001721/. DOI: 10.3390/cancers15225460.
[6]
LIANG Q H, LI W F, LIANG C S, et al. Clinical value of enhanced T2WI-FLAIR sequence in the diagnosis of intracranial micrometastases[J]. J Med Theory Pract, 2025, 38(8): 1365-1367. DOI: 10.19381/j.issn.1001-7585.2025.08.036.
[7]
YANG A, FINKELSTEIN M, KOO C, et al. Impact of deep learning image reconstruction methods on MRI throughput[J/OL]. Radiol Artif Intell, 2024, 6(3): e230181 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/38506618/. DOI: 10.1148/ryai.230181.
[8]
LIN D J, JOHNSON P M, KNOLL F, et al. Artificial intelligence for MR image reconstruction: an overview for clinicians[J]. J Magn Reson Imaging, 2021, 53(4): 1015-1028. DOI: 10.1002/jmri.27078.
[9]
CHEN Z H, ZHANG M Y, WANG M J, et al. Research progress of synthetic MRI in clinical diseases[J]. Chin J Magn Reson Imaging, 2025, 16(7): 227-234. DOI: 10.12015/issn.1674-8034.2025.07.036.
[10]
XIE Y X, TAO H Y, LI X W, et al. Prospective comparison of standard and deep learning-reconstructed turbo spin-echo MRI of the shoulder[J/OL]. Radiology, 2024, 310(1): e231405 [2025-09-04]. https://pubmed.ncbi.nlm.nih.gov/38193842/. DOI: 10.1148/radiol.231405.
[11]
WALTER S S, VOSSHENRICH J, CANTARELLI RODRIGUES T, et al. Deep learning superresolution for simultaneous multislice parallel imaging-accelerated knee MRI using arthroscopy validation[J/OL]. Radiology, 2025, 314(1): e241249 [2025-09-04]. https://pubmed.ncbi.nlm.nih.gov/39873603/. DOI: 10.1148/radiol.241249.
[12]
BAE K, LEE J, JUNG Y, et al. Deep learning reconstruction for zero echo time lung magnetic resonance imaging: impact on image quality and lesion detection[J/OL]. Clin Radiol, 2024, 79(11): e1296-e1303 [2025-09-04]. https://pubmed.ncbi.nlm.nih.gov/39112100/. DOI: 10.1016/j.crad.2024.07.011.
[13]
FENG W M, ZHU L, XIA Y H, et al. Deep learning-based reconstruction: a reliability assessment in preoperative magnetic resonance imaging for primary rectal cancer[J]. Quant Imaging Med Surg, 2024, 14(12): 8927-8941. DOI: 10.21037/qims-24-907.
[14]
LEE K L, KESSLER D A, DEZONIE S, et al. Assessment of deep learning-based reconstruction on T2-weighted and diffusion-weighted prostate MRI image quality[J/OL]. Eur J Radiol, 2023, 166: 111017 [2025-09-04]. https://pubmed.ncbi.nlm.nih.gov/37541181/. DOI: 10.1016/j.ejrad.2023.111017.
[15]
KOCH K M, SHERAFATI M, ARPINAR V E, et al. Analysis and evaluation of a deep learning reconstruction approach with denoising for orthopedic MRI[J/OL]. Radiol Artif Intell, 2021, 3(6): e200278 [2025-09-04]. https://pubmed.ncbi.nlm.nih.gov/34870214/. DOI: 10.1148/ryai.2021200278.
[16]
GIGANTI F, KASIVISVANATHAN V, KIRKHAM A, et al. Prostate MRI quality: a critical review of the last 5 years and the role of the PI-QUAL score[J/OL]. Br J Radiol, 2022, 95(1131): 20210415 [2025-09-04]. https://pubmed.ncbi.nlm.nih.gov/34233502/. DOI: 10.1259/bjr.20210415.
[17]
Chinese Medical Association Radiology Society Quality Management and Security Management Group, Chinese Medical Association Radiology Society Magnetic Resonance Imaging Group. China expert consensus on safety management of magnetic resonance imaging[J]. Chin J Radiol, 2017, 51(10): 725-731. DOI: 10.3760/j.issn.1005-1201.2017.10.003.
[18]
GRAVES M J. 3 T: the good, the bad and the ugly[J/OL]. Br J Radiol, 2022, 95(1130): 20210708 [2025-09-04]. https://pubmed.ncbi.nlm.nih.gov/34705565/. DOI: 10.1259/bjr.20210708.
[19]
KAUFMANN T J, SMITS M, BOXERMAN J, et al. Consensus recommendations for a standardized brain tumor imaging protocol for clinical trials in brain metastases[J]. Neuro Oncol, 2020, 22(6): 757-772. DOI: 10.1093/neuonc/noaa030.
[20]
LI H Y, ZHANG M, LV X Y, et al. Investigation research on the configuration status and application of radiographic imaging equipment in China[J]. J Clin Radiol, 2024, 43(4): 669-674. DOI: 10.13437/j.cnki.jcr.2024.04.035.
[21]
ZHOU P H, LIU H B, LIN H, et al. Application of deep learning-based image reconstruction technology in 5.0T MRI for nasopharyngeal carcinoma[J]. Chin J Med Imag, 2025, 33(7): 694-699. DOI: 10.3969/j.issn.1005-5185.2025.07.002.
[22]
FU L Y, LIANG Y G, CHEN Z Q, et al. Quality control test of MRI system and action limits setting based on American college of radiology(ACR) standards[J]. China Med Devices, 2018, 33(10): 7-11. DOI: 10.3969/j.issn.1674-1633.2018.10.002.
[23]
REN J, LI Y, LIU F S, et al. Comparison of a deep learning-accelerated T2-weighted turbo spin echo sequence and its conventional counterpart for female pelvic MRI: reduced acquisition times and improved image quality[J/OL]. Insights Imaging, 2022, 13(1): 193 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/36512158/. DOI: 10.1186/s13244-022-01321-5.
[24]
HUANG H Y, MO J Y, DING Z G, et al. Deep learning to simulate contrast-enhanced MRI for evaluating suspected prostate cancer[J/OL]. Radiology, 2025, 314(1): e240238 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/39807983/. DOI: 10.1148/radiol.240238.
[25]
KIM S H, CHOI Y H, LEE J S, et al. Deep learning reconstruction in pediatric brain MRI: comparison of image quality with conventional T2-weighted MRI[J]. Neuroradiology, 2023, 65(1): 207-214. DOI: 10.1007/s00234-022-03053-1.
[26]
WU H, WAN Y Y, TANG M, et al. Real-time denoising of Brillouin optical time domain analyzer with high data fidelity using convolutional neural networks[J]. J Light Technol, 2019, 37(11): 2648-2653. DOI: 10.1109/JLT.2018.2876909.
[27]
HERRMANN J, KOERZDOERFER G, NICKEL D, et al. Feasibility and implementation of a deep learning MR reconstruction for TSE sequences in musculoskeletal imaging[J/OL]. Diagnostics, 2021, 11(8): 1484 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/34441418/. DOI: 10.3390/diagnostics11081484.
[28]
SUN K C, WANG Q, SHEN D G. Joint cross-attention network with deep modality prior for fast MRI reconstruction[J]. IEEE Trans Med Imaging, 2024, 43(1): 558-569. DOI: 10.1109/TMI.2023.3314008.
[29]
ZENG G L, DIBELLA E V. Iterative versus non-iterative image reconstruction methods for sparse magnetic resonance imaging[J]. J Radiol Imaging, 2020, 4(5): 30-39. DOI: 10.14312/2399-8172.2020-5.
[30]
PAROT V, SING-LONG C, LIZAMA C, et al. Application of the fractional Fourier transform to image reconstruction in MRI[J]. Magn Reson Med, 2012, 68(1): 17-29. DOI: 10.1002/mrm.23190.
[31]
MONTALT-TORDERA J, MUTHURANGU V, HAUPTMANN A, et al. Machine learning in magnetic resonance imaging: image reconstruction[J/OL]. Phys Med, 2021, 83: 79-87 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/33721701/. DOI: 10.1016/j.ejmp.2021.02.020.
[32]
HYUN C M, KIM H P, LEE S M, et al. Deep learning for undersampled MRI reconstruction[J/OL]. Phys Med Biol, 2018, 63(13): 135007 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/29787383/. DOI: 10.1088/1361-6560/aac71a.
[33]
KIDOH M, SHINODA K, KITAJIMA M, et al. Deep learning based noise reduction for brain MR imaging: tests on phantoms and healthy volunteers[J]. Magn Reson Med Sci, 2020, 19(3): 195-206. DOI: 10.2463/mrms.mp.2019-0018.
[34]
RUFF C, HAUSER T K, RODER C, et al. Multidisciplinary, clinical assessment of accelerated deep-learning MRI protocols at 1.5 T and 3 T after intracranial tumor surgery and their influence on residual tumor perception[J/OL]. Diagnostics, 2025, 15(15): 1982 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/40804946/. DOI: 10.3390/diagnostics15151982.
[35]
RIEDERER S J, BORISCH E A, FROEMMING A T, et al. Comparison of model-based versus deep learning-based image reconstruction for thin-slice T2-weighted spin-echo prostate MRI[J]. Abdom Radiol, 2024, 49(8): 2921-2931. DOI: 10.1007/s00261-024-04256-1.
[36]
WANG G, YE J C, MUELLER K, et al. Image reconstruction is a new frontier of machine learning[J]. IEEE Trans Med Imaging, 2018, 37(6): 1289-1296. DOI: 10.1109/TMI.2018.2833635.
[37]
KABASAWA H, KIRYU S. Pulse sequences and reconstruction in fast MR imaging of the liver[J]. Magn Reson Med Sci, 2023, 22(2): 176-190. DOI: 10.2463/mrms.rev.2022-0114.
[38]
JASPAN O N, FLEYSHER R, LIPTON M L. Compressed sensing MRI: a review of the clinical literature[J/OL]. Br J Radiol, 2015, 88(1056): 20150487 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/26402216/. DOI: 10.1259/bjr.20150487.
[39]
YOON S, PARK S H, HAN D. Uncover this tech term: compressed sensing magnetic resonance imaging[J]. Korean J Radiol, 2023, 24(12): 1293-1302. DOI: 10.3348/kjr.2023.0743.
[40]
WRIGHT K L, HAMILTON J I, GRISWOLD M A, et al. Non-Cartesian parallel imaging reconstruction[J]. J Magn Reson Imaging, 2014, 40(5): 1022-1040. DOI: 10.1002/jmri.24521.
[41]
AKAI H, YASAKA K, SUGAWARA H, et al. Commercially available deep-learning-reconstruction of MR imaging of the knee at 1.5T has higher image quality than conventionally-reconstructed imaging at 3T: a normal volunteer study[J]. Magn Reson Med Sci, 2023, 22(3): 353-360. DOI: 10.2463/mrms.mp.2022-0020.
[42]
TAJIMA T, AKAI H, YASAKA K, et al. Usefulness of deep learning-based noise reduction for 1.5 T MRI brain images[J/OL]. Clin Radiol, 2023, 78(1): e13-e21 [2025-09-03]. https://pubmed.ncbi.nlm.nih.gov/36116967/. DOI: 10.1016/j.crad.2022.08.127.

PREV Dynamic DTI assessment of white matter evolution and its correlation with neurobehavioral function after hypoxic-ischemic brain injury in neonatal rats​
NEXT A challenging case report of IgG4-related sclerosing cholangitis with literature review
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn