Share：
Share this content in WeChat
X
[Chinese] [PDF] 617 8
• Clinical Article •
Prediction of lower-grade glioma IDH-1 mutation status using a combined model of radiomics and transformer deep learning features based on multi-parametric MRI of intratumoral and peritumoral edema
DOU Yue  LIU Yuanqing  LI Yongjun 

Cite this article as: DOU Y, LIU Y Q, LI Y J. Prediction of lower-grade glioma IDH-1 mutation status using a combined model of radiomics and transformer deep learning features based on multi-parametric MRI of intratumoral and peritumoral edema[J]. Chin J Magn Reson Imaging, 2025, 16(9): 46-52, 59. DOI:10.12015/issn.1674-8034.2025.09.008.


[Abstract] Objective To develop a combined model based on multiparametric MRI, radiomics, and deep learning techniques to predict isocitrate dehydrogenase gene (IDH-1) mutation status with lower-grade gliomas (LGGs) in patients.Materials and Methods Clinical, imaging, and pathological data were retrospectively collected from patients with pathologically confirmed LGGs. Based on multiparametric MRI, a predictive model for IDH-1 mutation status was constructed by combining radiomic features and deep learning features extracted from the 2.5D-CrossFormer deep learning model. Through feature selection, application of machine learning algorithms, and integration with clinical variables, a clinical-radiomics-deep learning nomogram model was developed.Results A total of 186 patients were included, with 79 IDH-1-positive cases and 107 IDH-1-negative cases. A total of 10 530 radiomic features and 32 deep learning features were extracted. After screening and feature dimensionality reduction, 20 radiomics-deep learning features were retained. Among various machine learning models, the LightGBM-based deep radiomics model performed best, with an area under the curve (AUC) of 0.94 in the training group and 0.86 in the validation group. The nomogram model constructed by combining clinical variables achieved an AUC of 0.97 in the training group, significantly outperforming the radiomics model and clinical model, and also demonstrated good predictive performance in the validation group.Conclusions Based on multiparametric MRI, radiomics, and deep learning techniques, this study successfully constructed a combined model incorporating intratumoral and peritumoral edema features to predict the IDH-1 mutation status in LGGs. This model exhibits high diagnostic accuracy and has the potential to provide important imaging evidence for the formulation of treatment plans and prognosis assessment in LGGs patients.
[Keywords] glioma；isocitrate dehydrogenase gene mutation；magnetic resonance imaging；deep learning；radiomics

DOU Yue   LIU Yuanqing   LI Yongjun*  

Department of Radiology, the First Affiliated Hospital of Soochow University, Soochow 215006, China

Corresponding author: LI Y J, E-mail: liyongjun1026@126.com

Conflicts of interest   None.

Received  2025-03-10
Accepted  2025-09-03
DOI: 10.12015/issn.1674-8034.2025.09.008
Cite this article as: DOU Y, LIU Y Q, LI Y J. Prediction of lower-grade glioma IDH-1 mutation status using a combined model of radiomics and transformer deep learning features based on multi-parametric MRI of intratumoral and peritumoral edema[J]. Chin J Magn Reson Imaging, 2025, 16(9): 46-52, 59. DOI:10.12015/issn.1674-8034.2025.09.008.

[1]
BRAT D J, VERHAAK R G W, ALDAPE K D, et al. Comprehensive, Integrative Genomic Analysis of Diffuse Lower-Grade Gliomas[J]. N Engl J Med, 2015, 372(26): 2481-2498. DOI: 10.1056/NEJMoa1402121.
[2]
LOUIS D N, PERRY A, WESSELING P, et al. The 2021 WHO Classification of Tumors of the Central Nervous System: a summary[J]. Neuro Oncol, 2021, 23(8): 1231-1251. DOI: 10.1093/neuonc/noab106.
[3]
ALGHAMRI M S, THALLA R, AVVARI R P, et al. Tumor mutational burden predicts survival in patients with low-grade gliomas expressing mutated IDH1[J/OL]. Neuro Oncol Adv, 2020, 2(1): vdaa042 [2025-03-10]. https://pubmed.ncbi.nlm.nih.gov/32642696/. DOI: 10.1093/noajnl/vdaa042.
[4]
HU C, WANG K, DAMON C, et al. ATRX loss promotes immunosuppressive mechanisms in IDH1 mutant glioma[J]. Neuro Oncol, 2022, 24(6): 888-900. DOI: 10.1093/neuonc/noab292.
[5]
YAN H, ZHANG H. Advances in MRI for diagnosis and treatment of lower-grade gliomas based on IDH and 1p/19q classification[J]. Chin J Magn Reson Imaging, 2023, 14(4): 137-141. DOI: 10.12015/issn.1674-8034.2023.04.024.
[6]
IUS T, POLANO M, BO M DAL, et al. Development of Clinical-Radiomics Nomogram for Predicting Post-Surgery Functional Improvement in High-Grade Glioma Patients[J/OL]. Cancers, 2025, 17(5): 758 [2025-03-10]. https://www.mdpi.com/2072-6694/17/5/758. DOI: 10.3390/cancers17050758.
[7]
WEI H H, YANG Y, FU F F, et al. Diagnosis and prognosis prediction of glioma based on multimodal MRI radiomics and deep learning[J]. Chin J Magn Reson Imaging, 2023, 14(5): 175-180. DOI: 10.12015/issn.1674-8034.2023.05.031.
[8]
LIANG Q, JING H, SHAO Y, et al. Artificial Intelligence Imaging for Predicting High-risk Molecular Markers of Gliomas[J]. Clin Neuroradiol, 2024, 34(1): 33-43. DOI: 10.1007/s00062-023-01375-y.
[9]
YANG X, HU C, XING Z, et al. Prediction of Ki-67 labeling index, ATRX mutation, and MGMT promoter methylation status in IDH-mutant astrocytoma by morphological MRI, SWI, DWI, and DSC-PWI[J]. Eur Radiol, 2023, 33(10): 7003-7014. DOI: 10.1007/s00330-023-09695-w.
[10]
CHOI Y S, BAE S, CHANG J H, et al. Fully automated hybrid approach to predict the IDH mutation status of gliomas via deep learning and radiomics[J]. Neuro Oncol, 2021, 23(2): 304-313. DOI: 10.1093/neuonc/noaa177.
[11]
ZHANG L, WANG R, GAO J, et al. A novel MRI-based deep learning networks combined with attention mechanism for predicting CDKN2A/B homozygous deletion status in IDH-mutant astrocytoma[J]. Eur Radiol, 2024, 34(1): 391-399. DOI: 10.1007/s00330-023-09944-y.
[12]
JEON Y H, CHOI K S, LEE K H, et al. Deep learning-based quantification of T2-FLAIR mismatch sign: extending IDH mutation prediction in adult-type diffuse lower-grade glioma[J/OL]. Eur Radiol, 2025[2025-03-10]. https://link.springer.com/article/10.1007/s00330-025-11475-7. DOI: 10.1007/s00330-025-11475-7.
[13]
VAN DER VOORT S R, INCEKARA F, WIJNENGA M M J, et al. Combined molecular subtyping, grading, and segmentation of glioma using multi-task deep learning[J]. Neuro Oncol, 2023, 25(2): 279-289. DOI: 10.1093/neuonc/noac166.
[14]
LI D, HU W, MA L, et al. Deep learning radiomics nomograms predict Isocitrate dehydrogenase (IDH) genotypes in brain glioma: A multicenter study[J/OL]. Magn Reson Imaging, 2025, 117: 110314 [2025-03-10]. https://www.sciencedirect.com/science/article/pii/S0730725X24001174. DOI: 10.1016/j.mri.2024.110314.
[15]
GÓMEZ VECCHIO T, NEIMANTAITE A, THURIN E, et al. Clinical application of machine-based deep learning in patients with radiologically presumed adult-type diffuse glioma grades 2 or 3[J/OL]. Neurooncol Adv, 2024, 6(1): vdae192 [2025-03-10]. https://academic.oup.com/noa/article/6/1/vdae192/7691502. DOI: 10.1093/noajnl/vdae192.
[16]
WU S, ZHANG X, RUI W, et al. A nomogram strategy for identifying the subclassification of IDH mutation and ATRX expression loss in lower-grade gliomas[J]. Eur Radiol, 2022, 32(5): 3187-3198. DOI: 10.1007/s00330-021-08444-1.
[17]
HOU H, YU J, DIAO Y, et al. Diagnostic performance of multiparametric nonenhanced magnetic resonance imaging (MRI) in grading glioma and correlating IDH mutation status[J/OL]. Clin Radiol, 2025, 82: 106791 [2025-03-10]. https://pubmed.ncbi.nlm.nih.gov/39837107/. DOI: 10.1016/j.crad.2024.106791.
[18]
WANG R, WANG S Y, ZHANG H P, et al. Application value of DCE-MRI in tumor body, peritumoral edema area in grading diffuse glioma[J]. Chin J Magn Reson Imaging, 2021, 12(6): 88-91. DOI: 10.12015/issn.1674-8034.2021.06.017.
[19]
TRUONG N C D, BANGALORE YOGANANDA C G, WAGNER B C, et al. Two-Stage Training Framework Using Multicontrast MRI Radiomics for IDH Mutation Status Prediction in Glioma.[J/OL]. Radiol Artif Intell, 2024, 6(4): e230218 [2025-03-10]. https://pubs.rsna.org/doi/10.1148/ryai.230218. DOI: 10.1148/ryai.230218.
[20]
REUSS D E, SAHM F, SCHRIMPF D, et al. ATRX and IDH1-R132H immunohistochemistry with subsequent copy number analysis and IDH sequencing as a basis for an "integrated" diagnostic approach for adult astrocytoma, oligodendroglioma and glioblastoma[J]. Acta Neuropathol, 2015, 129(1): 133-146. DOI: 10.1007/s00401-014-1370-3.
[21]
WANG W, CHEN W, QIU Q, et al. CrossFormer++: A Versatile Vision Transformer Hinging on Cross-Scale Attention[J]. IEEE Trans Pattern Anal Mach Intell, 2024, 46(5): 3123-3136. DOI: 10.1109/TPAMI.2023.3341806.
[22]
JIANG J, ZHANG X L, ZHOU J L. Research progress on isocitrate dehydrogenase genotype and imaging of glioma[J]. Chin J Magn Reson Imaging, 2021, 12(5): 103-106. DOI: 10.12015/issn.1674-8034.2021.05.025.
[23]
HARTMANN C, MEYER J, BALSS J, et al. Type and frequency of IDH1 and IDH2 mutations are related to astrocytic and oligodendroglial differentiation and age: a study of 1, 010 diffuse gliomas[J]. Acta Neuropathol, 2009, 118(4): 469-474. DOI: 10.1007/s00401-009-0561-9.
[24]
ZHANG B, ZHOU Q, XUE C, et al. Predicting telomerase reverse transcriptase promoter mutation status in glioblastoma by whole-tumor multi-sequence magnetic resonance texture analysis[J/OL]. Magn Reson Imaging, 2025, 118: 110360 [2025-03-10]. https://https://www.sciencedirect.com/science/article/abs/pii/S0730725X2500044X?via%3Dihub. DOI: 10.1016/j.mri.2025.110360.
[25]
WANG Y, GAO A, YANG H, et al. Using partially shared radiomics features to simultaneously identify isocitrate dehydrogenase mutation status and epilepsy in glioma patients from MRI images[J/OL]. Sci Rep, 2025, 15(1): 3591 [2025-03-10]. https://www.nature.com/articles/s41598-025-87778-y. DOI: 10.1038/s41598-025-87778-y.
[26]
TANG W, DUAN J Y, YU Z Y, et al. Value of enhanced MRI radiomics in predicting IDH-1 genotype in gliomas[J]. Chin J Magn Reson Imaging, 2022, 13(5): 111-114. DOI: 10.12015/issn.1674-8034.2022.05.020.
[27]
ZHAO S S, XIN Y K, ZHANG K, et al. Prediction of IDH-1 mutation status in WHO grade Ⅱ and Ⅲ gliomas by radiomics combined with T1-weighted contrast-enhanced image[J]. Chinese Journal of General Practice, 2023, 21(12): 2106-2110. DOI: 10.16766/j.cnki.issn.1674-4152.003301.
[28]
SHA Y J, WANG X C, TAN Y, et al. The value of predicting the subtype of IDH mutation combining with MGMT promoter methylation in lower grade gliomas by radiomics based on preoperative MRI[J]. Chin J Magn Reson Imaging, 2022, 13(7): 6-11. DOI: 10.12015/issn.1674-8034.2022.07.002.
[29]
ZHANG L, PAN H, LIU Z, et al. Multicenter clinical radiomics-integrated model based on [18F]FDG PET and multi-modal MRI predict ATRX mutation status in IDH-mutant lower-grade gliomas[J]. Eur Radiol, 2023, 33(2): 872-883. DOI: 10.1007/s00330-022-09043-4.
[30]
ZERWECK L, KLOSE U, WÜRTEMBERGER U, et al. Preoperative Adult-Type Diffuse Glioma Subtype Prediction with Dynamic Contrast-Enhanced MR Imaging and Diffusion Weighted Imaging in Tumor Cores and Peritumoral Tissue—A Standardized Multicenter Study[J/OL]. Diagnostics, 2025, 15(5): 532 [2025-03-10]. https://www.mdpi.com/2075-4418/15/5/532. DOI: 10.3390/diagnostics15050532.
[31]
CLUCERU J, INTERIAN Y, PHILLIPS J J, et al. Improving the noninvasive classification of glioma genetic subtype with deep learning and diffusion-weighted imaging[J]. Neuro Oncol, 2022, 24(4): 639-652. DOI: 10.1093/neuonc/noab238.
[32]
ZERWECK L, WÜRTEMBERGER U, KLOSE U, et al. Performance Comparison of Diffusion Kurtosis Imaging (DKI), Neurite Orientation Dispersion and Density Imaging (NODDI), and Diffusion Microstructure Imaging (DMI) in Predicting Adult-Type Glioma Subtype—A Pilot Study[J/OL]. Cancers, 2025, 17(5): 876 [2025-03-10]. https://pubmed.ncbi.nlm.nih.gov/40075723/. DOI: 10.3390/cancers17050876.
[33]
NI C, LIN R, YAO D, et al. Magnetic resonance diffusion-derived vessel density (DDVD) as a valuable tissue perfusion biomarker for isocitrate dehydrogenase genotyping in diffuse gliomas[J/OL]. BMC Med Imaging, 2025, 25(1): 79 [2025-03-10]. https://pubmed.ncbi.nlm.nih.gov/40050759/. DOI: 10.1186/s12880-025-01605-4.
[34]
DELFANTI R L, PICCIONI D E, HANDWERKER J, et al. Imaging correlates for the 2016 update on WHO classification of grade II/III gliomas: implications for IDH, 1p/19q and ATRX status[J]. J Neuro Oncol, 2017, 135(3): 601-609. DOI: 10.1007/s11060-017-2613-7.
[35]
BYEON Y, PARK Y W, LEE S, et al. Interpretable multimodal transformer for prediction of molecular subtypes and grades in adult-type diffuse gliomas[J/OL]. NPJ Digit Med, 2025, 8(1): 140 [2025-03-10]. https://pubmed.ncbi.nlm.nih.gov/40044878/. DOI: 10.1038/s41746-025-01530-4.
[36]
YANG X L, QIN D L, ZHANG H. Advances in deep learning-based radiomics for precision diagnosis and treatment of glioma[J]. Chin J Radiol, 2023, 57(8): 931-935. DOI: 10.3760/cma.j.cn112149-20220929-0078.
[37]
LIU Z, XU X, ZHANG W, et al. A fusion model integrating magnetic resonance imaging radiomics and deep learning features for predicting alpha-thalassemia X-linked intellectual disability mutation status in isocitrate dehydrogenase-mutant high-grade astrocytoma: a multicenter study[J]. Quant Imaging Med Surg, 2024, 14(1): 251-263. DOI: 10.21037/qims-23-807.

PREV The feasibility of fluid-suppressed amide proton transfer-weighted imaging in suppressing the fluid components of post-treatment gliomas
NEXT The value of MRI histogram analysis in predicting the long-term therapeutic effect after surgery for growth hormone-secreting pituitary neuroendocrine tumors
  


LINK


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