Share:
Share this content in WeChat
X
• Clinical Article •
MRI-Based multi-sequence 2.5D fusion deep learning model for preoperative prediction of lymphovascular space invasion in cervical cancer
QIN Fengying  WEI Yuze  LUAN Yiyi  TIAN Mingke  SUN Xin  WEN Feng  DONG Yue 

DOI:10.12015/issn.1674-8034.2026.11.012.


[Abstract] Objective To develop a deep learning (DL) model based on MRI multi-sequence 2.5D fused images for the preoperative prediction of lymphovascular space invasion (LVSI) in cervical cancer, and to evaluate its generalizability through internal cross-validation and external independent validation.Materials and Methods A total of 580 patients with cervical cancer who underwent radical hysterectomy from two centers were retrospectively enrolled and divided into a training set (n = 471, Center 1) and an external independent test set (n = 109, Center 2) according to the center of origin. Ten-fold cross-validation was used for model selection and internal evaluation within the training set. Axial images at the maximum tumor section were extracted from contrast-enhanced T1-weighted imaging (CE-T1WI) and T2-weighted imaging (T2WI), respectively. After isotropic resampling, coronal and sagittal images were obtained via multiplanar reconstruction, and three-channel 2.5D fusion images were constructed as input for the 2.5D DL model. DL features were extracted using a pretrained DenseNet169 model, and feature selection was performed via weighted elastic net logistic regression to build the 2.5D DL model. Independent clinical predictors were identified by univariable logistic regression analysis to construct a clinical model, and subsequently combined with the DL-Score to develop a nomogram. A 2D DL model and a radiomics model served as ablation controls. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), and comparisons between models were performed using the DeLong test.Results The 2.5D DL model selected 48 key features (62.5% from CE-T1WI). It achieved a 10-fold cross-validation AUC of 0.812 and an AUC of 0.806 (95% CI: 0.726 to 0.886) in the external test set. Univariable logistic regression identified deep stromal invasion (≥ outer 1/3) and maximum lymph node short-axis diameter as independent predictors of LVSI. The standalone clinical model, 2D dual-sequence fusion model, and radiomics model yielded external test AUCs of 0.706, 0.694, and 0.777, respectively, all lower than that of the 2.5D DL model. The combined model achieved an external test AUC of 0.833 (95% CI: 0.758 to 0.907), with sensitivity improving from 34.8% (DL alone) to 60.9%, although the difference in AUC did not reach statistical significance (P = 0.333). Multi-threshold analysis showed that at a threshold of 0.30, the combined model attained a sensitivity of 80.4% and a negative predictive value of 82.7%. Calibration curves and decision curve analysis demonstrated good calibration and higher net clinical benefit for the combined model.Conclusions The 2.5D DL model based on MRI multi-sequence 2.5D fusion images can preoperatively predict LVSI in cervical cancer, with acceptable generalizability demonstrated through internal cross-validation and external independent validation. The integration of clinical variables markedly improves model sensitivity and clinical net benefit, providing a new imaging tool for preoperative noninvasive LVSI assessment in cervical cancer.
[Keywords] cervical cancer;lymphovascular space invasion;deep learning;2.5D fusion images;nomogram

QIN Fengying1, 2   WEI Yuze2   LUAN Yiyi2   TIAN Mingke2   SUN Xin2   WEN Feng3   DONG Yue2*  

1 Dalian Medical University, Dalian 116044, China

2 Department of Radiology, Cancer Hospital of Dalian University of Technology (Liaoning Cancer Hospital & Institute), Shenyang 110042, China

3 Department of Radiology, Shengjing Hospital of China Medical University, Shenyang 110004, China

Corresponding author: DONG Y, E-mail: dyy1026@sina.com

Conflicts of interest   None.

Received  2026-06-17
Accepted  2026-09-18
DOI: 10.12015/issn.1674-8034.2026.11.012
DOI:10.12015/issn.1674-8034.2026.11.012.

[1]
WU J, JIN Q Y, ZHANG Y M, et al. Global burden of cervical cancer: current estimates, temporal trend and future projections based on the GLOBOCAN 2022[J]. J Natl Cancer Cent, 2025, 5(3): 322-329. DOI: 10.1016/j.jncc.2024.11.006.
[2]
BRAY F, LAVERSANNE M, SUNG H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2024, 74(3): 229-263. DOI: 10.3322/caac.21834.
[3]
STOLNICU S, TALIA K L, BANET N, et al. Lymphovascular space invasion in cervical cancer: a review of clinical significance and pathologic issues[J]. Int J Gynecol Pathol, 2026, 45(2): 152-161. DOI: 10.1097/PGP.0000000000001135.
[4]
XU M C, CAO C H, WU P, et al. Advances in cervical cancer: current insights and future directions[J]. Cancer Commun (Lond), 2025, 45(2): 77-109. DOI: 10.1002/cac2.12629.
[5]
HUANG Y, WEN W B, LI X D, et al. Prognostic value of lymphovascular space invasion in stage IA to IIB cervical cancer: A meta-analysis[J/OL]. Medicine, 2023, 102(15): e33547 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/37058045/. DOI: 10.1097/MD.0000000000033547.
[6]
WENZEL H H B, HARDIE A N, MONCADA-TORRES A, et al. A federated approach to identify women with early-stage cervical cancer at low risk of lymph node metastases[J]. Eur J Cancer, 2023, 185: 61-68. DOI: 10.1016/j.ejca.2023.02.021.
[7]
ABU-RUSTUM N R, CAMPOS S M, AMARNATH S, et al. Cervical Cancer, Version 2.2026, NCCN Clinical Practice Guidelines In Oncology[J/OL]. J Natl Compr Canc Netw, 2025, 23(12): 549-573 [2026-08-01]. DOI: 10.6004/jnccn.2025.0057.
[8]
LIN B J, LONG X F, WU Z X, et al. Predicting lymph-vascular space invasion in cervical cancer based on MR-T2WI with deep learning and radiomic features combined with clinical features[J]. Chin J Magn Reson Imaging, 2024, 15(3): 130-136. DOI: 10.12015/issn.1674-8034.2024.03.021.
[9]
BOEMI S, PAVAN M, SIENA R, et al. Locally advanced cervical cancer: multiparametric MRI in gynecologic oncology and precision medicine[J/OL]. Diagnostics (Basel), 2025, 15(22): 2858 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/41300883/. DOI: 10.3390/diagnostics15222858.
[10]
ZHANG H, TENG C, YAO Y, et al. MRI-based radiomics models for noninvasive evaluation of lymphovascular space invasion in cervical cancer: a systematic review and meta-analysis[J/OL]. Clin Radiol, 2024, 79(11): e1372-e1382 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/39183137/. DOI: 10.1016/j.crad.2024.07.018.
[11]
ZHAO M L, LI Z, GU X W, et al. The role of radiomics for predicting of lymph-vascular space invasion in cervical cancer patients based on artificial intelligence: a systematic review and meta-analysis[J/OL]. J Gynecol Oncol, 2025, 36(2): e26 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/39058366/. DOI: 10.3802/jgo.2025.36.e26.
[12]
WU L Y, LI S H, LI S F, et al. Preoperative magnetic resonance imaging-radiomics in cervical cancer: a systematic review and meta-analysis[J/OL]. Front Oncol, 2024, 14: 1416378 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/39026971/. DOI: 10.3389/fonc.2024.1416378.
[13]
YANG C S, WU M, ZHANG J C, et al. Radiomics based on MRI in predicting lymphovascular space invasion of cervical cancer: a meta-analysis[J/OL]. Front Oncol, 2024, 14: 1425078 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/39484029/. DOI: 10.3389/fonc.2024.1425078.
[14]
WANG Y X, HU Z T, WANG H Z. The clinical implications and interpretability of computational medical imaging (radiomics) in brain tumors[J/OL]. Insights Imaging, 2025, 16(1): 77 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/40159380/. DOI: 10.1186/s13244-025-01950-6.
[15]
ZHANG Y F, QIN Q. Prospects and challenges of deep learning in gynecologic malignancies[J/OL]. Front Oncol, 2025, 15: 1592078 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/41256318/. DOI: 10.3389/fonc.2025.1592078.
[16]
SUN S H, ZHANG J, MOU Y N, et al. Research progress and current status of artificial intelligence based on MR diffusion imaging technology in evaluating cervical cancer[J]. Chin J Magn Reson Imaging, 2025, 16(12): 220-226. DOI: 10.12015/issn.1674-8034.2025.12.032.
[17]
SHE L Z, LI Y F, WANG H Y, et al. Imaging-based AI for predicting lymphovascular space invasion in cervical cancer: systematic review and meta-analysis[J/OL]. J Med Internet Res, 2025, 27: e71091 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/40523236/. DOI: 10.2196/71091.
[18]
SULEMAN M U, MURSALEEN M, KHALIL U, et al. Assessing the generalizability of artificial intelligence in radiology: a systematic review of performance across different clinical settings[J]. Ann Med Surg, 2025, 87(12): 8803-8811. DOI: 10.1097/ms9.0000000000004166.
[19]
FOTOPOULOS D, LADAKIS I, FILOS D, et al. Explainable AI in cancer imaging: scoping review of methods, modalities, and clinical integration[J/OL]. J Med Internet Res, 2026, 28: e80645 [2026-08-01]. https://doi.org/10.2196/80645. DOI: 10.2196/80645.
[20]
BAHETI B, PATI S, MENZE B, et al. Leveraging 2D deep learning ImageNet-trained models forNative 3D medical image analysis[C]//Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Cham: Springer, 2023: 68-79. DOI: 10.1007/978-3-031-33842-7_6.
[21]
BHATLA N, AOKI D, SHARMA D N, et al. Cancer of the cervix uteri[J]. Int J Gynecol Obstet, 2018, 143(S2): 22-36. DOI: 10.1002/ijgo.12611.
[22]
ABU-RUSTUM N R, YASHAR C M, AREND R, et al. NCCN guidelines® insights: cervical cancer, version 1.2024[J]. J Natl Compr Canc Netw, 2023, 21(12): 1224-1233. DOI: 10.6004/jnccn.2023.0062.
[23]
CIBULA D, PÖTTER R, PLANCHAMP F, et al. The European society of gynaecological oncology/European society for radiotherapy and oncology/European society of pathology guidelines for the management of patients with cervical cancer[J]. Virchows Arch, 2018, 472(6): 919-936. DOI: 10.1007/s00428-018-2362-9.
[24]
EISENHAUER E A, THERASSE P, BOGAERTS J, et al. New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1)[J]. Eur J Cancer, 2009, 45(2): 228-247. DOI: 10.1016/j.ejca.2008.10.026.
[25]
CHOI H J, JU W, MYUNG S K, et al. Diagnostic performance of computer tomography, magnetic resonance imaging, and positron emission tomography or positron emission tomography/computer tomography for detection of metastatic lymph nodes in patients with cervical cancer: Meta-analysis[J]. Cancer Sci, 2010, 101(6): 1471-1479. DOI: 10.1111/j.1349-7006.2010.01532.x.
[26]
DONG J, ZHANG C, ZHANG Y, et al. Precise diagnosis of pediatric posterior cranial Fossa neoplasms based on 2.5D MRI deep learning[J/OL]. Front Oncol, 2025, 15: 1700694 [2026-08-01]. https://doi.org/10.3389/fonc.2025.1700694. DOI: 10.3389/fonc.2025.1700694.
[27]
MAYERHOEFER M E, MATERKA A, LANGS G, et al. Introduction to radiomics[J]. J Nucl Med, 2020, 61(4): 488-495. DOI: 10.2967/jnumed.118.222893.
[28]
CARMELIET P, JAIN R K. Molecular mechanisms and clinical applications of angiogenesis[J]. Nature, 2011, 473(7347): 298-307. DOI: 10.1038/nature10144.
[29]
LIU B, GAO H, ZHOU F, et al. Dynamic contrast-enhanced magnetic resonance imaging in cervical cancer: correlation between quantitative parameters and molecular markers hypoxia-inducible factors-1-alpha, vascular endothelial growth factor, and Ki-67[J/OL]. Clin Radiol, 2024, 79(6): e826-e833 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/38582634/. DOI: 10.1016/j.crad.2024.01.039.
[30]
HUANG G, CUI Y Q, WANG P, et al. Multi-parametric magnetic resonance imaging-based radiomics analysis of cervical cancer for preoperative prediction of lymphovascular space invasion[J/OL]. Front Oncol, 2021, 11: 663370 [2026-08-01]. https://pubmed.ncbi.nlm.nih.gov/35096556/. DOI: 10.3389/fonc.2021.663370.
[31]
LUO S G, GUO Y, YE Y Q, et al. Prediction of cervical cancer lymph node metastasis based on multisequence magnetic resonance imaging radiomics and deep learning features: a dual-center study[J/OL]. Sci Rep, 2025, 15: 29259 [2026-08-01]. https://doi.org/10.1038/s41598-025-13781-y. DOI: 10.1038/s41598-025-13781-y.
[32]
WANG S X, LIU X W, WU Y, et al. Habitat-based radiomics enhances the ability to predict lymphovascular space invasion in cervical cancer: a multi-center study[J/OL]. Front Oncol, 2023, 13: 1252074 [2026-08-01]. https://doi.org/10.3389/fonc.2023.1252074. DOI: 10.3389/fonc.2023.1252074.
[33]
WU Y, WANG S X, CHEN Y Q, et al. A multicenter study on preoperative assessment of lymphovascular space invasion in early-stage cervical cancer based on multimodal MR radiomics[J]. J Magn Reson Imaging, 2023, 58(5): 1638-1648. DOI: 10.1002/jmri.28676.
[34]
LI S Y, CHEN Y S, HUANG X Z, et al. Nomogram prediction of the lymph-vascular space invasion in cervical cancer: comparison of 2009 and 2018 staging systems[J/OL]. Front Oncol, 2025, 15: 1505512 [2026-08-01]. https://doi.org/10.3389/fonc.2025.1505512. DOI: 10.3389/fonc.2025.1505512.
[35]
YAN H W, HUANG G T, YANG Z H, et al. Machine learning-based multiparametric magnetic resonance imaging radiomics model for preoperative predicting the deep stromal invasion in patients with early cervical cancer[J]. J Imaging Inform Med, 2024, 37(1): 230-246. DOI: 10.1007/s10278-023-00906-w.
[36]
MANGANARO L, LAKHMAN Y, BHARWANI N, et al. Staging, recurrence and follow-up of uterine cervical cancer using MRI: Updated Guidelines of the European Society of Urogenital Radiology after revised FIGO staging 2018[J]. Eur Radiol, 2021, 31(10): 7802-7816. DOI: 10.1007/s00330-020-07632-9.
[37]
QIN F Y, SUN X Y, TIAN M K, et al. Prediction of lymph node metastasis in operable cervical cancer using clinical parameters and deep learning with MRI data: a multicentre study[J/OL]. Insights Imaging, 2024, 15(1): 56 [2026-08-01]. https://doi.org/10.1186/s13244-024-01618-7. DOI: 10.1186/s13244-024-01618-7.

PREV The value of DWI-based virtual magnetic resonance elastography combined with the magnetic resonance severity index for early risk stratification in acute pancreatitis
NEXT Advances in brain structural and functional MRI in self-limited epilepsy with centrotemporal spikes: from focal discharges to brain network remodeling
  



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