Back to all papers

Preoperative prediction of lymphovascular space invasion in early-stage cervical cancer using a deep learning-based 3D super-resolution MRI radiomics model.

November 15, 2026pubmed logopapers

Authors

Guo Y,Ye Y,Mu Q,Huang W,Luo S

Affiliations (3)

  • Department of Radiology, The First People's Hospital of Guangyuan, Guangyuan, Sichuan, China.
  • Department of Radiology, The First People's Hospital of Guangyuan, Guangyuan, Sichuan, China. [email protected].
  • Department of Radiology, The First People's Hospital of Guangyuan, Guangyuan, Sichuan, China. [email protected].

Abstract

Lymphovascular space invasion (LVSI) is a critical prognostic factor guiding adjuvant therapy in early-stage cervical cancer, yet its preoperative assessment is often limited by the spatial resolution of standard MRI. This study aimed to investigate the value of a deep learning-based 3D super-resolution (SR) MRI radiomics model in predicting LVSI and to evaluate how SR reconstruction impacts image quality and radiomic feature robustness. This retrospective study included 117 patients with early-stage cervical cancer (training cohort: n = 81; validation cohort: n = 36). A 3D Super-Resolution Generative Adversarial Network (SRGAN) reconstructed T2-SPAIR and DWI sequences into super-resolution images (SRI) with a 4× upsampling factor. Radiomic features were extracted, and an SRI-based multimodal fusion model was developed and systematically compared with original image (OI)-based and clinical models using the Area Under the Curve (AUC) and Decision Curve Analysis (DCA). SR reconstruction significantly improved the signal-to-noise ratio (SNR) for both DWI (P = 0.012) and T2-SPAIR (P = 0.030). For DWI, SRI preserved intratumoral texture heterogeneity, leading to improved diagnostic performance (Validation AUC: SRI-DWI 0.864 vs. OI-DWI 0.727, P = 0.041). The final SRI-Fusion model achieved the highest diagnostic accuracy (Validation AUC: 0.934), significantly outperforming the OI-Fusion model (AUC: 0.828, P = 0.044) and the clinical model (AUC: 0.626, P = 0.012). SRGAN-based reconstruction effectively improves MRI quality and feature robustness, particularly in DWI. The SRI-based multimodal radiomics model demonstrates superior predictive performance for LVSI in early-stage cervical cancer compared to conventional imaging and clinical models.

Topics

Deep LearningUterine Cervical NeoplasmsMagnetic Resonance ImagingImaging, Three-DimensionalJournal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.