PET-CT radiomics and artificial intelligence for predicting lymph node metastasis in cervical cancer: a systematic
Xiaoyu Liu1,2, Junqiang Lei3,4
1First Clinical Medical College, Lanzhou University, Lanzhou, 730000, Gansu, China.
Purpose:
Preoperative identification of lymph node metastasis (LNM) in cervical cancer is crucial for guiding therapeutic strategies but remains clinically challenging. This study represents the first systematic meta-analysis to quantitatively evaluate the diagnostic efficacy of PET/CT-based radiomics and artificial intelligence (AI) models in predicting LNM, providing a robust synthesis of current evidence.
Methods:
Comprehensive searches were conducted across PubMed, Embase, Web of Science, Cochrane Library, and three major Chinese databases for studies published through April 2, 2025. Methodological quality was rigorously assessed using the Radiomics Quality Score (RQS) and QUADAS-2 tools. Pooled sensitivity, specificity, and the area under the summary receiver operating characteristic curve (AUC) were synthesized via random-effects models.
Results:
Nine studies involving one thousand two hundred sixty-eight patients met the inclusion criteria. The radiomics-AI models demonstrated high diagnostic accuracy, with a pooled sensitivity of 0.90 (95% CI 0.84-0.93), specificity of 0.88 (95% CI 0.74-0.95), and an AUC of 0.92 (95% CI 0.90-0.94). Subgroup analyses indicated that semi-automated segmentation and larger sample sizes were associated with improved diagnostic stability. Despite promising results, clinical translation is hindered by modest methodological quality (mean RQS: 10.5). Standardized workflows and prospective validation remain essential to ensure model reliability.
Conclusion:
PET/CT-based radiomics models show strong potential for the non-invasive prediction of LNM in cervical cancer, offering clinically meaningful diagnostic accuracy. Nevertheless, their translation into routine clinical practice is currently limited by methodological heterogeneity and lack of high-quality studies. Future research should prioritize standardized workflows and multicenter prospective studies to facilitate clinical implementation.
