Multiparametric MRI-based radiomics for preoperative prediction of parametrial invasion in early-stage cervical cancer

Yang Chongshuang¹Li Man²XIN YI³Wang Lin¹Kuang Guangxian¹Zhang Chunfang¹Yao Benyong¹Qin Zhihong¹Shi Tianliang^1*Jiang Qiang^1*

• ¹Department of Radiology, Tongren People's Hospital,, Tongren, China
• ²Shanghai United Imaging Healthcare Co., Ltd., Shanghai, China
• ³Affiliated Hospital of North China University of Science and Technology, Tangshan, Hebei Province, China

Objective: To evaluate the performance of radiomics based on Multiparametric magnetic resonance imaging (MRI) for the preoperative prediction of parametrial invasion (PMI) in cervical cancer (CC).: This retrospective study included 110 consecutive patients with FIGO stage IB-IIA CC. Patients were randomly divided into training and testing cohort in an 8:2 ratio. The region of interest (ROI) was manually delineated.. Radiomics features were extracted separately from T2 weighted imaging (T2WI), diffusion weighted imaging (DWI), apparent diffusion coefficient (ADC), contrast-enhanced T1-weighted imaging (T1C). Feature selection was performed using the correlation coefficient, recursive feature cancellation and the Least absolute shrinkage and selection operator algorithm. Radiomics models based on single-sequence, dual-sequence and multi-sequence combinations were then constructed. Model performance was assessed using receiver operating characteristic (ROC) curve analysis. The DeLong test was used to compare the area under the curve (AUC), supplemented by net reclassification improvement and comprehensive discrimination improvement measures.Results: A total of 2264 radiomics features were initially extracted. After feature selection, 7, 10, 6 and 8 valid features were retained from T1C, T2WI, ADC and DWI sequence, respectively. Fifteen radiomics models were developed, including 4 single-sequence models, 6 double-sequence models and 5 multi-sequence models. All models showed good classification performance for PMI in both training and testing cohorts, with an AUCs ranging from 0.842 to 1.000 in the training cohort and from 0.755 to 0.917 in the testing cohort. Among them, the T1C+ADC+DWI model demonstrated the best diagnostic performance, significantly outperforming all other models (p ＜ 0.05), with the highest AUC in both training and testing cohorts (training: 1.000, testing: 0.917).Radiomics based on multiparametric MRI can effectively predict PMI status in patients with early-stage CC, offering valuable support for individualized treatment planning and clinical decision-making.