Abstract: OBJECTIVE: Radiomics and machine learning show great potential in oncology prediction and prognosis. This study aimed to develop a machine learning model integrating contrast-enhanced CT radiomics features and clinical characteristics to predict lymph node metastasis in cervical squamous cell carcinoma(CSCC) patients. METHODS: In this retrospective study, patients with pathologically confirmed CSCC who underwent radical hysterectomy at the Cancer Hospital of Shantou University Medical College between January 2016 and December 2021 were enrolled. Preoperative contrast-enhanced CT images obtained within three weeks before surgery were collected. A Radscore was calculated by delineating tumor volumes and extracting/selecting radiomic features. Clinical characteristics were also collected. Four machine learning modelsLogistic regression, LDA, SVM, and Naïve Bayes-were built using the clinical features and Radscore to identify the optimal predictive model. RESULTS: The Naive Bayes model demonstrated the best and most stable overall performance,achieving an AUC of 0.957. The AUCs for the other models were 0.953 for SVM,0.943 for Logistic regression, and 0.941 for LDA. On the test set, the Naive Bayes model also achieved excellent accuracy(0.819), sensitivity(0.915), specificity(0.727), and an F1-score(0.889). SHAP analysis identified CA-125 as the most important predictor in the model's decision-making. CONCLUSION: We successfully developed and validated a high-performance Naïve Bayes model for predicting lymph node metastasis risk in CSCC patients. SHAP interpretability analysis further confirmed CA-125,CEA,FIGO stage,and Radscore as key predictors,enhancing the model's transparency and clinical credibility.

Key words: cervical cancer, radiomics, lymph node metastasis, machine learning