Objective To investigate the value of a combined model constructed by transvaginal ultra⁃ sound radiomics and clinical risk factors and its nomogram in predicting preterm birth in late pregnancy. Methods One hundred and seventeen pregnant women from 28 + 0 to 36 + 6 weeks of gestation who attended regular maternity checkups at our hospital from January 2017 to December 2022 were selected as observation subjects，and transvaginal ultrasound cervical standard section images were collected and analyzed for radiomics features，and ultrasound radiomics models were constructed by removing redundant features through dimensionality reduction screening ， using logistic regression，and combining clinical data in the training group The nomogram of the combined models. To compare the efficacy of each model in predicting preterm birth. Results A total of 14 ultrasound radiomics features were screened to construct an ultrasound radiomics prediction model for assessing the risk of preterm delivery in late pregnancy with a model training set AUC of 0.841［95%CI（0.761，0.921）］and a validation set of 0.824［95%CI（0.616，1.00）］. The combined model was established by combining age and cervical length，and the AUC of the model training set was 0.908［95%CI（0.850，0.966）］and the validation set was 0.892［95% CI （0.693，0.100）］. The decision curves showed that the clinical utility of the combined model was better than that of the ultrasound radiomics model within the 0.2 ~ 0.8 threshold. Calibration curves showed that the combined model was more accurate than the ultrasound radiomics model. Conclusion The radiomics based on ultrasound features and clinical characteristics can predict the risk of preterm birth，which can provide a reference basis for clinical diagnosis and treatment.