The predictive value of ultrasound radiomics combined with clinical indicators for spontaneous preterm birth: a nested case-control study

Background Premature birth is a major cause of perinatal death and long-term health complications, but existing methods that can predict preterm birth have limitations during the third trimester. This study aimed to develop a combined model that integrates ultrasonic imaging omics with the anterior uterocervical angle, cervical length, and maternal age for predicting spontaneous preterm birth in late pregnancy. Methods This nested case-control study included 272 singleton pregnant women who underwent transvaginal ultrasound at 28–32 weeks. Radiomic features were extracted from cervical ultrasound images, and a radiomics score was calculated using least absolute shrinkage and selection operator (LASSO) regression. Logistic regression was used to develop clinical, radiomics, and combined models. Model performance was evaluated by the area under the receiver operating characteristic curve, the calibration curve, and decision curve analysis. Results The combined model exhibited the best predictive performance, with a training area under the receiver operating characteristic curve of 0.907 and a validation area under the receiver operating characteristic curve of 0.830. In the validation set, however, its performance was similar to that of the radiomics‑only model, suggesting that the clinical variables may have contributed less than expected to the combined model’s efficacy. The anterior uterocervical angle, cervical length, maternal age, and radiomics score were independent predictors. The combined model demonstrated good calibration and provided a positive net benefit across a reasonably broad threshold probability range (0.01–0.99 in the training set and 0.02–0.99 in the validation set). Conclusion Ultrasound radiomics has a promising value for predicting spontaneous preterm birth in the third trimester and its incorporation with clinical indicators may improve the predictive performance in clinical practice. These findings require validation in large multicenter prospective cohort studies before clinical translation.