García Ocaña, María Inmaculada2024-05-202024-05-202019-09-26https://hdl.handle.net/20.500.14468/14155This work explores the use of radiomics and machine learning to extract relevant biomarkers from ultrasound (US) images that can be used in obstetric practice. Two clinical applications are studied: the prediction of induction of labor (IOL) failure based on clinical data and US images obtained prior to IOL, and the estimation of risk of preterm birth based on routinely US images acquired in the 20th week of pregnancy. Several machine learning classifiers and feature selection techniques are tested and the results are compared. The best model for the prediction of IOL failure was a random forest that model obtained an AUC of 0.75, with 69% sensitivity and 71% specificity. The best model for the prediction of preterm birth was a random forest that obtained an AUC of 0.77 AUC, with 71% sensitivity and 69% specificity . These preliminary results suggest that features obtained from US images can be used to estimate risks in these two obstetric problems. Transvaginal US is cheap, widely available at hospitals, and performed routinely. Therefore these method can be easily implemented in clinical practice and help practitioners choose a most personalized treatment for each patient, improving the outcomes. Further validation with a largest and more diverse dataset is needed, especially to assess how the image analysis methods work with images from different US vendors.esAtribución-NoComercial-SinDerivadas 4.0 Internacionalinfo:eu-repo/semantics/openAccesstesis de maestría