Prediction of failure of induction of labor from ultrasound images using radiomic features

Fecha: 17.10.2019


Abstract

Induction of labor (IOL) is a very common procedure in current obstetrics; about 20% of women who undergo IOL at term pregnancy end up needing a cesarean section (C-section). The standard method to assess the risk of C-section, known as Bishop Score, is subjective and inconsistent. Thus, in this paper a novel method to predict the failure of IOL is presented, based on the analysis of B-mode transvaginal ultrasound (US) images. Advanced radiomic analyses from these images are combined with sonographic measurements (e.g. cervical length, cervical angle) and clinical data from a total of 182 patients to generate the predictive model. Different machine learning methods are compared, achieving a maximum AUC of 0.75, with 69% sensitivity and 71% specificity when using a Random Forest classifier. These preliminary results suggest that features obtained from US images can be used to estimate the risk of IOL failure, providing the practitioners with an objective method to choose the most personalized treatment for each patient.

BIB_text

@Article {
title = {Prediction of failure of induction of labor from ultrasound images using radiomic features},
pages = {153-160},
keywds = {
Induction of labor; Machine learning; Radiomics; Ultrasound
}
abstract = {

Induction of labor (IOL) is a very common procedure in current obstetrics; about 20% of women who undergo IOL at term pregnancy end up needing a cesarean section (C-section). The standard method to assess the risk of C-section, known as Bishop Score, is subjective and inconsistent. Thus, in this paper a novel method to predict the failure of IOL is presented, based on the analysis of B-mode transvaginal ultrasound (US) images. Advanced radiomic analyses from these images are combined with sonographic measurements (e.g. cervical length, cervical angle) and clinical data from a total of 182 patients to generate the predictive model. Different machine learning methods are compared, achieving a maximum AUC of 0.75, with 69% sensitivity and 71% specificity when using a Random Forest classifier. These preliminary results suggest that features obtained from US images can be used to estimate the risk of IOL failure, providing the practitioners with an objective method to choose the most personalized treatment for each patient.


}
date = {2019-10-17},
}
Vicomtech

Parque Científico y Tecnológico de Gipuzkoa,
Paseo Mikeletegi 57,
20009 Donostia / San Sebastián (España)

+(34) 943 309 230

close overlay