Quantitative analysis of cervical texture by ultrasound in mid-pregnancy and association with spontaneous preterm birth

Objective New tools are required to improve the identification of women who are at increased risk for spontaneous preterm birth (sPTB). Quantitative analysis of tissue texture on ultrasound has been used to extract robust features from the ultrasound image to detect subtle changes in its microstructure. This may be applied to the cervix. The aim of this study was to determine if there is an association between quantitative analysis of cervical texture (CTx) on mid-trimester ultrasound and sPTB < 37 + 0weeks' gestation. Methods This was a single-center nested case-control study of a prospective cohort of 677 consecutive women with singleton pregnancy assessed between 19 + 0 and 24 + 6weeks' gestation. Women at increased risk for sPTB were included unless they received treatment to prevent sPTB. Women who delivered < 37 + 0weeks (sPTB) were considered as cases and were matched in a 1: 10 ratio with randomly selected contemporary controls who delivered at term. For each woman, one ultrasound image of the cervix was obtained for which quality was assessed, cervical length (CL) measured offline and a region of interest in the midportion of the anterior cervical lip delineated for use in local binary patterns analysis of CTx. A learning algorithm was developed to obtain the combination of CTx features best associated with sPTB based on feature transformation and discriminant analysis regression. The ability of the learning algorithm to predict sPTB was evaluated using a leave-one-out cross-validation technique, which produced a CTx-based score for each participant. Receiver-operating characteristics (ROC) curves were produced and sensitivity, specificity, and positive and negative likelihood ratios were calculated for the optimal cut-off based on the ROC curve. The results were compared with those obtained for CL. Investigators studying the images were blinded to pregnancy outcome at all times. Results Images from 310 women (27 cases and 283 controls) were of sufficient quality and included in the study. Median CTx-based score was significantly lower in cases compared with controls (-1.01 vs -0.07, P <= 0.0001). CTx-based score maintained its significant association with sPTB after adjusting for possible confounders (history of sPTB, conization or Mullerian malformation, and CL< 25mm). CTx-based score was a better predictor of sPTB (AUC, 0.77; 95% CI, 0.66-0.87) than was CL (AUC, 0.60; 95% CI, 0.47-0.72) (P = 0.03). Median CL was similar for cases and controls (37.7 vs 38.6mm, P = 0.26), although cases were more likely to have CL < 25mm (18.5% vs 0.4%, P < 0.001). Conclusion Quantitative analysis of CTx enables the extraction of information relevant to sPTB from ultrasound images to generate a CTx-based score that is associated independently with sPTB. Copyright (C) 2017 ISUOG. Published by John Wiley & Sons Ltd.