Radiomic Analysis of Transvaginal Ultrasound Cervical Images for Prediction of Preterm Birth

Preterm birth (PTB) is the leading cause of mortality in neonates and children under five years around the world. Identification of women at high risk of PTB at early pregnancy stages is fundamental for devising intervention strategies. In this work, we investigate the use of radiomic features (RF) extracted from transvaginal ultrasound (TVUS) cervical images for PTB risk assessment. Radiomic features correspond to computerized texture descriptors extracted from medical images. In this retrospective study conducted on a sample of 253 women, from which 28 resulted in PTB, we perform preprocessing and feature extraction on TVUS images to assess the potential of RF as imaging biomarkers for PTB risk assessment. We evaluate performance using the area under the ROC curve (AUC) estimated with leave-one-out cross validation. For comparison, we consider known imaging biomarkers of PTB, namely the cervical length (CL) and the cervical consistency index (CCI). Experimental results yielded AUCs of 0.61, 0.68, and 0.70 for CL, CCI, and RF, respectively. When combining CL and CCI with RF, the AUC increased to 0.71 and 0.74. This highlights the potential of radiomic analysis in PTB risk assessment.