(770) The Efficacy of Deep Learning Based Automated Cervical Length Measurement for Predicting Preterm Birth

Objective: Cervical length measurement of less than 2.5 cm in pregnant women between 16 weeks 0 days and 24 weeks 6 days is commonly used to predict preterm labor. However, conventional measurements are highly operator-dependent and often underestimate actual cervical length, which can lead to overtreatment. This study aimed to compare conventional cervical length measurement with deep learning-based automated measurement to determine any differences in preterm prediction.

Study Design: The study included 197 women out of 1270 who received antenatal care and delivery at Severance Hospital from 2019 to 2023. Participants had cervical lengths measured conventionally as 2.5 cm or less within the specified gestational period. Those with preterm labor, multiple pregnancies, or who underwent cerclage were excluded. Cervical length was measured using traditional and deep learning methods. Deep learning program utilized a semantic image decomposition network to divide images into pre-cervix, cervix, and post-cervix regions for cervical canal identification. The canal was assessed at six points for distance calculation. Participants were divided into Group A (cervical length of 2.5 cm or less by both methods) and Group B (less than 2.5 cm by conventional but more than 2.5 cm by deep learning). The primary endpoint was preterm birth, defined as delivery before 37 weeks.

Results: Group A had an average cervical length of 2.32 cm (+-0.159), while Group B's average was 2.73 cm (+-0.317). The preterm birth rate was 16.2% (9/152) in Group A and 4.8% (21/197) in Group B. Group B had a significantly lower preterm birth rate (P < 0.05), suggesting that cervical tracing provides longer cervical length measurements, improving preterm birth prediction accuracy.

Conclusion: Cervical length measurement through tracing offers more accurate preterm birth predictions than conventional methods. Deep learning for CL measurement can also reduce patient discomfort, shorten examination times, and provide operator independence. Integrating deep learning AI for cervical length measurement has important implications for future practices.