EP01.05: Prenatal estimation of fetal femur bone mineral density using an artificial intelligence‐based algorithm and relationship with birthweight: a prospective validation study

To evaluate the relationship between the bone mineral density (BMD) of the fetal femur estimated by means of an artificial intelligence (AI)-based algorithm and the birthweight. Single-centre prospective study conducted at a tertiary maternity unit in Italy between November 2023 and February 2024 and including a non-selected cohort of singleton pregnancies undergoing routine prenatal care assessment at 35-37 weeks. In such cases a standard image of the fetal femur was acquired using the same US machine and probe and preset offline processed using a previously developed AI-based algorithm allowing the automated quantification of the pixels contained in the femoral area together with their relative brightness with the aim to estimate the BMD of the fetal femur. For the purposes of the study the output of the algorithm was evaluated in relation to the birthweight and the postnatal diagnosis of small-for-gestational age (SGA). 141 cases were included, of whom 18 (12.9%) had a postnatal diagnosis of SGA. A negative correlation between the algorithm output and birthweight and birthweight centile was demonstrated after adjusting for femur length(FL) (-0.282, p < 0.01 and -0.223, p = 0.01, respectively), FL+gain (-0.351, p < 0.01 and -0.278, p < 0.01, respectively), FL+gain+maternal BMI (-0.376, p < 0.01 and -0.297, p < 0.01, respectively) and also for gain+maternal BMI (-0.221, p = 0.02 and -0.202, p = 0.03, respectively). The mean algorithm output adjusted for FL, FL+gain and FL+gain+maternal BMI was higher in fetuses with confirmed SGA (p < 0.01 for all). The area-under-the curve for SGA was 0.748, 95% CI(0.596-0.899), p < 0.01 for the algorithm output adjusted for FL, 0.746,95% CI(0.608-0.883), p < 0.01 for the algorithm output adjusted for FL+gain and 0.733,95% CI(0.592-0.873), p < 0.01 for the algorithm output adjusted for FL+gain+maternal BMI. After adjustment for potential confounders the developed AI-based algorithm is associated with birthweight and with the postnatal diagnosis of SGA.