Highresolution Net for Estimation of Fetal Growth Parameters from First Trimester Ultrasound Images

Crown-rump length (CRL) from ultrasound is the the standards for estimating gestational age in the first trimester, however it must be measured manually. This work introduce Highresolution Net (HRNet), a deep learning model for fully automated fetus contour recognition and CRL measuring using ultrasound. The architecture integrates multi-scale contextual information as well as fine detail to ensure exact localization. Highresolution Net detects the fetus shape with 0.926 Dice similarity across 3205 images. The CRL measuring error is 0.67mm compared to experts. Derived gestational ages allow for delivery date prediction within 7 days in 96% of cases. Automated CRL quantification and pregnancy date in the first trimester may reduce interobserver variability and enhance workflow. A reliable measurement of early fetal growth will improve surveillance and anomaly prediction.