A regression framework to head-circumference delineation from US fetal images

• Head-circumference (HC) has an important role in the evaluation of fetal growth. We propose automatic algorithm to directly delineate HC US images. The approach is based on a regression CNN trained with distance field. A mean absolute difference = 1.9 mm was obtained testing HC18 Challenge dataset. Measuring head-circumference length from ultrasound (US) images crucial clinical task assess fetus To lower intra- and inter-operator variability measuring, several computer-assisted solutions have been proposed years. Recently, large number deep-learning approaches addressing problem delineation through segmentation whole head via convolutional neural networks (CNNs). Since edge-delineation problem, we different strategy CNNs. framework consists region-proposal for localization centering, accurately HC. first exploiting transfer learning, while training fields. tested dataset, which 999 335 1.90 ( ± 1.76) Dice similarity coefficient 97.75 1.32) % were achieved, overcoming literature. experimental results showed effectiveness framework, proving its potential supporting clinicians during practice.