Automatic segmentation of ventricular volume by 3D ultrasonography in post haemorrhagic ventricular dilatation among preterm infants

Abstract To train, evaluate, and validate the application of a deep learning framework in three-dimensional ultrasound (3D US) for automatic segmentation ventricular volume preterm infants with post haemorrhagic dilatation (PHVD). We trained 2D convolutional neural network (CNN) from 3D US PHVD. The method was validated Dice similarity coefficient (DSC) intra-class (ICC) compared to manual segmentation. mean birth weight included patients 1233.1 g (SD 309.4) gestational age 28.1 weeks 1.6). A total 152 serial 10 PHVD were analysed. 230 ventricles manually segmented. Of these, 108 used training CNN 122 validating methodology global agreement versus automated measures validation data (n = 122) excellent an ICC 0.944 (0.874–0.971). 0.8 (± 0.01). based estimation through software developed improves accuracy reduces processing time needed using VOCAL. should be considered promising tool help deepen our current understanding complex evolution