Rapid 3D reconstruction in fetal ultrasound imaging using artificial intelligence and medical 3D printing

Congenital heart disease (CHD) has been one of the most serious problems in newborns.Forfetalhearthealthcare,3Dmodelingandprintingtechnologyhavebeen adopted in the diagnosis of CHD during antenatal care. However, the development of 3D printing techniques and their clinical applications have been hindered by the manual processing of ultrasound (US) volume data in clinical practice. To overcome this problem, we present an interactive semi-automatic method based on deep learning that uses manual processing results from expert sonographers for training. The accuracy, interpretability, and variability of the performances were evaluated on the validation set. The results demonstrated that compared with a physician with less than 3 years of experience, a better Faster-region-based convolutional neural network-based threshold was achieved using our proposed fetal heart reconstruction technique (FRT), with enhanced performance based on the outflow tract view and three-vessel view. No significant difference was found among the clinical parameters, in proportion, measured from the model rebuilt using FRT and US volume data. Furthermore, the reconstruction time of the fetal heart blood pool model was reduced from approximately 5 h to 5 min. Our results indicate that deep learning has the ability to process US data accurately, representing an important step towards the reconstruction of the fetal heart digital model, which is critical for advancing clinical diagnosis and treatment of CHD during pregnancy.