Utilização de Redes Neurais Completamente Convolucionais para identificação e medição de crânios fetais

Ultrasonography is a non-invasive and nonradioactive imaging technique frequently used in the monitoring of fetal development during pregnancy. The precise assessment is important for the analysis of fetal growth, diagnosis of malformations and possible congenital diseases, thus ensuring the well-being of the mother and the fetus during pregnancy. However, for accurate measurement of anatomical structures of the fetus, specialized knowledge of the obstetrician is necessary. In addition to being a tedious and time-consuming process, contour extraction is influenced by his experience. In this sense, this work proposes the adaptation of the fully convolutional neural network V-Net for segmentation of fetal skulls present in two-dimensional ultrasound images, in addition to the construction of an algorithm for calculating the circumference of them. Preliminary results suggest that the methodology proposed can be automatically used as an auxiliary tool in the monitoring of fetal development.