Fetal echocardiographic image segmentation using neural networks

This paper discusses supervised and unsupervised neural network approaches to fetal echocardiographic image segmentation. The obtained results were compared with images segmented by a known unsupervised clustering technique (i.e. k-means). The visual aspect of the segmented images was evaluated with respect to its visual quality by an expert. A subset of the segmented images showed sufficient detail of the internal heart anatomy to allow medical diagnosis. The visual observation was matched closely by our unsupervised image segmentation approach, using the modified Hubert index.