Сегментация сердца по данным магнитно-резонансной томографии методами глубокого обучения

Abstract

Manual heart segmentation from MRI data takes considerable time even with a highly qualified expert. Automatic segmentation by deep neural networks may shorten the time needed to diagnose of a single patient and speed up the payback of expensive equipment. In our work, we have analyzed various modifications of the U-NET deep neural network architecture for automatic segmentation of the left ventricular myocardium, the internal volume of the left and right ventricle according to MRI data into the end of diastole or systole.