Semantic segmentation of lung radiographs using U-net type neural network

In this paper, we proposed semantic segmentation of lung radiographs using Transfer learning. A network of the U-net type is used, in which the weights of the EfficientNet network, pre-trained on the ImageNet set, were taken as an encoder. As a result of additional training of the network for 50 epochs, segmentation accuracy was achieved above 90% for the light and rib classes, and 80% for the heart class.