An Improved Method for General Purpose Medical Picture Signal Segmentation

It is recognized that if there is no large number of training samples, deep network training will not be successful. In this paper, data augmentation has played an important role in the network and training strategy, thus using the available samples which are annotated more sufficiently. The architecture reduces the semantic gap between the feature maps of encoder and decoder sub-networks [1]. There is no doubt that the feature maps from the decoder and encoder networks are semantically similar, making the learning task easier. Our research work focuses on general image segmentation, which includes heart, hippocampus, prostate, brain tumor et. al. with our optimized the loss function innovated network optimization module.