Brain stroke lesion segmentation using consistent perception generative adversarial network

The state-of-the-art deep learning methods have demonstrated impressive performance in segmentation tasks. However, the success of these depends on a large amount manually labeled masks, which are expensive and time-consuming to be collected. In this work, novel consistent perception generative adversarial network (CPGAN) is proposed for semi-supervised stroke lesion segmentation. CPGAN can reduce reliance fully samples. Specifically, similarity connection module (SCM) designed capture information multi-scale features. SCM selectively aggregate features at each position by weighted sum. Moreover, strategy introduced into model enhance effect brain prediction unlabeled data. Furthermore, an assistant constructed encourage discriminator learn meaningful feature representations often forgotten during training stage. employed jointly decide whether results real or fake. was evaluated Anatomical Tracings Lesions After Stroke (ATLAS). experimental demonstrate that achieves superior performance. task, using only two-fifths samples outperforms some approaches full