There is urgent need for accurate segmentation algorithms for infant brain magnetic resonance (MR) images, which is significant for the development of infant brain science in the future. However, the infant MR images brain segmentation is still an extremely challenged task since the fetus is in the process of myelin formation and maturation, its brain tissue is not yet fully developed. There is poor contrast between gray matter(GM) and white matter(WM) in brain tissues in both T1-weighted (T1w)

more »

... and T2weighted (T2w) images. To solve this problem, we proposed a Dense-Fuse network, which can fully use the ability of Dense Block feature multiplexing and information flow downward transfer, to preserve the detailed features in poor contrast between GW and WM. In addition, the model also uses feature maps fusion to achieve information complementary between different modal images, which is beneficial to make full use of the superior information between different modalities. We compared with other methods at the task of MICCAI iSeg-2017 Challenge. The results demonstrates that our model are very competitive and are the optimal one according to multiple indicators.