Segmentation of Corpus Callosum in Midsagittal using Convolutional Neural Networks with Anatomical Information

Abstract

Corpus Callosum (CC), the largest white matter structure, is connector between the two cerebral hemispheres in human brain. Structural features of CC such as shape and size have been used to study various neurological diseases.Robust segmentation of CC in midsagittal plane is critical in the qualitative studies. In this paper, we introduced a convolutional neural networks (CNN) with anatomical information for CC segmentation. Our method showed better segmentation performance (mean Dice index: 95.44±0.9859) than other methods (mean Dice index: 95.22±1.2532). We concluded that anatomical information integrated in CNN improve the segmentation performance significantly.