Combining Sequence Learning and Dilated Convolution Network for Hippocampus Segmentation
Chao Jia, Changrun Jia, Jianjing Wei · 2021
In the volume image of brain MRI, the volume of hippocampus is small, the boundary between hippocampus and surrounding tissue is fuzzy. In order to improve the accuracy of hippocampal segmentation, a new three-dimensional convolutional network is proposed which combines sequence learning and dilated convolution. The channel number of the convolution layer in the network is distributed in a downward increasing way, which effectively reduces the size of the parameters. Secondly, three-dimensional dilated convolution is used as the cascade convolution operation, which effectively combines the deep and shallow features of brain MRI, expands the receptive field of convolution layer and obtains multi-scale feature information, thus improving the segmentation performance of the network and greatly improving the segmentation accuracy. Experiments are carried out on the ADNI dataset, using dice similarity coefficient as evaluation indexes, and the accuracy reaches 89.23%. Experimental results show that the proposed network model makes full use of the three-dimensional spatial information of brain MRI images, and has better feature expression ability, thus greatly improving the segmentation accuracy.