Abstract：The unclear structure of intracranial hemorrhage,the presence of artifacts and other brain tissue noise have a serious influence on the segmentation task.To solve this problem,in order to improve the performance of intracranial hemorrhage segmentation,we propose a segmentation method for intracranial hemorrhage that fuses dense connections and attention mechanisms.In the encoder part of the full convolutional network,dense blocks are introduced for intracranial hemorrhage feature extraction,as not all the features extracted from the encoders are useful for segmentation.For this reason,we propose to incorporate an attention mechanism including a spatial and a channel attention,to architecture to re-weight the feature representation spatially and channel-wise to capture rich contextual relationships for better feature representation.In addition,the focal Tversky loss function is introduced to deal with small intracranial hemorrhage segmentation.The experimental results demonstrate the proposed method can achieve an accurate and rapid segmentation on intracranial hemorrhage segmentation.The obtained Precision,Sensitivity and Dice are 89.3%,87.58% and 88.28%,respectively.
2021, Vol. 42 
