Such problems as outlier and low edge segmentation accuracy easily appear in the existing convolutional neural networks for segmenting medical images.Therefore,we adopt an edge predictor module based on the attention mechanism,the variational representation of activation function,and the geodesic active contour model to develop a medical image segmentation network.Then,we design an algorithm to train the end-to-end network.The experimental results on two common datasets demonstrate that compared with other segmentation methods,the proposed method can extract more edge information and hold higher accuracy.
关键词
卷积神经网络/医学图像分割/测地活动轮廓/边缘预测/注意力机制
Key words
convolutional neural network/medical image segmentation/geodesic active contour/edge predictor/attention mechanism
维普
万方数据