Rethinking Pooling Operation for Liver and Liver-Tumor Segmentations

Deep convolutional neural networks (DCNNs) have been widely used in medical image segmentation due to their excellent feature learning ability. In these DCNNs, the pooling operation is usually for down-sampling, which can gradually reduce resolution and thus expands receptive field of convolution kernel. Although has above advantages, it inevitably causes information loss during down-sampling process. This paper proposes an effective weighted address problem loss. First, we set up a window with learnable parameters, then update parameters training Secondly, use improve full-scale skip connection enhance multi-scale fusion. We evaluated on two public benchmark datasets, LiTS2017 CHAOS. The experimental results show that proposed effectively network performance accuracy liver liver-tumor segmentation.