A Multiscale-Based Adjustable Convolutional Neural Network for Multiple Organ Segmentation

Figure 1

The architecture of the proposed network. The accuracy-complexity adjustment module (ACAM) is applied all-through the encoder and decoder. and represent the resolution and channels of the feature map, respectively. is a hyperparameter controlling the ratio of the low frequency along the channel dimension. The multiscale aggregation module (MAM) is added after the encoder to enrich feature representation.