Jin Yang, Daniel S. Marcus, and Aristeidis Sotiras
U-Net has been widely used for segmenting abdominal or- gans, achieving promising performance. However, when it is used for multi-organ segmentation, first, it may be limited in exploiting global long-range contextual information due to the implementation of stan- dard convolutions. Second, the use of spatial-wise downsampling (e.g., max pooling or strided convolutions) in the encoding path may lead to the loss of deformable or discriminative details. Third, features up- sampled from the higher level are concatenated with those that perse- vered via skip connections. However, repeated downsampling and up- sampling operations lead to misalignments between them and their con- catenation degrades segmentation performance. To address these limita- tions, we propose Dynamically Calibrated Convolution (DCC), Dynam- ically Calibrated Downsampling (DCD), and Dynamically Calibrated Upsampling (DCU) modules, respectively. The DCC module can utilize global inter-dependencies between spatial and channel features to cali- brate these features adaptively. The DCD module enables networks to adaptively preserve deformable or discriminative features during down- sampling. The DCU module can dynamically align and calibrate up- sampled features to eliminate misalignments before concatenations. We integrated the proposed modules into a standard U-Net, resulting in a new architecture, termed Dynamic U-Net. This architectural design enables U-Net to dynamically adjust features for different organs. We evaluated Dynamic U-Net in two abdominal multi-organ segmentation benchmarks. Dynamic U-Net achieved statistically improved segmenta- tion accuracy compared with standard U-Net. Our code is available at https://github.com/sotiraslab/DynamicUNet.