融合双解码和全局注意力上采样模块的胰腺囊性肿瘤分割网络

戴俊龙, 何聪, 武杰, 边云

Pancreatic Cystic Neoplasms Segmentation Network Combining Dual Decoding and Global Attention Upsampling Modules

Dai Junlong, He Cong, Wu Jie, Bian Yun

表2 不同文献中的模型和本文模型的Dice相似系数对比

Table 2 Comparison of Dice similarity coefficients between the model in different literature and the model in this paper

模型 数据来源 数据类型 Dice/% VA-3DUNet[21] NIH 3D(CT) 86.19±9.03 Attention U-Net[9] CT-150 3D(CT) 84.00±8.70 Cascaded multitask 3-D fully convolutional networks[11] 82(CT) 3D(CT) 85.9±5.10 HSSN[12] MRI 3D(MRI) 85.7±2.3 PanKNet[13] Pancreas-MRI 3D(MRI) 77.46±8.62 ConResNet[22] NIH 3D(CT) 86.06 PNet-MSA[8] MRI-79 2D(MRI) 80.7±7.40 Spatial aggregation of holistically-nested convolutional neural networks[23] NIH 2D(CT) 81.27±6.27 deep convolutional LSTM neural network[24] MRI-79 2D(MRI) 80.5±6.70 RSTN[25] NIH 2D(CT) 84.50±4.97 TransUnet[26] MICCAI 2015 2D(CT) 73.96 DGANet（本文模型） 长海医院 2D(MRI) 86.28±2.77