基于KAN和U-Net网络的颌面结构全景分割方法

蔡智,周正东,袁晓曦,杨泽毅,袁梦瑶

A panoramic segmentation method for maxillofacial structures based on KAN and U-Net networks

Zhi CAI,Zhengdong ZHOU,Xiaoxi YUAN,Zeyi YANG,Mengyao YUAN

表 1 不同的方法Dice系数比较

Tab.1 Comparison of Dice coefficients of different methods

网络 Dice/% 均值（不计牙髓） 颌骨 IAC 上颌窦 咽喉 牙齿 修复体 牙髓 CNNs nnU-Net[20] 70.92 90.31 71.34 64.81 95.66 73.17 29.50 — nnU-Net ResEnc[20] 74.16 91.77 73.01 65.71 95.26 76.48 37.10 — Transf. TransU-Net[20] 70.32 90.33 81.96 59.69 87.89 72.46 27.68 — nnFormer[20] 76.79 91.50 72.28 75.11 90.85 79.37 18.95 — UNETR++[20] 71.43 92.38 68.87 74.51 91.60 73.70 26.21 — Mamba UMamba[20] 85.05 90.05 85.26 77.02 92.18 86.58 43.89 — VMamba[20] 73.13 90.75 60.62 73.20 88.23 75.99 14.86 — Swin-UMamba[20] 79.64 94.29 72.64 87.75 93.05 80.71 25.30 — 3-Stage 3D UKAN_SE 86.51 94.95 82.62 76.84 95.96 89.49 80.49 86.09 PMFSNet3D 87.96 95.94 84.99 73.80 95.96 88.82 80.27 85.10 本研究模型 88.87 95.95 84.22 82.09 95.96 89.51 81.18 87.55