基于轻量高频Transformer与特征互补融合的视网膜血管分割

梁礼明,王成斌,钟奕,陈林俊,吴健

Retinal vessel segmentation based on lightweight high-frequency Transformer and feature complementary fusion

Liming LIANG,Chengbin WANG,Yi ZHONG,Linjun CHEN,Jian WU

表 2 所提算法与先进算法在3个数据集上的血管分割结果对比

Tab.2 Comparison of vessel segmentation results of proposed algorithm and state-of-the-art algorithms on three datasets

数据集 模型 ACC/% SE/% SP/% AUC/% F1/% DRIVE SFIT-Net[27] 97.07 81.59 98.55 98.75 82.97 PA-Net[28] 95.82 82.84 98.07 98.33 83.93 DAE-Former[29] 95.92 79.28 98.46 97.80 83.73 MSM-TDE[30] 96.66 84.92 97.23 97.80 79.30 BINet[31] 96.06 86.92 97.37 — 84.25 MSTP-Net[32] 96.91 83.68 98.18 — 82.58 DAU-Net[33] 95.85 81.55 98.15 98.18 82.99 LFF-Net 97.12 80.23 98.74 98.83 82.99 STARE SFIT-Net[27] 97.50 82.18 98.92 99.10 83.37 PA-Net[28] 97.09 88.13 98.05 99.08 85.61 DAE-Former[29] 97.06 82.66 98.66 98.97 84.78 MSM-TDE[30] 97.26 86.90 98.22 98.09 83.70 BINet[31] 96.16 82.76 97.76 — 81.33 MSTP-Net[32] 97.61 86.03 98.58 — 84.68 DAU-Net[33] 97.12 85.80 98.43 99.08 86.20 LFF-Net 97.62 80.48 99.02 99.12 83.73 CHASE_DB1 SFIT-Net[27] 97.53 82.19 98.56 98.81 80.76 PA-Net[28] 96.77 85.70 97.79 98.75 83.08 DAE-Former[29] 96.60 83.28 97.92 98.70 81.61 MSM-TDE[30] 96.67 86.02 97.53 96.45 78.05 BINet[31] 96.04 83.93 97.34 — 80.47 MSTP-Net[32] 97.45 84.85 98.30 — 80.74 DAU-Net[33] 97.00 83.64 98.35 98.94 84.99 LFF-Net 97.65 81.30 98.75 98.99 81.37