四川大学学报（自然科学版）2026，Vol.63Issue(1)：46-57,12.DOI:10.19907/j.0490-6756.250266
BiF-SegNet:双分支特征融合的腹部多器官图像分割方法
BiF-SegNet:Bidirectional feature fusion network for abdominal multi-organ segmentation
摘要
Abstract
Abdominal multi-organ segmentation remains challenging due to large variations in organ shape,uneven spatial distribution,and inter-subject anatomical differences.Existing methods often face a trade-off between accurate boundary modeling,multi-scale feature integration,and computational efficiency,and still struggle when segmenting small,blurred,or highly variable organs.To address these limitations,we pro-pose BiF-SegNet,a Bidirectional Feature Fusion segmentation network designed to capture local fine-grained details and global contextual information.The network consists of two parallel encoders:a Local En-coder(LE)using pixel-difference convolutions to enhance boundary and edge features,and a Global Encoder(GE)based on Transformer with an efficient spatial attention reduction module,maintaining global semantic modeling while reducing computational complexity.Furthermore,a Multi-Attention Feature Fusion(MAF)module achieves deep complementary integration between local and global features,improving representation and robustness.Experiments were conducted on two publicly available multi-organ datasets and one cardiac dataset.The results demonstrate that the proposed method outperforms state-of-the-art approaches in terms of average Dice and HD95,while maintaining a lightweight design with only 26.74 M parameters and 6.85 G FLOPs,achieving an excellent balance between accuracy and efficiency.关键词
多器官图像分割/特征融合/像素差异卷积/空间缩减注意力Key words
multi-organ image segmentation/feature fusion/pixel difference convolution/spatial reduction attention分类
信息技术与安全科学引用本文复制引用
黄惠涛,穆楠,李晓宁..BiF-SegNet:双分支特征融合的腹部多器官图像分割方法[J].四川大学学报（自然科学版）,2026,63(1):46-57,12.基金项目
国家自然科学基金(62006165) （62006165）
