Research map / Architectures and efficiency / Attention and transformers
Topology-Reinforced Swin Transformer for Medical Image Analysis
Pengfei Gu, Huimin Li, Guangyu Meng, Hao Zheng, Haoteng Tang, Bin Fu, Danny Z Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a topology-reinforced Swin Transformer that explicitly represents and incorporates multi-scale topological patterns (connected components, loops) into hierarchical ViT features for medical image analysis.
Problem
Hierarchical Vision Transformers such as Swin Transformer lack an explicit architectural scheme to represent and incorporate multi-scale topological structure (connected components, loops) in medical images, from microaneurysm-scale lesions to organ-level boundaries.
Approach
Adds an explicit topology-aware mechanism to the Swin Transformer hierarchy so that multi-scale topological features (connected components and loops) are represented and fed into the network; exact mechanism not stated in the abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Band-Aware H0/H1 residual refinement (ablation) | Topo-SwinV2-B / Topo-Swin-UNet · CBIS-DDSM / Synapse | accuracy / Dice: 73.58 -> 75.62 acc; 81.03 -> 82.19 Dice | Improved |
| inject topology as attention-logit bias rather than concatenation (ablation) | TRiST · CBIS-DDSM | accuracy: 72.73 -> 75.62 | Improved |
| inject topology as attention bias rather than residual gating (ablation) | TRiST · Synapse | Dice: 80.96 -> 82.19 | Improved |
| progressive band routing instead of uniform short band (ablation) | TRiST · Synapse | Dice: 77.19 -> 82.19 | Improved |
| progressive band routing instead of uniform medium band (ablation) | TRiST · CBIS-DDSM | AUC: 82.19 -> 82.01 | Mixed |
| compute topology natively at all four stages instead of stage 1 only (ablation) | TRiST · CBIS-DDSM | accuracy: 74.01 -> 75.62 | Improved |
| compute PH natively per stage instead of downsampling stage-1 topology (ablation) | TRiST · Synapse | HD95: 22.05 -> 11.10 | Improved |
| add topology representation to Swin-UNet | Swin-UNet · Synapse | Dice: 79.76 -> 82.19 | Improved |
| add topology to SwinV2-B | SwinV2-B · RetinaMNIST | accuracy: 67.67 -> 70.84 | Improved |
| 3D methods versus topology-enhanced 2D model | Topo-Swin-UNet · Synapse | Dice: 82.19 vs 86.62 (UNETR++) | Hurt |
| choice of topology descriptor (PI, BC, PS, PL) (ablation) | Topo-SwinV2-B · classification datasets | accuracy | Mixed |
Experimental setup
Builds on: Swin Transformer
Closest papers by meaning
- TopoAgent: An Agentic Framework for Automated Topology Learning in Medical Imaging (ECCV 2026 · similarity 0.90)
- Conformable Convolution for Topologically Constrained Learning of Complex Anatomical Structures (AAAI 2026 · similarity 0.89)
- TopoCL: Topological Contrastive Learning for Medical Imaging (CVPR 2026 · similarity 0.89)
- TopoDiffusionNet: A Topology-aware Diffusion Model (ICLR 2025 · similarity 0.86)
- Unsupervised Domain Adaptation for Anatomical Structure Detection in Ultrasound Images (ICML 2024 · similarity 0.85)
- STREAM: Stochastic Riemannian Flow Matching with Anisotropic Decoder for Digital Histopathology Image Generation (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
