Research map / Graph learning / Knowledge and temporal graphs
XFibrosis: Explicit Vessel-Fiber Modeling for Fibrosis Staging from Liver Pathology Images
Read the paperCheck your idea against 108,967 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 XFibrosis, which represents liver biopsy tiles as vessel-induced primal graphs and fiber-induced dual graphs and processes them with a primal-dual graph convolution for fibrosis staging.
Problem
Fibrosis staging from liver pathology images (NAFLD) where existing deep learning methods model tissue patches but not the explicit relationship between vessels and fibers, limiting accuracy and interpretability.
Approach
Build primal graphs from vessel key points (Delaunay triangulation) and their primal-dual dual graphs from fiber key points, then apply a primal-dual GCN that propagates within each graph and between them via edge-enhanced convolution, with TopK-style tile selection for WSI-level pooling.
Key results
- On Liver-Fibrosis-SR, XFibrosis reaches 84.00% AUC and 53.63% accuracy, versus 82.40% AUC for the next best (IBMIL).
- On FibrosisMT-v1.0, XFibrosis reaches a mean AUC of 92.82% and accuracy of 64.40%, versus 84.45% AUC for IBMIL.
- Adding PDGCN to PDEmbedding raises AUC from 80.14% to 84.00% and accuracy from 48.77% to 53.63% on Liver-Fibrosis-SR.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use explicit vessel-fiber graphs and PDGCN instead of IBMIL | Fibrosis staging · Liver-Fibrosis-SR | AUC: 82.40±2.99 -> 84.00±3.59 | Helped |
| use explicit vessel-fiber graphs and PDGCN instead of IBMIL | Fibrosis staging · FibrosisMT-v1.0 | AUC: 84.45±5.82 -> 92.82±7.42 | Helped |
| add PDGCN to PDEmbedding (ablation) | XFibrosis · Liver-Fibrosis-SR | AUC: 80.14±3.15 -> 84.00±3.59 | Helped |
| add PDGCN to PDEmbedding (ablation) | XFibrosis · Liver-Fibrosis-SR | Accuracy: 48.77±4.63 -> 53.63±7.97 | Helped |
| replace PDGCN with GEN (ablation) | XFibrosis · Liver-Fibrosis-SR | AUC: 84.00±3.59 -> 82.85±2.39 | Hurt |
| replace PDGCN with GATv2 (ablation) | XFibrosis · Liver-Fibrosis-SR | AUC: 84.00±3.59 -> 82.98±1.75 | Hurt |
| replace PDGCN with PDN (ablation) | XFibrosis · Liver-Fibrosis-SR | AUC: 84.00±3.59 -> 82.14±3.14 | Hurt |
| replace PDGCN with Transformer (ablation) | XFibrosis · Liver-Fibrosis-SR | AUC: 84.00±3.59 -> 81.10±2.44 | Hurt |
Limitations
- Evaluated on only two small datasets (132 and 47 WSIs)
- Relies on mean-shift vessel/fiber segmentation whose quality bounds graph construction
- Only vessels and fibers modeled; other tissue or cellular components not considered
Experimental setup
Datasets: Liver-Fibrosis-SR FibrosisMT-v1.0
Benchmarks: Liver-Fibrosis-SR FibrosisMT-v1.0
Metrics: AUC accuracy
Baselines: ReMix DTFD IBMIL GATv2 ARMA HGNN GNN-Fibrosis GEN PDN Transformer-based graph conv
Builds on: primal-dual graph theory Delaunay triangulation TopK pooling GATv2 ReMix
Closest papers by meaning
- vesselFM: A Foundation Model for Universal 3D Blood Vessel Segmentation (CVPR 2025 · similarity 0.83)
- Dynamic Entity-Masked Graph Diffusion Model for Histopathology Image Representation Learning (AAAI 2025 · similarity 0.83)
- STREAM: Stochastic Riemannian Flow Matching with Anisotropic Decoder for Digital Histopathology Image Generation (NeurIPS 2026 · similarity 0.81)
- MEATRD: Multimodal Anomalous Tissue Region Detection Enhanced with Spatial Transcriptomics (AAAI 2025 · similarity 0.81)
- Semantic and Visual Crop-Guided Diffusion Models for Heterogeneous Tissue Synthesis in Histopathology (NeurIPS 2025 · similarity 0.80)
- Leveraging Tumor Heterogeneity: Heterogeneous Graph Representation Learning for Cancer Survival Prediction in Whole Slide Images (NeurIPS 2024 · similarity 0.80)
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.