AI Paper Atlas

Research map / Graph learning / Graph self-supervised learning

VesselTok: Tokenizing Vessel-like 3D Biomedical Graph Representations for Reconstruction and Generation

Chinmay Prabhakar, Bastian Wittmann, Tamaz Amiranashvili, Paul Büschl, Ezequiel De la Rosa, Julian McGinnis, Benedikt Wiestler, Bjoern Menze, Suprosanna Shit

ECCV 2026 · Graph learning › Graph self-supervised learning · approach: Diffusion-based graph generation · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
tokenize vessel centerline occupancy instead of surface points (VesselTok vs 3DShape2VecSet/Hunyuan3D)VAE tokenizer · six anatomiesclDice / CD / |Δβ0|: e.g. COSTA clDice 59.65 -> 77.26; ATM 90.08 -> 96.33Improved
VesselTok vs 3DShape2VecSet on loop errorVesselTok · HiPas / PARSE / Pulmonary-AV|Δβ1|: HiPas 28.65 -> 38.42; PARSE 32.31 -> 34.06; Pulm 5.04 -> 5.52Hurt
VesselTok vs vessel-specific VesselGPTVesselTok · ATMclDice: 77.09 -> 96.61Improved
VesselTok vs VesselGPT on connected componentsVesselTok · ATM|Δβ0|: 0.05 -> 0.07Hurt
apply VesselTok to unseen anatomies and scalesVesselTok · TopCoW / renal / synthetic cutsclDice: TopCoW 97.96 -> 99.42; RV 79.66 -> 88.86Improved
use VesselTok tokens for latent diffusionEDM · generationFID: cond 73.58 -> 43.13; uncond 146.57 -> 96.87Improved
use VesselTok latent for infill vs autodecoderVesselTok · ATM infillclDice: 83.49 -> 88.13Improved
increase number of tokens K from 512 to 768 (ablation)VesselTok · ATMclDice: 96.61 -> 97.13; compression 7.03 -> 4.68No clear effect
reduce tokens K from 512 to 256 or fewer (ablation)VesselTok · ATMclDice: 96.61 -> 80.35 (256); 12.29 (128); 8.42 (64)Hurt
increase channel dim C from 4 to 16 (ablation)VesselTok · ATMclDice: 96.61 -> 96.32; compression 7.03 -> 1.76No clear effect
reduce channel dim C from 4 to 2 (ablation)VesselTok · ATMclDice: 96.61 -> 7.08Hurt
increase pseudo-radius r from 0.008 to 0.032 (ablation)VesselTok · ATMclDice loss vs beta1 error: clDice loss ~0.28 -> ~0.007; beta1 error ~4 -> ~16Mixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy