Paper Overview
- Field: Computer Vision (CV) / Clinical AI
- Authors: Yunsung Chung, Yingshuo Liu, Abboud F. Hassan, Han Feng, Mary M. Maleckar, Nassir Marrouche, Jihun Hamm
- Posted: 2026-08-13
- arXiv: 2608.13518
- AUROC: 0.756
- AUPRC: 0.777
- Scar-extent MAE: 2.971 percentage points (without follow-up MRI intensities at inference)
- Problem: Post-operative recovery is irregular and asynchronous, breaking the common one-step baseline-to-endpoint prediction paradigm.
- Method: A world-model framework with a 3D spatial latent state that is dynamically updated by procedural context, static covariates, elapsed time, and peri-event physiological embeddings.
- Training signal: Follow-up imaging serves only as a training-time latent forecasting objective; inference does not require follow-up MRI intensities.
- Clinical application: Atrial fibrillation (AF) ablation with a 90-day recovery window.
- Dataset: DECAAF-II, evaluated via repeated internal cross-validation.
- Results: AUROC 0.756, AUPRC 0.777 for recurrence; scar-extent MAE 2.971 percentage points.
- Capabilities: Multi-horizon recurrence-risk queries and retrospective editing of blanking-period records using the learned latent state.
Chinese Abstract (Translation)
Many clinical prediction models treat post-intervention outcomes as a one-step mapping from baseline measurements to a future endpoint. However, recovery after a procedure often unfolds as an irregular trajectory: clinical observations, medication changes, repeat interventions, and physiological measurements are recorded asynchronously and can change risk assessment over time.
We propose an intervention-aware clinical world model that represents each patient with a structured latent state and evolves it through time-ordered post-intervention events. The model first encodes baseline imaging into a 3D spatial latent state. It then updates this state using procedural context, static covariates, elapsed time, and peri-event physiological embeddings. Follow-up imaging provides training-only supervision through a latent forecasting objective.
We apply the framework to atrial fibrillation ablation. During the 90-day recovery window, irregular post-procedure records provide clinically meaningful evidence for long-term recurrence risk. In repeated internal cross-validation on DECAAF-II, our model achieves:
The learned state supports recurrence-risk queries at different horizons and retrospective input editing of blanking-period records.
Original Abstract
Many clinical prediction models treat post-intervention outcomes as a one-step mapping from baseline measurements to a future endpoint. However, recovery after a procedure often unfolds as an irregular trajectory: clinical observations, medication changes, repeat interventions, and physiological measurements are recorded asynchronously and can change risk assessment over time. We propose an intervention-aware clinical world model that represents each patient with a structured latent state and evolves it through time-ordered post-intervention events. The model first encodes baseline imaging into a 3D spatial latent state. It then updates this state using procedural context, static covariates, elapsed time, and peri-event physiological embeddings. Follow-up imaging provides training-only supervision through a latent forecasting objective. We apply the framework to atrial fibrillation ablation. During the 90-day recovery window, irregular post-procedure records provide clinically meaningful evidence for long-term recurrence risk. In repeated internal cross-validation on DECAAF-II, our model achieves AUROC 0.756 and AUPRC 0.777 for recurrence prediction. It also achieves a scar-extent MAE of 2.971 percentage points without requiring follow-up MRI intensities at inference. The learned state supports recurrence-risk queries at different horizons and retrospective input editing of blanking-period records.