X-WIN: Building Chest Radiograph World Model via Predictive Sensing

arXiv 25.11 2025 47.1 method, application

TLDR

X-WIN builds a chest radiograph world model by learning to predict 2D projections from CT, improving downstream tasks.

Reasoning

The paper presents a novel method for distilling 3D anatomical knowledge into a 2D world model, with strong empirical results. However, it is domain-specific and lacks interactive or video capabilities, limiting generalizability.

Read-first score

Read-first score 47.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 25.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation,experiment,metric

Topical relevance 42%

35.7

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 399.

Keyword Scores

world model

generative world model

world dynamics prediction

world simulator

interactive world model

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Distilling volumetric knowledge from chest CT into a CXR world model by learning to predict 2D projections in latent space
Affinity-guided contrastive alignment loss to capture correlated information across projections from the same volume
Incorporating real CXRs via masked image modeling and a domain classifier to align representations of real and simulated CXRs

Methodology

X-WIN learns a world model that internalizes 3D anatomical structure from CT scans by predicting 2D projections in latent space. It uses an affinity-guided contrastive alignment loss to leverage mutual similarities across projections, and incorporates real CXRs through masked image modeling and a domain classifier to encourage statistically similar representations. Evaluation is performed via linear probing and few-shot fine-tuning on downstream tasks.

Key Results

X-WIN outperforms existing foundation models on diverse downstream tasks using linear probing and few-shot fine-tuning, and demonstrates the ability to render 2D projections for reconstructing a 3D CT volume.

Resources

Paper