Xucheng (David) Wang

Papers

Robot learning, embodied AI, and clinical intelligence.

* equal contribution · † co-second author

Figure available upon publication
Science · In press

Abandonment of Safe Medical Advice by Large Language Models Under Patient Resistance

Reddy A, Kim TM, McCoy L, Ramaswamy A, Wang X, Eiss B, Ganti A, Acholonu-Fomin N, Fortenko A, Sanapala S, Kim SE, Gupta S, Nadkarni G, Shah NH, Topol E, Luo L, Rajpurkar P

More info will be available.

Paper will be available on the publication date.
Looping SimRetract demonstration showing robotic retractor take-over
Machine Learning for Health (ML4H 2026)

SimRetract: Zero-Shot Sim-to-Real Robotic Assistance for Retractor Take-Over in Open Surgery

Wang X, Yang Z, Kim SE, Hardy R, Sampson MP, Reddy A, Pal A, Zhang X, Luo L, Rajpurkar P

Learns retractor take-over entirely from simulated task demonstrations and transfers zero-shot to the real robot while preserving the surgeon-established retraction.

Paper coming soonVideo
Clinical world model interface, hospital and patient engines, and agent tools
Under review at Nature, following first revision

Towards Dynamic Healthcare Intelligence with a Clinical World Model

Co-second author

An LLM-driven clinical world model for interactive simulation and agent evaluation in emergency-department settings, with a focus on patient prioritization, demographic equity, and care management under resource constraints.

Paper coming soon
Figure showing the open-surgery suture-following task
Machine Learning for Healthcare Conference (MLHC 2026)

Imitation Learning for Robot Assistance in Open Surgery: A Multi-Policy Evaluation on Suture Following

Wang X, Yang Z, Zhang X, Kim SE, Hardy R, Rajpurkar P

Evaluates multiple imitation-learning policies for coordinated surgeon–robot manipulation and tests the resulting system in a live running-suture trial.

PaperCode
Figure 1 example from the ReXSonoVQA benchmark
Conference on Health, Inference, and Learning (CHIL 2026)

ReXSonoVQA: A Video QA Benchmark for Procedure-Centric Ultrasound Understanding

Wang X, Zhang X, Kim SE, Pal A, Rajpurkar P

Tests whether video-language models can understand ultrasound procedures, including action goals, troubleshooting, and causal planning across six clinical domains.

PaperCode
Figure 1 overview of the SurgeCount bimanual system and execution graph
Machine Learning for Health (ML4H 2026)

SurgeCount: A Composition Harness for Autonomous Bimanual Surgical Gauze Counting

Kim SE*, Zhang X*, Wang X*, et al., Rajpurkar P

Composes learned manipulation skills with a learned stopping mechanism and VLM supervision to execute an ordered surgical safety protocol with verification and recovery.

Paper coming soon
Figure 1 overview of Any2all training and multimodal sampling
Abstract at ISMRM 2025 · Under review at IEEE Transactions on Computational Imaging (TCI)

A Unified Framework for Multimodal Image Reconstruction and Synthesis Using Denoising Diffusion Models

Gan W*, Wang X*, et al., Kamilov US

Frames multimodal reconstruction and synthesis as one inverse problem and develops inference-time diffusion samplers for flexible MRI, CT, and PET recovery.

PaperISMRM 2025 abstract
Figure 2 showing the bladder-aware PET segmentation pipeline and spatially weighted loss
Radiation Research Society (RRS 2025) · Poster presentation

Bladder-Aware Marginal Supervision for Automatic Segmentation of Cervical Tumors in 3D FDG-PET

Wang X, Inkman M, Mazur T, Schwarz JK, Waters MR, Zhang J

Develops bladder-aware supervision that suppresses false positives caused by physiologic uptake in metabolically active regions and nearby confounding organs.

Paper coming soon
Front page of the interactive OPO and lung-transplant dashboard
American Journal of Transplantation · 2026

Donor Care Unit Availability and Organ Procurement Organization Performance in Lung Transplantation

Yang Z, Liu CR, Wang X, et al., Puri V

Studies donor-care-unit availability and OPO performance and accompanies an interactive dashboard for recipient outcomes, donor utilization, and geospatial organ flow.

PaperDashboard
Figure 1 overview of the ECGDreamSuite evaluation framework
Machine Learning for Health (ML4H 2026)

ECGDreamSuite: A Benchmark for ECG Waveform Generation in Patient World Models

Chen E, Banerjee O, Pal A, Wang X, et al., Rajpurkar P

Evaluates generated ECGs for waveform validity, patient-state fidelity, immediate treatment response, and longer-horizon temporal consistency.

Paper coming soon
Figure 1 overview of the radiation-oncology retrieval-augmented generation workflow
Radiation Research Society (RRS 2025) · Poster presentation

Development of a RAG-Based Expert LLM for Clinical Support in Radiation Oncology

Liu T, Wang X, Inkman M, Hong JC, Waters MR, Zhang J

Builds an evidence-grounded multi-agent RAG system for radiation-oncology reasoning, with expert evaluation and ablations across clinical knowledge sources.

Paper