The Trauma THOMPSON Challenge 2025¶
Trauma TeleHelper for Operational Medical Procedure Support and Offline Network
Artificial Intelligence (AI) holds immense promise for enhancing emergency medical care in austere and disconnected environments, where critical resources and trained personnel are scarce. The Trauma THOMPSON Challenge 2025 aims to drive innovation in AI-assisted remote instruction systems that can provide real-time guidance to frontline providers during high-stakes and lifesaving procedures. Leveraging a large-scale, egocentric video dataset of unscripted emergency interventions, this challenge supports the development of AI models capable of recognizing medical actions, anticipating next steps, detecting tools, tracking hand motion, assessing realism, and answering clinical questions.
Participants will tackle a series of clinically relevant tasks that reflect real-world challenges in emergency medicine, including just-in-time procedures using makeshift tools. Baseline methods are provided to encourage broad engagement and reproducibility. The overarching goal is to catalyze research in activity recognition, procedural understanding, and multimodal learning for emergency care. Through this challenge, we aim to foster AI solutions that improve procedural accuracy, reduce human error, and ultimately save lives in settings where traditional medical support is unavailable.
About 🎯¶
The Trauma THOMPSON Challenge 2025 introduces a multimodal benchmark for advancing AI in emergency and humanitarian medicine, featuring:
🩺 Egocentric Emergency Procedure Dataset: Unscripted footage of lifesaving procedures performed in both standard and just-in-time settings, capturing hand movements, tool usage, and complex action sequences.
🧠 5 Clinically Critical Tasks: Including action recognition and action anticipation, hand tracking, tool detection, realism assessment, and visual question answering, designed to simulate the challenges of high-stakes and low-resource environments.
🛠️ Just-in-Time (JIT) Scenarios: Emphasizing care in disconnected and resource-scarce situations where clinicians need to use common objects to perform essential interventions.
📊 Sample Models & Evaluation: Sample codes provided to assist development and evaluation. See our GitHub for sample codes and instructions to get started.
Participation 💻¶
Join us to advance AI for emergency medicine through the Trauma THOMPSON Challenge 2025. Whether you're a student, researcher, or industry expert, your work can directly to life-saving innovations in critical care. Tackle real-world clinical scenarios and develop AI solutions that assist medical responders in austere and high-pressure environments. Get started in three easy steps:
-
Register: Sign up on the Grand Challenge platform and click “Join” to participate in the challenge.
-
Build Your Solution: Access datasets, sample models, and development guidelines to train and evaluate your algorithms.
-
Submit and Compete: Submit your models for tasks such as action recognition, anticipation, hand tracking, realism detection, and medical VQA. Compete for prizes and recognition at MICCAI 2025. Full details on submission formats and evaluation criteria are available on the Submission Instructions page.
The proceeding from the Trauma THOMPSON Challenge 2023 is available here: https://link.springer.com/book/10.1007/978-3-031-71626-3