In trauma surgery, every second counts. A wrong decision, or even a moment of hesitation, can be life or death. Now, Eddie Zhang, a rising senior at The Harker School in San Jose, is working with the Edwardson School of Industrial Engineering to develop an AI system that remembers surgical procedures like an experienced surgeon, helping first responders make critical decisions when expert medical help isn't available. Zhang will present his work at the 2025 Advanced Concepts for Intelligent Vision Systems (ACIVS) Conference in Tokyo, Japan– placing him on the global stage alongside seasoned researchers from around the world with his paper titled "Context-Aware Vision Language Model for Action Recognition."

 

Zhang's journey into medical AI began last summer when he reached out to James H. and Barbara H. Greene Professor Juan Wachs at the Edwardson School of Industrial Engineering, expressing interest in Wachs' advanced research in medical robotics and computer vision. Dr. Wachs connected Zhang with PhD student Yupeng Zhuo, who leads The Trauma THOMPSON—a project focused on developing an AI-based copilot to assist first responders in providing rapid humanitarian aid. For the past year, Zhang has been working primarily on the action recognition and anticipation components of the project.

 

"Although I don't have any connections to the medical field, I've always been interested in computer vision and image processing. Thus, when Yupeng introduced me to what he was working on in our first meeting, I was really excited to work with him on the Trauma THOMPSON project," Zhang commented. Eventually, Zhang made a crucial discovery about surgical workflows: many of the actions performed in surgical procedures were performed in series, meaning that one action would constantly be performed after the other. Zhang said, “Current methods of action recognition don’t actually account for these patterns, which inspired me to write my paper.”

 

Zhang's paper, accepted through ACIVS's rigorous peer-review process and soon to be published by Springer in the Lecture Notes in Computer Science (LNCS) series, introduces a groundbreaking approach to medical AI guidance systems. The work proposes "a memory-augmented Vision-Language Model (VLM) for action recognition and action anticipation to assist in medical decision-making and provide real-time intraoperative guidance."

 

Working alongside Zhuo, Zhang developed a sophisticated vision language model that combines a vision transformer encoder with an LLM decoder. The key innovation lies in augmenting the LLM with a memory of past actions performed by first responders. This enhancement allows the model to not only recognize current surgical actions but also understand the procedural context, enabling it to provide more accurate guidance on next steps.

 

Zhang's research specifically targets the unique challenges of trauma surgery in "austere and resource-constrained environments" where there is "limited equipment, and the lack of available medical expertise." The architecture he helped develop features a sophisticated two-module system: "The vision module is based on the MViTv2 transformer, and its outputs are fed into the language module."

 

What distinguishes their approach is the system's ability to remember procedural flow. "We augment the language module with memory of previous actions through concatenating all past actions recognized or anticipated by the vision module." This memory-aware methodology reflects real-world surgical patterns, where "medical actions are often consecutive, with one action consistently performed after another."

 

The language model, built on FlanT5, leverages this contextual understanding to enhance decision-making accuracy. Training occurred on the open-source Trauma THOMPSON dataset, which comprises "unscripted egocentric videos of emergency trauma surgery procedures."

 

 

The impact of Zhang's contributions is evident in the model's performance metrics. The system "achieves a top1 accuracy of 69.11% on action recognition and 62.13% on action anticipation, outperforming the results reported by previous literature."

 

Despite being only 17, Eddie demonstrates remarkable intellectual maturity and ambition. His passion for understanding complex systems began in elementary school robotics, where he progressed from block-based coding to mastering C++ and Java, eventually discovering machine learning. He described being "amazed at the depth and breadth of what modern-day computers and algorithms could now do," which ultimately inspired him to pursue ML research and reach out to Dr. Wachs.

 

This July, Eddie will travel to Japan with his parents to present both a poster and deliver a research talk at ACIVS. "The poster sessions would be a good opportunity to engage in more in-depth and specific technical discussions, while the presentations would be a good opportunity to give a summary of the work and present to a broader audience," he noted.

 

While Zhang has previously participated in local research conferences, ACIVS represents his first international presentation opportunity. "I'm really excited to share our work, meet new people from all over the world, as well as learn about what others are researching currently," said Zhang. "I am a little nervous about presenting my work to a larger audience, but nonetheless I think this will be a great experience for me early on in my research journey."

 

Eddie's achievements also reflect The Harker School's commitment to student research excellence. "Harker has been very supportive of students doing their own research outside of school, and also has a really well-run research program," he said. He plans to present his work at Harker's annual science research symposium next year.

 

When not immersed in research, Zhang maintains diverse interests. "Beyond academics, I'm also interested in resolving ongoing environmental challenges, and in my spare time, I like to play tennis and travel around the world with my family," said Zhang.

 

Looking toward the future, Zhang remains committed to advancing this research field. "I'd love to continue working with Dr. Wachs in the future on cutting-edge medical robots and computer vision. There are a lot of different areas that we can explore, and I'm excited to see where the Trauma THOMPSON project will go in the upcoming years."

 

 

Zhang's contributions extend beyond his individual research paper. The Trauma THOMPSON team is currently organizing a challenge through the Medical Image Computing and Computer Assisted Intervention Society (MICCAI). "We encourage everyone to participate in and submit to the challenge," Eddie shared. The competition can be found at  t3challenge25.grand-challenge.org.

 

From early robotics competitions to cutting-edge AI research, Eddie Zhang's story exemplifies what can happen when passion meets opportunity– regardless of age. His work not only advances the field of medical AI but also demonstrates the remarkable potential of young researchers to contribute meaningfully to solving real-world challenges.