2025 Military Health System Research Symposium (MHSRS)

August 4-7, 2025 in Orlando, Florida

Kitware is proud to return to the Military Health System Research Symposium (MHSRS), the premier scientific meeting dedicated to addressing the unique medical needs of the Warfighter. This year, we’re exhibiting (Booth 518) and presenting five accepted posters that showcase our ongoing contributions to advancing military medical research and technology.

Our work spans a wide range of critical topics—including computational physiology, medical AI, digital twins, synthetic data generation, autonomous systems, and responsible AI. These efforts are tightly aligned with the Department of Defense’s mission to improve Warfighter performance, medical readiness, and return-to-duty outcomes.

Driving Innovation for Government Programs Using Open Source Tools

Kitware develops and maintains a suite of open source platforms designed to accelerate research, enable reproducibility, and support mission-critical applications. One such tool is the Pulse Physiology Engine, a digital twin simulation platform for modeling dynamic human physiology. Pulse is widely used in the MHSRS community for generating synthetic patient data, evaluating clinical interventions, and training AI systems in resource-limited triage scenarios.

This year, we’re also highlighting several ongoing projects that demonstrate our broader capabilities in medical AI, physiological modeling, and injury prevention:

• DARPA ITM: Kitware is a key performer on the DARPA In the Moment (ITM) program, which aims to develop AI systems that can make ethically and operationally sound medical triage decisions in high-stakes, resource-constrained environments. Our team is focused on generating synthetic physiology datasets that reflect real-world injury combinations, treatment responses, and physiological trajectories using physics-based simulation. This synthetic data serves as a scalable, privacy-preserving resource to train and validate AI-driven decision support systems for frontline combat care.
• DARPA GOLDEVAC: Kitware is participating in DARPA’s GOLDEVAC (GOLDen hour extended EVACuation) program as part of a collaborative team developing next-generation resuscitation and oxygenation technologies for use at the point of injury during prolonged evacuation. The program investigates whether a single intravascular device can provide continuous gas exchange and fluid resuscitation without the complications associated with current extracorporeal methods. Kitware supports this mission by using Pulse to accurately model and evaluate system performance under operational conditions. GOLDEVAC has the potential to revolutionize battlefield care and serve as a medical force multiplier, enhancing the DoD’s ability to stabilize and evacuate casualties in large-scale or resource-constrained scenarios.
• AFRL: Advancing Mechanical Ventilation Through Simulation: Kitware is part of a team collaborating with AFRL to advance a prototype all-in-one advanced simulator for training in mechanical ventilation. The technology enables clinicians to deliver care in accordance with current and emerging best practices, as well as comply with ARDSNET and Joint Trauma Clinical Practice Guidelines. Each scenario provides relevant clinical data for the clinician to manage the patient, and allows them to make changes and respond with appropriate physiological responses. It provides real-time feedback and debriefs after each scenario.

Visit us at Booth 518 to discuss these projects in greater detail.

Papers and Posters

Improved Hemorrhage and Resuscitation Simulations Validated with Swine Data

Poster | Poster Session 1 on Tuesday 05 Aug. 2025 1000-1200
Authors: Rachel B. Clipp¹, Ph.D., Jeffery B. Webb¹, Aaron Bray¹, Matthew D. Bacchetta, M.D.², Rei Ukita, Ph.D.², Steve Keller, M.D., Ph.D.^3
1Kitware, Inc.
²Vanderbilt University Medical Center
³Johns Hopkins School of Medicine

Kitware enhanced the Pulse Physiology Engine to simulate the long-term physiological effects of severe hemorrhage, including metabolic shifts and tissue perfusion changes. Validated against experimental pig data, the model accurately captured key indicators such as cardiac output and blood pressure across extended resuscitation timelines. With less than 10% error across multiple outputs, this open source model provides a reliable, reproducible resource for evaluating resuscitation protocols like fluid therapy, vasopressors, and ECMO. These capabilities support improved clinical decision-making and offer program managers a robust framework for testing interventions in prolonged care and high-fidelity triage scenarios.

Advancing Life Support with an Open Source ECMO Simulation Model

Poster | Poster Session 1 on Tuesday 05 Aug. 2025 1000-1200
Authors: Jeffery B. Webb¹, Aaron Bray¹, Rachel B. Clipp, Ph.D.¹, Rei Ukita, Ph.D.², Matthew Bacchetta, M.D.², Steve Keller, M.D., Ph.D.^3
1Kitware, Inc.
²Vanderbilt University Medical Center
³Johns Hopkins School of Medicine

Kitware extended the Pulse Physiology Engine to simulate ECMO’s complex hemodynamic and substance transport processes, enabling digital twin modeling of respiratory and circulatory support. The framework integrates key device parameters and simulates multi-organ interactions to evaluate intervention strategies. Validated against empirical data, the model maintained oxygen saturation above 90% while supporting systemic perfusion in multitrauma scenarios. This open source simulation platform supports data-driven ECMO optimization, informs device design, and lays the groundwork for automated control systems, synthetic data generation, and clinical extensions—advancing readiness for respiratory failure in prolonged and resource-constrained care settings.

Aligning a Large Language Model to protocol-based medical triage decision-making

Poster | Poster Session 2 on Tuesday 05 Aug. 2025 1530-1730
Authors: Emily Veenhuis, Aaron Bray, David Joy, Jadie Adams, Ph.D., Rachel B. Clipp, Ph.D., Jeffrey B. Webb, Brian Hu, Ph.D., and Arslan Basharat, Ph.D.

As part of DARPA’s In the Moment program, Kitware evaluated the ALIGN system, which uses large language models (LLMs) to automate medical triage tagging based on Key Decision-Making Attributes (KDMAs). We tested ALIGN’s alignment with START, SALT, and BCD Sieve protocols using a gold standard dataset of 10,000+ synthetic casualties. Results showed over 90% accuracy in LLM-generated tags across all three protocols. This work demonstrates the potential for LLMs to generate protocol-aligned medical decisions with interpretable reasoning, supporting scalable, human-trusted triage in high-pressure or resource-limited environments.

Evaluation of Triage Tagging Protocols Using a Synthetic Dataset Representative of Battlefield Injury Profiles

Poster | Poster Session 2 on Tuesday 05 Aug. 2025 1530-1730
Authors: Rachel B. Clipp, Ph.D., Jeffery B. Webb, Aaron Bray, Emily Veenhuis, Brian Hu, Ph.D., and Arslan Basharat, Ph.D.

Kitware generated a large, open synthetic dataset of over 10,000 battlefield-representative casualties to evaluate triage tagging protocols (START, SALT, BCD Sieve) across time and treatment conditions. Using the Pulse Physiology Engine, we simulated physiological trajectories and assigned tags every 5 minutes post-injury. Results showed that Minimal and Expectant classifications remained stable, while Immediate and Delayed tags varied more significantly over time and intervention. This structured dataset provides a critical resource for assessing algorithm effectiveness, supporting development of automated tagging systems, and enabling future AI applications, including training large language models (LLMs) for triage decision-making.

Large Language Model Regression for Fine-grained and Aligned Decision-Making in Medical Triage

Poster | Poster Session 2 on Tuesday 05 Aug. 2025 1530-1730
Authors: Jadie Adams, Ph.D., Emily Veenhuis, David Joy, Aaron Bray, Brian Hu, Ph.D., and Arslan Basharat, Ph.D.

Kitware developed a regression-based large language model (LLM) approach to improve alignment with human values in complex medical triage decisions. Using ethically ambiguous battlefield scenarios from DARPA’s In the Moment (ITM) dataset, we trained LLMs to match trusted human decision-makers across multiple Key Decision-Making Attributes (KDMAs), including moral judgment and quality of life. The method achieved over 31% higher alignment accuracy than baseline techniques while reducing sensitivity to bias and improving interpretability. This approach enables more traceable, equitable, and controllable AI decision-making in high-stakes environments—supporting the development of responsible AI systems for operational triage support.

Working with Kitware

Kitware’s interdisciplinary team of software engineers, machine learning experts, and medical imaging researchers brings decades of experience supporting government health agencies with open source software and custom solutions. We work collaboratively across the research and development lifecycle—from proof-of-concept studies to deployment of HIPAA-compliant and FDA-ready systems.

Whether you’re looking to augment your team with domain-specific AI expertise or build scalable medical software using open source tools, Kitware is ready to help.

Visit Booth 518 at MHSRS 2025 or contact us to explore partnership opportunities.