| dc.description.abstract | Currently, military combat medics are tasked with stabilizing casualties that arise during an operation, and preparing them to be evacuated to higher echelons of care. While doing so, medics have minimal time to record what care is being administered and communicate the status of the patient to other stakeholders, particularly as severity of injury increases. This often results in medical evacuation and surgical assets receiving information which is inaccurate, missing vital elements, and/or communicated so closely to the reception of the casualty that it does not yield any benefit in preparation.
This project assess the feasibility, and proposes a design for a novel machine learning-enabled tool to autonomously detect and communicate casualty status information as a redundancy mechanism to ensure accurate, comprehensive, and timely medical information reaches higher echelons of care.
This thesis specifically details a system design, establishes data collection protocols, and begins prototyping the machine-learning based components of the tool. Design interviews were conducted with a variety of end-users and key stakeholders of the medic-enabling tool in order to inform and construct a system design. The design proposed here is specifically tailored for US Special Operations Command combat medics, to whom this tool would be most applicable. As a proof of concept for this design, a data collection protocol for egocentric perspective video footage from combat medical training exercises was created. Accuracy baselines for state of the art computer vision and speech detection algorithms were established to assess tool feasibility and as guidance for future development. Preliminary proof of concept results are also used to inform design considerations for future work.
While further algorithm development, design refinement, integration planning, and system testing will be required to field this tool, this thesis lays the groundwork for a novel and potentially life-saving capability. | |
