Investigating Methods for Performance Overdrive (IMPROVE)
A well-trained medical corps is necessary for readiness, resilience, and reliability. Simulation-based medical skill training, both initial and refresher training, require systematic, objective, high quality trainee evaluation and feedback. Currently, evaluation and feedback are based on the trainer’s mental model of the activity – where a discrepancy between the learner’s performance and the trainer’s mental model of performance is fed back by the trainer to the learner to improve performance.
Unfortunately, the learner’s training, evaluation, and feedback are usually performed by the same instructors who teach the training courses. There are at least five problems with the current approach. First, “unequal ability;” not all trainers are equally good at evaluation and feedback. Second, “lack of qualifications;” trainers may be called upon to evaluate and provide feedback on training they are not competent to perform. Third, “idiosyncratic evaluation;” evaluation is subjective, different trainers can make different assessments of the same trainee performance. Fourth, “conscious and unconscious bias;” evaluation can be influenced by factors other than the trainee’s objective performance. Fifth, “paucity of instructors;” there are too few qualified instructors, this is a choke-point on training. To overcome these problems, the Government would like to develop IMPROVE, an advanced training and evaluation system.
The field of the high-speed videography of movement and its analysis by deep learning algorithms has progressed to a point where we can now capture and analyze sophisticated behaviors in real time. The Government would like to apply this burgeoning area of research to skill training. This program focuses on the development of an automated 3D videography system that records trainee skill performance, compares trainee performance to a trained deep learning (e.g., a convolutional neural network) model of the performance, and provides systematic feedback to the trainee regarding the mismatch between the expected (model) performance and the trainee’s actual performance (Project I: Train-for-gain).
In order to apply this system to Military Health System (MHS) simulation-based training, the current training simulation-based activities and how trainees are evaluated needs to be determined. Furthermore, the simulation-based training activities that would be most amenable to the machine learning system needs to be selected. Generally, the MHS has two main simulation domains: point-of-injury (POI) and hospital-based medicine (HBM). POI simulation is primarily aimed at first responders (combat lifesavers, medics, corpsman, technicians) who are close to the battle action. Assessments in most POI simulations consist of the direct observation of learners’ performances by Tactical Combat Casualty Care (TCCC) instructors using either global judgments or checklists. HBM simulation is primarily aimed at physicians and nurses within a medical treatment facility (MTF), usually a hospital, although simulation in field hospitals and large deck ships is also part of this domain. Assessment is by direct observation of learners’ performance by instructors and usually consists of checklists of performance items successfully completed by the trainee.
Once it is possible to accurately model behavior, the team can begin to model the neural processes that give rise to the behavior. The mapping of the neural processes that generate behavior is a highly developed area of neuroscience. The Government would like to understand the relationship between neural function and behavior in order to tailor training to the neural processes that are responsible for learning, so that we can optimize training methods and performance. The goal of this project is to spatio-temporally associate neural processes with learning specific tasks (Task 2: Brain-to-train). The research project award recipients were selected from the Offerors who responded to MTEC’s Request for Project Proposals (20-05-IMPROVE).
AI Agents for Investigating Methods for Performance Overdrive (IMPROVE)
Project Team: Rensselaer Polytechnic Institute
Award Amount: $2.21M (additional cost share = $105K)
Project Duration: 18 months
Project Objective: In this project, we propose a comprehensive, novel AI-based paradigm (IMPROVE) for automated and objective assessment, training, and timely guidance of both POI and HBM skills, that adapts to the trainee’s performance and needs. To accomplish its goals, IMPROVE utilizes real-time 3D videography-based data and combines the power of three collaborative artificially intelligent agents to evaluate and train the learner: an expert agent, a learner agent, and a tutor agent. While the base period focuses on 3D videography data, it also builds the fundamental science of brain-based behavior mapping and its encapsulation in a deep learning BrainNET. This knowledge will be used in option years to enhance performance through noninvasive neuromodulation.
Markerless Biomechanics To Investigate Performance
Project Team: Southwest Research Institute
Award Amount: $1.07M
Project Duration: 18 months
Project Objective: The objective of the proposed IMPROVE program is to modify the current Southwest Research Institute (SwRI) markerless motion capture and biomechanics system to track, record, and analyze medical professional trainees during simulated medical training. The underlying artificial intelligence engine will be modified to quantitatively compare trainee performance to exemplar behavior patterns for the selected training task. Quantitative feedback will be provided to the trainee indicating specific behavior patterns required for task improvement.
