Maya Trutschl was still a senior at Caddo Parish Magnet High School in Shreveport, Louisiana, when she built a patient-monitoring system for less than $130 in parts: a thermal-sensing camera paired with machine-learning software meant to track how a bedridden patient is positioned and to flag when their skin is at risk of breaking down. She didn’t stop at a science-fair table. She took the device into a hospital intensive care unit and ran it against real patients for more than 150 hours.
That combination is what separates the project from a typical classroom prototype. Thermal-sensing hardware is cheap and widely available, and pressure-injury prediction has been an active research area for years. What’s unusual is that a high school student ran that hardware on an ICU floor long enough to generate a real accuracy number. Most comparable projects test only in a lab, with volunteers or mannequins.
Reading body position from heat, not touch
The positioning component works by reading heat patterns radiating off a patient’s body and skin surface. It needs no pressure-sensing mats, wearable straps, or cameras requiring clear sightlines and even lighting. A model trained on that thermal data infers how someone is lying — on their back, side, or another position — which matters because prolonged pressure on the same spot, especially over bony areas like the hips, heels, and tailbone, is the mechanical cause of most bedsores.
Over the 150-plus hours of real-patient testing, Trutschl has said the system tracked patient positioning with more than 99 percent accuracy. That figure describes how reliably the thermal readings matched actual body position during the testing window. It says nothing about how well the system prevented any specific injury. Coverage of the project has not disclosed how many individual patients contributed to that testing period, whether the hours were continuous or spread across shifts, or the consent and hospital-oversight process that allowed a student project onto an active ICU floor — details that would normally accompany a clinical study but are standard to omit from a science-fair or scholarship write-up.
A second model built to flag pressure-ulcer risk before it happens
Positioning is only half of Trutschl’s project. A separate machine-learning model was trained to predict which patients are at elevated risk of developing a pressure injury in the first place, using the MIMIC-IV critical care database, a large, de-identified dataset of real intensive-care records maintained for research use. Within that dataset, pressure-ulcer cases were vastly outnumbered by patients without one — roughly 3,900 positive cases against about 90,000 negative outcomes — so Trutschl applied SMOTE oversampling alongside undersampling to correct the imbalance before testing the data against ten different classification algorithms to find the best-performing one.
That is a defensible technical approach to a genuinely hard problem in medical machine learning, where rare-outcome datasets routinely produce models that look accurate on paper simply by predicting the majority class. The publicly available accounts of the project don’t include a single, specific accuracy figure for this risk-prediction model comparable to the 99-percent positioning number. Any percentage attached to the risk-prediction side should be treated as unconfirmed until Trutschl’s full methodology is published or independently reviewed.
From a Louisiana classroom to national science stages
The project has been through more than one round of development. At the 2025 Regeneron International Science and Engineering Fair, Trutschl won a First Award in the Embedded Systems category for research titled Development of a Novel Solution for Mitigation of Debilitating Decubitus (Pressure) Ulcers, Year Two — the “Year Two” in the title signals a project rebuilt and retested across successive versions, not a single prototype built once.
Trutschl was later named a 2026 Regeneron Science Talent Search Scholar for a retitled version of the work, Automated Prediction and Prevention of Pressure Injuries Using Machine Learning and Thermal Sensing: A Clinical Validation Study. That distinction placed her among 300 scholars selected from more than 2,600 applicants nationwide, and the only Louisiana student chosen that cycle. She has since been named a 2026 Davidson Fellow in the engineering category, carrying a $25,000 scholarship, and was recognized at the National STEM Festival. She is headed to MIT this fall, where she has said she is considering aerospace engineering or computer science.
Why pressure injuries remain a stubborn, costly problem
Bedsores, known clinically as pressure injuries or pressure ulcers, form when sustained pressure cuts off blood flow to skin and underlying tissue, most often in patients who cannot easily reposition themselves. The National Pressure Injury Advisory Panel estimates roughly 2.5 million cases occur annually in the United States, with incidence running between 10 and 20 percent in critical care units specifically, and puts total acute-care treatment costs at close to $26.8 billion a year, with advanced-stage cases individually costing $75,000 to $150,000 to treat.
Hospitals currently manage this risk mainly through periodic nursing assessments using scoring tools such as the Braden Scale, which rates factors like mobility, moisture, and nutrition on a fixed schedule rather than continuously. A cheap sensor that can watch positioning between those scheduled checks, and flag rising risk earlier, addresses a real gap in that workflow — which is part of why a student project tackling it drew attention from judges at ISEF, the Talent Search, and the Davidson Institute.
A promising prototype, not a hospital-ready device
Trutschl’s system is a student research project entered in science competitions. It is not a commercially available or regulatory-cleared medical device. No peer-reviewed clinical trial describing the work has surfaced in the sources reviewed for this article, and the hospital, patient count, and ethical-oversight arrangements behind the 150-plus hours of ICU testing have not been made public in the coverage available. Readers should not treat a thermal camera and a laptop as a substitute for standard nursing pressure-injury protocols, and the more than 99 percent positioning accuracy reported from that testing window is an early, promising result from one testing period. It is not yet a settled clinical benchmark.
Here’s what’s documented: a high school senior built working hardware for under $130, ran it against real ICU patients long enough to generate a meaningful accuracy figure, and had the underlying methodology reviewed favorably by judges at three separate national science competitions before heading to MIT. Whether the approach holds up under formal clinical study, with a published patient count, control group, and peer review, is the next test the project has not yet faced.