A Real Time Multi-Sensor Wearable for Early Hypothermia and Heat Strain Detection

[Add Text Here]https://www.youtube.com/watch?v=1JSGa7q85_0

Hypothermia is a dangerous but often overlooked health problem. Each year, about 20,000 people die directly from exposure to extreme cold, and many more suffer long-term health effects. In total, about 4.6 million deaths worldwide are linked to cold-related conditions [1,2]. Early symptoms, such as shivering, weakness, and a slow pulse, which can resemble normal cold reactions, are often not recognized until they become severe [3].

Heat-related conditions are also a major concern. Each year, approximately 489,000 people die due to extreme heat [4]. In some cases, young children and elderly individuals are accidentally left in enclosed spaces, such as vehicles, where temperatures can rise quickly and become dangerous.

High-risk groups include homeless individuals, infants, and the elderly [4]. Detecting hypothermia or heat strain early is difficult because these conditions depend on core body temperature rather than skin temperature [5,6]. However, real-time medical devices for measuring core temperature are often invasive, expensive, and not accessible to most people [7]. This lack of accessible tools contributes to the high number of deaths caused by extreme temperatures.

To address these problems, my objective became clear:

1. Create a wearable device that is small and convenient

2. Make it affordable and accessible

3. Ensure it is non-invasive

4. Maintain a high level of accuracy

5. Integrate early detection

This device can estimate core body temperature and monitor activity levels to detect early warning signs of both hypothermia and heat strain, especially in vulnerable populations.

To address the challenge of non-invasive core body temperature estimation, a multi-sensor fusion approach was developed based on the relationships found in earlier studies. A review of the literature identified two high-impact studies, which demonstrate that core temperature can be estimated from heart rate (HR), skin temperature (Tskin), and activity level [8,9]. Based on these findings, a multi-stage wearable device was designed, implemented, and tested to detect early signs of hypothermia and heat strain.

The system consists of three sensing components:

•     Heart rate sensor: MAX30102 photoplethysmography (PPG) sensor

•     Skin temperature sensor: NTC thermistor

•     Motion sensor: integrated into the microcontroller (Arduino Nano 33 BLE)

An Arduino Nano 33 BLE serves as the central processing unit, handling data acquisition and AI-based activity. The full system design and assembly are shown in Figures 5 and 6. Heart rate is measured near the subclavian artery (left of the sternum), while the thermistor is positioned under the clavicle. The accelerometer is located on the chest via the microcontroller board. These placements were selected to balance measurement accuracy and user comfort (Figure 7).

Core temperature is estimated using a linear model derived from prior research [8]: Tcore = a•HR + b•Tskin + c, where a and b are experimentally determined coefficients. The parameter c represents an adaptive baseline that accounts for variations in metabolic heat production and activity level. During hypothermia, reduced activity and metabolic suppression lead to a decrease in c. In contrast, during heat strain, increased metabolic heat production causes c to rise.

The device integrates heart rate, skin temperature, and activity level to estimate core body temperature. The microcontroller processes these inputs, selects appropriate model parameters based on conditions, and estimates core temperature using real-time embedded processing. The system then classifies risk level and provides early warning signals for hypothermia and heat strain.

The developed system continuously monitors physiological signals and analyzes them in real time to detect dangerous thermal conditions (Figure 8). Firstly, sensors measure heart rate, skin temperature, and movement. Using these values, the system estimates the user’s core body temperature. A C++ program controls the sensors and the microcontroller, ensuring they work together correctly to collect and process data.

The processed data is transmitted to a Python-based adaptive AI system for personalized analysis, which uses artificial intelligence to personalize detection for each user. During an initial learning phase, the AI collects data over time to establish a personalized physiological baseline, including their core temperature and activity level. The system compares the current core temperature to the user’s baseline to quantify how much it deviates from their normal physiological state. Small changes are considered normal, while larger changes indicate higher levels of risk. Based on this, the system determines whether the user is at risk of hypothermia or heat strain and then classifies the risk as normal, mild, moderate, or severe. Finally, the result is sent back to the Arduino, which alerts the user through buzzers, LEDs, and Bluetooth notifications (Figure 9).

The system uses calculated core temperature to classify risk levels. Core temperature is calculated using the formula Tcore = 0.0100 × HR + 0.0837 × Tskin + c [8], where c changes, depending on the condition. A normal state is defined as Tcore > 35°C with c = 33.5.

For hypothermia detection:

Mild: 34-35°C, where Tskin < 33°C with c = 33

Moderate: 32-34°C, where Tskin < 30°C, heart rate < 65, and movement < 1.0, with c = 32.2

Severe: < 32°C, where Tskin < 27°C, heart rate < 50, and movement < 0.5, with c = 30.0

For heat strain detection:

Mild: 37-38°C with c = 33.8

Moderate: 38-39°C with c = 34.3

Severe: > 39°C with c = 35.0

Metabolic heat production increases in the heat strain state, which causes the value of c to increase.

To test accuracy, 100 trials were conducted under both indoor (20°C) and outdoor (0°C) conditions to evaluate system accuracy.  A digital oral thermometer (Femometer DMT-427) was used to measure body temperature as a reference standard, given that oral temperature is a reliable approximation of core body temperature [10]. Simultaneously, my device measured heart rate, skin temperature, and activity level to estimate core temperature. Each condition included 50 trials, and the results were compared with the thermometer readings. The results (shown in Figures 10 and 11) show that the temperatures from my device are very close to the thermometer values. The device achieved an average bias of only -0.11°C (~-0.3%) (indoor: -0.1758°C, -0.49%; outdoor: -0.044°C, -0.12%), demonstrating that accurate core temperature estimation is possible using a low-cost, non-invasive, and multi-sensor approach. Accuracy may be further improved by adjusting the value of c in the future. Mild, moderate, and severe conditions were not tested in person but simulated (Figure 12).

Every year, millions of people die from hypothermia and heat strain. Currently, monitoring devices are far too expensive, invasive, and unsanitary; because of this, I created a device that addresses a critical gap in the early detection of hypothermia and heat strain by enabling non-invasive, real-time monitoring of core body temperature.

This device is simple, affordable, and easy to use. It collects data such as heart rate, skin temperature, and activity level using sensors, then uses that data to estimate core body temperature. A C++ program controls the system and makes sure everything works properly. It also uses artificial intelligence to learn each user’s normal condition and compare it to current data. This means the system can detect small changes early and warn the user before the situation becomes dangerous.

The device gives early warnings through LEDs, buzzers, and Bluetooth notifications, helping prevent serious health issues or even death. With a prototype cost of approximatley $28, which can be reduced to $13 in large-scale production. By combining low cost, personalization, and real-time monitoring, this device has the potential to make life-saving technology accessible to people around the world.

Currently, I am integrating the system with a mobile application to enable real-time monitoring and alerting. I am also developing a cloud system to track patterns over multiple days, improving early detection of hypothermia and heat strain. In emergencies, the system can send risk notifications and alerts to contacts or medical services. In the future, I aim to miniaturize the device and develop a fully wireless sensor system (like a CGM) to improve wearability and conduct real-world validation with medical or research institutions to upgrade performance.

I have multiple people I would like to thank!

Thank you to Youth Science Canada for providing me with an opportunity to explore STEM.

Thank you to my friends who graciously supported me through this journey.

Lastly, thank you to my family for facilitating the amount of mess I created at home and for supporting and funding my project!

[1] Gasparrini, A., Guo, Y., Hashizume, M., Lavigne, E., Zanobetti, A., Schwartz, J., Tobias, A., Tong, S., Rocklöv, J., Forsberg, B., Leone, M., De Sario, M., Bell, M. L., Guo, Y. L. L., Wu, C. F., Kan, H., Yi, S. M., Coelho, M. S. Z. S., Saldiva, P. H. N., & Armstrong, B. (2015). Mortality risk attributable to high and low ambient temperature: A multicountry observational study. The Lancet, 386(9991), 369–375. https://doi.org/10.1016/S0140-6736(14)62114-0

[2] Zhao, Q., Guo, Y., Ye, T., Gasparrini, A., Tong, S., Overcenco, A., Urban, A., Schneider, A., Entezari, A., Vicedo-Cabrera, A. M., Zanobetti, A., Analitis, A., Zeka, A., Tobias, A., Nunes, B., Madureira, J., Forsberg, B., Åström, C., Ragettli, M. S., & Kan, H. (2021). Global, regional, and national burden of mortality associated with non-optimal ambient temperatures from 2000 to 2019: A three-stage modelling study. The Lancet Planetary Health, 5(7), e415–e425. https://doi.org/10.1016/S2542-5196(21)00081-4

[3] Centers for Disease Control and Prevention. (2024). Recognizing hypothermia. https://www.cdc.gov/natural-disasters/psa-toolkit/recognizing-hypothermia.html

[4] World Health Organization. (2023). Heat and health. https://www.who.int/news-room/fact-sheets/detail/climate-change-heat-and-health

[5] Mayo Clinic. (2023). Hypothermia. https://www.mayoclinic.org/diseases-conditions/hypothermia/symptoms-causes/syc-20352682

[6] Centers for Disease Control and Prevention. (2022). Heat stress. https://www.cdc.gov/niosh/topics/heatstress/

[7] https://coach.jove.com/microcourse/nursing/measuring-and-interpreting-body-temperature-as-a-vital-sign/what-is-assessing-body-temperature-rectal?utm

[8] Eggenberger, P., MacRae, B. A., Kemp, S., Bürgisser, M., Rossi, R. M., & Annaheim, S. (2018). Prediction of core body temperature based on skin temperature, heat flux, and heart rate under different exercise and clothing conditions in the heat in young adult males. Frontiers in Physiology, 9, 1780. https://doi.org/10.3389/fphys.2018.01780

[9] Laxminarayan, S., Rakesh, V., Oyama, T., Kazman, J. B., Yanovich, R., Ketko, I., Epstein, Y., Morrison, S., & Reifman, J. (2018). Individualized estimation of human core body temperature using noninvasive measurements. Journal of Applied Physiology, 124(6), 1387–1402. https://doi.org/10.1152/japplphysiol.00837.2017

[10] Eisenkraft, A., Goldstein, N., Fons, M., Tabi, M., Sherman, A. D., Ben Ishay, A., Merin, R., & Nachman, D. (2023). Comparing body temperature measurements using the double sensor method within a wearable device with oral and core body temperature measurements using medical grade thermometers - A short report. Frontiers in Physiology, 14, 1279314. https://doi.org/10.3389/fphys.2023.1279314

[11] OpenAI. (2026). ChatGPT (Mar 20 version) [Large language model]. https://chatgpt.com/