Contact

Core Body Temperature Algorithm

calera | Core Body Temperature Algorithm

This short article sheds more light on the Core Body Temperature Algorithm.

Unlike conventional thermometers, the CORE Body Temperature Monitoring technology is worn externally and indirectly calculates the core body temperature using temperature and heat transfer measurements.This approach relies on a combination of classical models and ‘machine learning’ and relies on the data from thousands of data-points which is used to create the algorithm.The algorithm has the task of converting the raw sensor signals which are captured by CORE into the core body temperature value.

How was the Core Body Temperature algorithm developed?

There are four main components and this is a process that has been conducted over the course of six years and is continuing.

  1. Reference core body temperature measurements. Accurate core body temperature measurements can be captured from ingestible electronic pill (e-pill), rectal prob, heart catheter and bladder catheter. These each have tolerances and traits which require a consistent testing approach. For practical purposes, BodyCap brand e-pills are used primarily as the reference thermometer for comparison testing for CORE with a testing protocol to accommodate for temperature variations as they pass through the body.
  2. Accurate external sensors. In this case, the external sensors need to be able to collect accurate and reliable data and be able to accommodate expected variations or usage situations. Example of sensors are the heat transfer sensor, skin temperature, air temperature and other vital sensors.
  3. Trustworthy measurement data. Simultaneous collection of accurate reference measurements and the external sensor data across all of the different usage scenarios.
  4. Deep knowledge of thermophysiology combined within a Machine Learning framework. The data is processed to create the algorithm. The Machine Learning processes all of the data and systematically scans the collected data for patterns and correlations

As a result, the software learns from the data to establish the algorithm that is able to then automatically decode raw sensor data in all imaginable combinations and provide core body temperature values that closely match the results from conventional thermometers.

Accuracy & Statistics of the Core Body Temperature algorithm

To compare the results of any measurement systems with one another, Statistical Measures are used to quantify their ‘mutual agreement’.The following statistical measures are commonly used:

Bias

This is the difference between the average of the algorithm output and the average of the reference temperature. Bias is a good measure for assessing overall over- or underestimation however does not give any indication on how closely the calculated temperature follows the reference temperature.

Standard Deviation

This is the standard deviation of the differences between algorithm output and temperature reference (which is the amount of variation in the data). It describes how “scattered” the calculated temperature is compared to the reference temperature although ignores systematic offset errors. Each of this alone have a limited value so are combined and expressed in terms of “limits of agreement” which is graphically represented in a Bland-Altman plot.

The accuracy of the algorithm technology

The accuracy of the algorithm technology

*BodyCap E-celsius performance
 

**Braun Thermoscan Pro 4000
 

***Omron Flex Temp Smart
 

****Mogensen, C.B., Vilhelmsen, M.B., Jepsen, J. et al. Ear measurement of temperature is only useful for screening for fever in an adult emergency department. BMC Emerg Med 18, 51 (2018). https://doi.org/10.1186/s12873-018-0202-5
 

The accuracy of CORE was determined by comparing the output of the CORE algorithm based on skin temperature and energy transfer and the reading of an indigestible radio pill (E-Celsius Pill from Body Cap).
 

As described in ISO 80601-2-56_2017 this accuracy determination is defined as the clinical accuracy.
 

The laboratory accuracy of a temperature sensor is determined using a stirred fluid bath or an infrared calibrator, which does not necessarily reflect the suitability or accuracy for determining core body temperature during intended use. A key figure for quantifying clinical accuracy is the 95% confidence interval, also listed in the table above, and as illustrated in our use case available here.

Core Body Temperature algorithm explained

CORE body temperature monitor has two core body temperature algorithms built-in to calculate accurate core body temperature values under different physiological and environmental conditions: the Everyday Living mode and the Intense Endurance Sports Activity mode

Everyday Living Algorithm

When a heart rate monitor is not paired (or not in range), CORE uses a Low Activity Mode Algorithm – referred to as the Everyday Living 24/7 algorithm, which is easily understood as low impact activity, resting and sleeping.

The algorithm exclusively works using temperature and heat transfer signals from the sensor and perform the most “straightforward” approach to calculate core body temperature.

However, as it only uses these ‘thermal’ values alone, it is subjected to some specific limitations.

Thermal Inertia

Sudden changes in core temperature require time to diffuse to the surface of the body and to become detectable by CORE. This adds a lag which can be between 5 to 30 minutes. This delay is barely noticeable during slow-changing, continuous 24 hour thermal behaviour. However it becomes a serious constraint for athletes who require immediate feedback during their sporting activity.

Very cold ambient conditions

One of CORE’s strength is the ability to compensate for fluctuations in ambient conditions. However, the lower the ambient temperature, the more difficult it is to fully compensate for these fluctuations. Optimal accuracy is achieved when skin temperature is at 34°C or higher and below this skin temperature, accuracy starts to decrease.

In this example with a high-intensity sports activity of outdoor cycling, the Everyday living Algorithm (left) reacts slowly and the fast-changing athletic intensity creates discrepancies. The Intense Endurance Sports Activity Algorithm (right) reacts faster and with better accuracy.

In this example with a high-intensity sports activity of outdoor cycling, the Everyday living Algorithm (left) reacts slowly and the fast-changing athletic intensity creates discrepancies. The Intense Endurance Sports Activity Algorithm (right) reacts faster and with better accuracy.

Intense Endurance Sports Activity Algorithm

To enable this mode, a heart rate monitor is paired to the CORE Body Temperature Monitor. When CORE is operating and detects the heart rate monitor’s signal, it switches to the Intense Endurance Sports Activity Algorithm. CORE can be paired with numerous heart rate monitors and it will then use the signal from the first heart rate monitor it detects.
 

This Algorithm has been developed to overcome the problem of thermal inertia and enable faster core body temperature calculation. The heart rate reacts extremely quickly to changes in physical activity and can therefore be used to anticipate changes in the core body temperature.
 

This additional signal stream also reduces the uncertainty in low ambient temperature conditions, mentioned above, enabling accurate core body temperature prediction even during outdoor sports like cycling in which the environmental temperature and thus skin temperature is subject to greater change.
 

Validated Use Cases: Intense Endurance Sports Activity Mode (Indoor, Outdoor, Intense Intervals)

Validated Use Cases: Intense Endurance Sports Activity Mode (Indoor, Outdoor, Intense Intervals)

Drawbacks of high activity mode

With the advantages, it would be fair to ask why a Low Activity Mode is required at all. There are some important aspects to the Intense Endurance Sports Activity Algorithm which help provide answers:

  1. Quality of the heart rate signal: The algorithm relies on accurate raw signals. Inaccurate or buggy heart rate monitors, frequent dropouts etc. will compromise the accuracy of the heart rate signal and spoil the accuracy of the temperature calculation.
  2. Switching artifacts: If the heart rate monitor signal drops-out, CORE will automatically switch to Low Activity Mode. Unreliable heart rate monitors will cause CORE to constantly switch back and forth between algorithm modes, leading to a fragmented and inaccurate core body temperature calculations.
  3. Fitness-level and demography: The High Activity Mode Algorithm has been tailored to physically active adults. The heart rate profile of unfit, elderly or very young people can differ greatly and lead to systematic deviations in these demographic groups.
  4. Physiological extremes: Any physiological condition that causes unpredictable or abnormal heart rate behavior like sickness, fever, external or internal heat sources or even sleep, will mislead the high activity algorithm and result in less accurate prediction than by using the purely thermal low activity algorithm.
  5. Power consumption: In this mode, the CORE Body Temperature Monitor is always seeking the heart rate and this increases power consumption.

In the above example calculating core body temperature while sleeping, the Low Activity Mode (left) is more accurate against the reference e-pill compared to the High Activity Mode algorithm.

CORE Uses Cases Validated & Under-validation

Human thermoregulation as well as the conditions and behavior can vary significantly across different activities so for each activity and scenario, we validate CORE to ensure that it is able to provide accurate data. This begins with the comparison testing in order to first gather data that will train the algorithm. Within each activity there is usually a range of ‘use-cases’ and in the trials we need reliable testing protocols (for both the reference thermometer and CORE), we need to gather numerous data-sets and we also seek to identify and gather data for extreme situations, as an example very cold and very hot weather for outdoor cycling.

Use Cases currently under validation<br>&nbsp;

All of the data is then fed into the algorithm where testing and optimisation continues until it is eventually validated for this activity. The following overview shows the status of validation for various activities. Certain of the use cases displayed here are still under validation. We are actively working on improving the algorithm to allow using CORE on these scenarios.

Use Cases currently under validation
 

All of the data is then fed into the algorithm where testing and optimisation continues until it is eventually validated for this activity. The following overview shows the status of validation for various activities. Certain of the use cases displayed here are still under validation. We are actively working on improving the algorithm to allow using CORE on these scenarios.

Use Cases currently under validation<br>&nbsp;

To learn more about the full product line for the various use cases, click <a href="/en">here</a>. For any of those solutions, our team of experts will accompany you in your project.

Use Cases currently under validation
 

To learn more about the full product line for the various use cases, click here. For any of those solutions, our team of experts will accompany you in your project.

Recommended readings to go further

  • Discover all the Use Cases related to our technology, here.
  • Explore our product line, here.
  • Discover all the peer-reviewed publications using our technology, here.