If we want to evaluate the skin temperature in humans we can use infrared thermography with different methods, as well as several metrics.
An experimented thermographer might be able to perform an immediate analysis using the qualitative method, which gives us the possibility to examine the thermal image interpreting the colours. It is fast and very intuitive, but risky, because it is based on the subjective interpretation of the technician that reads colours that might be modified easily using the scale, so quite easy to underestimate or overestimate a colour spot.
On the other hand, we have the quantitative method, based on the radiometric data contained within the pixels of the thermal image, allowing us to carry out a reproducible, reliable and comparable analysis through a software. When using the quantitative method, the main challenge that we face is the variability of the skin temperature due to the influence factors, which forces us not to focus on absolute temperatures (for instance: this knee is 28,5ºC). Among relative temperatures, thermal asymmetry is one of the most solid and used metrics nowadays.
Thermography applications have been discussed in other posts to improve understanding of the tool.
The human body is designed to maintain a balance, in bio-medical sciences this concept is known as homeostasis. Thermoregulation is one of the main systems ruled by this principle. That is why, authors as Uematsu (1988) has showed in asymptomatic normal individuals that “the degree of thermal asymmetry between opposite sides of the body (AT) is very small”, with values under 0.38ºC. The thermal differences between bilateral regions of interest (ROI), with maximum or average temperatures have been shown as a valid method in several studies (Formenti et al. 2018)
That is why we use thermal asymmetries from the first image. In ThermoHuman software, we created a classification scale highlighting with different colours thermal asymmetries above 0.3ºC, so it is very intuitive to spot areas that are not in a thermal balance with a simple glance.
Thermal asymmetry (showed as Asymmetry in ThermoHuman, figure 1) compares the average skin temperature of one ROI with the bilateral one. We recommend using these metrics mainly if you are analyzing a subject for the first time or in the following evaluations. In this example, we can observe the first evaluation of an athlete, who practices a collective sport that involves high-intensity movements. ThermoHuman avatars show us significant asymmetries (above 0.3ºC) on the left ankle and foot, and on the posterior right thigh area that might be produced by that demand of repeated efforts of his sport.
Figure 1. ThermoHuman thermograms and avatar showing average thermal asymmetries
In a recent investigation, from the ThermoHuman group, it has been seen that analyzing 950 healthy athletes without pain, the general difference in the asymmetry metrics is 0.004ºC ±0.66ºC (see figure 2). That is, a high level of homeostasis is expected in all regions when the athlete is healthy (Escamilla-Galindo et al. 2022).
Figure 2. Extracted from the ECSS 2022 conference presentation with normative values for body regions.
The following video shows the oral communication presented at the ECSS 2022 with the investigation of normal temperature values between regions in 950 athletes:
Video 1. Summary of the oral comunication presented at the ECSS 2022.
The analysis of the maximum thermal asymmetries (showed as Maximum Asymmetry in ThermoHuman) is an interesting alternative. It follows the same principle as average thermal asymmetry, but comparing the maximum temperature data within the ROI with its bilateral area. Both metrics (maximum and average thermal asymmetries) are validated and show similar results.
This metric is especially useful when we find pathologies that generate a localized and significant increase of skin temperature (what we know as a hot spot) but that does not affect the entire ROI and might not create an alarm using average thermal asymmetry. In the next example, we see the case of a patient affected by human papilloma virus (visible on figure 4, on the right first metatarsal joint). On the left (figure 3), we observe no significant average thermal asymmetry on that ROI, while if we analyse it with the maximum asymmetry we obtain a very relevant alarm as we see in right avatar.
Figure 3. Average thermal asymmetry
Figure 4. Maximum asymmetry
“Not all asymmetries mean injury”, this is one of the quotes that we use the most. Neutralized asymmetry is one of the metrics that helps us to identify better if a thermal asymmetry is relevant or might not.
Although Uematsu (1988) and other authors showed, in some cases, we find thermal asymmetries that are above 0.3ºC (even much more) on areas with no pain, previous injury or affected by influence factors. With just one image, it is hard to differentiate, but if we have the chance of assessing someone frequently we might observe consistent and repeated alarms in some ROIs.
This metric requires several evaluations to establish, as long as there is no pain or injury, a localized an individual threshold based on the historical average of each ROI. That is what we know as building an individualized thermal profile. An example of the usefulness of this metric is to neutralize the asymmetries generated by an asymmetric sport such as tennis or judo, where the forearm of the grip will present a hyperthermic asymmetry produced by adaptation to the sport itself (Arnaiz-Lastras et al. 2011), as we can see on figure 5:
Figure 5. Image from Arnaiz-Lastras et al. 2011
In the following example, we see the tracking report of a subject with a consistent asymmetry alarm on his left knee (figure 6). Since he is pain-free and uninjured, when we choose the neutralized asymmetry option (figure 7), the alarm on the knee just appears during the sixth evaluation, when it is actually significant.
Figure 6. Average asymmetry
Figure 7. Neutralized asymmetry
Using only thermal asymmetries (whether average, maximum, or neutralized) has two main limitations:
The Coefficient of Variation (CV) solves these issues and complements the use of thermal asymmetries. Similar to neutralized asymmetry, the CV requires several thermograms to build a consistent thermal profile over time.
This metric is calculated independently for each ROI, taking into account only its current temperature and comparing it with its historical temperature. Since absolute temperature changes are usually correlated across the whole body—and are also influenced by external factors—it is common that all ROIs show a similar behavior in a given session.
Because of this, the coefficient of variation is one of the key metrics for analyzing fatigue with the software. It is especially useful if all sessions are performed under similar conditions (same place and similar environmental temperature)..
Figure 8. Average asymmetry
Figure 9. Coefficient of variation
This is a key factor to understand the thermoregulatory behavior of tissues, since the CV allows us to determine over time whether an asymmetry results from one ROI heating up, the opposite ROI cooling down, or both tendencies simultaneously.
For instance, in the case of a muscle injury, an asymmetry usually appears—but sometimes in the opposite region. Why? Because ThermoHuman highlights the warmer ROI, which does not necessarily mean the problem is located there. (See our publication on the importance of hyperthermia and hypothermia.)
The CV can reveal if the injured ROI is actually cooling down instead. In the following case of a right calf muscle injury, the neutralized asymmetry is shown first (Figure 10), followed by the CV (Figure 11), which clearly highlights a decrease in the injured area (take a look of our publication about the importance of hyperthermia an hypothermia).
Figure 10. Neutralized asymmetry of a subject with a muscle right calf injury
Figure 11. Coefficient of variation of the same subject, showing a decrease on the injured area
In some cases, the results of the Coefficient of Variation (CV) can show extreme tendencies. For example, in Figure 12 we can see drastic cold and warm changes between consecutive evaluations. This often occurs when subjects are analyzed over long time intervals or under very different environmental conditions (e.g., warmer or colder room temperatures), which can create global changes across the whole body.
Figure 12. Tracking report showing drastic changes in the coefficient of variation
To reduce this bias, we use the Softened Coefficient of Variation (SCV). This metric incorporates the global and historical temperature of the protocol in order to minimize the impact of global factors. Instead of relying only on absolute temperature, the SCV also subtracts the protocol’s average temperature. By doing so, global trends are removed, and the analysis emphasizes those ROIs that have undergone significant local changes.
Because of this adjustment, the SCV is not especially recommended for fatigue analysis, as the global contribution is removed. In practice, ROI values in this metric are often distributed from red to blue within the same avatar—some ROIs appear below the global temperature, while others are above it.
The main use of the SCV is to detect anomaly patterns and to support the analysis of laterality. However, it is important to note that, due to the subtraction of global contribution, some noise may appear in the results, caused by ROIs other than the one being analyzed.
The Normalized Coefficient of Variation (NCV) is calculated in the same way as the Coefficient of Variation (CV), with one key difference: instead of using absolute temperatures, it uses normalized temperatures.
The normalized temperature is an estimation based on the background temperature, which reduces the impact of ambient conditions across different sessions. To achieve this, we apply a regression model that estimates the temperature each ROI would have if the room temperature were fixed at 23.5 ºC (the most stable value in our database), instead of the actual ambient temperature. These normalized values are then used to compare each ROI against its historical record.
This metric is especially useful when sessions are carried out under different ambient conditions, since it minimizes environmental influence. For this reason, the NCV should be used instead of the Coefficient of Variation whenever ambient temperature is not constant between sessions.
This is a classification of the injury prevention mode that allows categorizing individuals based on the number and severity of asymmetries on a scale from 0 to 100 (with 0 being the lowest risk value and 100 being the highest risk value). (See Figure 13)
Figure 13. Prevention mode with the calculation of the TRI
For this metric, the number of regions with asymmetry and the degree of significance based on the scale described in the asymmetry section are taken into account. With the aim of identifying profiles with a greater thermal imbalance, it has been observed through research on the homogeneity and stability of asymmetries that healthy bodies tend to be in balance.
Therefore, this global metric aims to provide an indication in the prevention mode. However, it is worth noting that if an injury is present, this index may not be as relevant. This is because, even if the severity is very high in the ROI (Region of Interest), as it also depends on the number of alarms, we may only find asymmetry in the region where the injury is located. This will result in the individual not appearing at the top of this index.
This classification in the fatigue mode allows categorizing individuals based on the temperature trend over time. This metric calculates the temperature variation over time to identify whether an individual is warming up, cooling down, or maintaining a relatively constant temperature in general. (See Figure 14)
Figure 14. Fatigue mode sorted by TSI to visualize the trend.
This allows for the identification of those who respond in a normal or abnormal manner to exercise. The scale ranges from +100 to -100 based on the variation in temperature compared to previous measurements, indicating the deviation from the normal temperature trend.
Therefore, if an individual appears warmer, it suggests a response to exercise and/or activity, whereas if an individual appears cooler, it is necessary to investigate the underlying factors contributing to that response, which could be due to prolonged inactivity or an excessive response from the system related to more central/metabolic fatigue.
In order to display the maximum temperature of the face it is necessary to carry out an upper body protocol from the front (AP protocol). Remember not to make any of these common evaluation mistakes so that the photo can be processed. Automatically, the software will segment and calculate the temperature of the face within that protocol and display it in the "hoover", that is, the last of the data on the poster that appears when placing the mouse over a region of interest in the avatar view (see figure 13):
Figure 13. Evaluation of a subject for the upper body protocol (anterior and posterior parts). By placing the cursor on the ROI of the chest, we obtain a reference in the last row for the temperature of the face, in this case 35.2ºC.
As can be seen in Figure 13, the temperature of the face can be displayed by placing the cursor over any ROI, in this case the ROI of the chest was chosen, and in the last line the temperature of the face appears with a value of 35.2 ºC.
This data allows us to visualize a region of special interest due to its relationship with the body's core temperature, its stability in repeated evaluations and the ability to generate proportions with other regions. We recall that the maximum temperature of the face is related to febrile processes and allows us to relate the internal temperature with the temperature of the inner corner of the eye through these metrics (Zhou et al. 2020; Pascoe et al. 2010; Mercer et al. 2009).
In Figure 14 we can see the region of the face where temperature is measured in feverish states. The difference between core temperature (Tcore) and skin temperature (Tskin) is considered to be ±0.5 ºC, so if the measurement exceeds 39 ºC, it is considered dangerous.
Figure 14. Region of the inner canthus of the eye, where the temperature is extracted for thermal analysis (A) and measurement comparison with thermography and with an axillary contact thermometer of two people, one with fever (B) and the other without fever (C).
Isotherms are a common functionality in the field of thermography and even thermal imaging cameras include a function that allows the most opposite ranges of the scale to be displayed. At ThermoHuman we have included this possibility from the selection of avatars.
It is possible to select in the group report or in the individual follow-up report the isotherm functionality to display, on a scale of 1 Z score, 1.5 Z score and 2 score Z metrics above and below the standard deviation, the most opposite values of the scale (see figure 15):
Figure 15. Scale of isotherms on the individual follow-up report of a soccer player. The areas in red indicate the deviation from the scale above. Blue areas indicate deviation below the scale.
With this functionality, abnormal patterns that are related to hyperthermic and/or hypothermic areas are quickly displayed. In addition, this visualization allows to see those individual patterns that are repeated over time as a result of the athletes' condition and that allow a more advanced analysis (Barcelos et al. 2014). It is an advanced scale of thermograms that highlights the most hypothermic and the most hyperthermic.
Another functionality that ThermoHuman has is the ability to modify the scale within the software itself to establish a more adapted vision to the needs. This scale change affects all selected protocols in both group and follow-up reports. This fact is noteworthy, since once the scale is modified and the modification is accepted, the changes, being modifiable again, will not be reversible at the first visualization.
ThermoHuman, at first when it segments and analyzes, generates an image with a color scale based on the maximum and minimum values of the image itself that allows optimization of the display range, maximizing the contrast between the highest and lowest values. temperature drops. By optimizing the range of each image, the scales are different in the historical series of a person or a group.
For this reason, this way of viewing does not allow a comparison between the images analyzed at different times, because in two different images the same color does not represent the same temperature value.
Hence the need to create thermograms with a constant color scale to allow viewing all thermograms with the same scale defined by the user, so that in these compared views the thermal evolution can be easily seen, at the expense of being able to lose contrast within the same image.
When we get into a group report or a follow-up report and access this functionality we can see that when we select a protocol and decide to make a scale change, it is applied to all the images with that protocol selected (see figure 16):
Figure 16. Rescaling for the previous leg protocol, rescaling to make all images the same with a scale of 24.8 for the cold range and 31.1 for the hot range. In the individual follow-up report of the same player in figure 9.
This modification allows us to quickly observe on which days there has been a qualitative change in temperature towards both hot and cold regions, since it homogenizes the temperatures of the scale for all the images of the different days. In addition, it is especially useful in group reports, since when viewing the different individuals, it is qualitatively appreciated who is colder and/or hotter in the evaluation of the same day. Finally, it is an advanced analysis that allows you to find hyperthermic and/or hypothermic regions and foci in a similar way to isotherms.
For example, at a qualitative level it seems that the player on January 31 and February 8 was generally colder, while on February 3 she was warmer.
The latest functionality within the software is the inclusion of the alarm system. Under an advanced calculation, with an equation that takes into account seven variables (from the player's injury history, the asymmetry of the region to the epidemiology of the sport), a series of traffic light alarms are calculated (yellow, orange and red) and are placed as labels in the tooltip, the area above the evaluations. When selecting the alarm view, the legend on the side informs us of the meaning of each color (see image 17):
Image 17. Scale of alarms in the alarm view if the cursor is placed over the number, a letter appears that informs of the meaning.
Table 1 shows the explanation for each value of the alarm system scale:
Table 1. Alarm levels and intervention proposals.
If the software detects that the player has an abnormality in one of the regions that may pose a risk of injury, a flag will appear for that region based on the appropriate equation. Let us remember that the software learns from the data that we provide, therefore if we add the previous pain regions or previous injuries, the alarm metrics will have a greater statistical power. Similarly, we need to feed the influencer data to the software. If we do not do it correctly, it can be the case of having a serious alarm in a case in which there is a bad segmentation, an ice application has been carried out, treatment... that generates a significant asymmetry).
This alarm system metrics allows us to screen the regions with the greatest relevance in an individual within the individual report, follow-up or in the group reports, to filter more quickly those evaluations that are of greater importance (see figure 18):
Figure 18. Player monitoring report with the evolution of her alarms placed as labels above the thermograms.
As we can see in the example of figure 19, the most significant day is February 3, in which red and orange labels appear that indicate that the right knee and leg in general have an abnormal thermal pattern and that there would be to choose an intervention strategy to solve it. Here you have an example:
Figure 19. ThermoHuman tracking report with "empty" sessions and reminders in some of them
The logic of the alarm system is to offer a first level of approximation to improve the interpretation of the ThermoHuman software and that these labels serve to offer different levels of intervention depending on the urgency and the level of alert shown by the alarm system to the staff of the team.
All metrics requiring several evaluations, thus using historical averages (neutralized asymmetry and both coefficient of variations) might show you this alarm on the tracking report:
It means that the values of this session are not used in the historical metrics calculation because it has been indicated that some ROI has pain, injury and/or the presence of influence factors. Therefore you might not have an avatar or the results are based on calculations of other “clean” sessions.
If you want to include those session in the global calculation just click on the following funnel button:
So you will be able to see those metrics including sessions with pain, injury an/or influence factors on figure 20.
Figure 20. Same report but activating the funnel button to include sessions with pain, injury and / or influence factors
In conclusion, thermography allows us to use metrics from the first evaluation. In this sense:
Infrared thermography is a solution that allows us to get relevant information from the first moment. Obviously, the more evaluations we have the better, because besides ThermoHuman will give us the possibility of richer and complementary analysis using all these metrics.
Uematsu, S., Edwin, D. H., Jankel, W. R., Kozikowski, J., & Trattner, M. (1988). Quantification of thermal asymmetry. Part 1: Normal values and reproducibility. J Neurosurg, 69(4), 552-555.
Formenti, D., Ludwig, N., Rossi, A., Trecroci, A., Alberti, G., Gargano, M., . . . Caumo, A. (2018). Is the maximum value in the region of interest a reliable indicator of skin temperature? Infrared Physics & Technology, 94, 299-304.
Arnaiz Lastras, J., Fernández Cuevas, I., Gómez Carmona, P. M., Sillero Quintana, M., García de la Concepción, M. Á., & Piñonosa Cano, S. (2011, 6th-9th july). Pilot study to determinate thermal asymmetries in judokas. Paper presented at the 16th Annual Congress of the European College of Sport Sciences ECSS, Liverpool, United Kingdom.
If you have any questions or would like to make a comment, do not hesitate to write to us. We will be happy to read you.
ThermoHuman has had the support of the Funds of the European Union and the Community of Madrid through the Operational Programme on Youth Employment. Likewise, ThermoHuman within the framework of the Export Initiation Program of ICEX NEXT, had the support of ICEX and the co-financing of the European Regional Development Fund (ERDF).
Thermohuman has received funding from the Centre for the Development of Industrial Technology (CDTI), in participation with the European Regional Development Fund (ERDF), for the R+D activities involved in creating a new tool, based on thermography, for the prediction and prevention of rheumatoid arthritis. See project detail.
