The use of infrared thermography in the measurement and characterization of brown adipose tissue activation

ABSTRACT

Interest in brown adipose tissue has increased in recent years as a potential target for novel obesity, diabetes and metabolic disease treatments. One of the significant limitations to rapid progress has been the difficulty in measuring brown adipose tissue activity, especially in humans. Infrared thermography (IRT) is being increasingly recognized as a valid and complementary method to standard imaging modalities, such as positron emission tomography–computed tomography (PET/CT). In contrast to PET/CT, it is non-invasive, cheap and quick, allowing, for the first time, the possibility of large studies of brown adipose tissue (BAT) on healthy populations and children. Variations in study protocols and analysis methods currently limit direct comparison between studies but IRT following appropriate BAT stimulation consistently shows a change in supraclavicular skin temperature and a close association with results from BAT measurements from other methods.

Brown adipose tissue

BAT was first documented in 1551 by the naturalist Conrad Gessner, who described tissue found in the interscapular region of the marmot as “neither fat nor flesh – but something in between”.¹ Having initially been identified solely by its color (due to its granular, mitochondria-rich cytoplasm), BAT was only conclusively demonstrated as being a separate pathological entity to white adipose tissue (WAT) in the early 2000's – where it was shown that bone morphogenic proteins were involved in the differentiation of adipocytes into white or brown adipose tissue.² BAT is unique in that it contains uncoupling protein 1 (UCP1) on the inner surface of its mitochondrial membrane which, when activated, allows rapid dissipation of energy by the free flow of electrons, producing heat³ and playing a pivotal role in thermoregulation, particularly in infancy. There are also discrete regions of UCP1 containing cells within WAT, that have been defined as beige adipocytes,^4–7 although these may have the capacity to utilize other modes of uncoupling.⁸

Historically, BAT was thought to disappear with age, however the persistence into adulthood of BAT in small depots, at sites similar to those found in infants, was noted by the pathologist Heaton in the 1970s.⁹ It is likely that further insights were apparent at this time but were largely neglected.¹⁰ The largest BAT depots are found surrounding the vasculature of the neck,¹¹ thought to play a role in thermoregulation of the blood to and from the brain.¹² It is the activity of these depots, noticed on PET/CT imaging of adults,^13,14 which promoted the resurgence of interest into BAT activity early in the millennium. The estimated volume of BAT varies depending on the conditions and quantification methods,¹¹ which are not standardized.¹⁵

BAT activity is controlled by the action of norepinephrine (from the sympathetic nervous system) on β₃-adrenoreceptors,¹⁶ and is inhibited by vagal nerve stimulation¹⁷ (from the parasympathetic nervous system). Norepinephrine has a number of actions on BAT to increase its activity; promoting proliferation of preadipocytes, differentiation of mature adipocytes,¹⁸ directly upregulating the expression of genes coding for UCP1,¹⁹ increasing mitochondrial mass²⁰ and preventing apoptosis.²¹ The activation of BAT through this pathway can be acute (i.e. in response to abrupt cold exposure or a meal²²) or can be a result of enhanced BAT recruitment, increasing thermogenic capacity over a period of time.

As implied above, there are many stimuli to BAT activation (whether acute or chronic). The most documented is the activation of BAT in response to cold, which probably underpins the evolutionary survival of mammals in the neonatal period. Extreme cold exposure results in the production of heat primarily through shivering, and although BAT is also activated, it contributes to overall energy expenditure far less.²³ At lesser degrees of cooling, BAT is the main source of heat production. Over a period of time, following repeated cold exposure, or prolonged acclimation to a low temperature (over weeks), the recruitment of BAT can result in non-shivering thermogenesis to maintain a stable body temperature (adaptive thermogenesis).^24,25

BAT is also activated through diet. As early as the nineteenth century it was noted that when humans over-ate, weight gain did not mirror the additional calories ingested,²⁶ suggesting an increase of energy expenditure in response to food.²⁷ The hypothesis that this was due to BAT activation was supported when studies on UCP1 knock-out mice showed an increase weight gain compared to wild-type mice housed at the same temperature.²⁸

In addition to environmental factors, the activity of BAT is also affected by different drugs, mainly those that impact the sympathetic system, such as cocaine and amphetamines²⁹ as well as serotonergic drugs.³⁰ Other drugs have been studied to assess their ability to increase BAT activity through other means. For example, thiazolidinediones have been shown to induce “browning” of WAT in mouse models,³¹ whilst raised triiodothyronine has been shown to be associated with higher activity of BAT, both in cell cultures (through increased upregulation of UCP-1 expression and mitochondrial biogenesis³²) and in clinical practice (as demonstrated by an increased basal metabolic rate and non-shivering thermogenesis in response to thyroxine treatment post thyroidectomy³³). Although a relatively wide range of different factors have been found to impact on the activity of BAT, the best way to measure that activity in a research environment is contentious. The pros and cons of different modalities (including thermal imaging) in humans are discussed below.

Assessing BAT activity

There are many methods of assessing the activity of BAT that have been utilized experimentally. Broadly speaking they can be divided into direct and indirect measures of activity.

Direct examination of BAT requires biopsy samples (usually taken from the supraclavicular (SCV) depot under CT or PET/CT guidance following activation through cooling). There are a number of different direct methods of assessing activity including the measurements of gene expression markers (UCP1 mRNA,³⁴ lipoprotein lipase mRNA³⁵), gene protein products (UCP1 or BAT lipid content), or measures of mitochondrial respiration.^36,37 The requirement of biopsy samples for all these methods limits the use of direct measures of BAT activity to animal models or small human studies.

Indirect measures of BAT activity are more widely used in clinical research, although which method is optimal remains a matter of debate. The principal function of BAT is to produce heat, and as such, calorimetry may be considered the most appropriate method of measuring the heat production from BAT. This is usually performed through indirect calorimetry that measures gas exchange, i.e. oxygen consumption and carbon dioxide production, calculating energy expenditure using the Weir equation (EE(J) = 15.818VO₂(l/min) + 5.176VCO₂(l/min)). Indirect calorimetry has therefore been utilized as a method of measuring an increase in cold-induced energy expenditure in numerous human studies.^38,39 However, this rise in metabolic rate in response to cold does not necessarily indicate an increase in BAT activity. A number of studies have shown that UCP1 knock-out mice maintain the ability to increase their metabolic rate in response to cold – indicating a lack of specificity to BAT.^40,41 The use of indirect calorimetry alone as a measure of BAT activity may not be sufficient.

PET/CT

The revival in interest in BAT physiology over the last 10 years followed the demonstration that the symmetrical uptake seen on PET/CT¹³ would resolve if the participant was kept warm^14,42 and the proof that this was due to BAT present in adult humans.^43–47 PET/CT, specifically ¹⁸F-FDG PET/CT, has since been considered the gold standard for imaging and measuring BAT in vivo in humans.¹⁸F-FDG PET/CT relies on the high glucose uptake rate of activated brown adipocytes. PET detects the radioactivity of the decaying ¹⁸F-FDG in cells, whilst CT measures the density of tissue (distinguishing BAT from WAT).

The utility of PET/CT is limited most significantly by the necessary exposure to relatively high levels of ionizing radiation (∼8mSv). Studies of healthy individuals are therefore ethically limited to small numbers and, although some studies have employed repeated PET/CT BAT measurements on the same individuals, these studies are particularly limited to very small numbers at most,^48,49 reducing confidence in the results. In addition, the amount of BAT identified depends on external factors such as PET resolution, the region of interest (ROI) chosen and the PET threshold criteria and may have been being significantly underestimated.¹¹ Larger studies can be undertaken by looking at images taken for clinical indications but conditions and protocols are optimized to minimize BAT identification so these studies find much smaller prevalence of BAT^46,50 than those that use cooling protocols,^45,47 and cannot necessarily be extrapolated to healthy populations. Other practical limitations of PET/CT include the availability and cost of equipment and scanning time. A PET/CT scanner costs upwards of $100,000; a single research scan for BAT quantification typically costs around $450 and takes 1.5 hours from administration of ¹⁸F-FDG to completion.

In addition to the practical concerns, there are also methodological questions. The most common tracer used in PET/CT imaging of BAT is ¹⁸F-FDG. Whether glucose uptake is a good surrogate measure of BAT activity remains to be answered. Although BAT⁵¹ and beige cells⁵² do utilize circulating glucose for oxidation, ¹⁸F-FDG uptake is unaffected even in UCP1 KO mice where BAT thermogenesis is diminished.⁵³ Further to the concerns around substrate uptake being equated with cellular activity, PET/CT is also not suitable for studying the effect of a meal. Following a meal, insulin release causes glucose uptake by muscle reducing the contrast with BAT uptake, making interpretation difficult⁵⁴ and risking under-estimation of BAT.⁵⁵ Since the field of BAT research is largely ultimately looking towards novel obesity and diabetes interventions, the inability to study the effects of meals is a significant limitation that should not be overlooked.

The non-destructive, non-invasive properties of infrared (IR) and, being part of the electromagnetic spectrum, its ability to travel in a vacuum, have meant it has found a use in a wide range of disciplines including conservation and heritage,⁵⁶ astronomy,⁵⁷ mineralogy⁵⁸ and engineering.^59,60 IRT was first used for medical purposes in the 1950s for the identification of breast malignancy.⁶¹ While mammography and ultrasound overtook IRT in this particular application, it has since then gone on to be used for a wide range of medical uses⁶² including sports medicine,⁶³ arthritis,^64,65 and a renewed interest in malignancy.^66,67

Background of IRT

IRT is the process of constructing an image of an object's temperature by first measuring the IR radiation being emitted, then converting that radiometric data to a temperature and, finally, displaying the temperature data as an image. The IR radiation emitted is related to the temperature of the body with long-wavelength IR (8–15μm) emitted by objects whose temperature is between -80°C and 89°C, including the human body.⁶⁸ Planck's Law can be used to convert radiometric data to temperature data when accounting for emissivity of the body being measured.^68,69 In addition, adjustment must be made for any attenuation of the signal prior to it reaching the detector, for instance due to water vapor in the atmosphere (a function of humidity and distance).⁶⁸ The variables used in the conversion are listed in Table 1 along with an outline of their effect.

Table 1.

Variables which influence conversion of radiometric to thermal data.

Variable	Effect
Relative humidity	Water vapor is the main atmospheric absorber of IR in the LWIR range. High humidity therefore results in less IR reaching the sensor.
Object Distance	Increasing the distance between the object and the camera increases the optical path length and hence the attenuation due to any water vapor present. For the distances typically used in measuring BAT (<2m) the effect will be minimal.
Emissivity (ε)	Emissivity is the proportion of radiation emitted by an object compared to an ideal black body. Emissivity of human skin is 0.98.
Atmospheric temperature	Atmospheric temperature is required to calculate the incident radiation that the object is exposed to (that is, the amount of energy available to be absorbed by the body).
Reflected temperature	Reflected temperature is used to estimate the proportion of the radiation arriving at the sensor that is from background radiation, i.e. not emitted from the object. For imaging biological objects, emissivity is high (0.95–0.98) and so background reflected radiation is relatively low.
External IR window compensation	This is not applicable unless the camera is imaging an object from which it is segregated. This may be desirable in, for instance, the imaging of newborn infants in an incubator.

BAT: brown adipose tissue; IR; infrared; LW: long-wavelength

IRT of BAT

IRT offers a complementary detection method for BAT activity and can overcome some of the drawbacks of PET/CT. The non-invasive nature of thermal imaging and its avoidance of ionizing radiation make it highly suitable for BAT measurements in healthy cohorts or children and young people. Children as young as three years old can participate in short imaging protocols.^70,71 The flexible and non-harmful nature of IRT make serial measurements of large numbers of individuals become feasible and ethical, allowing greater power through paired, rather than independent, analysis.

IRT requires very little equipment and good quality measurements can be made if sufficient care is taken in the set up. Changes in SCV temperature (T_SCV) can be measured by five minutes of stimulation,⁷⁰ although further changes are seen with prolonged cooling.⁷² Environmental temperature will affect results⁷⁰ but this need not prevent studies being conducted outside the laboratory environment and they have successfully been undertaken in people's homes⁷⁰ and at locations within schools.⁷¹

In contrast to PET/CT, IRT measures heat directly which is the key outcome of thermogenesis and has been used successfully to study meal effects.^73–75 The practical and ethical freedoms of IRT compared to PET/CT thereby confer the ability to create large datasets which will provide increased confidence in results as well as allow others to check that results are reproducible.

IRT was used to study BAT as early as the late 1970s to show a thermogenic response to ephedrine in an individual²⁷ although this was subsequently found to only measure changes in blood flow. As techniques developed, BAT hotspots were consistently located, especially in rodents, and some quantification was attempted but was not compared to other measures⁷⁶ and was limited by the low spatial resolution of the cameras available.⁷⁷ Following the confirmation in 2009 of BAT in human adults,^44–47 IRT was re-examined as a potential technique to measure BAT activation.^70,74,78 Rodent studies demonstrate consistently strong associations between BAT activity measured using thermal imaging and other methods such as 18F-FDG uptake⁷⁹ or metabolic demand,^80,81 whether using paired analysis^79,80 or comparing between groups.⁸¹ However, it remains to be shown whether these results would translate to humans who have a much lower body surface area:volume ratio, tend to live at thermoneutrality⁸² and have significantly different metabolic demands.^83,84

As the largest and most active BAT depot in humans,^11,85 the SCV region has been the area most often studied using IRT. It occupies a superficial location below the subcutaneous adipose tissue in the SCV fossa within the lateral cervical region of the neck making it highly suitable for this modality.^85,86 In early life, there is an active interscapular depot^9,87 which was felt to represent the closest human depot to classical BAT as found in rodents^85,88 and is similarly amenable to thermal imaging⁷⁶ but which rapidly declines in prevalence with age.⁹ However, mice also possess a classical BAT depot in the SCV region with a gene profile similar to that of human SCV BAT.⁸⁹

Lee et al. showed that IRT had promise as a method for measuring BAT activity in humans, demonstrating that cold stimulation resulted in increased relative sparing of the SCV temperature, at least in some individuals,⁷⁴ and it was clear that the SCV “hotspot” was projected in the anatomical area close to BAT on PET/CT. Early studies showed a specific and reproducible warming of the SCV region following cool stimulus, consistent with BAT activation^70,71 but did not directly compare IRT with PET/CT in the same individual.^70,74 Paired studies failed to show a correlation between SCV temperature and BAT volume⁴⁹ or positive BAT status on PET/CT⁵⁰ but only analyzed single images at one⁵⁰ or three time-points.⁴⁹ The use of single images does not capture the dynamic nature of the changes seen on IRT^70,72 and is vulnerable to bias from external and environmental factors.⁹⁰ Moreover, the use of clinically indicated PET/CT scans, which typically use protocols to reduce BAT “artefact” (e.g. by keeping the room warm) resulted in less than 10% of participants being defined as BAT positive.⁵⁰ In contrast, protocols designed specifically to measure BAT activity demonstrate positive scans in the majority of patients.⁴⁹ Furthermore, studies analyzed thermal images either using triangles to approximate the SCV region⁵⁰ or using circles to define the SCV hotspot, which requires a subjective decision about the precise placement.⁴⁹ It has now been confirmed that BAT on PET/CT and the hotspot on IRT co-localize closely and that IRT measurements are strongly correlated with measurements on BAT activity on PET/CT.⁷²

Despite these promising findings, caution must be exercised so as not to underestimate the complexity of the relationship between skin surface temperature and BAT heat output, reflected in recent efforts to standardize the collection and analysis of data.⁹⁰ The key function of BAT is efficient heat production to enable thermoregulation.³ Consistent with this, BAT depots are richly vascularized and are generally located near to major vasculature.^86,91 The location of the major BAT depot within the SCV fossa is presumably to contribute to cerebral blood temperature,¹⁵ consistent with its proximity to carotid arteries. Heat is transported from the activated tissue around the body and specifically caudally, but sufficient heat is transmitted to the surface of the skin to be detected.

Heat that is radiated to the external environment must first be conducted through the subcutaneous adipose and the dermal tissues. Many factors, other than BAT activity, may therefore alter the heat signal measured using IRT. Adipose tissue is highly insulating⁹² and increased adiposity reduces skin temperature.⁹³ Body mass index (BMI) is inversely associated with SCV temperature in children⁷¹ and to what extent this an effect of attenuation of the heat signal or reduced BAT activity is yet to be fully elicited. The percentage of body fat in any given anatomical area is inversely correlated with skin temperature in that area and subscapular skinfold thickness specifically is negatively associated with lower local skin temperature.⁹⁴ The insulative and dissipating effect of adiposity can be corrected for either by calculating general indices of body composition, such as BMI,⁵⁰ or by specifically measuring skinfold thickness⁹⁴ close to the ROI, for instance subscapular. At the population level skinfold thickness and BMI are strongly correlated^95–98 but the variation between individuals^96–98 may mean a direct measure is necessary.

On the other hand, BAT is not the only local source of heat in the neck and, notably, increased blood flow in the carotid arteries may be expected to confound IRT measurements. Relative SCV temperature increases even in the absence of cardiovascular changes (Law et al., unpublished) but images from some individuals show “hot spots” which appear to follow the path of the carotid vessels (Figure 1). Further work is needed to establish even more rigorous methodologies to isolate thermal changes related to BAT activity, but there is no evidence currently that this is a significant issue at a practical level.

Figure 1.

(a) Original thermal image (b) Thermal image after processing.

Thermal image of a 7-year-old boy. (a) false color image and (b) grey-scale image with the SCV ROIs outlined in blue and “hotspot” (upper decile of values) in red. The participant's right ROI contains two distinct red areas of which the medial one (arrow) close to the projection of the carotid artery. ROI: region of interest; SCV: supraclavicular.

IRT is also limited to measuring SCV and interscapular BAT depots due to their convenient anatomical location. While the interscapular depot in the newborn is classical BAT, it is argued that the molecular signature of the supraclavicular depot in adults is more similar to beige depots in rodents.^5,85,88 Whether inducible beige depots, located within WAT, possess a greater or lesser combined potential to improve metabolic health than classical BAT depots remains unanswered.^6,99 Their diffuse nature, however, prevents them from being measured using thermal imaging or PET/CT currently.^100,101 If the supraclavicular depot is truly beige, changes in its activation may mirror browning in other potential beige locations, allowing measurements of its activity to be used as a surrogate for whole body changes.

Study protocols using IRT

Standardization of IRT protocols remains poor and is a significant barrier to comparing results. Prior to commencing imaging measurements, it is essential that the participant is fully rested and acclimatized. Activity can affect skin perfusion.^102,103 Acclimatization is usually reported (Table 2 in supplementary data) but duration varies from 5 minutes⁷⁰ to 2 hours.¹⁰⁴ Changes in environmental temperature can likewise affect recordings⁷⁰ and room temperatures are consistently reported. Except for when participant cooling is being achieved by reduced air temperature, they are usually within the thermoneutral zone (approximately 22–24°C, in the absence of wind or airflow, for adult participants, in light cotton shorts and t-shirt¹⁰⁵ but affected by individual factors such as adiposity, age and possibly gender;^82,105,106 the effect of variation in one of more of these factors limits direct comparison between studies).

To maximize the IR radiation measured by the thermal camera, the camera should be perpendicular to the ROI and the ROI should be as large as possible within the field of view. The externally rotated shallow angle of the SCV fossa from the coronal plane means that the optimal camera position would be rotated down and inwards. This would mean only one side could be measured by a single camera and the position would be difficult to reproducible reliably. An acceptable compromise of positioning the camera level with the participant at the level of the larynx is adopted in most studies^{49,50,70–74,104,107–110} but it should be realized that this is at the expense of reduction in the IR radiation being captured by the camera and, therefore, the calculated temperature will be below the true temperature (Figure 2). The camera should be moved along an imaginary line perpendicular to the coronal plane and passing through the larynx to ensure that the deltoid muscles are close to the border of the field of view but not extending beyond it so that the ROI occupies a maximal area while maintaining the necessary anatomical landmarks. In addition, the participant should maintain a standard body position with the head in a neutral position and the shoulders abducted and this should be consistent between participants.

Figure 2.

Representation of infrared radiation emitted by the supraclavicular region.

Radiation from the supraclavicular region (red triangle) is emitted maximally perpendicular to the region in the direction of a camera at position A (dark red arrows), which is elevated and rotated relative to the camera at position B. Position B is on a reproducible imaginary line that passes through the larynx and is perpendicular to the coronal plane (black dotted line). A camera at position B will underestimate the true skin temperature of the supraclavicular region due to reduced infrared radiation being measured (light red arrows).

Acute stimulation can be achieved in several ways including cooling,^{49,70,71,74,75,104,109–111} pharmacological agents,²⁷ diet⁷³ or mental stress.¹⁰⁷ Conversely, relaxation can inactivate BAT.¹¹² Due to its core function of thermogenesis, the most commonly used methods of stimulation involve cooling. This may be achieved be either decreasing the air temperature,^49,74,104 using water either in direct contact with an extremity,^{70,71,75,104,110,111} or using a water-cooled blanket or vest.¹⁰⁹ Water temperature from 5°C^110,111 to 15–20°C has been used.^{70,71,74,75,104} Ice-water (4–5°C) is significantly below the temperature at which shivering thermogenesis is activated²⁴ and exposure will lead to maximal SNS activation due to pain.¹¹³ Even when shivering is not observed,¹¹⁰ its absence should be confirmed by both participant report and electromyograph measurements as well.^15,38,114 Shorter protocols may not allow significant time to measure activation and potentially fail to find a change in activation where one exists.¹¹¹ More acceptable cooling protocols detect changes within five minutes.^70,71,75

Several studies measuring BAT activity with PET/CT have utilized individualized protocols to maximize cold stimulation^{24,55,115–117} and attempt to correct for factors such as obesity. To date individualized protocols have not been used in IRT studies. While PET/CT measures activity at a given timepoint (usually relative to a non-BAT tissue such as muscle), much of the information from IRT is in the change over time (i.e. from resting, in response to stimulation). Individualized protocols are usually determined on the day of imaging, as they may vary from day to day, but cooling beyond the onset of shivering on the same day would risk affecting resting IRT measurements. Nevertheless, particularly with the interest in measurements on obese participants, measures to correct for adiposity will be increasingly important and require further consideration.

Analysis of thermal images

Like the variation seen in imaging protocols, analysis of the resulting images is highly heterogenous (Table 2 in supplementary data). This is exacerbated by the fact that thermal imaging cameras will usually save radiometric images in a proprietary file format that encourages the use of the manufacturer's own software. Apart from reducing the ease of collaborations, this also risks data in images taken using obsolete formats rapidly becoming inaccessible. For example, FLIR, a major thermal imaging supplier, currently stores its raw thermal data within the metadata of a joint photographic experts group (JPEG) file of a false color image that is representative of the thermal data contained within (Figure 3A). While the pseudocolored image is readily accessible in image processing software, the raw radiometric data cannot easily be obtained without the use of FLIR™ proprietary software. The radiometric data stored in the file, along with other stored variables (Table 1) allow the image to be viewed as temperature information when accessed using the proprietary software (Figure 3B). In this environment, basic analysis and manipulation of the image is possible, such as identifying minimum, maximum and average temperatures within user-defined areas, changing scales and color maps and altering the value of recorded variables. Depending on the software being used, images may also be saved in an alternative, more accessible format such as comma-separated values (.csv) or MATLAB™ (.mat) files, where the pixel data is now stored as temperature values for further analysis.

Figure 3.

Thermal image saved by a FLIR™ camera (a) viewed with standard image-viewing software and (b) opened in FLIR™ ResearcherIR™.

(A) The JPEG false-color image is static and overlaid with crosshairs, FLIR™ logo and any other information on the screen of the camera when captured. Thermal values can be estimated by reference to the colourmap, but cannot be precisely calculated. (B) The same image with FLIR™ ResearcherIR™ is interactive and ROIs can be defined (e.g. cyan circle) from which basic data such as minimum, maximum and average temperature can be derived (white box at bottom). JPEG: joint photographic experts group; ROI: region of interest.

An alternative, if using FLIR™ cameras, is to utilize their software development kit that contains a compiled binary code file (a MATLAB™ .mex file) that allows radiometric data to be read directly into MATLAB™ without prior conversion to another format. This enables many still images and video formats to be read, including historic ones. Finally, images can be stored in a futureproof, openly accessible format such as a 16-bit portable network graphics format.⁷² Conversion of large numbers of images can be undertaken automatically and rapidly and the resulting images can be easily imported into any programming environment and converted to temperature data with the required flexibility to modify recorded variables.⁷²

Once the image has been opened within the desired environment the ROI needs to be identified and analyzed. Original methods using large rectangular ROIs⁷⁰ were superseded by careful manual approximation of the contour of the shoulder between the clavicle and sternocleidomastoid muscle⁷¹ (Figure 4A), approximating the lateral cervical region within which the SCV fossa is situated using anatomical landmarks identifiable on the thermal image, and which are therefore reproducible. Data was exported from proprietary software as CSV files to calculate the average value of the BAT “hotspot” within this larger area.^70,71 These methods, however, preclude analysis of large numbers of images which is desirable for many studies. To simplify the identification of the ROI, studies have used circles^{49,74,104,108} (Figure 4B), triangles^50,110 (Figure 4C) or even squares¹¹¹ (Figure 4D) to define the SCV area. Unfortunately this relies on a variable degree of subjective placement and/or poor approximation to the contour of the neck,¹¹⁰ and may result in apparent inclusion of the background region.¹¹¹ To simplify the calculation of the metric, basic measures of a whole ROI can be made using the manufacturers’ software (Figure 4B). However, the average^{50,104,110,111} necessarily assumes BAT contributes to the whole ROI and the maximum^49,50,104 relies on the temperature at a single point.

Figure 4.

Examples of different methods used to analyze thermal images.

(a) Identification of SCV ROI (cyan line) using FLIR™ ThermaCAM™ Researcher Professional polygon tool (reproduced using the methodology from ref. 62. The process shown here for the left side would be repeated for the right. The ROI is then exported as a CSV file for further analysis. (b) Two thermal images using triangles to approximate the ROI. Subjective placement can result in a variable ROI being selected and reduced reproducibility. Unusually the hotspot in the upper image appears to have a maximal temperature of approximately 40°C. © Institute of Physics and Engineering in Medicine. Reproduced by permission of IOP Publishing. All rights reserved.¹¹⁰ (c) Thermal image from a child with a diagnosis of hypothyroidism. A wide temperature range on the color-scale reduces contrast in the SCV ROI impeding interpretation of image. Reference area (black rectangle) taken across jugular and sternal region. Reproduced by permission of Oxford University Press. ¹⁰⁸ (d) Square SCV and reference ROI which include background of image. © 2017 The Obesity Society. Reproduced by permission of John Wiley and Sons.¹¹¹ (e) Circular ROI used for SCV and reference ROI. Reproduced under CC-BY license.¹⁰⁴ (f) Thermal image analyzed in MATLAB™ with using semi-automated process to define SCV ROI, SCV hotspot and reference ROI. Reproduced under CC-BY license. ⁷² CSV: comma-separated values; ROI: region of interest; SCV: supraclavicular.

Improved computational methodologies have started to try and address these shortcomings. The ROI can now be calculated automatically with the user identifying just the apices,⁷² reducing the subjective placement of the ROI outline and allowing close approximation to the shoulder contour (Figure 4E). This method is currently used to calculate the median value of the upper decile of points within the ROI (equivalent to the 95^th percentile). While a range of values have been used,^{70,71,73,107,109} measurements made using 10% correlate well with BAT activity measured by PET/CT.⁷² An alternative option using a seeding approach has been suggested⁷⁵ which has the benefit of identifying a discrete single hotspot but requires a subjective decision about a key parameter (T_t). This is currently limited by the computational power required and the increased complexity does not appear to offer a significant advantage over a percentile approach in practice at the moment.

Options for IRT outcome measures include resting temperature (T_r), stimulated temperature (T_s), change in temperature (ΔT = T_r – T_s), with each measure able to be calculated as an absolute value (i.e. T_SCV) or relative to a reference point (T_rel = T_SCV – T_ref). For the purposes of this review, ΔT refers to the change from resting, but it is used by some to mean the difference between the SCV and reference,⁴⁹ here called T_rel.

Reduced T_SCV in cohorts with increasing BMI⁹³ is due to a combination of reduced BAT^115,118 and increased insulation.^93,94 This compound effect of obesity can be moderated by using ΔT, rather than T_r or T_s, and by comparing the SCV ROI to a non-BAT area on the same individual (i.e. T_rel). This approach effectively uses the participant as their own control and may explain why the best correlation with PET/CT is ΔT_rel (r² = 0.583; p = 0.027),⁷² rather than T_r or T_s, and why T_rel offers a better prediction of BAT status on PET/CT than T_SCV.⁴⁹ Given the significantly different components of BAT function being measured by these two methods (glucose uptake verses thermogenic capacity), better agreement is unlikely.

Different approaches to identifying a reference value have been attempted, but these can broadly be grouped into two general approaches: either a non-BAT region on the thermal image is used,^{49,50,73,74,104,108–111} or a measure of skin temperature is made by direct contact methods (thermocouples or iButtons™^71,107). Using a non-BAT region on the thermal image has the advantage that it can be added or altered retrospectively if the relevant area was imaged originally. However, it is worth noting the relative scarcity of female participants (Table 2 in supplementary data) and that the frequent use of a reference point including the chest area is less likely to be acceptable for females. In addition, use of contact measurement allows mean skin temperature to be calculated over a larger anatomical distribution.^107,119

While this approach reduces the effect of differences in obesity (as well as gender, age and other variables), T_SCV, either at rest or stimulated, may be an important measure of effective BAT activity in itself. Theoretically, a participant who has very active BAT, even at rest, may not increase their T_SCV by much in response to stimulation as there is a maximal threshold temperature.⁷³ Consequently, it may be difficult to distinguish this from someone with inactive unresponsive BAT (Figure 5). Where the intention is to take serial measurements to look at changes in BAT over time in the same individual, whether in longitudinal or interventional studies, comparison of the change in T_r or T_s from baseline would avoid the complexity in interpreting ΔT and, without the ionizing radiation and cost limitations of PET/CT, such study designs are now feasible.

Figure 5.

Simulated data from a mock IRT study.

Participant A (blue) has little active BAT – they therefore have a low resting SCV temperature(T_SCV,r) and little rise (ΔT_SCV). Participant B (yellow) has activatable BAT which shows reduced activity at rest (low T_SCV,r) but a large response to the stimulation and therefore a large ΔT_SCV. Finally, participant C (red) has highly activated BAT; they show a small ΔT because their BAT is already maximally activated. Since resting temperature is affected by factors other than BAT activity, IRT may not be able to distinguish maximally activated BAT from inactivated BAT. BAT: brown adipose tissue; SCV: supraclavicular; T_r: resting temperature; ΔT: change in temperature from resting.

Summary

Developments in IRT over the last six years have established it as a significant and validated tool for the measurement of BAT activation in humans as well as other animals. The ready availability of affordable equipment and lack of ionizing radiation make possible for the first-time large studies on cohorts not previously accessible and lead to novel insights into BAT function and physiology. However, significant challenges remain to ensure reproducibility and consensus statements on imaging acquisition and analysis are urgently required to allow results to be compared.

Abbreviations

BAT

Brown Adipose Tissue

BMI

Body mass index

IRT

Infrared thermography

JPEG

Joint photographic experts group

PET/CT

Positron emission tomography–computed tomography

ROI

Region of interest

SCV

Supraclavicular

T

Temperature

T_r

Resting temperature

T_ref

Reference temperature

T_rel

Relative temperature

T_s

Stimulated temperature

T_SCV

Supraclavicular temperature

UCP1

Uncoupling protein 1

WAT

White adipose tissue

ΔT

Change from resting temperature

Biographies

James Law is an Assistant Professor in Child Health at the University of Nottingham. Alongside his clinical interest in pediatric endocrinology, James’ research interest in obesity and adiposity currently focuses on brown adipose tissue, its interaction with the endocrine system and its measurement using infrared thermography.

Jane Chalmers is a clinical research fellow at the Gastrointestinal and Liver disorders theme of the NIHR Nottingham Biomedical Research Centre. Jane's main research interest is in the role of brown adipose tissue activity in relation to the metabolic syndrome, in particular non-alcoholic fatty liver disease.

David Morris is an Assistant Professor in the Faculty of Engineering at the University of Nottingham. His research interests lie in automating the processing of biomedical data and the development of novel sensors that can obtain this data.

Lindsay Robinson is a Clinical Research Fellow in Child Health at the University of Nottingham. Her research interest is brown adipose tissue development and its quantification by thermal imaging.

Helen Budge is a Clinical Professor in Neonatal Medicine at the University of Nottingham, where she is Deputy Head of the School of Medicine. Her research interests include the lasting effects of early life on physiology, development and metabolism.

Michael Symonds is a Professor of Developmental Physiology at the University of Nottingham. He has wide-ranging research interests which include adipose tissue development and its role in body weight regulation and metabolic homeostasis.

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.