Using free, open-source software, we have developed an analytical approach to quantify total and regional brown adipose tissue (BAT) volume and metabolic activity of BAT using 18F-FDG PET/CT.
In endothermic animals, brown adipose tissue (BAT) is activated to produce heat for defending body temperature in response to cold. BAT's ability to expend energy has made it a potential target for novel therapies to ameliorate obesity and associated metabolic disorders in humans. Though this tissue has been well studied in small animals, BAT's thermogenic capacity in humans remains largely unknown due to the difficulties of measuring its volume, activity, and distribution. Identifying and quantifying active human BAT is commonly performed using 18F-Fluorodeoxyglucose (18F-FDG) positron emission tomography and computed tomography (PET/CT) scans following cold-exposure or pharmacological activation. Here we describe a detailed image-analysis approach to quantify total-body human BAT from 18F-FDG PET/CT scans using an open-source software. We demonstrate the drawing of user-specified regions of interest to identify metabolically active adipose tissue while avoiding common non-BAT tissues, to measure BAT volume and activity, and to further characterize its anatomical distribution. Although this rigorous approach is time-consuming, we believe it will ultimately provide a foundation to develop future automated BAT quantification algorithms.
The increasing prevalence of obesity worldwide1 has prompted an investigation into novel therapeutics to prevent and ameliorate obesity and its associated complications. Obesity is due in part to excess energy stored in white adipose tissue (WAT) in the form of triglycerides2. Brown adipose tissue (BAT) differs from WAT most notably due to its higher mitochondrial content, smaller and multilocular lipid droplets, distinct anatomical distribution, greater sympathetic innervation, and heat generating ability. Although BAT was once thought to exist only in small mammals and newborn infants, the presence of functional BAT was confirmed in adult humans in 20093,4,5. The thermogenic capacity of human BAT is not yet known, but extensive study in small animals has shown that non-shivering thermogenesis can constitute up to 60% of their metabolism during cold-exposure6. As a result, human BAT is now being explored as a target for the treatment and prevention of obesity and related disorders7. Several clinical studies have shown that BAT thermogenesis correlates with increased glucose uptake and energy expenditure upon activation by mild cold exposure8,9,10. Yet, BAT's contribution to cold-induced thermogenesis remains controversial11,12,13,14, with much debate centered around how to quantify human BAT15. To better understand if BAT thermogenesis can be harnessed to combat obesity, it is critical to have an accurate measurement of its volume and metabolic activity.
Obtaining precise measurements of BAT is challenging due to BAT's unique anatomical distribution in humans. BAT is distributed within the white adipose depots in the neck, thorax, and abdomen in sites that are inaccessible to uncomplicated biopsies14. Autopsies have been used to characterize BAT anatomically16, but are infeasible for most research laboratories doing large studies and cannot provide longitudinal or functional information. Since BAT has a similar density to WAT and can occur in narrow fascial layers or in small pockets interspersed with WAT16, it is difficult to identify using a single, conventional imaging technique. This heterogeneity also makes automatic quantification of BAT more difficult than quantification of homogenous structures such as the liver17.
To overcome these challenges, BAT volume and activity are commonly quantified by coupling computed tomography (CT) and positron emission tomography (PET). The radiolabeled glucose analog 18F-Fluourodeoxyglucose (18F-FDG) is the most widely used tracer to study BAT metabolic activity18. Adipose tissue can be differentiated from other tissue and air based on density information provided by the CT image in Hounsfield units (HU). PET images show the amount of 18F-FDG taken up into a volume of tissue in units of standardized uptake values (SUV). Active BAT can be separated from tissue with insignificant tracer uptake, including WAT and inactive BAT, by co-registering PET images with corresponding CT scans and choosing an appropriate SUV threshold.
Through this paper, we aim to provide a step-by-step approach with an instructional video that can be used by clinical researchers to quantify human BAT using 18F-FDG PET/CT scans. This image analysis technique is ideally used after subject(s) have been exposed to cold or treated with pharmacological BAT stimulants. Specifically, we demonstrate to users on how to construct regions of interest (ROIs) while minimizing false positives using a free, open-source image-processing software (ImageJ) with a specific plug-in (petctviewer.org). The result of this approach can be used to study BAT volume, activity (glucose uptake), and anatomical distribution in individual study subjects.
All PET/CT images shown in this manuscript were obtained from participants in National Institutes of Health protocol no. 12-DK-0097 (ClinicalTrials.gov identifier NCT01568671). All participants provided written informed consent, and all experiments were approved by the Institutional Review Board of the National Institute of Diabetes and Digestive and Kidney Diseases.
1. Software Installation
2. Loading PET/CT Images
3. Navigating the PET/CT Viewer Plug-in
4. Quantifying whole-body BAT
5. Quality Assurance
6. Segmenting BAT into Individual Depots
Note: The following section is focused only on quantifying regional depots of BAT17. The steps are not necessary to obtain whole body BAT volume and activity.
BAT is quantified through a series of post image-acquisition processing steps as shown in Figure 1. PET and CT thresholds are used to identify voxels that are metabolically active and have the density of adipose tissue. However, some voxels meeting these criteria can occur in anatomical locations not likely to contain BAT. To avoid these false positives, PET, CT, and anatomical information must all be considered when drawing ROIs (Figure 2). Several common regions to include and avoid when quantifying whole body BAT in cold-stimulated subjects are shown in Figure 2, such as metabolically active cervical BAT vs. salivary glands, vocal chords, and thyroid (Figure 2A and 2B); supraclavicular BAT vs. shivering muscle near borders of air and solid tissue (e.g. intercostal muscles) (Figure 2C); and abdominal BAT vs. the calyces of the kidneys as they clear labeled glucose (Figure 2D). After the ROI's of each axial slice are compiled, BAT depots can be segmented in the sagittal plane to examine intra-/inter-individual differences in regional BAT activation (Figure 3).
Figure 1. Schematic Flow of the Image Processing Steps. First, PET images and corresponding CT images are uploaded into the PET/CT plug-in (A). After axial ROIs are drawn on each PET/CT slice (B), each voxel meeting both PET and CT criteria are identified in blue (C). A mask is generated from these BAT-identified voxels (D), which is substituted for the original corrected PET scan (E), and depots are segmented in the sagittal view (F). Please click here to view a larger version of this figure.
Figure 2. Axial BAT Region-of-Interest Selection and Common Areas To Avoid in Multiple BAT Depots. Axial slices from a fused PET/CT image (columns 1 and 2) and a maximal intensity projection image (MIP, column 3) with green lines to denote slice height from a scan acquired following cold-stimulation. Green ROIs are drawnaround areas with adipose tissue density, high FDG uptake, and anatomical locations likely to contain active BAT in columns 1 and 2. Anatomical areas unlikely to contain BAT are highlighted in red in column 2. Voxels meeting the BAT PET and CT criteria are confirmed by ImageJ and highlighted in blue. Examples are taken from the (A) anterior cervical depot, (B) cervical depot at the level of the thyroid, (C) Supraclavicular/Axillary depot nearby shivering skeletal muscle (i.e., Intercostals), and (D) the Abdominal depot at the level of the ureters of the kidneys. Please click here to view a larger version of this figure.
Figure 3. Regional Segmentation of Seven BAT Depots in the Sagittal View. Following the generation of a "BAT mask" image containing only PET voxels previously identified as active BAT, the following regions can be separated with ROIs drawn in the sagittal plane: (A) Cervical (C3-C7), (B) Supraclavicular (C7-T3, excluding vertebrae), (C) Axillary (T3-T7, excluding vertebrae), (D) Mediastinal (anterior mediastinum), (E) Paraspinal (T1-T12, from the anterior edge of the vertebrae to the spinous processes), (F) Abdominal (T12-L3, retroperitoneal), and (G) Dorsocervical (fat depot distinct and posterior to the paraspinal depot; near the cervical region). The composite image with all regions appears in (H). Please click here to view a larger version of this figure.
Since confirmation of functional BAT in adult humans, there has been great interest in understanding the role of BAT in human physiology. However, because this thermogenic tissue is often found in narrow fascial planes, interspersed within white fat, and surrounding other organs, it is challenging to quantify. In 2016, a consensus document was published by an International BAT expert panel with recommendations for reporting relevant participant characteristics, criteria for subject preparation, and a protocol for acquiring PET/CT images21. The panel also identified the need for more consistency in the processing of PET/CT for BAT quantification, noting that methods to identify BAT have varied widely and, in most cases, only limited detail of the BAT quantification procedure is provided. Consequently, while reports of within study reproducibility are high22,23,24, appreciably different BAT volume and activity has been reported by groups using different quantification methods, even when participants are of similar age, sex, and BMI25,26. These inconsistencies make comparing findings difficult, and have led to a controversy over the amount of BAT in the adult human15.
An inherent limitation of PET/CT image processing is the inclusion of voxels that meet both PET and CT criteria but are in anatomical locations that correspond to structures other than BAT. Perfect co-registration of PET and CT images is nearly impossible due to differences in resolution and subject motion during scans. As a consequence, structures bordering air or bone and regions of high tracer uptake are often incorrectly identified as active BAT. To limit inclusion of false positive voxels, one should apply PET and CT criteria only within the ROIs that users construct. But current approaches to quantify BAT with user-specified ROIs or automated analyses differ in the amount of user involvement and knowledge they require. We have shown that using a single, two-dimensional user-defined coronal ROI applied to the entire stack of images may be more prone to including false positive areas19. Several groups have developed automated methods to quantify BAT that are capable of rapidly processing large datasets without much user input. However, these methods either fail to include all potential BAT-containing regions, particularly in the lower body27, or incur relatively high rates of false positives28 and false negatives26. Since the volume of human BAT is generally low (<600 mL, or <2% of total body mass), small absolute errors in quantification may lead to large relative differences.
The more rigorous approach described by this study of drawing ROIs on each axial PET-CT slice allows the detection of BAT in narrow fascial layers while providing more confidence that false positives have been excluded. This yields a detailed quantification in each individial, rather than a binary assessment of BAT's presence or absence29. Therefore, it may be more suitable for controlled experiments in small sample sizes intending to study BAT physiology and/or effects from interventions. Furthermore, the ability to define region-specific BAT depots may provide more insight into BAT's functional relevance and developmental origin. We believe these quantitative measures are important not only for comparison across the field, but also to better estimate BAT's contribution to energy metabolism and thermoregulation in adult humans.
Several anatomical features of BAT will help users of our method limit inclusion of false positive voxels. BAT is typically found in continuous and symmetric fascial layers. Thus, while drawing and refining an ROI, examining the superior and inferior axial slices for continuity and symmetry of the selected adipose tissue can help users maximize inclusion of adipose tissue while minimizing inclusion of skeletal muscle, bone, and other obvious non-BAT structures. Active BAT is also rarely present in subcutaneous adipose depots, so we advise users to avoid these areas when constructing ROIs. As noted in the protocol, BAT is distributed in several distinct anatomical regions, including the cervical, dorsocervical, supraclavicular, axillary, mediastinal, paraspinal, and abdominal depots. These depots are distributed such that one axial slice may contain more than BAT from multiple depots. For instance, an axial slice in the thoracic region can contain BAT from the mediastinal depot (proximal and anterior), paraspinal depot (proximal and posterior, along the spine), and axillary depot (lateral and near the mid-antero-posterior line). Knowledge of these depots can help users create ROIs in the various regions of the body, since they occur in pre-described locations are largely contiguous, as described in our protocol. However, because we encourage users to draw only one ROI per slice to avoid ROI overlap, the additional steps of generating a BAT mask and drawing sagittal ROIs is required to separate the previously-identified BAT voxels into the distinct regional depots if information of BAT distribution is desired, i.e., separating mediastinal, paraspinal, and axillary BAT detected in the same axial ROI into depots based on sagital location (Figure 3).
The PET/CT viewer software can also be used to quantify the activity of tissues other than BAT, for instance shivering skeletal muscle, which also major plays a role cold induced thermogenesis19, or various areas of the brain or liver that have been suggested as reference tissues for PET/CT analysis21. However, these tissues will have densities and anatomical distributions that differ from BAT and are outside the focus of our current protocol. We direct readers to the consensus document for greater detail on these subjects21. Finally, we advise all users to continually update ImageJ and visit petctviewer.org for Plug-in updates and software assistance.
Though we believe that this rigorous method is more precise than automated methods26,28 and methods that use a simplified, single ROI to estimate total BAT volume9,30, it is not without limitations. There is no ideal method to non-invasively quantify BAT in humans, and 18F-FDG represents only glucose uptake, which is not the same as glucose metabolism11. However, even though other radioactive tracers have been used31,32,33, 18F-FDG is the most prominent tracer used to study human BAT. Thus, developing standardized methods to analyze 18F-FDG PET/CT images will continue to be impactful in the study of human BAT physiology for the foreseeable future.
The method we propose, creating an ROI on each BAT-containing axial slice while avoiding common problem areas, is labor intensive and requires the user to have some knowledge of underlying anatomy. It is also possible that the stringent ROI selection may introduce false negatives, since some BAT-containing depots may be avoided. Drawing ROIs on every axial slice of the fused PET/CT image allows for careful discrimination between adipose tissue and neighboring metabolically active tissues and/or regions impacted by spill over and partial volume effects34. However, the time it takes to complete analysis of a single scan can range from three to eight hours, with the possibility of shortening the time frame with practice and experience. Various machine learning approaches may be able to reduce the labor and expertise required to accomplish this task. However, creating a more automated method that can accurately detect BAT and is robust to false positives created by current imaging limitations will require a large dataset with individuals of varied body composition and BAT distribution. We hope that this method can be used to produce a detailed BAT atlas that may serve as a template for more sophisticated big data approaches.
In conclusion, we demonstrated a step-by-step image analysis approach to quantify human brown adipose tissue volume, activity, and distribution using cold-induced FDG PET/CT scans. The critical steps include 1) continuously and sequentially analyzing axial ROIs and 2) assessing relevant BAT depots by their anatomical location while avoiding other metabolically active tissues. This rigorous quantification approach can be used by investigators in the field to study BAT physiology and serve as reference standard for developing automated human BAT quantitation approaches in the future.
We would like to thank all of the study volunteers, nursing and clinical staff, and the dieticians of the NIH Clinical Center for their participation in our cold exposure studies and care provided during the inpatient stays. We'd also like to thank Dr. Bill Dieckmann for all of his assistance with the acquisition and distribution of the PET-CT images for our studies. This work was supported by Intramural Research Program of the National Institute of Diabetes and Digestive and Kidney Diseases Grants Z01 DK071014 (to K.Y.C.) and DK075116-02 (to A.M.C.).
Name | Company | Catalog Number | Comments |
ImageJ/Fiji Software | ImageJ | https://imagej.net/Fiji/Downloads | Open Source Software |
This article has been published
Video Coming Soon
ABOUT JoVE
Copyright © 2024 MyJoVE Corporation. All rights reserved