Quantitative analysis of skeletal muscle by computed tomography imaging—State of the art

Abstract

The radiological assessment of muscle properties—size, mass, density (also termed radiodensity), composition, and adipose tissue infiltration—is fundamental in muscle diseases. More recently, it also became obvious that muscle atrophy, also termed muscle wasting, is caused by or associated with many other diseases or conditions, such as inactivity, malnutrition, chronic obstructive pulmonary disorder, cancer-associated cachexia, diabetes, renal and cardiac failure, and sarcopenia and even potentially with osteoporotic hip fracture. Several techniques have been developed to quantify muscle morphology and function. This review is dedicated to quantitative computed tomography (CT) of skeletal muscle and only includes a brief comparison with magnetic resonance imaging. Strengths and limitations of CT techniques are discussed in detail, including CT scanner calibration, acquisition and reconstruction protocols, and the various quantitative parameters that can be measured with CT, starting from simple volume measures to advanced parameters describing the adipose tissue distribution within muscle. Finally, the use of CT in sarcopenia and cachexia and the relevance of muscle parameters for the assessment of osteoporotic fracture illustrate the application of CT in two emerging areas of medical interest.

Introduction

The quantitative assessment of muscle properties is of increasing interest in musculoskeletal research and routine. Historically, computed tomography (CT) and magnetic resonance imaging (MRI) have been used in the diagnosis and monitoring of muscle diseases such as myopathies and muscular dystrophies [1], [2], [3]. Muscle imaging has also been used to assess muscle atrophy or muscle wasting caused by or associated with inactivity, denervation, fasting, malnutrition, chronic obstructive pulmonary disorder, cancer-associated cachexia, diabetes, renal failure, cardiac failure, Cushing syndrome, sepsis, burns, and trauma [4]. More recently, the paradigm of the bone–muscle unit [5], [6] has focussed musculoskeletal research on quantitative muscle assessments such as muscle volume, CT muscle density and content, and distribution of adipose tissue. A new field of application is sarcopenia, where muscle function and perhaps muscle mass are important parameters to measure [7], [8].

In skeletal muscle, lipids are either stored as adipocytes, with fat imaging characteristics, either in between muscle groups (perimuscular adipose tissue), or as interstitial (intramuscular) adipose tissue inside muscles (extramyocellular lipids) or as intramyocellular lipid (IMCL) droplets, which are not visually identified as fat in CT or MRI but modify the imaging appearance of muscle tissue [9]. The combination of intramuscular and perimuscular adipose tissue is typically denoted as intermuscular adipose tissue (IMAT) (Figure 1). Increased amounts of intermuscular adipose tissue correlate with cardiovascular risk [10], while the presence of IMCL is a risk factor for of insulin resistance [11]. Increased baseline extramyocellular lipid in isolated supraspinatus muscle tears is associated with surgical repair failure at 6-month follow-up [12].

Figure 1.

Muscle–lipid system illustrated for the thigh. The blue contour denotes the fascia that separates subcutaneous tissue from the intrafascia region. The green contour outlines a single muscle also denoted as a muscle organ in this contribution to distinguish it from muscle tissue that includes intramyocellular lipids but excludes intramuscular AT (red), which is composed of extramyocellular lipids. Intermuscular AT is the combination of perimuscular AT (yellow) and intramuscular AT (red).

AT = adipose tissue.

The two standard techniques available for body composition parameters quantification are dual X-ray absorptiometry (DXA), which provides lean and fat mass assessment, and bioelectrical impedance, which estimates fat-free and fat masses. However, both bioelectrical impedance and DXA cannot provide a spatially resolved distribution of muscle and adipose tissue. This is the domain of CT and MRI. While this review is focussed on CT, a brief comparison between MRI and CT may help the reader to put both techniques into perspective.

MRI provides better soft tissue contrast than CT (Figure 2), but standard spin-echo T1-weighted sequences only provide a qualitative assessment of fat, which appears white, compared with muscle, which in this sequence is dark. The extend of larger agglomerations of adipose tissue can be measured, but the true fat content of muscle cannot be determined from T1-weighted images because the grey values of the muscle voxels do not scale in a known way with the fat content. MR spectroscopy (MRS) and Dixon sequences [13], [14] are MR techniques originally developed to quantify the fat content of the liver, which, from the perspective of MRI, is a very homogenous organ. MRS is the only imaging method that allows for a detailed analysis of the acquired MR spectrum and, more specifically, for the separation of intracellular and extracellular lipids [15], [16]. However, with MRS, only a very small volume of interest, a so-called spectroscopy voxel with a volume about 1 cm³, is analysed. This works well for the liver but for muscle, such a small volume may not be representative of the overall fat distribution, especially in elderly and diseased individuals with a high and inhomogeneous muscle fat infiltration. A solution could be a multivoxel MRS protocol [17], but applications in muscle have rarely been reported.

Figure 2.

T1-weighted MRI (left) and a low radiation exposure CT (right) image of the thigh. The CT image contains an in-scan calibration phantom to obtain a scan with specific CT values of water. The slice position does not match exactly, and the slice thickness differs.

MRI = magnetic resonance imaging; CT = computed tomography.

MR Dixon sequences provide a scalable map of the fat fraction from 1 to 100%, in which a grey value of 1 corresponds to a fat fraction of 0.1% and a grey value of 1000 corresponds to a fat fraction of 100%. A quantitative map of the muscle fat fraction can be obtained once the muscle has been segmented. Several variants of Dixon sequences exist. However, accuracy errors of the muscle fat faction may be rather large and depend on the specific implementation on a given scanner. Also, many implementations are dedicated to liver but not to muscle imaging [18], [19].

In contrast to MRI, CT is faster, more widely accessible, and cheaper. Radiation exposure of arms or legs is low (<0.5 mSv), while volumetric muscle measurements of the trunk are associated with higher exposure. For example, using a low-dose protocol, exposure for a single slice scan at the height of L3 is about 0.1–0.2 mSv depending on the slice thickness and about 2 mSv for a volumetric scan at the level of L1-4, which is still rather moderate when compared with an annual background radiation of about 2.5 mSv. One advantage of CT is the possibility to quantify the muscle density also termed as muscle attenuation or muscle radiation attenuation, which linearly depends on the muscle fat content. The muscle volume may be measured either with MRI or CT. However, the spatial resolution of CT images is higher. Typical MR parameters are a slice thickness of 3 mm and an in-plane pixel size of 0.5 mm, whereas for CT, a slice thickness of 1 mm and an in-plane pixel size of 300 microns are state of the art.

In this contribution, we will first review CT imaging protocols including methods for muscle density calibration and segmentation methods used in quantitative muscle research. An overview of basic and advanced parameters that can be quantified in a CT image and their potential limitations will be given. Finally, the use of muscle parameters in clinical routine and research will be discussed.

Methods of quantitative CT imaging

CT imaging techniques

Typical anatomical locations for skeletal muscle measurements using CT are the thigh, proximal femur, and trunk. CT acquisition and reconstruction parameters vary widely across studies. Most CT muscle studies are performed on whole-body clinical CT scanners. A spiral CT scan or single slices of the targeted volume are acquired. These scanners are widely available, and protocols can easily be tailored to the specific location. Dedicated peripheral CT scanners have been used in some studies to image muscle and adipose tissue of arms or legs [20], [21]. As these scanners lack flexibility and scan times are high, this review is restricted to the whole-body equipment.

Although CT studies on muscle assessment appeared not long after the introduction of the first CT scanner in the mid-1970s, little has been published on acquisition protocols for muscle measurements. For the thigh and upper femur, the following settings have been reported: tube voltage of 100–120 kV, exposure: 40–200 mAs, slice thickness 1–10 mm and a soft to medium reconstruction kernel [22], [23], [24], [25]. Thicker slices reduce noise and are adequate when the muscle volume or density is of primary interest. For a more advanced analysis of the adipose tissue distribution, a higher spatial resolution is required, and a slice thickness of ≤2 mm is preferable. In the thigh and proximal femur, scans usually cover 10–20 cm, but in some studies, the complete upper or lower limb was scanned [26], [27], while in others, a single 10-mm-thick slice was acquired at mid-thigh [22], [28]. Several studies applied a retrospective analysis of existing CT scans, often initially collected in epidemiological or cross-sectional studies, to measure the bone mineral density (BMD) and assess fracture risk [25], [28], [29], [30].

Studies dedicated to the assessment of the paraspinal muscles typically assessed the complete lumbar spine, acquiring either a spiral scan or single 3- to 5-mm-thick slices at the level of the intervertebral disks [31], [32], [33], [34], [35]. Another application of CT is the evaluation of cachexia in patients with cancer, where the volume of the subcutaneous and visceral fat together with the of the abdominal and paraspinal muscles is measured. For this purpose, analysis is often limited to a single 5- to 10-mm-thick CT slice at the level of L3 [36], [37], [38], [39]. In many other instances, existing CT scans from clinical routine are analysed.

Muscle analysis and quantitative parameters

The most straightforward quantitative measurements are the muscle area from a single slice or muscle volume from a stack of slices covering a whole muscle. These measurements require a prior segmentation. However, muscle segmentation is a difficult step, due to the low soft tissue contrast, especially at the trunk where muscles and surrounding organs have similar CT values. Therefore, the most simplistic segmentation approach is a manual outline of the muscles [23], [34], [40], an acceptable technique as long as the number of slices to be processed is small. The advantage of a manual segmentation is the use of expert knowledge to outline the muscle border, although different experts will never fully agree on a specific segmentation. However, when a larger stack of CT images needs to be segmented, as with spiral scanning, a more automated segmentation approach is required.

Most of the literature on muscle segmentation is dedicated to MRI, whereas reports using CT are sparse [25], [27], [41], [42], [43], [44], [45], [46], [47]. In several clinical studies and research projects, commercially or publically available software tools such as Slice-O-Matic (http://www.tomovision.com/index.html), OsiriX (http://www.osirix-viewer.com), or ImageJ (https://imagej.nih.gov/ij/) have been used. OsiriX and ImageJ do not provide dedicated muscle segmentation algorithms, but standard segmentation tools such as thresholding, contouring regions, or volumes of similar signal intensity that define muscle, fat, or organs in combination with morphological operations can be easily combined with corrective operator interaction. Three studies [48], [49], [50], in which the three software tools were compared, reported excellent agreement between paraspinal muscle of cross-sectional area (CSA) measurements of abdominal and paraspinal muscles, despite the use of a simple threshold or even fully manual segmentation.

In the thigh, usually first the outer margin, i.e., the skin, then the fascia, and finally the individual muscles within the fascia are successively segmented [25]. In the trunk, statistical shape or atlas-based approaches [41], [51] have been proposed to facilitate the segmentation of abdominal and paraspinal muscles, which is complicated by the presence of organs and visceral fat. Atlas-based approaches, which require prior training with a gold standard, seem to be the method of choice to separate individual muscles, for example, the quadratus lumborum, multifidus, and sacrospinalis within the paraspinal muscle group. Similar segmentation techniques have been suggested for the hip and thigh muscles [27].

Analysis of entire muscles requires scanning of a largevolume. However, most often, only a small range (several centimetres) of a muscle is scanned to reduce the radiation dose. In this case, area and volume measurements need to be normalized, for example, to the square of the body height [24], [39], [52], [53], to be representative regarding fatty infiltration assessment. The trunk muscle area at the L4-5 level or 5 cm above is highly correlated with the total body skeletal muscle volume (r² > 0.8) [54]. In the appendicular skeleton, radiation exposure is much lower than that in the trunk and larger volumes can be scanned.

The assessment of the muscle density requires a calibration step but not necessarily muscle segmentation. Representative muscle density values may be obtained by geometrically defined volumes of interest (VOIs) placed in the muscle of interest such as a circle for a slice-based analysis or a sphere or cylinder for a volume-based analysis. A detailed overview of studies in which a geometrical VOI was used to determine the muscle density has been provided in Ref. [55]. So far, in the vast majority of studies measuring the muscle density, muscle was first segmented and the muscle area was obtained as a primary outcome.

It must be understood that the muscle density is not a physical density measured in mg/cm³, but a ‘CT density’ measured in Hounsfield units. The term muscle attenuation is used synonymously [56], but the term CT density will be used in the present contribution. By default, each CT scanner is calibrated to water:

$$\text{CT}\text{number}=\frac{\mu -{\mu }_w}{{\mu }_w}·1000\left[\text{HU}\right]$$

where μ is the linear absorption coefficient of the material under investigation, and μ_w is the attenuation coefficient of distilled water at room temperature. In a calibrated scanner, the CT numbers for water and air are 0 and -1000, respectively. For water-equivalent materials, i.e., for materials with comparable mass absorption coefficients such as muscle, CT values quantify percentage density differences to water. A CT value of 1 HU quantifies a density difference to water of 0.1%. The physical density of skeletal muscle, including IMCLs, is about 1.055 g/cm³ [57], [58], [59] compared with 1 g/cm³ for water. Thus, for a properly calibrated scanner, the CT value of skeletal muscle also termed as muscle tissue is about 50–60 HU. Published values of IMCL content in human skeletal muscle as measured by MRS are between 6% and 14% [60] compared with 2%–9% for chemical analysis [57] and 2%–5% for histochemical analysis [61]. Additional infiltration of extracellular lipids—or adipose tissue within muscle—will decrease the muscle density value below 50 HU.

Owing to CT scanner instabilities, the water value often deviates from 0 HU. Figure 3 shows CT values measured in the European Spine Phantom (ESP), which was scanned on 41 scanner units. The ESP is a reference standard used in bone densitometry, in which soft tissue is made of a water-equivalent plastic. Each scan was acquired with the same phantom on a different CT scanner, and 17 different CT scanner models from four manufacturers were tested. All scanners were regularly calibrated according to the local hospital and manufacturer standard operating procedures. In the ESP, a cylindrical VOI positioned in the water-equivalent soft tissue region anterior to the vertebrae was analysed. In 15% of the scanner units (6 of 41), CT values differed by 5 or more HU from 0. In the thigh, a difference of 5 HU in the water value causes a 15% difference in the muscle density [25]. Thus, for accurate muscle density measurements the use of a calibration phantom should be considered. This phantom can either be scanned with the subject (simultaneous calibration). An example is shown in Figure 2. Alternatively the phantom can be scanned before or after the patient (asynchronous calibration).

Figure 3.

CT values of the water-equivalent soft tissue material of the same European Spine Phantom measured anteriorly to the vertebrae from 41 different CT scanners.

CT = computed tomography.

CT values are used not only to determine the muscle density but also to segment muscle and adipose tissue. Furthermore, they play a central role in estimating the adipose tissue content within muscle. Table 1 lists the published CT values used for these three different applications and the terminology used in the corresponding reference for muscle and adipose tissue parameters. It shows that even for a given application, for example, to determine the tissue composition, CT values vary significantly, which has also been noted in an earlier review [55]. One source of confusion in comparing CT values of muscle and adipose tissue among previous works is the great variability in terminology. For example, the term muscle is typically used to denote the organ muscle, which includes adipose tissue, whereas the term muscle tissue includes intramyocellular lipids (ICML) but excludes adipose tissue. Thus, the CT value of muscle will vary considerably depending on its adipose tissue content. Some authors, however, use muscle tissue synonymously with muscle [24]. CT values for muscle tissue are higher than those of muscle. Pure muscle and high-density muscle have been used to denote the muscle density in young healthy individuals [23], [25], although it can be speculated that some amount of intermuscular and intramuscular adipose tissue is present even in young healthy individuals, and therefore, that the term pure muscle may not be fully appropriate. However, the average muscle density of young people may serve as an upper fix point for muscle density calibration using CT. The terminology is even more diverse for fat/adipose tissue.

Table 1.

CT values used for segmentation, tissue composition, and the measurement of adipose tissue infiltration of muscle.

Reference	Terminology	CT value (HU)	Comment
Segmentation
Mitsiopoulos 1998 [109]	Skeletal muscle	−29 to 150
Mitsiopoulos 1998 [109], Irving 2007 [49], Mühlberg 2017 [25]	Adipose tissue	−190 to −30
Lönn 1994 [67]	Adipose tissue	−190 to −30
	Muscle incl. skin/visceral organs	−29 to 151
Fuchs 2018 [110]	Muscle density	−29 to 150
Edmunds 2016 [26]	Fat	−200 to 10
	LDM	−9 to 40
	NDM	41 to 70
Van der Werf 2018 [24]	Muscle tissue	−29 to 150
Neander 1997 [111]	Muscle	10 to 100
Strandberg 2010 [22]	Muscle	1 to 101
Tissue composition
Goodpaster 2000 [101]	Skeletal muscle	0 to 100
Margadant 2016 [38]	Muscle tissue	60
Mourtzakis 2008 [39]	Skeletal muscle	−29 to 150
	Subcutaneous and intramuscular adipose tissue	−190 to −30
	Visceral adipose tissue	−150 to −50
Heymsfield 1997 [112]	Adipose tissue	−90 to −30
	Nonadipose tissue lean	−29 to 151
Muscle fat infiltration
Goodpaster 2000 [101]	LDM	0 to 29
	NDM	30 to 100
Lang 2008 [29], Frank-Wilson 2018 [62]	Fat, lean		HU thresholds to separate lean and fat mass derived from in-scan calibration phantom
Gargiulo 2014 [113]	Fat	−200 to −6
	Loose connective tissue	−5 to 20
	LDM	20 to 40
	Normal muscle	40 to 200
Daguet 2011 [23]	Pure fat	−108	Internal calibration – . fat
	Pure muscle	60	Internal calibration – psoas muscle of young healthy individuals
Mühlberg 2017 [25]	Pure adipose tissue	≈ −100	Internal calibration – thigh subcut. adipose tissue
	HDM	35	Internal calibration – psoas muscle of young healthy individuals
Hahn 2016 [106]	Muscle	30 to 80
	Fat	−190 to −30

HDM = high-density muscle; LDM = low-density muscle; NDM = normal-density muscle; subcut. = subcutaneous.

The CT value range from −29 to 150 originally used by Mitsiopoulos and even earlier by Lönn has been used in many subsequent studies as can be seen in the table.

Differences in CT values shown in Table 1 are probably not very relevant when used as thresholds for segmentation of muscle, subcutaneous, or visceral compartments. Automated techniques may use fixed thresholds to initiate the segmentation but are usually refined without using absolute thresholds. In semiautomatic procedures, initial threshold-based contours guide the operator and are corrected if required. In contrast to segmentation, differences in CT values have a much larger impact on the quantification of the adipose tissue content within a single muscle or within a muscle group contained by the perimuscular deep fascia such as the paraspinal muscles or the whole thigh muscles. Such an entity is often analysed as a whole to avoid a rather difficult segmentation of individual muscles. Therefore, the separation of intermuscular and intramuscular adipose tissue components is of interest. Obviously, the use of absolute thresholds is problematic. They are not consistent across the literature (Table 1) and do not take into account deviations from the ideal water calibration of the CT scanner. Therefore, some studies have suggested techniques based on in-scan calibration phantoms [29], [62], internal calibration [23], or both [25].

For a more detailed characterisation of the adipose tissue infiltration, a frequency analysis (Figure 4) of the density values [23], [55] and of its spatial distribution has been suggested [25]. Based on the histogram of the CT values of the segmented muscle, five different bins were distinguished [23]: pure muscle, pure fat and voxels containing 25%, 50%, and 75% adipose tissue according to the muscle density, respectively. This is the quantitative analogue to the Goutallier score [63] used to qualitatively score fatty muscle degeneration.

Figure 4.

Histogram of density values on a CT slice of the left glutaeus maximus muscle (inset). Red pixels, with a density between 130 and 190 Hounsfield units (HU), were classified as pure muscle. Deep navy blue pixels, with a density of ≤75 HU, were classified as pure fat. Between the −75 and 130 HU boundaries, brown pixels were counted as being composed of 75% muscle and 25% fat, blue–green pixels as 50% muscle and 50% fat, and medium blue pixels as 25% muscle and 75% fat (Figure 2 from Ref. [23]). The adipose tissue content of the individual voxels of the glutaeus maximus is not presented in colour.

CT = computed tomography.

The histogram analysis requires two values, one for pure adipose tissue and one for pure muscle. In Ref. [25], the value for adipose tissue was obtained by internal calibration based on the distribution of subcutaneous adipose tissue corrected for potential inclusion of water, oedema, and vessels. The value for pure muscle was obtained from a group of young active athletes. For each CT scan, this high-density muscle value was individually corrected for potential water calibration offsets using an in-scan calibration phantom [25]. If the CT scanner is stable, the in-scan calibration phantom can be replaced by asynchronous calibration, which means that phantom and subject scans are typically not performed at the same time [64].

A further step is the quantification of the spatial distribution of adipose tissue infiltration and the characterisation of muscle structures termed as washed out or moth eaten (Figure 5) in radiological scoring systems [65]. For this purpose, we have recently used a number of different descriptors that quantify the texture [25]. They can either be applied to the CT images directly or after image binarisation. While details are beyond the scope of this review, Figure 7 illustrates the principle, assuming four different muscle–fat compositions, where for simplicity, muscle in red has a value of 1 and fat in white has a value of 0, corresponding to a binary image. Texture parameters may help to better distinguish different structural patterns than CT density (or in case of the simplified simulation, relative volume).

Figure 5.

The Mercuri score for grading the muscle appearance in neuromuscular diseases [65] and two CT images demonstrating moth-eaten and washed-out muscle structures.

CT = computed tomography.

Figure 7.

Four simulated binary muscle (red)–fat (white) distributions where muscle has a value of 1 and fat, of 0. The table shows the results of three different texture parameters plus relative volume. It shows that, for example, the relative volume differs only slightly between the two leftmost structures, while local inhomogeneity discriminates much better. By default, a periodic continuation of the 5 × 5 pattern was assumed for calculations except when marked with an asterisk (*), where zero boundary conditions were used, i.e., all pixels outside the pattern were white; values marked (^$) appeared independent of the boundary type used.

Muscle mass in gram is not a primary parameter of CT. It has been estimated by multiplying the muscle volume by a density of 1.06 g/cm³ [66], [67] or 1.04 [68]. Accuracy of this approach is questionable as the degree of muscle fat infiltration is neglected. A better approach would be to calculate the physical density of each voxel from its CT number, which as described previously, denotes the percentage density difference from water. Nevertheless, correlation coefficients of r² ≥ 0.96 have been reported in healthy volunteers [68], [69] between muscle mass in the thigh estimated by CT, even with the simplified technique, and fat-free mass (FFM) measured by DXA. Correlation between the muscle area measured by CT and FFM was r² = 0.74. Another study of premenopausal lean and obese women reported correlations between the CT muscle area and DXA FFM at the thigh of r² = 0.59 and r² = 0.58, respectively [70].

Applications of quantitative muscle imaging by CT

Several recent systematic reviews and meta-analyses impressively demonstrated that muscle imaging by quantitative CT plays an important role in a large variety of diseases affecting organs such as the lung [71], liver [72], [73], or abdomen [74], [75]. Further systematic reviews were dedicated to muscle characteristics in cerebral palsy [76], vascular claudication [77], low back pain [78], and muscle fatigue in old age [79]. CT imaging is used to assess muscle diseases, age-related muscle deterioration (sarcopenia) and muscle wasting (cachexia), other chronic diseases associated with or causing muscle weakness, after organ transplantation specifically and surgery in general; in body composition studies to assess efficacy of exercise or nutritional intervention; and to predict the risk of osteoporotic fracture or the impact of fracture on muscle. Obviously, it is beyond the scope of this review to cover all these applications in relevant details, in particular, as other imaging techniques—radiography, MRI, and DXA—are often used as well. Therefore, the following two sections have been selected as example applications.

Sarcopenia and cachexia

“Cachexia and sarcopenia present several analogies in both the pathogenic mechanisms and the clinical picture. The loss of muscle mass and strength is a hallmark of these two clinical conditions. Although frequently overlapping and often indistinguishable, especially in old individuals, these two conditions should be considered as distinct clinical entities” [80]. The first definition of sarcopenia proposed in 1988 [81] was completely based on appendicular lean mass, determined from a whole-body DXA scan. Later definitions of sarcopenia included impaired muscle strength and/or physical performance [82], [83], [84], [85]. None of these definitions is based on CT, although specific thresholds of the muscle size assessed by CT have been proposed for the diagnosis of sarcopenia [86]. Yet, the controversy on which imaging modality should be preferred is still ongoing [87], [88]. One limitation to the use of CT is its lower degree of standardisation when compared with DXA, but also controversies concerning which of the muscle parameters assessed by CT should be used. DXA only provides lean mass, and in sarcopenia research, more studies used DXA than CT. Similarly to the use of DXA for routine BMD measurements, DXA may be preferred for sarcopenia in daily clinical routine. However, CT and MRI provide some muscle measurements, such as fatty infiltration, an important factor of muscle degeneration, which are not available for DXA and are key techniques to address the pathophysiology of sarcopenia and mechanisms of interventions. Several epidemiological studies such as Osteoporotic Fractures in Men, AGES-Reykjavik, and Health, Aging, and Body Composition (Health ABC) included CT of the proximal femur and upper shaft, a valuable source for sarcopenia research largely unexplored so far. Also existing CT scans obtained for other diagnostic purposes may be retrospectively analysed to obtain muscle characteristics [89], [90], [91].

Cachexia has been defined as “a complex metabolic syndrome associated with underlying illness and characterised by the loss of muscle mass with or without the loss of fat mass”. A quick search for cachexia and cancer in PubMed resulted in 1250 entries. The prominent feature of cachexia is “weight loss in adults” [92]. In addition, a more specific definition has been agreed on for cancer cachexia [93]. For patients with cancer, CT imaging is a key diagnosis tool for follow-up diagnosis and treatment monitoring [39]. The potential for opportunistic muscle imaging in such patients has already been emphasised a decade ago [36]. A typical imaging technique in cancer cachexia is the analysis of a single CT at the level of L3. In addition to the muscle volume, density and intramuscular adipose tissue, the volume of subcutaneous and visceral fat volume is determined (Figure 6). However, longitudinal measurements of visceral fat on a single slice are problematic because of motion, in particular that affecting the visceral organs, which can significantly change the local amount of visceral fat.

Figure 6.

The left CT image at the level of L3; the right segmentation: subcutaneous adipose tissue (blue), paraspinal muscle (brown), abdominal muscle (orange), intramuscular adipose tissue (green), visceral fat (yellow), and visceral organs (grey).

CT = computed tomography.

Osteoporotic fracture

BMD is an important risk factor for osteoporotic fracture, but the prediction of osteoporotic fracture remains challenging [94]. In particular, progress remains limited for the prediction whether an individual will sustain a hip fracture within the next 5 or 10 years. In contrast to vertebral fractures, almost all osteoporotic hip fractures are caused by a fall, which is associated with muscle strength [95], [96], [97], [98]. Age-related loss of skeletal muscle mass is associated with the loss of bone mass [99], [100], and low muscle attenuation is associated with obesity and impaired muscle function [101]. So far, only a few studies have investigated the impact of soft tissue parameters on hip fracture. In the Health ABC study, decreases in DXA-derived subcutaneous fat thickness significantly increased hip fracture risk after adjustment for hip areal BMD (aBMD) in men [hazard ratio (HR) 1.44 confidence interval (CI; 1.02–2.02)] and women [HR 1.39 (CI 1.07; 1.82)]. An increase in appendicular lean mass in the leg decreased hip fracture risk in men (HR 0.65 [CI 0.46; 0.92)] but not in women, even after adjustment for the body height [102]. The association between lean mass and fracture was confirmed in the Framingham (leg lean mass) [103], Epidémiologie de l’Ostéoporose (whole-body lean mass) [104], and (WHI; appendicular lean mass) [105] studies, all of which included women only. After adjustment for hip aBMD, fat mass did not significantly increase hip fracture risk in the Epidémiologie de l’Ostéoporose (whole-body fat mass) and WHI (appendicular fat mass) studies. It was not separately assessed in the Framingham study.

In contrast, CT studies showed a significant contribution of muscle characteristics to hip fracture risk. In the pelvis [29], the CSA of total fat and of the extensor and adductor muscles, as well as CT attenuation of the adductor muscles remained significantly lower in fractured individuals after adjustment for age and body mass index. Compared with DXA aBMD alone, fracture discrimination was improved by the inclusion of the soft tissue descriptors in a multivariate model. In the Health ABC study [28], CT attenuation of the thigh muscle predicted hip fracture with a relative risk (RR) of 1.6, which remained significant with an RR of 1.4 after adjustment for aBMD. The RR for aBMD alone was not reported in this study.

In the Osteoporotic Fractures in Men study, composite descriptors combining the stress strain index or bending strength of bone with the muscle CSA were determined by peripheral QCT (pQCT) of the lower limb. These descriptors predicted hip fracture with a hazard ratio (HR) of up to 1.6 that remained significant after adjustment for aBMD of the spine and total hip (HR up to 1.2) [30], but areal total hip BMD alone resulted in an HR of 2.1 for hip fracture prediction. The other two studies [106], [107] did not specifically target hip fracture prediction or discrimination.

In the retrospective analysis of the European Femur Fracture Study, the relative volume of adipose tissue in the upper thigh and 12 soft tissue texture parameters characterising the muscle–lipid distribution (Figure 8) were significant discriminators of femur fracture. After combination with the standard BMD in a multivariate model, four parameters remained significant discriminators: BMD and cortical thickness representing the bone strength and the relative volume of adipose tissue representing mechanical protection against the hip fracture by a larger cushion and the surface density of the adipocyte distribution within muscle representing muscle degeneration [108]. Area under the curve values was 0.85 for a combined soft tissue model, 0.84 for a combined bone model containing BMD and cortical thickness, and 0.92 for a model combining soft tissue and bone parameters, indicating that soft tissue parameters discriminated proximal femur fracture and BMD and that a combined approach was providing the best discrimination (Figure 9).

Figure 8.

The CT image of the thigh from an elderly individual of the EFFECT study. Left: segmented fascia separating subcutaneous adipose tissue (yellow), muscle (red), and perimuscular adipose tissue (uncoloured); center: muscle tissue; and right: distribution of high-density muscle (red) and intermuscular and intramuscular adipose tissue (yellow). Uncoloured voxels contain mixtures of muscle and adipose tissue.

CT = computed tomography; EFFECT = European Femur Fracture Study.

Figure 9.

ROC curves of three different models used for hip fracture discrimination of the EFFECT study [108].

BMD = bone mineral density; EFFECT = European Femur Fracture Study. ROC = receiver operator characteristics.

Future directions

Quantitative CT to assess the muscle volume and density is an established technique used routinely in a large variety of applications. Automated or at least semiautomatic segmentation algorithms must be further developed, in particular to analyse larger body parts instead of single slices. The assumption of a well calibrated scanner, i.e., a CT value of zero for water, is misleading and can lead to larger accuracy errors in the determination of the CT muscle density. The use of phantoms should be considered. Standardisation of the terminology is also required.

The analysis of the muscle lipid distribution which is still in its infancy may give new insights into the pathophysiology of sarcopenia and cachexia and could in particular improve risk prediction of osteoporotic hip fracture because it takes into account to some extent the risk of falls through the muscle status. In sarcopenia, CT should be regarded as a complementary imaging method to DXA that in combination with MRI may also improve our understanding on intervention and may better explain effects on functional muscle outcomes than simple lean mass measurements.

Funding

None.

Conflict of interest

None declared.

Appendix A. Supplementary data

The following is the Supplementary data to this article: