J Appl Physiol 103: 1428–1435, 2007.
First published July 12, 2007; doi:10.1152/japplphysiol.01163.2006.
Innovative Methodology
Modeling upper and lower limb muscle volume by bioelectrical
impedance analysis
Alexander Stahn,1,2 Elmarie Terblanche,1 and Günther Strobel2
1
Department of Sport Science, Stellenbosch University, Stellenbosch, South Africa; and 2Institute of Sports Medicine,
University Hospital Charité, Campus Benjamin Franklin, Free University of Berlin, D-Berlin, Germany
Submitted 5 October 2006; accepted in final form 9 July 2007
of skeletal muscle mass (SMM) is
imperative in health and disease. For instance, the severity of
protein energy malnutrition has been shown to be reliably
indicated by SMM (17). Furthermore, in the critically ill
patient, SMM is a major physiological reserve. Catabolic
stress, anorexia, and immobilization worsen the nutritional
status so that the amount and function of SMM become a
determinant of survival (15). In older subjects, the age-related
loss of SMM, sarcopenia, is a common occurrence (20). Since
sarcopenia has been associated with physical frailty, an
increased occurrence of falls, decreased independence, and
subsequently increased health costs (20, 27), the continuous
monitoring of SMM in these people is crucial. Furthermore,
physiotherapists, sport scientists, exercise physiologists, and
related health professionals rely on the quantification of SMM
as a measure to determine an individual’s progress during
rehabilitation and training (19, 21).
Several techniques, such as whole body counting with neutron activation, dual-energy X-ray absorptiometry, computerized tomography (CT), and magnetic resonance imaging
(MRI), have been shown to be reliable and valid indirect
measures of muscle mass. However, all of these methods are
expensive and time consuming, and they require sophisticated
personnel to perform and evaluate the measurements. Moreover, those methods reported to be most accurate and precise
(CT and MRI) involve considerable radiation exposure or
extremely expensive equipment that is limited to specialized
research units.
During the past decade, a technique known as bioelectrical
impedance analysis (BIA) has gained recognition as an affordable, noninvasive, easy-to-operate, and fast alternative to assess body composition. Janssen et al. (19) reported correlation
coefficients between SMM determined by BIA and SMM
measured by MRI exceeding 0.88 and standard errors of the
estimate (SEE) of ⬃9% in a multiethnic sample of 158 women
and 230 men. Although it seems as if BIA is sufficiently
accurate to estimate SMM in epidemiological studies, it remains questionable if BIA is also suitable for the determination
of SMM in individuals.
Current BIA models have, however, a number of drawbacks. First and foremost, the prediction algorithms are
statistically derived rather than based on a causal physiological relationship and therefore are highly population
specific. Second, the conventional whole body electrode
arrangement assumes that the body consists of one concentric cylinder. Furthermore, measured impedance is assumed
to result from tissue volumes arranged in parallel. Tissues
are not, however, homogeneous conductors but rather are
composed of chemicals and cellular constituents that may or
may not be modeled as parallel resistance circuits. Moreover, the most popular frequency to obtain impedance measurements (50 kHz) is expected to be far too low to penetrate
the cells completely, and as a result the prediction is based
on a measure of the extracellular space. Finally, tissue is
believed to be isotropic, although SMM is anisotropic as a
result of the muscle fibers. Additionally, fluid distribution
between the intra- and extracellular spaces may vary between individuals and affect resistance independently from
total tissue volume. To overcome some of these drawbacks,
Salinari et al. (34, 35) suggested a very promising, although
more complex, electrical and geometric BIA model for
Address for reprint requests and other correspondence: A. Stahn, Dept. of
Sport Science, Stellenbosch University, Private Bag X1, 7602 Stellenbosch,
South Africa (e-mail: [email protected]).
The costs of publication of this article were defrayed in part by the payment
of page charges. The article must therefore be hereby marked “advertisement”
in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
magnetic resonance imaging; bioimpedance; body composition; skeletal muscle mass; segmental
THE ACCURATE MEASUREMENT
1428
8750-7587/07 $8.00 Copyright © 2007 the American Physiological Society
http://www. jap.org
Downloaded from http://jap.physiology.org/ by 10.220.32.246 on June 18, 2017
Stahn A, Terblanche E, Strobel G. Modeling upper and lower
limb muscle volume by bioelectrical impedance analysis. J Appl
Physiol 103: 1428–1435, 2007. First published July 12, 2007;
doi:10.1152/japplphysiol.01163.2006.—Most studies employing bioelectrical impedance analysis (BIA) for estimating appendicular skeletal muscle mass using descriptive BIA models rely on statistical
rather than biophysical principles. The aim of the present study was to
evaluate the feasibility of estimating arm and leg muscle volume
(MV) based on multiple bioimpedance measurements and using a
recently proposed mathematical model and to compare this technique
to conventional segmental BIA at high and low frequencies. MV of
the arm and leg, respectively, was determined in 15 young, healthy,
active men [age 22 ⫾ 2 (SD) yr, total body fat 15.6 ⫾ 5.1%] by
magnetic resonance imaging (MRI) and BIA using a conventional and
new bioimpedance model. MRI-determined MV for leg and arm was
6,268 ⫾ 1,099 and 1,173 ⫾ 172 cm3, respectively. Estimated MV by
the new BIA model [leg: 6,294 ⫾ 1,155 cm3 (50 kHz), 6,278 ⫾ 1,103
cm3 (500 kHz); arm: 1,216 ⫾ 172 cm3 (50 kHz), 1,155 ⫾ 157 cm3
(500 kHz)] was not statistically different from MRI-determined MV
(leg: P⫽ 0.958; arm: P⫽ 0.188). The new BIA model was superior to
conventional BIA and performed best at 500 kHz for estimating leg
MV as indicated by the lower relative total error [new: 3.6% (500
kHz), 5.2% (50 kHz); conventional: 7.6% (500 kHz) and 8.3% (50
kHz)]. In contrast, the new BIA model, both at 50 and 500 kHz, did
not improve the accuracy for estimating arm MV [new: 10.8% (500
kHz), 10.6% (50 kHz); conventional: 11.8% (500 kHz), 11.4% (50
kHz)]. It was concluded that modeling of multiple BIA measurements
has advantages for the determination of lower limb muscle volume in
healthy, active adult men.
Innovative Methodology
1429
DETERMINING LIMB MUSCLE VOLUME BY BIOIMPEDANCE ANALYSIS
determining lower appendicular SMM based on multiple
bioimpedance measurements at 50 kHz, which does not rely
on regression-derived prediction algorithms.
The aim of the present study was to test whether this model
could be extended to quantifying upper appendicular mass and,
second, whether the model can be improved by employing
current frequencies higher than 50 kHz. Specifically, we compared upper and lower appendicular muscle volume (MV) in
young adult men as determined by MRI with different modelderived estimates by BIA.
METHODS
J Appl Physiol • VOL
CSA m (x) ⫽
␳m
dR/dx
(1)
where ␳m is the resistivity of muscle (in ⍀ 䡠 cm) and R is the resistance
(in ⍀) of any specific site. ␳m was assumed to be 149 ⍀ 䡠 cm (1) and
124 ⍀ 䡠 cm (38) for the leg at 50 and 500 kHz, respectively, and 109
⍀ 䡠 cm and 89 ⍀ 䡠 cm for the arm at 50 and 500 kHz, respectively (see
DISCUSSION). To solve Eq. 1 for CSAm, resistance data were fitted by
a weighted sum of two cumulative Gaussian functions plus a straight
line using a weighted least-squares index (TableCurve 2D, V. 5.01,
Systat Software, Point Richmond, CA)
冦
R̂(x) ⫽ ␣1
冤 冢 冣冥 冧
冦 兰 冤 冢 冣冥 冧
2
x
1
␴1 冑2␲
兰
exp ⫺0.5
x ⫺ ␮1
␴1
dx ⫺ 0.5
⫺⬁
2
x
⫹ ␣2
1
exp ⫺0.5
␴2 冑2␲
x ⫺ ␮2
␴2
dx ⫹ ␣3x
(2)
⫺⬁
where R̂ (in ⍀) is the estimated resistance, ␣1 (in ⍀), ␣2 (in ⍀), and
␣3 (in ⍀/cm) refer to the weighting factors, ␴1 (in cm) and ␴2 (in cm)
to the variances (i.e., transition heights), and ␮1 (in cm) and ␮2 (in
cm) to the means (i.e., transition centers). ␮1 was set to 0 to force the
function through the origin. Finally, the total limb MV was computed
as the integral over x of the estimated cross-sectional area of muscle.
Additionally, total segmental resistance indexes for the arm and leg at
50 kHz and 500 kHz, respectively, were calculated as l2/R, where l
refers to total segmental length determined by MRI and R to corresponding total segmental resistance.
MRI. A whole body 1.5-T scanner (Magnetom Symphony, Siemens
Medical Solutions, Erlangen, Germany) was employed to acquire
transaxial images of the right arm and leg using a T1-weighted,
spin-echo sequence with a 363-ms repetition time and a 17-ms
echo-time. The images consisted of a 50 cm field of view and a 512 ⫻
352 pixel matrix. Subjects were lying in a prone position with their
103 • OCTOBER 2007 •
www.jap.org
Downloaded from http://jap.physiology.org/ by 10.220.32.246 on June 18, 2017
Subjects. Fifteen apparently healthy Caucasian men aged 19 to 25
yr were recruited via the University’s Internet Bulletin and public
notices on University campus of the Faculty of Health Sciences.
Inclusion into the study required the absence of taking any medication, any orthopedic problems, any amputations, and/or any diseases
per medical history and physical examination. All of the participants
were physically active and exercised for 30 – 60 min several times per
week. The mean self-reported physical activity rating (PA-R; dimensionless) was 5.6 ⫾ 1.9 (14). After the purpose, procedures, and
known risks of the tests had been explained, each participant gave
written informed consent. All procedures were approved by the Ethics
Committee for Human Research of Stellenbosch University.
Experimental design. Testing was performed at Stellenbosch MediClinic between 8.00 A.M. and 2.00 P.M. in a thermally controlled
room (20 –22°C ambient temperature and 55– 60% relative humidity).
To ensure normohydration, subjects were advised to drink at least 2
liters of water per day during the week preceding the measurements.
They were instructed to refrain from alcohol and physical exercise for
48 h and to refrain from smoking or drinking coffee on the day of the
test. First, data for anthropometry and BIA were obtained. After a
5-min break, during which subjects where expected to walk leisurely
to equilibrate any fluid shifts caused by lying supine, they completed
an MRI scan of the extremities.
Anthropometry. Body mass was measured barefoot, after voiding,
in minimal clothes on a calibrated electronic digital scale to the
nearest 100 g. Standing height was measured barefoot to the nearest
0.1 cm using a stadiometer. Body fat mass was derived from body
mass and fat-free mass (FFM), which was estimated using the following regression equation (37): FFM (kg) ⫽ ⫺10.68 ⫹ 0.65 ⫻
l2/R50 ⫹ 0.26 ⫻ body mass ⫹ 0.02 ⫻ R, where l is standing height (in
cm), body mass is in kilograms, and R50 is resistance at 50 kHz (in ⍀).
This prediction formula was cross-validated in a multiethnic sample
of 669 men, diverse in age, varying in adiposity, and it was found that
the equation provides sufficiently accurate estimates of FFM (coefficient of determination R2 ⫽ 0.9, SEE ⫽ 3.9 kg).
Bioelectrical impedance measurement. Impedance measurements
were performed with a SEAC SFB3 multifrequency bioelectrical
impedance monitor (Impedimed, Eight Mile Plains, Queensland,
Australia). After calibrating the measurement unit according to the
guidelines of the manufacturer using an 11-resistor (10 –2,000 ⍀)
network and a resistor-resistor-capacitor circuit, a sinusoidal current
of 190 ␮A was applied. Impedances and phase angles were recorded
at 50 kHz to 500 kHz and subsequently downloaded to a computer
workstation using software provided by the manufacturer (v1.5. 2003,
Impedimed, Eight Mile Plains). Measurements were taken from the
right side of the body via a tetrapolar electrode arrangement following
standard procedures (26). Subjects, dressed in minimal clothes, were
lying in a supine position with arms and legs 10° and 20° abducted
from the body, respectively. To control fluid shifts caused by lying
supine after standing upright, subjects were lying supine for 15 min
before the commencement of the BIA measurements. Before electrode
application, skin hair was removed, and skin was rubbed with a 70%
alcohol solution and allowed to dry. Current-introducing electrodes
(2.0 cm ⫻ 2.0 cm) (Impedimed, Eight Mile Plains) were placed in the
middle on the dorsal surface of the right hand and foot just below the
metacarpal-phalangeal and metatarsal-phalangeal joints, respectively.
Whole body resistance was determined by positioning two voltagesensing electrodes (same type as current-introducing electrodes) on
the dorsal surfaces of the right wrist and ankle midline between the
styloid processes of the ulna and radius and midline between the
medial and lateral malleoli, respectively. To obtain a resistance profile
along the limb, electrodes were positioned midline on the anterior
surface of the limbs at 2-cm intervals, commencing after the voltagesensing electrode at the wrist and ankle used for the whole body
measurement, up to the proximal part of the femoral and humeral
head, respectively. For the arm, additional electrodes were placed
midline on the posterior surface of the arm corresponding to the
anterior electrode positions. Subsequently, resistance was determined
between the original electrode at the wrist and each electrode along
the leg and between the original electrode at the ankle and each
electrode along the arm. Measured resistance was finally subtracted
from whole body resistance to obtain the resistance profiles of the arm
and leg, respectively.
Computation of muscle cross-sectional area and segmental MV.
The determination of MV based on limb resistance profiles has been
suggested and thoroughly outlined by Salinari et al. (34, 35). Briefly,
assuming that the injected current is flowing in longitudinal direction
only and that the resistivity of muscle is much smaller than that of
other tissues, an estimate of muscle cross-sectional area (CSAm) at a
specific site (x) can be obtained from
Innovative Methodology
1430
DETERMINING LIMB MUSCLE VOLUME BY BIOIMPEDANCE ANALYSIS
RESULTS
Descriptive data for all subjects are summarized in Table
1. The 15 men ranged in age from 19 to 25 yr (mean 22.1
yr). Percent body fat obtained with conventional whole body
BIA was 7.5% to 24.8% (mean 15.6%), and self-reported
physical activity levels (modified PA-R) varied between 2
and 8 (mean 5.5).
Table 1. Descriptive characteristics of subjects
Age, yr
Height, cm
Body Mass, kg
Body Fat, %
Modified PA-R
22.1⫾1.6
182.2⫾5.4
81.9⫾9.8
15.6⫾5.1
5.5⫾1.9
Values are means ⫾ SD; n ⫽ 15 men. Modified PA-R, modified physical
activity rating (dimensionless).
J Appl Physiol • VOL
Fig. 1. Measured (circles) and curve-fitted (continuous lines) resistance (R)
profile of lower (A) and upper (B) limbs at 50 and 500 kHz of a single subject.
Figure 1 shows the measured and curve-fitted resistance
profile of the limbs from one subject. The goodness-of-fit was
excellent for all observations. Each model exceeded a coefficient of determination of 0.999, and the SEE was less than 1.
Mean-fitting parameters are presented in Table 2. Results for
the leg at 50 kHz are well in agreement with the findings by
Salinari et al. (34, 35). Highly significant differences between
the parameters for 50 and 500 kHz data were observed for all
parameters except for ␴1 (P ⫽ 0.104) and ␮2 (P ⫽ 0.654) for
the leg, and ␴1 (P ⫽ 0.071) and ␴2 (P ⫽ 0.966) for the arm.
Figure 2 illustrates the estimated and measured muscle CSA
from a single subject corresponding to the illustrated BIA data
presented in Fig. 1. Mean measured and estimated MV are
shown in Table 3. Results from repeated-measures ANOVA
confirmed that neither for the leg (P ⫽ 0.958) nor for the arm
(P ⫽ 0.188) were any significant differences between measured and estimated MV observed. TE was substantially and
significantly lower at 500 kHz compared with 50 kHz for leg
MV (332 vs. 217 cm3; P ⬍ 0.05). TE differences between 50
and 500 kHz for arm MV were minimal and not statistically
meaningful (120 vs. 123 cm3; P ⫽ 0.398). Agreement for
individual data is shown in Fig. 3. No systematic bias was
observed as none of the correlations between the differences
103 • OCTOBER 2007 •
www.jap.org
Downloaded from http://jap.physiology.org/ by 10.220.32.246 on June 18, 2017
arms stretched overhead and their hands and feet fastened with velcro
straps to prevent rotation. Arms and legs were also slightly elevated
using supporters to ensure that all limbs were parallel to the table.
After an initial phase of 15 min lying supine to control fluid shifts,
10-mm-thick consecutive transaxial images with a 20-mm interslice
gap were taken between the ankle (medial and lateral malleoli) and the
inferior border of the ischial tuberosity, as well as between the wrist
(styloid processes of the ulna and radius) and the most distal visible
part of the deltoid, respectively.
Following the scanning procedure, all data were transferred to a
workstation and analyzed with the National Institutes of Health image
analysis software program (Image J1.35). Muscle cross-sectional
areas and MVs of the right arm and leg, respectively, were determined
as follows. At first, thresholds for muscle (including intramuscular
vasculature and connective tissue) and adipose tissue (including skin)
were selected on the basis of the gray-level histograms of each image.
In addition to the peak of the background value, the remaining two
highest maxima were identified as the muscle and adipose tissue gray
levels. The lowest levels between the peaks of background and muscle
gray level, as well as between the peaks of muscle and adipose gray
level, were taken as the thresholds for muscle and adipose tissue,
respectively. Muscle tissue was then color-coded, and a semitransparent copy was superimposed on the original gray-level image to verify
the accuracy of the selected tissue area and, if necessary, corrected.
Muscle cross-sectional area in each image was then calculated as the
sum of the given pixel of the respective tissue area multiplied by the
individual pixel surface area. MV in each slice was obtained by
multiplying tissue area (cm2) by slice thickness. Finally, whole
segment skeletal MV (cm3) was computed by the two-column (parallel trapezium) model, as suggested by Shen et al. (36). To maintain
consistency, all images were read and analyzed by a single trained
investigator. Intraobserver differences for the estimation of whole
body scans by experienced investigators have been reported to be less
than 2% (13, 33).
Data analysis. Statistical analyses were carried out using the SPSS
software (version 12.00, SPSS, Chicago, IL). Descriptive data are
reported as means and SDs. Differences between methods were tested
for significance by repeated-measures ANOVA and t-tests, and
Wilcoxon signed-rank and Friedman tests where appropriate. Agreement between methods was assessed by Bland-Altman plots. Total
error (TE) was calculated as TE ⫽ [(Y⬘ ⫺ Y)2/n]0.5, where Y⬘ and Y
refer to estimated and measured MV (in cm3), respectively, and n to
sample size. MRI-determined MV was regressed on segmental impedance indexes of the arm and leg at 50 kHz and 500 kHz to
scrutinize the accuracy of conventional segmental BIA modeling in
the present sample. Calculation of bias, limits of agreement, and TE
for these models were based on PRESS-predicted MV to obtain less
sample-specific statistics, as no data set for cross-validation was
available (18). The level of significance was set at ␣ ⫽ 0.05.
Innovative Methodology
1431
DETERMINING LIMB MUSCLE VOLUME BY BIOIMPEDANCE ANALYSIS
Table 2. Fitting parameters of limb resistance profiles
Leg
50 kHz
500 kHz
Arm
50 kHz
500 kHz
␣1 , ⍀
␣ 2, ⍀
␣3, ⍀/cm
␴1, cm
␴2, cm
␮2, cm
188.1⫾25.4
161.1⫾20.1†
56.4⫾5.9
42.8⫾3.8†
0.8⫾0.1
0.7⫾0.1†
10.8⫾1.5
10.5⫾1.2
8.5⫾1.0
7.6⫾0.9†
36.7⫾2.1
36.8⫾2.2
109.1⫾19.0
83.3⫾15.7†
18.1⫾5.9
12.8⫾5.1†
2.2⫾0.4
2.0⫾0.3*
6.3⫾0.8
6.3⫾1.4
2.5⫾1.0
2.6⫾1.2
24.8⫾2.3
24.8⫾2.8*
Values are means ⫾ SD. Significantly different from 50 kHz: *P ⬍ 0.05, †P ⬍ 0.001.
were, however, still larger compared with ⫺450 cm3 to 430
cm3 (mean: ⫺9.8 cm3) at 500 kHz.
The relative difference between measured and estimated leg
MV was within 5% for 73 and 87% and within 10% for 93 and
100% of the subjects for 50 kHz and 500 kHz, respectively. For
arm MV, relative differences were within 5% for 47 and 47%,
within 10% for 80 and 60%, and within 15% for 87 and 87%
of the subjects for 50 kHz and 500 kHz, respectively. We also
evaluated the relative error on estimated muscle CSA, calculated as CSAError (%) ⫽ 100 ⫻ ¥兩CSAMRI ⫺ CSABIA兩/
¥CSAMRI for an individual subject. Mean CSAError at 50 kHz
(15.8% ⫾ 6.7%) was comparable to the observations by
Salinari et al. (34), but significantly greater than mean CSAError
at 500 kHz (11.1% ⫾ 2.4%, P ⬍ 0.05). The difference between
mean CSAError at 50 kHz and at 500 kHz for the arm was
negligible and nonsignificant (14.1% ⫾ 5.8% vs. 15.1% ⫾
4.6%, P ⫽ 0.307).
Finally, to assess the accuracy of conventional segmental
BIA in the present sample, arm and leg MV were regressed on
resistance indexes of the arm and leg, respectively. Results
from linear regression analysis are shown in Table 4. Despite
nonsignificant differences between measured and predicted
MV, TE for leg MV was considerably and significantly increased compared with the new BIA model [507 vs. 332 cm3
(P ⬍ 0.001) and 466 vs. 217 cm3 (P ⬍ 0.001) for 50 kHz and
500 kHz, respectively]. Accordingly, the larger prediction error
was also reflected by extended limits of agreement (⫺1,030
cm3 to 1,026 cm3 and ⫺949 cm3 to 940 cm3 for 50 kHz and
500 kHz, respectively). In contrast, TE for arm MV obtained
by the conventional segmental BIA model was similar and
nonsignificant compared with the new BIA model [128 vs. 120
cm3 (P ⫽ 0.316) and 112 vs. 123 cm3 (P ⫽ 0.184) for 50 kHz
and 500 kHz, respectively].
DISCUSSION
Fig. 2. MRI-measured (‚) and bioelectrical impedance analysis (BIA)-estimated muscle volume derived from limb resistance profiles at 50 kHz (continuous line) and 500 kHz (dotted line) for the leg (A) and arm (B) of a single
subject. Leg muscle volume was 7,747 cm3, 7,404 cm3, and 7,625 cm3 as
determined by MRI, BIA (50 kHz), and BIA (500 kHz), respectively. Arm
muscle volume was 1,251 cm3, 1,308 cm3, and 1,053 cm3 as determined by
MRI, BIA (50 kHz), and BIA (500 kHz), respectively. CSA, cross-sectional
area.
J Appl Physiol • VOL
The measurement of regional SMM by BIA is of increasing
interest (23, 25), and a number of investigators have demonstrated the potential of segmental BIA to determine segmental
muscle mass (2, 3, 7, 8, 10 –12, 16, 22, 24, 28 –32, 34, 35).
However, despite these promising findings, only few investigators have scrutinized the ability of explanatory rather than
descriptive BIA models, the latter of which rely on statistical
techniques instead of causal relationships, and thus often imply
the need of sample-specific regression equations. Brown et al.
(7) first introduced a technique for determining muscle CSA
solely based on measured impedance and assumed specific
resistivities of all underlying tissues. Fuller et al. (12) extended
this specific resistivity approach to a 20-cm thigh and a 10-cm
calf section in 16 healthy subjects. Recently, a very promising
103 • OCTOBER 2007 •
www.jap.org
Downloaded from http://jap.physiology.org/ by 10.220.32.246 on June 18, 2017
and means were significant (P ⫽ 0.352, 0.321, 0.615, and
0.712, for leg MV at 50 kHz, leg MV at 500 kHz, arm MV at
50 kHz, and arm MV at 500 kHz, respectively). Nearly all of
the differences were located within 1.96 SD except for one
obvious outlier of leg MV at 50 kHz. Removing the outlier
revealed its impact on the limits of agreement, which were
reduced to ⫺468 cm3 to 537 cm3 (mean: 34.7 cm3). The limits
Innovative Methodology
1432
DETERMINING LIMB MUSCLE VOLUME BY BIOIMPEDANCE ANALYSIS
Table 3. Muscle volume measured by MRI and estimated by BIA
Leg
50 kHz
500 kHz
Arm
50 kHz
500 kHz
l, cm
MVMRI, cm3
MVBIA, cm3
Bias, cm3
LoA, cm3
TE, cm3
72.8⫾6.0
6,268⫾1,099
6,294⫾1,155
6,278⫾1,103
⫺26.4
⫺9.8
⫺697 to 644
⫺450 to 430
332
217*
37.7⫾2.3
1,173⫾172
1,216⫾172
1,155⫾157
⫺43.7
17.8
⫺271 to 184
⫺228 to 264
120
123
Values are means ⫾ SD; l, length indicated by MRI; MVMRI, muscle volume measured by magnetic resonance imaging (MRI); MVBIA, muscle volume
estimated by bioelectric impedance analysis (BIA); LoA, limits of agreement calculated as mean bias ⫾ 1.96SD; TE, total error computed as TE ⫽
关(Y⬘ ⫺ Y)2/n兴0.5. *Significantly different from 50 kHz: P ⬍ 0.05.
Fig. 3. Bland-Altman plot (bias and 95% confidence intervals) of the difference between the methods against the mean of methods for leg and arm muscle
volume. 50 kHz: E, continuous lines; 500 kHz: F, dashed lines.
J Appl Physiol • VOL
0.5% at both 50 and 500 kHz. The %SEE was, however, less
pronounced at 500 kHz (3.6%) compared with 50 kHz (5.2%),
which was also reflected by smaller limits of agreement for 500
kHz (%SEE were nearly identical to %TE for all models and
will be treated interchangeably to avoid confusion). Removing
the apparent outlier in Fig. 3A narrowed the limits of agreement for the 50-kHz approach to that of the 500-kHz model
and reduced the %SEE to 3.8%. However, a similar proportional reduction was correspondingly observed for the 500-kHz
approach (SEE% ⫽ 2.7%). The slight advantage of the 500kHz model was also reflected by conventional segmental BIA
(SEE% ⫽ 8.3% vs. 7.6% for 50 and 500 kHz, respectively). In
contrast, the 1.6- (50 kHz) to 2.2-fold (500 kHz) difference
between conventional segmental and the new BIA model was
substantial and statistically significant (P ⬍ 0.05) and clearly
demonstrates the superiority of the new BIA model in active
young men, confirming the findings by Salinari et al. (35).
Arm MV was underestimated by a mean of 3.7% at 50 kHz
and overestimated by 1.5% at 500 kHz by the new BIA model.
Absolute SEE was lower than for leg MV (Table 2). The lower
absolute SEE did not, however, decrease proportionally to
lower arm MV, resulting in a considerably larger %SEE
compared with leg MV (10.6% and 10.8% for 50 and 500 kHz,
respectively). The new BIA model for arm MV did not outperform conventional segmental BIA models as estimations
simply based on resistance indexes produced similar errors
(11.4% and 11.8% for 50 and 500 kHz, respectively). Although
the size of these errors is in agreement with conventional
segmental BIA models for total arm muscle volume (mass)
reported by other investigators (6, 16), the poor performance
for arm MV deserves clarification. To avoid an even more
complex fitting equation, data collection for the arm was
limited to the most distal visible part of the deltoid. In turn, a
relatively small number of data points were obtained, jeopardizing a reliable curve fit. However, reducing the number of
parameters to be estimated, e.g., setting the value of ␮2 at the
measured level of the elbow as determined by anthropometry,
did not change the remaining parameter estimates. Given the
reliable fitting procedures, it was speculated that an irregular
current distribution might have affected the MV estimate of the
arm. We therefore investigated whether a combination of an
anterior-posterior electrode arrangement might reveal the full
potential of the technique. Accordingly, electrodes were placed
both on the anterior and posterior surface of the arm and
measurements of two electrodes opposite to each other were
averaged to obtain a more characteristic resistance profile of
the arm. A visual inspection of the resulting resistance profiles
revealed that they were nearly identical with their anterior
103 • OCTOBER 2007 •
www.jap.org
Downloaded from http://jap.physiology.org/ by 10.220.32.246 on June 18, 2017
technique for estimating total leg MV (ankle to approximately
inguinal crease) was introduced by Salinari et al. (34). This
approach underestimated MRI-determined MV by a mean of
1.7% and %SEE of 5.3% in six healthy adults. Until now,
application of this technique has been limited to lower appendicular SMM. Furthermore, according to the authors’ knowledge, no study has scrutinized the role of using frequencies
higher than 50 kHz, an area of inquiry that deserves particular
attention with regard to the study of regional BIA (25).
The present study showed that leg MV determined by this
new BIA model was underestimated by a mean of less than
Innovative Methodology
1433
DETERMINING LIMB MUSCLE VOLUME BY BIOIMPEDANCE ANALYSIS
Table 4. Linear regression analysis for predicting muscle volume from resistance index (l2/R)
Leg (50 kHz)
MVBIA ⫽ 254 ⫻l2/R ⫺ 419
Leg (500 kHz)
MVBIA ⫽ 224 ⫻l2/R ⫺ 735
Arm (50 kHz)
MVBIA ⫽ 103 ⫻l2/R ⫹ 221
Arm (500 kHz)
MVBIA ⫽ 96 ⫻ l2/R ⫹ 114
R2
SEE, cm3
Bias, cm3
LoA, cm3
TE, cm3
0.818
487
⫺2.1
⫺1,030 to 1,026
507
0.848
444
⫺4.1
⫺949 to 940
466
0.565
118
⫺6.8
⫺265 to 252
128
0.649
106
⫺5.0
⫺232 to 222
112
Values are means ⫾ SD; MVBIA, muscle volume estimated by bioelectric impedance analysis (BIA); SEE, standard error of the estimate calculated as
SDy ⫻ 关(Y⬘ ⫺ Y)2/(n ⫺ 1)兴0.5; LoA, limits of agreement calculated as mean bias ⫾ 1.96SD; TE, total error computed as TE ⫽ 关(Y⬘ ⫺ Y)2/n兴0.5; R,
resistance; R2, coefficient of determination.
J Appl Physiol • VOL
higher relative measurement errors, and it may well be possible
that the full potential of the new model lies within the estimation of lower limb MV, as tissue composition of the upper limb
is less dominated by muscle mass compared with the lower
limb. As a result, the resistance profile of the arm might not be
as distinctive as for the leg. In turn, the smoother behavior of
the arm resistance profile might also explain the estimation of
arm MV by regression analysis. Additionally, the contribution
of skin, tendons, ligaments, synovial fluid, and subcutaneous
adipose tissue as well as inter- and intramuscular fat to the flow
of current is assumed to be negligible. However, particularly in
regions with lower muscle mass, this assumption might be
violated. The impact of anatomic structure other than muscle
on resistance measurements might be less pronounced in the
leg because of its higher MV. Furthermore, irrespectively of
the lack of the improvement of using two electrodes at a single
measurement site, the assumption of a homogeneous current
distribution in longitudinal direction might nevertheless be
violated and crucial for estimating arm MV. Finally, it should
be stressed that the technical error for arm MV was as low as
120 cm3, and in light of the assumptions underlying the model
a further improvement seems questionable.
The most crucial aspect of the new bioimpedance model is
the magnitude of the resistivity constants. The true in vivo
value for skeletal muscle resistivity is a fundamental yet
unanswered question (25). Thus the derivation of the resistivities used in the present study deserves clarification. One of the
widely employed resistivity constants was determined by
Zheng et al. (38), who measured samples from rabbit, dog, cat,
monkey, and pig muscle in a test cell at selected frequencies
between 1 Hz and 1 MHz. Resistivity at 50 kHz varied between
116.8 and 133.2 ⍀䡠cm. According to the authors’ knowledge,
a mean resistivity value of 118.0 ⍀ 䡠cm based on the samples
from the five mammals was first cited by Brown et al. (7) and
used for the determination of upper arm muscle CSA. Several
investigators have since used this value for testing the validity
of BIA models (12, 34, 35). A close inspection of the original
study by Zheng et al. (38), however, reveals that the sample
sizes of the different mammals tested were highly variable
(range: 1–14 specimens per sample) and therefore demand the
calculation of a weighted mean specific resistivity, which
equals 124.1 ⍀䡠cm. It is impossible to verify at this stage
which value yields an estimation closer to the true resistivity,
and it is not intended to question the results of previous studies.
However, it should be noted that the choice between the two
resistivities can introduce a change of 4 – 6 percentage points in
103 • OCTOBER 2007 •
www.jap.org
Downloaded from http://jap.physiology.org/ by 10.220.32.246 on June 18, 2017
counterparts. Regressing averaged (anterior-posterior) resistance against anterior resistance for each subject confirmed
these observations. Slopes were not significantly different from
1, and intercepts were not significantly different from 0. All
coefficients of determination exceeded 0.999. Mean differences between averaged and anterior resistance were negligible
(0.87 and ⫺0.62 ⍀ for 50 and 500 kHz, respectively) and well
within the technical accuracy of the equipment. Accordingly,
no significant differences were observed for estimated MV
based on averaged and anterior resistance measurements alone
[1,216 vs. 1,201 cm3 (P ⫽ 0.590) and 1,155 vs. 1,148 cm3
(P ⫽ 0.632) for 50 and 500 kHz, respectively]. It was therefore
concluded that a combined anterior-posterior electrode arrangement does not improve estimated arm MV. It was not
tested whether additional electrode sites might yet enhance
measurement accuracy. However, in light of the present findings, it seems as if any irregularities of the arm’s current
distribution do not substantially confound the model, and it is
therefore unlikely that measurements at any additional electrode sites further improve the estimation of arm MV. Finally,
it was tested whether any anisotropic effects might have
confounded the estimation of arm MV. If a frequency of 500
kHz might have been too low to fully penetrate all muscle
cells, the new BIA model might be only advantageous with
respect to the statistical model by using resistance determined
at infinite frequency. Bioimpedance data for all subjects were
therefore reexamined using Cole-Cole plots to extrapolate
resistance at infinite frequency for each electrode site. Plotting
resistance against arm length showed that resistance profiles at
500 kHz and at infinite frequency were nearly identical. Correspondingly, no statistical difference was found between
mean arm MV determined at 500 kHz and infinity (1,155 vs.
1,144 cm3, P ⫽ 0.362). In conclusion, it was found that
extrapolated resistance at infinity does not improve arm MV
estimation. It should be noted that the higher technical error
inherent in determining resistance at infinite frequency might
have limited the approach. However, given the nearly identical
behavior of resistance profiles at 500 kHz and at infinity,
measurement errors did not seem to substantially affect the
approach. Thus it was concluded that anisotropic effects were
negligible and do not explain the poorer reconstruction of arm
MV with respect to the regression-based BIA model.
In light of these experimental findings, the authors suggest
that the simplifying hypotheses underlying the model explain
the high relative measurement error of arm MV with respect to
leg MV. Clearly, the lower arm MV will inherently lead to
Innovative Methodology
1434
DETERMINING LIMB MUSCLE VOLUME BY BIOIMPEDANCE ANALYSIS
1
Resistivity was calculated based on the quotient of MRI-determined
muscle volume and the integrated resistance over segment length.
J Appl Physiol • VOL
erence rather than experimental data and the discussion will
simply shift from sample-specific regression equations to sample-specific resistivity constants. Nevertheless, the new BIA
model produced some exceptional results for determining leg
MV in healthy, young men, confirming the findings by Salinari
et al. (34, 35).
In conclusion, the present study confirmed that the new
bioimpedance model is superior to conventional segmental
BIA for estimating lower limb MV. In contrast, estimation of
arm MV seems to benefit less from the new model as arm MV
was equally well predicted by conventional BIA modeling
based on simple linear regression analysis. However, it should
be noted that the estimation error of conventional segmental
BIA reported in the present study is likely to be too optimistic
as results from regression analysis were not cross-validated.
Furthermore, we conclude that irrespective of the equivalent
prediction accuracy, the new BIA model is superior to conventional segmental BIA since no sample-specific regression algorithm is needed. Using 500 kHz instead of 50 kHz slightly
improved the accuracy for lower limb but not upper limb MV.
Further research is suggested to assess the true benefit of using
higher frequencies than 50 kHz for the new bioimpedance
model in a large heterogeneous sample and its feasibility to
track changes in lower appendicular SMM following exercise
training programs.
ACKNOWLEDGMENTS
We thank Prof. Allan Scher from the Dept. of Radiology, Stellenbosch
University, and Prof. Johan Koeslag from the Dept. of Medical Physiology,
Stellenbosch University, for promoting the realization of the project. We also
thank Drs. Van Wageningen and Partners for performing the MRI scans and
acknowledge the outstanding technical contributions by Marie-Louise De
Villers, Annelise Coetzee, and Annelie le Roux in preparing and performing
the MRI scans.
Finally, we thank Impedimed for excellent technical service.
GRANTS
This work was supported by funds provided by the Harry Crossley Foundation.
REFERENCES
1. Aaron R, Huang M, Shiffman CA. Anisotropy of human muscle via
non-invasive impedance measurements. Phys Med Biol 42: 1245–1262,
1997.
2. Bartok C, Schoeller DA. Estimation of segmental muscle volume by
bioelectrical impedance spectroscopy. J Appl Physiol 96: 161–166, 2004.
3. Baumgartner RN, Ross R, Heymsfield SB. Does adipose tissue influence
bioelectric impedance in obese men and women? J Appl Physiol 84:
257–262, 1998.
4. Biggs J, Cha K, Horch K. Electrical resistivity of the upper arm and leg
yields good estimates of whole body fat. Physiol Meas 22: 365–376, 2001.
5. Bracco D, Berger MM, Revelly JP, Schutz Y, Frascarolo P, Chiolero
R. Segmental bioelectrical impedance analysis to assess perioperative
fluid changes. Crit Care Med 28: 2390 –2396, 2000.
6. Bracco D, Thiebaud D, Chiolero RL, Landry M, Burckhardt P,
Schutz Y. Segmental body composition assessed by bioelectrical impedance analysis and DEXA in humans. J Appl Physiol 81: 2580 –2587, 1996.
7. Brown B, Karatzas T, Nakielny R. Determination of upper arm muscle
and fat areas using electrical impedance methods: a comparative study.
Clin Phys Physiol Meas 9: 47–55, 1988.
8. Chumlea WC, Baumgartner RN, Roche AF. Specific resistivity used to
estimate fat-free mass from segmental body measures of bioelectric
impedance. Am J Clin Nutr 48: 7–15, 1988.
9. Elia M, Fuller NJ, Hardingham CR, Graves M, Screaton N, Dixon
AK, Ward LC. Modeling leg sections by bioelectrical impedance analysis, dual-energy X-ray absorptiometry, and anthropometry: assessing
103 • OCTOBER 2007 •
www.jap.org
Downloaded from http://jap.physiology.org/ by 10.220.32.246 on June 18, 2017
estimated segmental MV. Because of the above-mentioned
sample bias underlying the calculated resistivity of 118.0
⍀䡠 cm (precisely 118.7 ⍀䡠cm), the value of 124.1 ⍀䡠cm was
preferred in the present study. Nevertheless, we questioned
whether this value would be appropriate for estimating leg MV
as the resistivity of 124.1 ⍀䡠cm was obtained in vitro for
muscles in the longitudinal direction. In the leg, and particularly in the calf, the assumption of isotropy is violated as not all
muscles run in the same longitudinal direction, and their
orientation varies presumably between subjects (9). In fact, as
demonstrated by Aaron et al. (1), human leg muscle is highly
anisotropic. Given that at 50 kHz only a variable proportion of
total muscle volume is penetrated by the applied current flow,
the specific resistivity for 50 kHz obtained ex vivo for muscle
in longitudinal direction must be adjusted, i.e., increased for in
vivo applications. This is in agreement with the findings of
Aaron et al. (1), who reported a value of 149.2 ⍀䡠cm for
in vivo measurements of human thigh muscle resistivity at 50
kHz using a tetrapolar electrode arrangement. The size of this
value is well in accordance with leg muscle resistivity obtained
in the present study at 50 kHz (148.8 ⍀䡠cm).1 Furthermore,
this value has been shown to be superior for estimating lower
and upper leg MV to 118.0 ⍀䡠cm (12). On the other hand, at
500 kHz, cell membranes are highly short-circuited compared
with 50 kHz and nearly all muscle fibers should be penetrated
by the current and the leg muscle segment becomes purely
resistive, justifying the use of a resistivity constant obtained for
longitudinal muscle. For these reasons, we adopted a specific
resistivity of 149.0 ⍀䡠cm for leg MV at 50 kHz, and 124.1
⍀䡠cm for leg MV at 500 kHz.
Consequently, it would be expected to use the same constants for estimating arm MV. Numerous previous studies
have, however, indicated that resistivity is segment specific.
(4 – 6, 8, 10, 40). On the basis of four studies reporting
segmental resistivity for upper and lower limbs (4, 5, 8, 10), we
calculated a mean 1.4-fold (⫾0.2) lower arm muscle resistivity
compared with leg muscle resistivity. Similarly, a 1.46-fold
difference was computed for estimating intracellular water and
extracellular water of the arm and leg (39). These differences
are well in agreement with our own observations (1.43- and
1.34-fold difference at 50 and 500 kHz, respectively). The
reason for segment-specific resistivities can be traced to different composition of body tissues within segments and, second, to differences of geometric shape and volume that can
cause inhomogeneous current distributions. Accordingly, resistivity for the arm muscle was obtained by correcting leg
muscle resistivity by 1.4, producing arm muscle resistivities of
106 and 89 ⍀䡠cm for 50 and 500 kHz, respectively. These
constants were also in agreement with values obtained on the
basis of the present data (104 and 93 ⍀䡠 cm for 50 and 500 kHz,
respectively). We feel therefore confident that the predetermined specific resistivities of 106.0 and 89.0 ⍀䡠cm for arm
muscle and 149.0 and 124.1 ⍀䡠 cm for leg muscle were close to
their true in vivo equivalents in the present study.
It should be noted, however, that unless further research
establishes the true value for skeletal muscle resistivity, the
choice of resistivity will rather depend on investigators’ pref-
Innovative Methodology
DETERMINING LIMB MUSCLE VOLUME BY BIOIMPEDANCE ANALYSIS
10.
11.
12.
13.
14.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
J Appl Physiol • VOL
26. Lukaski HC, Johnson PE, Bolonchuk WW, Lykken GI. Assessment of
fat-free mass using bioelectrical impedance measurements of the human
body. Am J Clin Nutr 41: 810 – 817, 1985.
27. Marcell TJ. Sarcopenia: causes, consequences, preventions. J Gerontol A
Biol Sci Med Sci 58: M911–M916, 2003.
28. Miyatani M, Kanehisa H, Fukunaga T. Validity of bioelectrical impedance and ultrasonographic methods for estimating the muscle volume of
the upper arm. Eur J Appl Physiol 82: 391–396, 2000.
29. Miyatani M, Kanehisa H, Masuo Y, Ito M, Fukunaga T. Validity of
estimating limb muscle volume by bioelectrical impedance. J Appl Physiol
91: 386 –94, 2001.
30. Nunez C, Gallagher D, Grammes J, Baumgartner RN, Ross R, Wang
Z, Thornton J, Heymsfield SB. Bioimpedance analysis: potential for
measuring lower limb skeletal muscle mass. J Parenter Enteral Nutr 23:
96 –103, 1999.
31. Pietrobelli A, Morini P, Battistini N, Chiumello G, Nunez C,
Heymsfield SB. Appendicular skeletal muscle mass: prediction from
multiple frequency segmental bioimpedance analysis. Eur J Clin Nutr 52:
507–511, 1998.
32. Pietrobelli A, Nunez C, Zingaretti G, Battistini N, Morini P, Wang
ZM, Yasumura S, Heymsfield SB. Assessment by bioimpedance of
forearm cell mass: a new approach to calibration. Eur J Clin Nutr 56:
723–728, 2002.
33. Ross R, Pedwell H, Rissanen J. Effects of energy restriction and exercise
on skeletal muscle and adipose tissue in women as measured by magnetic
resonance imaging. Am J Clin Nutr 61: 1179 –1185, 1995.
34. Salinari S, Bertuzzi A, Mingrone G, Capristo E, Pietrobelli A,
Campioni P, Greco AV, Heymsfield SB. New bioimpedance model
accurately predicts lower limb muscle volume: validation by magnetic
resonance imaging. Am J Physiol Endocrinol Metab 282: E960 –E966,
2002.
35. Salinari S, Bertuzzi A, Mingrone G, Capristo E, Scarfone A, Greco
AV, Heymsfield SB. Bioimpedance analysis: a useful technique for
assessing appendicular lean soft tissue mass and distribution. J Appl
Physiol 94: 1552–1556, 2003.
36. Shen W, Wang Z, Tang H, Heshka S, Punyanitya M, Zhu S, Lei J,
Heymsfield SB. Volume estimates by imaging methods: model comparisons with visible woman as the reference. Obes Res 11: 217–225, 2003.
37. Sun SS, Chumlea WC, Heymsfield SB, Lukaski HC, Schoeller D,
Friedl K, Kuczmarski RJ, Flegal KM, Johnson CL, Hubbard VS.
Development of bioelectrical impedance analysis prediction equations for
body composition with the use of a multicomponent model for use in
epidemiologic surveys. Am J Clin Nutr 77: 331–340, 2003.
38. Zheng E, Shao S, Webster JG. Impedance of skeletal muscle from 1 Hz
to 1 MHz. IEEE Trans Biomed Eng 31: 477– 481, 1984.
39. Zhu F, Kuhlmann MK, Kaysen GA, Sarkar S, Kaitwatcharachai C,
Khilnani R, Stevens L, Leonard EF, Wang J, Heymsfield S, Levin
NW. Segment-specific resistivity improves body fluid volume estimates
from bioimpedance spectroscopy in hemodialysis patients. J Appl Physiol
100: 717–724, 2006.
40. Zhu F, Wystrychowski G, Kitzler T, Thijssen S, Kotanko P, Levin
NW. Application of bioimpedance techniques to peritoneal dialysis. Contrib Nephrol 150: 119 –128, 2006.
103 • OCTOBER 2007 •
www.jap.org
Downloaded from http://jap.physiology.org/ by 10.220.32.246 on June 18, 2017
15.
segmental muscle volume using magnetic resonance imaging as a reference. Ann NY Acad Sci 904: 298 –305, 2000.
Fuller NJ, Elia M. Potential use of bioelectrical impedance of the “whole
body” and of body segments for the assessment of body composition:
comparison with densitometry and anthropometry. Eur J Clin Nutr 43:
779 –791, 1989.
Fuller NJ, Fewtrell MS, Dewit O, Elia M, Wells JC. Segmental
bioelectrical impedance analysis in children aged 8 –12 y. 2. The assessment of regional body composition and muscle mass. Int J Obes Relat
Metab Disord 26: 692–700, 2002.
Fuller NJ, Hardingham CR, Graves M, Screaton N, Dixon AK, Ward
LC, Elia M. Predicting composition of leg sections with anthropometry
and bioelectrical impedance analysis, using magnetic resonance imaging
as reference. Clin Sci 96: 647– 657, 1999.
Gallagher D, Belmonte D, Deurenberg P, Wang Z, Krasnow N,
Pi-Sunyer FX, Heymsfield SB. Organ-tissue mass measurement allows
modeling of REE and metabolically active tissue mass. Am J Physiol
Endocrinol Metab 275: E249 –E258, 1998.
George JD, Stone WJ, Burkett LN. Non-exercise V̇O2 max estimation for
physically active college students. Med Sci Sports Exerc 29: 415– 423,
1997.
Griffiths RD. Muscle mass, survival, and the elderly ICU patient. Nutrition 12: 456 – 458, 1996.
Heymsfield SB, Gallagher D, Grammes J, Nunez C, Wang Z,
Pietrobelli A. Upper extremity skeletal muscle mass: potential of measurement with single frequency bioimpedance analysis. Appl Radiat Isot
49: 473– 474, 1998.
Heymsfield SB, McManus C, Stevens V, Smith J. Muscle mass: reliable
indicator of protein-energy malnutrition severity and outcome. Am J Clin
Nutr 35: 1192–1199, 1982.
Holiday DB, Ballard JE, McKeown BC. PRESS-related statistics: regression tools for cross-validation and case diagnostics. Med Sci Sports
Exerc 27: 612– 620, 1995.
Janssen I, Heymsfield SB, Baumgartner RN, Ross R. Estimation of
skeletal muscle mass by bioelectrical impedance analysis. J Appl Physiol
89: 465– 471, 2000.
Janssen I, Heymsfield SB, Ross R. Low relative skeletal muscle mass
(sarcopenia) in older persons is associated with functional impairment and
physical disability. J Am Geriatr Soc 50: 889 – 896, 2002.
Kasper CE, Talbot LA, Gaines JM. Skeletal muscle damage and
recovery. AACN Clin Issues 13: 237–247, 2002.
Kyle UG, Genton L, Hans D, Pichard C. Validation of a bioelectrical
impedance analysis equation to predict appendicular skeletal muscle mass
(ASMM). Clin Nutr 22: 537–543, 2003.
Lorenzo AD, Andreoli A. Segmental bioelectrical impedance analysis.
Curr Opin Clin Nutr Metab Care 6: 551–555, 2003.
Lukaski HC. Assessing regional muscle mass with segmental measurements of bioelectrical impedance in obese women during weight loss. Ann
NY Acad Sci 904: 154 –158, 2000.
Lukaski HC. Regional bioelectrical impedance analysis: applications in
health and medicine. Acta Diabetol 40, Suppl 1: S196 –S199, 2003.
1435