- Email: [email protected]
Manual segmentation of DXA scan images results in reliable upper and lower extremity soft and rigid tissue mass estimates
ARTICLE IN PRESS Journal of Biomechanics 42 (2009) 1138–1142
Contents lists available at ScienceDirect
Journal of Biomechanics journal homepage: www.elsevier.com/locate/jbiomech www.JBiomech.com
Short communication
Manual segmentation of DXA scan images results in reliable upper and lower extremity soft and rigid tissue mass estimates Timothy A. Burkhart a,b, Katherine L. Arthurs a, David M. Andrews a,b, a b
Department of Kinesiology, University of Windsor, Windsor, Ontario, Canada N9B 3P4 Department of Industrial and Manufacturing Engineering, University of Windsor, Windsor, Ontario, Canada
a r t i c l e in f o
a b s t r a c t
Article history: Accepted 18 February 2009
Quantification of segment soft and rigid tissue masses in living people is important for a variety of clinical and biomechanical research applications including wobbling mass modeling. Although Dualenergy X-ray Absorptiometry (DXA) is widely accepted as a valid method for this purpose, the reliability of manual segmentation from DXA scans using custom regions of interest (ROIs) has not been evaluated to date. Upper and lower extremity images of 100 healthy adults who underwent a full body DXA scan in the supine position were manually segmented by 3 measurers independently using custom ROIs. Actual tissue masses (fat mass, lean mass, bone mineral content) of the arm, arm with shoulder, forearm, forearm and hand, thigh, leg, and leg and foot segments were quantified bilaterally from the ROIs. There were significant differences between-measurers, however, percentage errors were relatively small overall (o1–5.98%). Intraclass correlation coefficients (ICCs) were very high between and withinmeasurers, ranging from 0.990 to 0.999 and 0.990 to 1.00 for the upper and lower extremities, respectively, suggesting excellent reliability. Between and within-measurer errors were comparable in general, and differences between the tissue types were small on average (maximum of 42 and 53 g for upper and lower extremities, respectively). These results suggest that manual segmentation of DXA images using ROIs is a reliable method of estimating soft and rigid tissues in living people. & 2009 Elsevier Ltd. All rights reserved.
Keywords: Reliability DXA Segmentation Upper extremity Lower extremity Tissue mass prediction
1. Introduction Accurate and reliable quantification of soft (muscle, fat) and rigid (bone mineral content) tissue masses in living people is very important for many biomechanical applications such as the modeling of human impacts using wobbling mass models (e.g. Gruber et al., 1998). Dual-energy X-ray Absorptiometry (DXA) has become a commonly adopted method for estimating soft and rigid tissue masses in-vivo (Cawkwell, 1998; Van Loan, 1998). Most DXA imaging analysis software packages provide automated regions of interest (ROI), but for limited clinical applications. Researchers can also obtain soft and rigid tissue mass information for other specific segments (Arthurs and Andrews, 2009; Durkin et al., 2002; Durkin and Dowling, 2003; Holmes et al., 2005) or sections of a segment (Clarke et al., 2004) by parceling them out via custom ROIs created by manually tracing polygons over the images. DXA is generally accepted as a valid method for quantifying body composition (Bracco et al., 1996; Fields and Goran, 2000; Fuller et al., 1992; Prevral et al., 2005). However, the reliability of segment tissue masses, including fat mass (FM), lean mass (LM), bone Corresponding author at: Department of Kinesiology, University of Windsor, Windsor, Ontario, Canada N9B 3P4. Tel.: +1 519 253 3000x2433; fax: +1 519 973 7056. E-mail address: [email protected] (D.M. Andrews).
0021-9290/$ - see front matter & 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.jbiomech.2009.02.017
mineral content (BMC), and wobbling mass (WM ¼ FM+LM), which result from the manual segmentation of DXA scan images using custom ROIs, has yet to be systematically documented between and within-measurers for a range of segments. Therefore, the main objective of this study was to quantify the between and withinmeasurer reliability of tissue mass estimates resulting from manual segmentation of the upper and lower extremities from DXA images.
2. Methods One hundred healthy adults (50 M, 50 F: mean age, mass, height of 21.876.2 years, 70.475.0 kg, 170.8719.4 cm, respectively) underwent a full body DXA scan (GE Lunar Prodigy Advance, System # DF+300481, GE Healthcare, Waukesha, Wisconsin) in a supine position with upper extremities by their sides. The mean (SD) BMI for all subjects was 24.0 (3.53) kg/m2 with a range of 17.0–34.0 kg/m2. These values are representative of people that vary from under weight to obese (Statistics Canada, 2003), and are therefore fairly indicative of the general population. The study was approved by the Research Ethics Boards of the University of Windsor and Windsor Regional Hospital and all participants provided written consent prior to participation. Using enCORE software (2006, GE Healthcare, v. 10.51.006), custom ROIs were traced manually over the DXA images to segment the upper and lower extremities (e.g. Fig. 1). ROI borders were adapted from Dempster (1955) and Clarys and Marfell-Jones (1986) (Table 1). Three trained measurers independently segmented each segment twice on the original scans (i.e. no ROIs or personal information remained on the scans to help the measurers), completing both extremities for all subjects before repeating (e.g. 100 upper limbs, then 100 lower limbs, then repeat).
ARTICLE IN PRESS T.A. Burkhart et al. / Journal of Biomechanics 42 (2009) 1138–1142
1139
Fig. 1. Sample regions of interest (ROI) on DXA scans showing the hard (bone) and soft (fat and muscle) tissue masses for an arm (a) arm with shoulder (b), and thigh (c) segment. The order in which the measurers segmented the upper and lower extremities was randomized. Fat mass, lean mass, bone mineral content and wobbling mass were quantified for each ROI. Measurement error was represented in two ways: as absolute between and within-measurer differences (g), and relative or percentage (%) error, expressed as the difference in tissue masses between trials (within-measurer error) or betweenmeasurers (between-measurer error), divided by the mean tissue masses of each segment (Burkhart et al., 2008). Intraclass correlation coefficients (ICCs) were used to quantify between and within-measurer reliability, and good to excellent reliability was accepted for ICCsX0.75 (Portney and Watkins, 2000). Finally, a Pearson correlation was performed to determine the relationship between subject body mass index (BMI) and the percentage error of the estimated tissue masses.
(Fig. 2), with maximum errors occurring in the forearm and hand segment (5.05%) and for BMC tissue (5.98%). In all cases, the mean between-measurer ICCs far exceeded the 0.75 criterion set as good to excellent reliability (0.990–0.999) (Table 2). Ninety one percent of the within-measurer errors were less than 5% (ranging from 1.9% to 6.2%). The largest segment and tissue mass errors were found for the arm with shoulder segment (4.45%) and fat mass (4.10%), respectively (Fig. 2). Overall, withinmeasurer ICCs were very high (ranging from 0.991 to 0.998). 3.2. Lower extremity
3. Results 3.1. Upper extremity The overall percentage errors across all segments and tissue types (1.12–5.98%) were relatively small between-measurers
Similar to the upper extremity, none of the between-measurer differences resulted in percentage errors greater than 5% (0.20–4.11%), with maximum errors occurring in the leg segment (2.24%) (Fig. 2). Averaged across segment, ICCs were very high and consistent, ranging from 0.995 (thigh) to 1.00 (leg and foot).
ARTICLE IN PRESS 1140
T.A. Burkhart et al. / Journal of Biomechanics 42 (2009) 1138–1142
Table 1 Description of the proximal/distal and medial/lateral borders that were used to define all regions of interest (ROIs) used by the measurers for manual segmentation (refer also to the sample images in Fig. 1).
Arm with shoulder
Forearm
Forearm and hand
Lower extremity Thigh
Leg
Leg and foot
Distally: a line drawn connecting the medial and lateral epicondyles of the humerus through the middle of the elbow joint space. Proximally: lines drawn around the humeral head through the shoulder joint space. Medially/laterally: all soft tissues included. Distally: a line drawn connecting the medial and lateral epicondyles of the humerus through the middle of the elbow joint space. Proximally: a horizontal line drawn above the soft tissue of the shoulder. Laterally: all lateral soft tissue included. Medially: a straight line drawn through the joint space just medial to the humeral head, starting near the convergence of arm and chest soft tissues in the axilla. Distally: an angled line drawn in the joint space connecting the ulnar and radial styloid processes through the wrist joint center. Proximally: an angled line drawn connecting the medial and lateral condyles of the humerus through the elbow joint center. Medially/laterally: all soft tissues included. Distally: a horizontal line drawn distal to all phalanges. Proximally: an angled line drawn connecting the medial and lateral condyles of the humerus through the elbow joint center. Medially/laterally: all soft tissues included. Distally: a horizontal line drawn parallel with the tibial plateau through the knee joint center. Proximally: an angled line drawn connecting the lateral aspect of the anterior superior iliac spine and the inferior ramus of the pubis, through the femoral head. Medially/laterally: all soft tissues included. Distally: an angled line drawn connecting the medial and lateral malleoli through the ankle joint center. Proximally: a horizontal line drawn parallel with the tibial plateau through the knee joint center. Medially/laterally: all soft tissues included. Distally: a horizontal line drawn distal to all phalanges. Proximally: a horizontal line drawn parallel with the tibial plateau through the knee joint center. Medially/laterally: all soft tissues included.
The largest within-measurer errors were found for the thigh segment (2.41%) and for fat mass (2.34%), respectively (Fig. 2). Collapsed across segment, the thigh segment had the lowest mean ICC at 0.996, while the leg and foot was found to have the highest mean ICC (0.998) (Table 3). Overall, there were relatively small differences in the magnitudes of the between and withinmeasurer errors (Fig. 2) and the ICCs for both the upper and lower extremities were similarly high in magnitude (Tables 2 and 3).
4. Discussion Manual segmentation of DXA scans resulted in soft and rigid tissue mass estimates from segments of both the upper and lower extremities that had relatively low errors between and withinmeasurers. The reliability of these measures was found to be very high for all tissue types and segments. Although the errors reported here were relatively small and the measurements very reliable, it is necessary to identify the sources
Percentage error (%)
Upper extremity Arm
Description
Between
7.0
Within
6.0 5.0 4.0 3.0 2.0 1.0 0.0 FM
LM
BMC
WM
Tissue 5.0
Percentage Error (%)
Segment ROI
8.0
Between Within
4.0 3.0 2.0 1.0 0.0 FM
LM
BMC
WM
Tissue Fig. 2. Mean (SD) between and within-measurer percentage (%) errors for each tissue type (FM ¼ fat mass; LM ¼ lean mass; BMC ¼ bone mineral content; WM ¼ wobbling mass ¼ FM+LM) of the upper (a) and lower (b) extremities. Percentage errors (%) are averaged across all segments.
Table 2 Summary of the mean segment tissue masses, and the mean absolute tissue differences between and within-measurers for the upper extremities. Tissue/segment
Mean (g)
Mean difference (g)
Mean (SEM) ICC
Between
Between
Within
Within
Fat mass (FM) Arm Arm with shoulder Forearm Forearm+hand
560 599 90 94
14 14 3 2
16 25 4 5
0.993 0.997 0.998 0.991
(14) (20) (4) (8)
0.998 0.995 0.996 0.994
(12) (22) (4) (8)
Lean mass (LM) Arm Arm with shoulder Forearm Forearm+hand
1695 1816 887 1111
37 52 41 24
50 75 22 22
0.999 0.996 0.994 0.998
(40) (54) (21) (19)
0.996 0.994 0.997 0.998
(42) (62) (20) (20)
BMC Arm Arm with shoulder Forearm Forearm+hand
92 99 73 97
6 6 5 3
3 5 2 3
0.993 0.992 0.991 0.996
(3) (4) (2) (2)
0.994 0.987 0.994 0.995
(3) (4) (2) (2)
Wobbling mass (WM) Arm Arm with shoulder Forearm Forearm+hand
2248 2416 978 1211
52 66 43 26
59 108 26 26
0.996 0.995 0.992 0.997
(52) (71) (24) (22)
0.996 0.992 0.996 0.998
(53) (80) (24) (24)
Mean (SEM) between-measurer ICCs for all tissue types and segments of the upper extremities are also included.
ARTICLE IN PRESS T.A. Burkhart et al. / Journal of Biomechanics 42 (2009) 1138–1142
Table 3 Summary of the mean segment tissue masses, and absolute tissue mass differences between and within-measurers for the lower extremities. Tissue/segment
Mean (g)
Mean difference (g)
Mean (SEM) ICC
Between
Between
Within
0.998 (76) 0.999 (9) 0.999 (14)
0.998 (68) 0.999 (11) 0.997 (17)
0.996 (129) 0.998 (20) 1.00 (83)
0.995 (125) 0.999 (22) 0.999 (19)
0.999 (3) 0.991 (3) 1.00 (2)
0.998 (3) 0.998 (3) 0.999 (2)
0.995 (194) 0.996 (25) 0.999 (22)
0.993 (183) 0.998 (30) 0.999 (27)
Within
Fat mass (FM) Thigh Leg Leg+foot
2813 664 736
59 9 7
79 11 19
Lean mass (LM) Thigh Leg Leg+foot
5415 2142 2611
122 42 9
156 25 21
262 214 293
1 8 1
3 3 2
Wobbling mass (WM) Thigh 8678 Leg 2790 Leg+foot 3564
180 50 10
233 33 31
BMC Thigh Leg Leg+foot
Mean (SEM) between-measurer ICCs for all tissue types and segments of the lower extremities are also included.
of error in an attempt to minimize between and within-measurer differences (Atkinson and Nevill, 1998; Burkhart et al., 2008; Olds, 2002). Two types of errors will be considered here; systematic bias and random error. Technical and procedural errors are both sources of systematic bias that may have contributed to measurement differences both between and within-measurers. The first source of systematic bias is DXA image resolution. Resolution quality may be affected by movement artifact that occurs when people make unwanted movements during the scanning process (Cawkwell, 1998). Slight movements could have impacted the quality of the images in this study, creating difficulties in accurately predicting soft tissue edges and joint spaces. However, this source of error was limited as much as possible by having very quick scans (5 min) and vigilant technicians. Another source of systematic error commonly cited is the level of expertise/training of the measurers (Atkinson and Nevill, 1998). While all three measurers received a standard amount of training, one measurer had less experience with this type of data analysis. While no research on expertise using DXA has specifically looked at manual segmentation, Lewiecki et al. (2006) comment on the importance of utilizing highly trained and qualified personnel to improve the quality of bone mineral density analysis. The most commonly cited source of random error and the largest contributor to measurement differences is biological variability (e.g. body composition, joint geometry) (Burkhart et al., 2008; Olds, 2002; Weir, 2005), which was found to be tissue-type dependent. Errors in the rigid tissue mass estimates (BMC) are caused in part by joint geometry and placement of the proximal and distal borders of the ROIs. The resulting images were two dimensional, providing a frontal view only. Therefore, when segmentation occurs through the joint centers, any bony landmarks on the posterior aspect of the segments were unseen to the measurer and may inconsistently be included in the ROI. This may explain why the mean percentage errors in the upper extremity were greater than those in the lower extremity where the joint space between the femoral condyles and the tibial plateau is generally wider and more easily discernable on the scans than the elbow joint space. Contrary to the rigid tissue mass errors, soft tissue measurement errors are more attributable to the placement of the medial
1141
and lateral borders of the ROIs. While a series of straight lines can be drawn around specific structures, a limitation of the imaging software used in this study is the inability to draw smooth curved lines. As a result, unwanted soft tissue may be included in a ROI when soft tissues from different body parts are pressed together in the image. This issue is exacerbated in individuals with greater body fat content and wider bodies, as less space would be available on the scanning bed. The upper extremity was most affected by this relationship, which further explains the greater percentage errors in upper extremity tissue masses when compared to the lower extremity. An analysis of the correlation between BMI and tissue error suggested that the wobbling mass estimates are more sensitive to the errors in the fat mass than lean mass, thereby confirming the main source of tissue misallocation between segments. The impact of factors such as body mass and overall body fat percentage on soft tissue mass estimation using manual ROI needs to be investigated further, since the range of body masses in this study was fairly limited compared to the general population. In summary, there were high ICCs and relatively low errors in tissue masses across the three measurers for all segments of the upper and lower extremities. These results clearly show that manual segmentation of DXA images using regions of interest is a robust and attractive method for determining the magnitudes of soft and rigid tissue masses of living people that can be used in biomechanical modeling research.
Conflict of interest There are no conflicts of interest associated with this research.
Acknowledgements Thanks to NSERC for funding, to the Diagnostic Imaging Department at Windsor Regional Hospital for the use of their facility, equipment, and technical support, and to Christopher Cindric and Alison Schinkel for their assistance with data analysis. References Arthurs, K., Andrews, D.M., 2009. Upper extremity soft and rigid tissue mass prediction using segment anthropometric measures and DXA. Journal of Biomechanics 42, 389–394. Atkinson, G., Nevill, A.M., 1998. Statistical methods for assessing measurement error (reliability) in variables relevant to sports medicine. Sports Medicine 26, 217–238. Bracco, D., Thiebaud, D., Chiolero, R.L., Lamdry, M., Burckhardt, P., Schutz, Y., 1996. Segmental body composition assessed by bioelectric impedance and DEXA in humans. Journal of Applied Physiology 81, 2580–2587. Burkhart, T.A., Arthurs, K.L., Andrews, D.M., 2008. Reliability of upper and lower anthropometric measurements and the effect on tissue mass predictions. Journal of Biomechanics 41, 1604–1610. Cawkwell, G.D., 1998. Movement artifact and dual X-ray absorptiometry. Journal of Clinical Densiometry 1, 141–147. Clarke, S., Wakely, C., Duddy, J., Watt, I., Ellington, K., Elson, C.J., Nikols, G., Kirwan, J.R., 2004. Dual-energy X-ray absorptiometry, applied to the assessment of tibial subchondral bone mineral density in osteoarthritis of the knee. Skeletal Radiology 33, 588–595. Clarys, J.P., Marfell-Jones, M.J., 1986. Anatomical Segmentation in humans and the prediction of segmental masses from intra-segmental anthropometry. Human Biology 58, 771–782. Dempster, W.T., 1955. Space requirements of the seated operator. WADC TR 55-159. Wright Peterson Air Force Base, Ohio. Durkin, J.L., Dowling, J.J., Andrews, D.M., 2002. The measurement of body segment inertial parameters using dual energy X-ray absorptiometry. Journal of Biomechanics 35, 1575–1580. Durkin, J.L., Dowling, J.J., 2003. Analysis of body segment parameter differences between four human populations and the estimation errors of four popular mathematical models. Journal of Biomechanical Engineering 125, 515–522. Fields, D.A., Goran, M.I., 2000. Body composition techniques and the fourcompartment model in children. Journal of Applied Physiology 89, 613–620.
ARTICLE IN PRESS 1142
T.A. Burkhart et al. / Journal of Biomechanics 42 (2009) 1138–1142
Fuller, N.J., Laskey, M.A., Elia, M., 1992. Assessment of the composition of major body regions by dual-energy X-ray absorptiometry (DEXA), with special reference to limb muscle mass. Clinical Physiology 12, 253–266. Gruber, K., Ruder, H., Denoth, J., Schneider, K., 1998. A comparative study of impact dynamics: wobbling mass model vs. rigid body models. Journal of Biomechanics 31, 439–444. Holmes, J.D., Andrews, D.M., Durkin, J.L., Dowling, J.J., 2005. Predicting in-vivo, soft tissue masses of the lower extremity using segment anthropometric measures and DXA. Journal of Applied Biomechanics 21, 371–382. Lewiecki, E.M., Binkley, N., Petak, S.M., 2006. DXA quality matters. Journal of Clinical Densiometry 9, 388–392. Olds, T., 2002. Five errors about error. Journal of Science and Medicine in Sport 5, 336–340. Portney, L.G., Watkins, M.P., 2000. Foundation of Clinical Research Applications to Practice, Second ed. Prentice-Hall Inc., New Jersey.
Prevral, S., Lu, Y., Gennant, H.K., Toschke, J.O., Shepherd, J.A., 2005. Towards standardization of dual X-ray absorptiometry (DXA) at the forearm. Calcified Tissue International 76, 348–354. Statistics Canada, 2003. Body Mass Index (BMI) by sex, household population aged 18 and over and excluding pregnant women, Canada, provinces, territories, health regions, and peer groups 2003. Retrieved on February 16, 2009, from /http://www.statscan.gc/pub182-221-x/00604/t/th/4150121-emg.htmS. Van Loan, M.D., 1998. Is dual-energy X-ray absorptiometry ready for prime time in the clinical evaluation of body composition? American Journal of Clinical Nutrition 68, 1155–1156. Weir, J.P., 2005. Quantifying test–retest reliability using the intraclass correlation coefficient and SEM. Journal of Strength and Conditioning Research 19, 231–240.
Contents lists available at ScienceDirect
Journal of Biomechanics journal homepage: www.elsevier.com/locate/jbiomech www.JBiomech.com
Short communication
Manual segmentation of DXA scan images results in reliable upper and lower extremity soft and rigid tissue mass estimates Timothy A. Burkhart a,b, Katherine L. Arthurs a, David M. Andrews a,b, a b
Department of Kinesiology, University of Windsor, Windsor, Ontario, Canada N9B 3P4 Department of Industrial and Manufacturing Engineering, University of Windsor, Windsor, Ontario, Canada
a r t i c l e in f o
a b s t r a c t
Article history: Accepted 18 February 2009
Quantification of segment soft and rigid tissue masses in living people is important for a variety of clinical and biomechanical research applications including wobbling mass modeling. Although Dualenergy X-ray Absorptiometry (DXA) is widely accepted as a valid method for this purpose, the reliability of manual segmentation from DXA scans using custom regions of interest (ROIs) has not been evaluated to date. Upper and lower extremity images of 100 healthy adults who underwent a full body DXA scan in the supine position were manually segmented by 3 measurers independently using custom ROIs. Actual tissue masses (fat mass, lean mass, bone mineral content) of the arm, arm with shoulder, forearm, forearm and hand, thigh, leg, and leg and foot segments were quantified bilaterally from the ROIs. There were significant differences between-measurers, however, percentage errors were relatively small overall (o1–5.98%). Intraclass correlation coefficients (ICCs) were very high between and withinmeasurers, ranging from 0.990 to 0.999 and 0.990 to 1.00 for the upper and lower extremities, respectively, suggesting excellent reliability. Between and within-measurer errors were comparable in general, and differences between the tissue types were small on average (maximum of 42 and 53 g for upper and lower extremities, respectively). These results suggest that manual segmentation of DXA images using ROIs is a reliable method of estimating soft and rigid tissues in living people. & 2009 Elsevier Ltd. All rights reserved.
Keywords: Reliability DXA Segmentation Upper extremity Lower extremity Tissue mass prediction
1. Introduction Accurate and reliable quantification of soft (muscle, fat) and rigid (bone mineral content) tissue masses in living people is very important for many biomechanical applications such as the modeling of human impacts using wobbling mass models (e.g. Gruber et al., 1998). Dual-energy X-ray Absorptiometry (DXA) has become a commonly adopted method for estimating soft and rigid tissue masses in-vivo (Cawkwell, 1998; Van Loan, 1998). Most DXA imaging analysis software packages provide automated regions of interest (ROI), but for limited clinical applications. Researchers can also obtain soft and rigid tissue mass information for other specific segments (Arthurs and Andrews, 2009; Durkin et al., 2002; Durkin and Dowling, 2003; Holmes et al., 2005) or sections of a segment (Clarke et al., 2004) by parceling them out via custom ROIs created by manually tracing polygons over the images. DXA is generally accepted as a valid method for quantifying body composition (Bracco et al., 1996; Fields and Goran, 2000; Fuller et al., 1992; Prevral et al., 2005). However, the reliability of segment tissue masses, including fat mass (FM), lean mass (LM), bone Corresponding author at: Department of Kinesiology, University of Windsor, Windsor, Ontario, Canada N9B 3P4. Tel.: +1 519 253 3000x2433; fax: +1 519 973 7056. E-mail address: [email protected] (D.M. Andrews).
0021-9290/$ - see front matter & 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.jbiomech.2009.02.017
mineral content (BMC), and wobbling mass (WM ¼ FM+LM), which result from the manual segmentation of DXA scan images using custom ROIs, has yet to be systematically documented between and within-measurers for a range of segments. Therefore, the main objective of this study was to quantify the between and withinmeasurer reliability of tissue mass estimates resulting from manual segmentation of the upper and lower extremities from DXA images.
2. Methods One hundred healthy adults (50 M, 50 F: mean age, mass, height of 21.876.2 years, 70.475.0 kg, 170.8719.4 cm, respectively) underwent a full body DXA scan (GE Lunar Prodigy Advance, System # DF+300481, GE Healthcare, Waukesha, Wisconsin) in a supine position with upper extremities by their sides. The mean (SD) BMI for all subjects was 24.0 (3.53) kg/m2 with a range of 17.0–34.0 kg/m2. These values are representative of people that vary from under weight to obese (Statistics Canada, 2003), and are therefore fairly indicative of the general population. The study was approved by the Research Ethics Boards of the University of Windsor and Windsor Regional Hospital and all participants provided written consent prior to participation. Using enCORE software (2006, GE Healthcare, v. 10.51.006), custom ROIs were traced manually over the DXA images to segment the upper and lower extremities (e.g. Fig. 1). ROI borders were adapted from Dempster (1955) and Clarys and Marfell-Jones (1986) (Table 1). Three trained measurers independently segmented each segment twice on the original scans (i.e. no ROIs or personal information remained on the scans to help the measurers), completing both extremities for all subjects before repeating (e.g. 100 upper limbs, then 100 lower limbs, then repeat).
ARTICLE IN PRESS T.A. Burkhart et al. / Journal of Biomechanics 42 (2009) 1138–1142
1139
Fig. 1. Sample regions of interest (ROI) on DXA scans showing the hard (bone) and soft (fat and muscle) tissue masses for an arm (a) arm with shoulder (b), and thigh (c) segment. The order in which the measurers segmented the upper and lower extremities was randomized. Fat mass, lean mass, bone mineral content and wobbling mass were quantified for each ROI. Measurement error was represented in two ways: as absolute between and within-measurer differences (g), and relative or percentage (%) error, expressed as the difference in tissue masses between trials (within-measurer error) or betweenmeasurers (between-measurer error), divided by the mean tissue masses of each segment (Burkhart et al., 2008). Intraclass correlation coefficients (ICCs) were used to quantify between and within-measurer reliability, and good to excellent reliability was accepted for ICCsX0.75 (Portney and Watkins, 2000). Finally, a Pearson correlation was performed to determine the relationship between subject body mass index (BMI) and the percentage error of the estimated tissue masses.
(Fig. 2), with maximum errors occurring in the forearm and hand segment (5.05%) and for BMC tissue (5.98%). In all cases, the mean between-measurer ICCs far exceeded the 0.75 criterion set as good to excellent reliability (0.990–0.999) (Table 2). Ninety one percent of the within-measurer errors were less than 5% (ranging from 1.9% to 6.2%). The largest segment and tissue mass errors were found for the arm with shoulder segment (4.45%) and fat mass (4.10%), respectively (Fig. 2). Overall, withinmeasurer ICCs were very high (ranging from 0.991 to 0.998). 3.2. Lower extremity
3. Results 3.1. Upper extremity The overall percentage errors across all segments and tissue types (1.12–5.98%) were relatively small between-measurers
Similar to the upper extremity, none of the between-measurer differences resulted in percentage errors greater than 5% (0.20–4.11%), with maximum errors occurring in the leg segment (2.24%) (Fig. 2). Averaged across segment, ICCs were very high and consistent, ranging from 0.995 (thigh) to 1.00 (leg and foot).
ARTICLE IN PRESS 1140
T.A. Burkhart et al. / Journal of Biomechanics 42 (2009) 1138–1142
Table 1 Description of the proximal/distal and medial/lateral borders that were used to define all regions of interest (ROIs) used by the measurers for manual segmentation (refer also to the sample images in Fig. 1).
Arm with shoulder
Forearm
Forearm and hand
Lower extremity Thigh
Leg
Leg and foot
Distally: a line drawn connecting the medial and lateral epicondyles of the humerus through the middle of the elbow joint space. Proximally: lines drawn around the humeral head through the shoulder joint space. Medially/laterally: all soft tissues included. Distally: a line drawn connecting the medial and lateral epicondyles of the humerus through the middle of the elbow joint space. Proximally: a horizontal line drawn above the soft tissue of the shoulder. Laterally: all lateral soft tissue included. Medially: a straight line drawn through the joint space just medial to the humeral head, starting near the convergence of arm and chest soft tissues in the axilla. Distally: an angled line drawn in the joint space connecting the ulnar and radial styloid processes through the wrist joint center. Proximally: an angled line drawn connecting the medial and lateral condyles of the humerus through the elbow joint center. Medially/laterally: all soft tissues included. Distally: a horizontal line drawn distal to all phalanges. Proximally: an angled line drawn connecting the medial and lateral condyles of the humerus through the elbow joint center. Medially/laterally: all soft tissues included. Distally: a horizontal line drawn parallel with the tibial plateau through the knee joint center. Proximally: an angled line drawn connecting the lateral aspect of the anterior superior iliac spine and the inferior ramus of the pubis, through the femoral head. Medially/laterally: all soft tissues included. Distally: an angled line drawn connecting the medial and lateral malleoli through the ankle joint center. Proximally: a horizontal line drawn parallel with the tibial plateau through the knee joint center. Medially/laterally: all soft tissues included. Distally: a horizontal line drawn distal to all phalanges. Proximally: a horizontal line drawn parallel with the tibial plateau through the knee joint center. Medially/laterally: all soft tissues included.
The largest within-measurer errors were found for the thigh segment (2.41%) and for fat mass (2.34%), respectively (Fig. 2). Collapsed across segment, the thigh segment had the lowest mean ICC at 0.996, while the leg and foot was found to have the highest mean ICC (0.998) (Table 3). Overall, there were relatively small differences in the magnitudes of the between and withinmeasurer errors (Fig. 2) and the ICCs for both the upper and lower extremities were similarly high in magnitude (Tables 2 and 3).
4. Discussion Manual segmentation of DXA scans resulted in soft and rigid tissue mass estimates from segments of both the upper and lower extremities that had relatively low errors between and withinmeasurers. The reliability of these measures was found to be very high for all tissue types and segments. Although the errors reported here were relatively small and the measurements very reliable, it is necessary to identify the sources
Percentage error (%)
Upper extremity Arm
Description
Between
7.0
Within
6.0 5.0 4.0 3.0 2.0 1.0 0.0 FM
LM
BMC
WM
Tissue 5.0
Percentage Error (%)
Segment ROI
8.0
Between Within
4.0 3.0 2.0 1.0 0.0 FM
LM
BMC
WM
Tissue Fig. 2. Mean (SD) between and within-measurer percentage (%) errors for each tissue type (FM ¼ fat mass; LM ¼ lean mass; BMC ¼ bone mineral content; WM ¼ wobbling mass ¼ FM+LM) of the upper (a) and lower (b) extremities. Percentage errors (%) are averaged across all segments.
Table 2 Summary of the mean segment tissue masses, and the mean absolute tissue differences between and within-measurers for the upper extremities. Tissue/segment
Mean (g)
Mean difference (g)
Mean (SEM) ICC
Between
Between
Within
Within
Fat mass (FM) Arm Arm with shoulder Forearm Forearm+hand
560 599 90 94
14 14 3 2
16 25 4 5
0.993 0.997 0.998 0.991
(14) (20) (4) (8)
0.998 0.995 0.996 0.994
(12) (22) (4) (8)
Lean mass (LM) Arm Arm with shoulder Forearm Forearm+hand
1695 1816 887 1111
37 52 41 24
50 75 22 22
0.999 0.996 0.994 0.998
(40) (54) (21) (19)
0.996 0.994 0.997 0.998
(42) (62) (20) (20)
BMC Arm Arm with shoulder Forearm Forearm+hand
92 99 73 97
6 6 5 3
3 5 2 3
0.993 0.992 0.991 0.996
(3) (4) (2) (2)
0.994 0.987 0.994 0.995
(3) (4) (2) (2)
Wobbling mass (WM) Arm Arm with shoulder Forearm Forearm+hand
2248 2416 978 1211
52 66 43 26
59 108 26 26
0.996 0.995 0.992 0.997
(52) (71) (24) (22)
0.996 0.992 0.996 0.998
(53) (80) (24) (24)
Mean (SEM) between-measurer ICCs for all tissue types and segments of the upper extremities are also included.
ARTICLE IN PRESS T.A. Burkhart et al. / Journal of Biomechanics 42 (2009) 1138–1142
Table 3 Summary of the mean segment tissue masses, and absolute tissue mass differences between and within-measurers for the lower extremities. Tissue/segment
Mean (g)
Mean difference (g)
Mean (SEM) ICC
Between
Between
Within
0.998 (76) 0.999 (9) 0.999 (14)
0.998 (68) 0.999 (11) 0.997 (17)
0.996 (129) 0.998 (20) 1.00 (83)
0.995 (125) 0.999 (22) 0.999 (19)
0.999 (3) 0.991 (3) 1.00 (2)
0.998 (3) 0.998 (3) 0.999 (2)
0.995 (194) 0.996 (25) 0.999 (22)
0.993 (183) 0.998 (30) 0.999 (27)
Within
Fat mass (FM) Thigh Leg Leg+foot
2813 664 736
59 9 7
79 11 19
Lean mass (LM) Thigh Leg Leg+foot
5415 2142 2611
122 42 9
156 25 21
262 214 293
1 8 1
3 3 2
Wobbling mass (WM) Thigh 8678 Leg 2790 Leg+foot 3564
180 50 10
233 33 31
BMC Thigh Leg Leg+foot
Mean (SEM) between-measurer ICCs for all tissue types and segments of the lower extremities are also included.
of error in an attempt to minimize between and within-measurer differences (Atkinson and Nevill, 1998; Burkhart et al., 2008; Olds, 2002). Two types of errors will be considered here; systematic bias and random error. Technical and procedural errors are both sources of systematic bias that may have contributed to measurement differences both between and within-measurers. The first source of systematic bias is DXA image resolution. Resolution quality may be affected by movement artifact that occurs when people make unwanted movements during the scanning process (Cawkwell, 1998). Slight movements could have impacted the quality of the images in this study, creating difficulties in accurately predicting soft tissue edges and joint spaces. However, this source of error was limited as much as possible by having very quick scans (5 min) and vigilant technicians. Another source of systematic error commonly cited is the level of expertise/training of the measurers (Atkinson and Nevill, 1998). While all three measurers received a standard amount of training, one measurer had less experience with this type of data analysis. While no research on expertise using DXA has specifically looked at manual segmentation, Lewiecki et al. (2006) comment on the importance of utilizing highly trained and qualified personnel to improve the quality of bone mineral density analysis. The most commonly cited source of random error and the largest contributor to measurement differences is biological variability (e.g. body composition, joint geometry) (Burkhart et al., 2008; Olds, 2002; Weir, 2005), which was found to be tissue-type dependent. Errors in the rigid tissue mass estimates (BMC) are caused in part by joint geometry and placement of the proximal and distal borders of the ROIs. The resulting images were two dimensional, providing a frontal view only. Therefore, when segmentation occurs through the joint centers, any bony landmarks on the posterior aspect of the segments were unseen to the measurer and may inconsistently be included in the ROI. This may explain why the mean percentage errors in the upper extremity were greater than those in the lower extremity where the joint space between the femoral condyles and the tibial plateau is generally wider and more easily discernable on the scans than the elbow joint space. Contrary to the rigid tissue mass errors, soft tissue measurement errors are more attributable to the placement of the medial
1141
and lateral borders of the ROIs. While a series of straight lines can be drawn around specific structures, a limitation of the imaging software used in this study is the inability to draw smooth curved lines. As a result, unwanted soft tissue may be included in a ROI when soft tissues from different body parts are pressed together in the image. This issue is exacerbated in individuals with greater body fat content and wider bodies, as less space would be available on the scanning bed. The upper extremity was most affected by this relationship, which further explains the greater percentage errors in upper extremity tissue masses when compared to the lower extremity. An analysis of the correlation between BMI and tissue error suggested that the wobbling mass estimates are more sensitive to the errors in the fat mass than lean mass, thereby confirming the main source of tissue misallocation between segments. The impact of factors such as body mass and overall body fat percentage on soft tissue mass estimation using manual ROI needs to be investigated further, since the range of body masses in this study was fairly limited compared to the general population. In summary, there were high ICCs and relatively low errors in tissue masses across the three measurers for all segments of the upper and lower extremities. These results clearly show that manual segmentation of DXA images using regions of interest is a robust and attractive method for determining the magnitudes of soft and rigid tissue masses of living people that can be used in biomechanical modeling research.
Conflict of interest There are no conflicts of interest associated with this research.
Acknowledgements Thanks to NSERC for funding, to the Diagnostic Imaging Department at Windsor Regional Hospital for the use of their facility, equipment, and technical support, and to Christopher Cindric and Alison Schinkel for their assistance with data analysis. References Arthurs, K., Andrews, D.M., 2009. Upper extremity soft and rigid tissue mass prediction using segment anthropometric measures and DXA. Journal of Biomechanics 42, 389–394. Atkinson, G., Nevill, A.M., 1998. Statistical methods for assessing measurement error (reliability) in variables relevant to sports medicine. Sports Medicine 26, 217–238. Bracco, D., Thiebaud, D., Chiolero, R.L., Lamdry, M., Burckhardt, P., Schutz, Y., 1996. Segmental body composition assessed by bioelectric impedance and DEXA in humans. Journal of Applied Physiology 81, 2580–2587. Burkhart, T.A., Arthurs, K.L., Andrews, D.M., 2008. Reliability of upper and lower anthropometric measurements and the effect on tissue mass predictions. Journal of Biomechanics 41, 1604–1610. Cawkwell, G.D., 1998. Movement artifact and dual X-ray absorptiometry. Journal of Clinical Densiometry 1, 141–147. Clarke, S., Wakely, C., Duddy, J., Watt, I., Ellington, K., Elson, C.J., Nikols, G., Kirwan, J.R., 2004. Dual-energy X-ray absorptiometry, applied to the assessment of tibial subchondral bone mineral density in osteoarthritis of the knee. Skeletal Radiology 33, 588–595. Clarys, J.P., Marfell-Jones, M.J., 1986. Anatomical Segmentation in humans and the prediction of segmental masses from intra-segmental anthropometry. Human Biology 58, 771–782. Dempster, W.T., 1955. Space requirements of the seated operator. WADC TR 55-159. Wright Peterson Air Force Base, Ohio. Durkin, J.L., Dowling, J.J., Andrews, D.M., 2002. The measurement of body segment inertial parameters using dual energy X-ray absorptiometry. Journal of Biomechanics 35, 1575–1580. Durkin, J.L., Dowling, J.J., 2003. Analysis of body segment parameter differences between four human populations and the estimation errors of four popular mathematical models. Journal of Biomechanical Engineering 125, 515–522. Fields, D.A., Goran, M.I., 2000. Body composition techniques and the fourcompartment model in children. Journal of Applied Physiology 89, 613–620.
ARTICLE IN PRESS 1142
T.A. Burkhart et al. / Journal of Biomechanics 42 (2009) 1138–1142
Fuller, N.J., Laskey, M.A., Elia, M., 1992. Assessment of the composition of major body regions by dual-energy X-ray absorptiometry (DEXA), with special reference to limb muscle mass. Clinical Physiology 12, 253–266. Gruber, K., Ruder, H., Denoth, J., Schneider, K., 1998. A comparative study of impact dynamics: wobbling mass model vs. rigid body models. Journal of Biomechanics 31, 439–444. Holmes, J.D., Andrews, D.M., Durkin, J.L., Dowling, J.J., 2005. Predicting in-vivo, soft tissue masses of the lower extremity using segment anthropometric measures and DXA. Journal of Applied Biomechanics 21, 371–382. Lewiecki, E.M., Binkley, N., Petak, S.M., 2006. DXA quality matters. Journal of Clinical Densiometry 9, 388–392. Olds, T., 2002. Five errors about error. Journal of Science and Medicine in Sport 5, 336–340. Portney, L.G., Watkins, M.P., 2000. Foundation of Clinical Research Applications to Practice, Second ed. Prentice-Hall Inc., New Jersey.
Prevral, S., Lu, Y., Gennant, H.K., Toschke, J.O., Shepherd, J.A., 2005. Towards standardization of dual X-ray absorptiometry (DXA) at the forearm. Calcified Tissue International 76, 348–354. Statistics Canada, 2003. Body Mass Index (BMI) by sex, household population aged 18 and over and excluding pregnant women, Canada, provinces, territories, health regions, and peer groups 2003. Retrieved on February 16, 2009, from /http://www.statscan.gc/pub182-221-x/00604/t/th/4150121-emg.htmS. Van Loan, M.D., 1998. Is dual-energy X-ray absorptiometry ready for prime time in the clinical evaluation of body composition? American Journal of Clinical Nutrition 68, 1155–1156. Weir, J.P., 2005. Quantifying test–retest reliability using the intraclass correlation coefficient and SEM. Journal of Strength and Conditioning Research 19, 231–240.