JOURNAL OF MAGNETIC RESONANCE IMAGING 40:239–247 (2014)
Technical Note
Manual Segmentation of Individual Muscles of the Quadriceps Femoris Using MRI: A Reappraisal Yoann Barnouin, MS,1 Gillian Butler-Browne, PhD,1 Thomas Voit, MD, PhD,1 €lle Leroux, MD,1 Anthony Behin, MD,1,2 David Reversat, MS,1 Noura Azzabou, PhD,1 Gae 3 Jamie S. McPhee, PhD, Pierre G. Carlier, MD, PhD,1,4 and Jean-Yves Hogrel, PhD1* Key Words: nuclear magnetic resonance imaging; segmentation; reliability; anatomical cross-sectional area; quadriceps femoris; muscle volume J. Magn. Reson. Imaging 2014;40:239–247. C 2013 Wiley Periodicals, Inc. V
Purpose: To propose a manual segmentation method for individual quadriceps femoris (QF) muscles and to test its reliability for muscle volume estimation. Materials and Methods: Images were acquired every 5 mm along the thigh using a 3T MRI scanner on 10 young (mean age: 25 years) and 10 older (mean age: 75 years) adults using a three-point 3D Dixon sequence. In each slice, anatomical cross-sectional areas of the individual quadriceps muscles of the dominant leg were outlined by two operators working independently. Differences between operators were assessed by means of Bland–Altman plots and intraclass correlation coefficients (ICC). This study was approved by the local Ethics Committee. Results: Precise delimitation of individual muscles along the femur often remains challenging, particularly near their insertion areas where some muscles may be partially or totally fused. There was, however, an excellent interoperator segmentation reliability despite a systematic significant difference between operators (ICC > 0.99), mainly due to delineation divergences. Considering all subjects and muscles, differences between operators were all lower than 4.4%. Conclusion: This work has demonstrated the excellent reliability of manual segmentation to assess crosssectional areas and therefore the volume of individual QF muscles using MRI. It may serve as a basis for a future segmentation consensus of the QF muscles.
1 Institut de Myologie, UPMC UM 76, INSERM U 974, CNRS UMR 7215, GH Piti e-Salp^ etrie`re, Paris, France. 2 AP-HP, Centre de R ef erence de Pathologies Neuromusculaire Paris Est, Institut de Myologie, GH Piti e-Salp^ etrie`re, Paris, France. 3 School of Healthcare Science, Manchester Metropolitan University, UK. 4 CEA, I2BM, MIRCen, IdM NMR Laboratory, GH Piti e-Salp^ etrie`re, Paris, France. Funded in part by the EU within the frame of the FP7 Project Myoage, contract No. 23576 and the Association Franc¸aise contre les Myopathies. *Address reprint requests to: J.-Y.H., Institut de Myologie, GH Piti eSalp^ etrie`re, 75651 Paris Cedex 13, France. E-mail: [email protected] Received November 9, 2012; Accepted July 14, 2013. DOI 10.1002/jmri.24370 View this article online at wileyonlinelibrary.com. C 2013 Wiley Periodicals, Inc. V
AS A POSSIBLE CAUSE of decreased muscle strength (along with a possible reduction of muscle quality), muscle size remains crucial to monitor in order to accurately assess changes in muscle quantity in response to exercise, ill health, immobilization, or aging (1). Computed tomography (2) and nuclear magnetic resonance imaging (MRI) (3) are considered reference standards for assessing muscle size because they generate detailed images and enable examination of all muscles acting on particular joints and 3D reconstruction, MRI having the advantage that it does not use ionizing radiation. Muscle volume is estimated from the acquisition of contiguous axial anatomical cross-sectional images, followed by offline analyses, which require time-consuming manual segmentation of each serial image using contour tracings of each individual muscle of interest (4). To reduce the cumbersome analysis and to speed up MRI-based muscle volume estimation, technicians have developed computerized semiautomated or fully automated segmentation methods (eg, 5–7). Even if these approaches have promising results, future work is warranted to ensure their routine use in variable anatomical situations taking into account within- and betweenindividual variability, particularly when muscles lie in close anatomical proximity, such as the quadriceps femoris (QF) muscles (8). The more laborious, but accurate, manual segmentation to measure muscle volume requires separate analysis of each individual muscle of interest in each serial axial-plane image, following the contours of the individual muscles set out by the muscle aponeurosis and fascia and omitting adipose tissue. In a typical adult, with a 5-mm interslice distance this would require analysis of about 100 images per subject for each individual muscle. In the QF, an additional challenge is to identify the borders between the four
239
240
individual muscles (rectus femoris [RF]; vastus intermedius [VI]; vastus medialis [VM] and vastus lateralis [VL]). For the RF and most of the time for the VM muscles, this is usually straightforward because their fascia and anatomical disposition are quite apparent. For the VL and VI, however, it may be much more difficult. Cadaver studies have shown that VI and VL are fused in some, but not all, individuals, particularly in the more proximal regions (in the deep portion of VL), with large individual variations in the amount of fusion (9). Even using high-resolution images, one may have difficulties in identifying the external borders for some muscles. Indeed, there is no consensus in the literature about which anatomical landmarks are correct to delineate individual QF muscles when they are apparently fused (10–13). Figure 1 illustrates such discrepancies that may occur between publications (Fig. 1a–d). The main objective of this work was to propose a manual segmentation method for individual QF muscles and to test its reliability for muscle volume estimation. A secondary objective aimed to compare the volume of individual QF muscles in a population of young and old subjects as a preliminary application.
Barnouin et al.
MATERIALS AND METHODS Experimental Subjects Twenty healthy volunteers (10 young and 10 older) who underwent muscle magnetic resonance imaging (MRI) investigation were selected from the population of subjects involved in the European project “MyoAge” (a collaborative EU FP7 project on understanding and combating age-related muscle weakness). Subjects were included in the study if they were found to fulfill the inclusion criteria set down by the project: free from any clinically relevant disease, not performing physical activity at a competition level, and not taking medications that are known to affect the study variables, such as corticosteroids, biotherapies, or nonsteroidal antiinflammatory drugs. Written informed consent was obtained before enrolment. This research was approved by the local Ethics Committee (CPP Paris-Ile de France VI, IB number: 2010-A006114-35). Image Acquisition Method Volumetric acquisitions of the two entire thighs were acquired on a 3T MRI scanner (Tim Trio, Siemens
Figure 1. Transverse MRI scans of the QF at 50% of femur length. VI: vastus intermedius, VL: vastus lateralis, VM: vastus medialis, RF: rectus femoris. Note the differences in VM and VI segmentation in a (10) and b (13) and in VI and VL segmentation in c (12) and d (11).
Manual Segmentation of Quadriceps Muscles
Healthcare, Erlangen, Germany) using a three-point Dixon gradient echo sequence (14), which allows decomposition of water and fat signals and differentiation of muscle and fat mass. To obtain a proton density weighting, a 10-msec repetition time and a 3 flip angle were chosen. The other acquisition parameters were 2.75/3.95/5.15 msec echo times, one excitation, 448 224 64 matrix, 448 224 320 mm3 field of view. The Dixon imaging technique was used because we were also interested in intramuscle fat quantification. In addition, for images where fat and water are out of phase (TE ¼ 3.95 msec), the separation between muscles is enhanced, which makes the segmentation task easier. Spin echo T1-weighted and T2-weighted images were also acquired but contours were less visible. A quadrature birdcage body coil was used for transmission and sets of phased-array receiver coils surrounded the thigh, with 3 6 flexible coils covering the segment and 3 6 coils embedded in the patient
241
table. No acceleration nor parallel imaging techniques were used. Two volumes were acquired to cover the anatomical structures, from iliac crests to the articular surface of the tibia. Subjects rested quietly in the magnet in a supine position with their legs and arms extended. Their heels were fixed on a nonmetallic support to avoid movements and to minimize compression of the lower limbs. All MRI data were transferred in DICOM format to a micro-PACS system and were analyzed using a standard image analysis software (Radionet v. 2.2.10., Scito, Paris, France). Muscle Segmentation Muscle anatomical cross sectional areas (ACSA) were measured on 5-mm thick contiguous axial out-ofphase images reconstructed from the 3D volumes. In each slice, ACSAs of the four QF muscles of the right leg were tracked independently by two operators (Y.B. and D.R.) trained by an experienced NMR
Figure 2. Axial out-of-phase Dixon images showing possible sources of segmentation errors for individual QF components. a: MR cross-sectional image at 50% of the femur length in a healthy woman age 27 years shows the caput breve of the biceps femoris which is included in hamstrings muscle and not in QF muscle. Images b–d are taken in the proximal part of QF. b: The iliacus muscle appears in the proximal part, whereas the VM begins to disappear. c: Some care must be taken not to segment the tensor fasciae latae as part of VL. d: When the sartorius starts to come closer to the RF, the aponeurosis between these two muscles is not readily distinguishable. Also, the central aponeurosis of the RF appears as a comma-shaped structure of comparable signal intensity located in the superficial portion of the cranial two-thirds of this bipennate muscle.
242
radiographer (30 years of experience, P.G.C.). All of the slices were outlined from the distal slice where the VM could first be visualized to the most proximal slice containing RF. Both operators were blinded to the results obtained by the other. Some of the reasons for discrepancies between studies, as exemplified in Fig. 1, are proposed in Fig.
Barnouin et al.
2. The presence of some visible intramuscular fascia instead of aponeurosis VI/VL (Fig. 2a) may result in an overestimation of VL volume and an underestimation of VI volume and consequently the relative contribution of individual muscles within the whole QF may be changed. Also, the caput breve of the biceps femoris should not be included in the VI/VL segmentation
Figure 3. Delimitation of the four components of the quadriceps femoris (RF, rectus femoris; VI, vastus intermedius; VL, vastus lateralis; and VM, vastus medialis) on axial MR images in a young woman (25 years old), where a represents axial images at 23%, b at 38%, c at 53%, d at 68%, e at 83%, and f 98% of femur length.
Manual Segmentation of Quadriceps Muscles
243
(Fig. 2a). This muscle belongs to the hamstrings but comes close to the QF in the first half of the thigh (from 25% to 60% of the femur length). In the proximal part of the thigh, the iliacus muscle appears where the VM fades (at about 80% of the femur length), which may give rise to some errors (Fig. 2b). At about the same level, the tensor fasciae latae appears on the upper exterior side of the VL (Fig. 2c). At about 60% of the femur length, the sartorius comes closer to the medial part of the RF. From 60% to 80% of the femur length, the aponeurosis between the sartorius and RF is difficult to distinguish (Fig. 2d) because the sartorius is running obliquely across the upper and anterior part of the thigh in an inferomedial direction. Referring to the central aponeurosis of this muscle may induce errors in the segmentation of the RF. Due to its bipennate pattern, a visible comma-shaped aponeurosis located in the superficial portion of the cranial two-thirds can lead to an underestimation of RF muscle volume if it is taken as an RF landmark. A segmentation procedure of QF is proposed in Fig. 3.
Muscle Volume Estimation The volume of muscle tissue per slice was calculated by multiplying the muscle tissue area by the interslice distance. The volumes of each of the QF components were calculated as the sum of all corresponding slice volumes. Visible fat or connective tissue was excluded as much as possible within the measurement region. Subcutaneous fat and bone were also eliminated from the regions of interest (ROIs). Each scan was accurately positioned relative to the femur length, which was measured as the distance from the bottom of the lateral condyle of the femur at the joint between the femur and the tibia (0% femur length) through to the top of the greater trochanter (100% femur length).
Bland–Altman plots in which the absolute difference between ACSA values of the two operators (operator A minus operator B) was plotted against the mean of the ACSA segmented by the two operators. Interoperator reproducibility from duplicate MRI measurements was assessed by using a two-tailed paired Student’s t-test. Intraclass correlation coefficients (ICCs, two-way random, absolute agreement, single measure) for ACSAs, volume measurements, and relative contribution of each QF component within the whole QF obtained by the two operators were completed to determine the reliability of measurements. The effect of age and gender on muscle volume was assessed using an analysis of variance (ANOVA). The level of significance was set to 0.05. RESULTS General Remarks A total of 5087 equidistant ACSAs were recorded along the length of the femur from 20 participants by two operators. The duration for the segmentation procedure was about 5 hours per subject. Table 1 presents the characteristics of the populations and the absolute and relative volume for each individual QF muscle and of the whole QF. Interoperator Reliability of Manual Segmentation Bland–Altman plots showed a systematic segmentation difference between operators confirmed statistically (Fig. 4 and Table 2), operator B obtaining higher ACSAs compared to operator A. ICCs ranged from 0.992 to 0.996, which underlines an excellent reliability of manual segmentation between the two operators. The mean relative difference between operators was lower than 4.4%.
Statistical Analyses
Interoperator Reliability of Muscle Volume Estimation
All data were analyzed with SPSS (v. 19, Chicago, IL). The data were first examined graphically using
The reliability was assessed for each individual QF component and for the whole QF (Table 3). Mean
Table 1 Subject Characteristics and Their Absolute Volumes of QF Muscles (cm3) and Corresponding Percentages (%) Young men n Age (year) Standing height (m) Body weight (kg) BMI (kg.m2) RF (cm3) (%) VI (cm3) (%) VL (cm3) (%) VM (cm3) (%) QF (cm3)
24.8 1.81 72.6 22.2 306.2 14.7 582.0 27.9 702.0 33.8 494.0 23.6 2084.2
5 6 6 6 6 6 6 6 6 6 6 6 6 6
2.6 0.05 6.0 2.6 61.6 1.8 90.2 1.8 76.4 1.1 85.8 1.3 287.3
Young women 5 24.7 6 1.67 6 59.1 6 21.1 6 204.7 6 14.1 6 417.3 6 28.8 6 485.5 6 33.4 6 344.2 6 23.7 6 1451.8 6
Values are means 6 SD; n: number of subjects; BMI: body mass index.
1.9 0.04 5.5 2.0 24.4 1.4 26.1 2.2 32.5 1.2 13.4 0.8 47.2
Old men 74.4 1.65 68.5 25.3 188.1 14.2 368.7 27.3 454.6 33.8 331.1 24.7 1342.5
4 6 6 6 6 6 6 6 6 6 6 6 6 6
4.1 0.03 4.0 2.3 35.4 2.6 79.3 1.5 83.0 1.9 54.1 0.9 227.0
Old women 74.8 1.56 58.2 23.9 120.6 12.6 258.7 27.2 323.9 33.9 250.7 26.3 953.9
6 6 6 6 6 6 6 6 6 6 6 6 6 6
2.8 0.02 5.3 2.2 14.9 1.4 13.4 1.0 29.7 1.3 20.6 1.4 60.3
244
Barnouin et al.
Figure 4. Bland–Altman plots depicting the difference between operators for ACSAs of a: RF, b: VI, c: VL, and d: VM.
interoperator difference results revealed that QF volume estimation was 3.1% lower in operator A compared with operator B (Fig. 5). As a consequence of recording higher ACSAs, the observed Bland–Altman plots and the negative mean interoperator differences suggest that operator B recorded higher volumes for each QF component in comparison to operator A (P < 0.0001 for RF, VL, VM, and whole QF), with the exception of the VI volume (P ¼ 0.114). For this last muscle, a mean interoperator difference of 3.5 cm3 representing 1.4% of mean volume was not significant in relation to the global VI volume. Notwithstanding these small but statistically significant systematic differences between operators, the ICCs observed for all muscle volume estimations were excellent, ranging from 0.988 to 0.997. Percentage differences in the contribution of each of the QF components between the two operators are presented in Table 4. Two-tailed paired t-test did not detect any significant difference in the estimation of the contribution of RF and VM in the whole QF between the two operators (P ¼ 0.052 and P ¼ 0.840, respectively) with a small mean interoperator differ-
ence (0.6% and 0.1% of mean percentages, respectively). However, statistical tests indicate significant interoperator differences in the estimation of the contribution of VI and VL in the whole QF between the two operators (P < 0.01) with a respective mean interoperator difference of 1.8% for VI and 1.3% for VL. ICC values were high for VL (0.836) and VI (0.880) and excellent for RF (0.995) and VM (0.951). Variations Among Groups of Subjects Gender and age had a significant effect on volume for each individual QF muscle and consequently for the whole QF (see Table 1). In our population, the diminution in QF volume with age was similar in both genders. Interestingly, the largest decrease with aging was observed in the RF volume compared with the other muscles, whereas VM volume seemed slightly less affected by aging. Standard deviations did not reveal large interindividual differences (less than 2.6%) within each group of subjects in the relative contribution of each individual muscle within the QF.
Table 2 Reliability of ACSA Estimates Between the Two Operators, Displayed as the Deviation of Values Recorded by Operator A From Those of Operator B Quadriceps muscles Number of ACSAs Mean interoperator difference (mm2) SD (mm2) P-value Mean difference (%) SD (%) ICC
RF
VI
VL
VM
1257 20.7 33.2
Technical Note
Manual Segmentation of Individual Muscles of the Quadriceps Femoris Using MRI: A Reappraisal Yoann Barnouin, MS,1 Gillian Butler-Browne, PhD,1 Thomas Voit, MD, PhD,1 €lle Leroux, MD,1 Anthony Behin, MD,1,2 David Reversat, MS,1 Noura Azzabou, PhD,1 Gae 3 Jamie S. McPhee, PhD, Pierre G. Carlier, MD, PhD,1,4 and Jean-Yves Hogrel, PhD1* Key Words: nuclear magnetic resonance imaging; segmentation; reliability; anatomical cross-sectional area; quadriceps femoris; muscle volume J. Magn. Reson. Imaging 2014;40:239–247. C 2013 Wiley Periodicals, Inc. V
Purpose: To propose a manual segmentation method for individual quadriceps femoris (QF) muscles and to test its reliability for muscle volume estimation. Materials and Methods: Images were acquired every 5 mm along the thigh using a 3T MRI scanner on 10 young (mean age: 25 years) and 10 older (mean age: 75 years) adults using a three-point 3D Dixon sequence. In each slice, anatomical cross-sectional areas of the individual quadriceps muscles of the dominant leg were outlined by two operators working independently. Differences between operators were assessed by means of Bland–Altman plots and intraclass correlation coefficients (ICC). This study was approved by the local Ethics Committee. Results: Precise delimitation of individual muscles along the femur often remains challenging, particularly near their insertion areas where some muscles may be partially or totally fused. There was, however, an excellent interoperator segmentation reliability despite a systematic significant difference between operators (ICC > 0.99), mainly due to delineation divergences. Considering all subjects and muscles, differences between operators were all lower than 4.4%. Conclusion: This work has demonstrated the excellent reliability of manual segmentation to assess crosssectional areas and therefore the volume of individual QF muscles using MRI. It may serve as a basis for a future segmentation consensus of the QF muscles.
1 Institut de Myologie, UPMC UM 76, INSERM U 974, CNRS UMR 7215, GH Piti e-Salp^ etrie`re, Paris, France. 2 AP-HP, Centre de R ef erence de Pathologies Neuromusculaire Paris Est, Institut de Myologie, GH Piti e-Salp^ etrie`re, Paris, France. 3 School of Healthcare Science, Manchester Metropolitan University, UK. 4 CEA, I2BM, MIRCen, IdM NMR Laboratory, GH Piti e-Salp^ etrie`re, Paris, France. Funded in part by the EU within the frame of the FP7 Project Myoage, contract No. 23576 and the Association Franc¸aise contre les Myopathies. *Address reprint requests to: J.-Y.H., Institut de Myologie, GH Piti eSalp^ etrie`re, 75651 Paris Cedex 13, France. E-mail: [email protected] Received November 9, 2012; Accepted July 14, 2013. DOI 10.1002/jmri.24370 View this article online at wileyonlinelibrary.com. C 2013 Wiley Periodicals, Inc. V
AS A POSSIBLE CAUSE of decreased muscle strength (along with a possible reduction of muscle quality), muscle size remains crucial to monitor in order to accurately assess changes in muscle quantity in response to exercise, ill health, immobilization, or aging (1). Computed tomography (2) and nuclear magnetic resonance imaging (MRI) (3) are considered reference standards for assessing muscle size because they generate detailed images and enable examination of all muscles acting on particular joints and 3D reconstruction, MRI having the advantage that it does not use ionizing radiation. Muscle volume is estimated from the acquisition of contiguous axial anatomical cross-sectional images, followed by offline analyses, which require time-consuming manual segmentation of each serial image using contour tracings of each individual muscle of interest (4). To reduce the cumbersome analysis and to speed up MRI-based muscle volume estimation, technicians have developed computerized semiautomated or fully automated segmentation methods (eg, 5–7). Even if these approaches have promising results, future work is warranted to ensure their routine use in variable anatomical situations taking into account within- and betweenindividual variability, particularly when muscles lie in close anatomical proximity, such as the quadriceps femoris (QF) muscles (8). The more laborious, but accurate, manual segmentation to measure muscle volume requires separate analysis of each individual muscle of interest in each serial axial-plane image, following the contours of the individual muscles set out by the muscle aponeurosis and fascia and omitting adipose tissue. In a typical adult, with a 5-mm interslice distance this would require analysis of about 100 images per subject for each individual muscle. In the QF, an additional challenge is to identify the borders between the four
239
240
individual muscles (rectus femoris [RF]; vastus intermedius [VI]; vastus medialis [VM] and vastus lateralis [VL]). For the RF and most of the time for the VM muscles, this is usually straightforward because their fascia and anatomical disposition are quite apparent. For the VL and VI, however, it may be much more difficult. Cadaver studies have shown that VI and VL are fused in some, but not all, individuals, particularly in the more proximal regions (in the deep portion of VL), with large individual variations in the amount of fusion (9). Even using high-resolution images, one may have difficulties in identifying the external borders for some muscles. Indeed, there is no consensus in the literature about which anatomical landmarks are correct to delineate individual QF muscles when they are apparently fused (10–13). Figure 1 illustrates such discrepancies that may occur between publications (Fig. 1a–d). The main objective of this work was to propose a manual segmentation method for individual QF muscles and to test its reliability for muscle volume estimation. A secondary objective aimed to compare the volume of individual QF muscles in a population of young and old subjects as a preliminary application.
Barnouin et al.
MATERIALS AND METHODS Experimental Subjects Twenty healthy volunteers (10 young and 10 older) who underwent muscle magnetic resonance imaging (MRI) investigation were selected from the population of subjects involved in the European project “MyoAge” (a collaborative EU FP7 project on understanding and combating age-related muscle weakness). Subjects were included in the study if they were found to fulfill the inclusion criteria set down by the project: free from any clinically relevant disease, not performing physical activity at a competition level, and not taking medications that are known to affect the study variables, such as corticosteroids, biotherapies, or nonsteroidal antiinflammatory drugs. Written informed consent was obtained before enrolment. This research was approved by the local Ethics Committee (CPP Paris-Ile de France VI, IB number: 2010-A006114-35). Image Acquisition Method Volumetric acquisitions of the two entire thighs were acquired on a 3T MRI scanner (Tim Trio, Siemens
Figure 1. Transverse MRI scans of the QF at 50% of femur length. VI: vastus intermedius, VL: vastus lateralis, VM: vastus medialis, RF: rectus femoris. Note the differences in VM and VI segmentation in a (10) and b (13) and in VI and VL segmentation in c (12) and d (11).
Manual Segmentation of Quadriceps Muscles
Healthcare, Erlangen, Germany) using a three-point Dixon gradient echo sequence (14), which allows decomposition of water and fat signals and differentiation of muscle and fat mass. To obtain a proton density weighting, a 10-msec repetition time and a 3 flip angle were chosen. The other acquisition parameters were 2.75/3.95/5.15 msec echo times, one excitation, 448 224 64 matrix, 448 224 320 mm3 field of view. The Dixon imaging technique was used because we were also interested in intramuscle fat quantification. In addition, for images where fat and water are out of phase (TE ¼ 3.95 msec), the separation between muscles is enhanced, which makes the segmentation task easier. Spin echo T1-weighted and T2-weighted images were also acquired but contours were less visible. A quadrature birdcage body coil was used for transmission and sets of phased-array receiver coils surrounded the thigh, with 3 6 flexible coils covering the segment and 3 6 coils embedded in the patient
241
table. No acceleration nor parallel imaging techniques were used. Two volumes were acquired to cover the anatomical structures, from iliac crests to the articular surface of the tibia. Subjects rested quietly in the magnet in a supine position with their legs and arms extended. Their heels were fixed on a nonmetallic support to avoid movements and to minimize compression of the lower limbs. All MRI data were transferred in DICOM format to a micro-PACS system and were analyzed using a standard image analysis software (Radionet v. 2.2.10., Scito, Paris, France). Muscle Segmentation Muscle anatomical cross sectional areas (ACSA) were measured on 5-mm thick contiguous axial out-ofphase images reconstructed from the 3D volumes. In each slice, ACSAs of the four QF muscles of the right leg were tracked independently by two operators (Y.B. and D.R.) trained by an experienced NMR
Figure 2. Axial out-of-phase Dixon images showing possible sources of segmentation errors for individual QF components. a: MR cross-sectional image at 50% of the femur length in a healthy woman age 27 years shows the caput breve of the biceps femoris which is included in hamstrings muscle and not in QF muscle. Images b–d are taken in the proximal part of QF. b: The iliacus muscle appears in the proximal part, whereas the VM begins to disappear. c: Some care must be taken not to segment the tensor fasciae latae as part of VL. d: When the sartorius starts to come closer to the RF, the aponeurosis between these two muscles is not readily distinguishable. Also, the central aponeurosis of the RF appears as a comma-shaped structure of comparable signal intensity located in the superficial portion of the cranial two-thirds of this bipennate muscle.
242
radiographer (30 years of experience, P.G.C.). All of the slices were outlined from the distal slice where the VM could first be visualized to the most proximal slice containing RF. Both operators were blinded to the results obtained by the other. Some of the reasons for discrepancies between studies, as exemplified in Fig. 1, are proposed in Fig.
Barnouin et al.
2. The presence of some visible intramuscular fascia instead of aponeurosis VI/VL (Fig. 2a) may result in an overestimation of VL volume and an underestimation of VI volume and consequently the relative contribution of individual muscles within the whole QF may be changed. Also, the caput breve of the biceps femoris should not be included in the VI/VL segmentation
Figure 3. Delimitation of the four components of the quadriceps femoris (RF, rectus femoris; VI, vastus intermedius; VL, vastus lateralis; and VM, vastus medialis) on axial MR images in a young woman (25 years old), where a represents axial images at 23%, b at 38%, c at 53%, d at 68%, e at 83%, and f 98% of femur length.
Manual Segmentation of Quadriceps Muscles
243
(Fig. 2a). This muscle belongs to the hamstrings but comes close to the QF in the first half of the thigh (from 25% to 60% of the femur length). In the proximal part of the thigh, the iliacus muscle appears where the VM fades (at about 80% of the femur length), which may give rise to some errors (Fig. 2b). At about the same level, the tensor fasciae latae appears on the upper exterior side of the VL (Fig. 2c). At about 60% of the femur length, the sartorius comes closer to the medial part of the RF. From 60% to 80% of the femur length, the aponeurosis between the sartorius and RF is difficult to distinguish (Fig. 2d) because the sartorius is running obliquely across the upper and anterior part of the thigh in an inferomedial direction. Referring to the central aponeurosis of this muscle may induce errors in the segmentation of the RF. Due to its bipennate pattern, a visible comma-shaped aponeurosis located in the superficial portion of the cranial two-thirds can lead to an underestimation of RF muscle volume if it is taken as an RF landmark. A segmentation procedure of QF is proposed in Fig. 3.
Muscle Volume Estimation The volume of muscle tissue per slice was calculated by multiplying the muscle tissue area by the interslice distance. The volumes of each of the QF components were calculated as the sum of all corresponding slice volumes. Visible fat or connective tissue was excluded as much as possible within the measurement region. Subcutaneous fat and bone were also eliminated from the regions of interest (ROIs). Each scan was accurately positioned relative to the femur length, which was measured as the distance from the bottom of the lateral condyle of the femur at the joint between the femur and the tibia (0% femur length) through to the top of the greater trochanter (100% femur length).
Bland–Altman plots in which the absolute difference between ACSA values of the two operators (operator A minus operator B) was plotted against the mean of the ACSA segmented by the two operators. Interoperator reproducibility from duplicate MRI measurements was assessed by using a two-tailed paired Student’s t-test. Intraclass correlation coefficients (ICCs, two-way random, absolute agreement, single measure) for ACSAs, volume measurements, and relative contribution of each QF component within the whole QF obtained by the two operators were completed to determine the reliability of measurements. The effect of age and gender on muscle volume was assessed using an analysis of variance (ANOVA). The level of significance was set to 0.05. RESULTS General Remarks A total of 5087 equidistant ACSAs were recorded along the length of the femur from 20 participants by two operators. The duration for the segmentation procedure was about 5 hours per subject. Table 1 presents the characteristics of the populations and the absolute and relative volume for each individual QF muscle and of the whole QF. Interoperator Reliability of Manual Segmentation Bland–Altman plots showed a systematic segmentation difference between operators confirmed statistically (Fig. 4 and Table 2), operator B obtaining higher ACSAs compared to operator A. ICCs ranged from 0.992 to 0.996, which underlines an excellent reliability of manual segmentation between the two operators. The mean relative difference between operators was lower than 4.4%.
Statistical Analyses
Interoperator Reliability of Muscle Volume Estimation
All data were analyzed with SPSS (v. 19, Chicago, IL). The data were first examined graphically using
The reliability was assessed for each individual QF component and for the whole QF (Table 3). Mean
Table 1 Subject Characteristics and Their Absolute Volumes of QF Muscles (cm3) and Corresponding Percentages (%) Young men n Age (year) Standing height (m) Body weight (kg) BMI (kg.m2) RF (cm3) (%) VI (cm3) (%) VL (cm3) (%) VM (cm3) (%) QF (cm3)
24.8 1.81 72.6 22.2 306.2 14.7 582.0 27.9 702.0 33.8 494.0 23.6 2084.2
5 6 6 6 6 6 6 6 6 6 6 6 6 6
2.6 0.05 6.0 2.6 61.6 1.8 90.2 1.8 76.4 1.1 85.8 1.3 287.3
Young women 5 24.7 6 1.67 6 59.1 6 21.1 6 204.7 6 14.1 6 417.3 6 28.8 6 485.5 6 33.4 6 344.2 6 23.7 6 1451.8 6
Values are means 6 SD; n: number of subjects; BMI: body mass index.
1.9 0.04 5.5 2.0 24.4 1.4 26.1 2.2 32.5 1.2 13.4 0.8 47.2
Old men 74.4 1.65 68.5 25.3 188.1 14.2 368.7 27.3 454.6 33.8 331.1 24.7 1342.5
4 6 6 6 6 6 6 6 6 6 6 6 6 6
4.1 0.03 4.0 2.3 35.4 2.6 79.3 1.5 83.0 1.9 54.1 0.9 227.0
Old women 74.8 1.56 58.2 23.9 120.6 12.6 258.7 27.2 323.9 33.9 250.7 26.3 953.9
6 6 6 6 6 6 6 6 6 6 6 6 6 6
2.8 0.02 5.3 2.2 14.9 1.4 13.4 1.0 29.7 1.3 20.6 1.4 60.3
244
Barnouin et al.
Figure 4. Bland–Altman plots depicting the difference between operators for ACSAs of a: RF, b: VI, c: VL, and d: VM.
interoperator difference results revealed that QF volume estimation was 3.1% lower in operator A compared with operator B (Fig. 5). As a consequence of recording higher ACSAs, the observed Bland–Altman plots and the negative mean interoperator differences suggest that operator B recorded higher volumes for each QF component in comparison to operator A (P < 0.0001 for RF, VL, VM, and whole QF), with the exception of the VI volume (P ¼ 0.114). For this last muscle, a mean interoperator difference of 3.5 cm3 representing 1.4% of mean volume was not significant in relation to the global VI volume. Notwithstanding these small but statistically significant systematic differences between operators, the ICCs observed for all muscle volume estimations were excellent, ranging from 0.988 to 0.997. Percentage differences in the contribution of each of the QF components between the two operators are presented in Table 4. Two-tailed paired t-test did not detect any significant difference in the estimation of the contribution of RF and VM in the whole QF between the two operators (P ¼ 0.052 and P ¼ 0.840, respectively) with a small mean interoperator differ-
ence (0.6% and 0.1% of mean percentages, respectively). However, statistical tests indicate significant interoperator differences in the estimation of the contribution of VI and VL in the whole QF between the two operators (P < 0.01) with a respective mean interoperator difference of 1.8% for VI and 1.3% for VL. ICC values were high for VL (0.836) and VI (0.880) and excellent for RF (0.995) and VM (0.951). Variations Among Groups of Subjects Gender and age had a significant effect on volume for each individual QF muscle and consequently for the whole QF (see Table 1). In our population, the diminution in QF volume with age was similar in both genders. Interestingly, the largest decrease with aging was observed in the RF volume compared with the other muscles, whereas VM volume seemed slightly less affected by aging. Standard deviations did not reveal large interindividual differences (less than 2.6%) within each group of subjects in the relative contribution of each individual muscle within the QF.
Table 2 Reliability of ACSA Estimates Between the Two Operators, Displayed as the Deviation of Values Recorded by Operator A From Those of Operator B Quadriceps muscles Number of ACSAs Mean interoperator difference (mm2) SD (mm2) P-value Mean difference (%) SD (%) ICC
RF
VI
VL
VM
1257 20.7 33.2
