Lipids (2008) 43:97–103 DOI 10.1007/s11745-007-3121-x
METHODS
Fourier Transform Near Infrared Spectroscopy: A Newly Developed, Non-Invasive Method To Measure Body Fat Non-invasive body fat content measurement using FT-NIR H. Azizian Æ J. K. G. Kramer Æ S. B. Heymsfield Æ S. Winsborough
Received: 25 July 2007 / Accepted: 7 September 2007 / Published online: 6 November 2007 Ó AOCS 2007
Abstract An FT-NIR technique is reported to provide a fast, accurate and low cost method of determining in-vivo human body fat content. The body fat content of 353 healthy subjects (154 males and 199 females) of various height, weight, and age were measured by FT-NIR and compared to 420 subjects investigated by magnetic resonance imaging (MRI). The procedure involved scanning each subject’s upper ear that provided a necessary reflectance surface and proved representative of the subject’s subcutaneous fat content. The average FT-NIR spectrum was compared to a reference mixture with known and similar fat content and composition to that of humans. The FT-NIR response was incorporated into an empirical equation using the ratio of subcutaneous to total body fat from MRI data, taking into account the subject’s gender, height, weight and age. The results on the two data sets were similar and demonstrated that the FT-NIR technique can be used to obtain a measure of the body fat content of individuals, similar to that using MRI. In addition, the H. Azizian (&) S. Winsborough NIR Technologies Inc., 1312 Fairmeadow Trail, Oakville, ON, Canada L6M 2M2 e-mail:
[email protected] J. K. G. Kramer Food Research Program, Agriculture and Agri-Food Canada, Guelph, ON, Canada S. B. Heymsfield Body Composition Laboratory and Weight Control Unit, St. Luke’s-Roosevelt Hospital Center, New York, NY 10025, USA Present Address: S. B. Heymsfield Merck and Co., Inc., Clinical Research, Metabolism, Whitehouse Station, NJ, USA
FT-NIR was used to more accurately monitor the fat content of sleep apnea patients. Keywords Body fat Fourier transform near infrared spectroscopy Obstructive sleep apnea Human Magnetic resonance imaging Obesity
Introduction According to a recent Statistics Canada Report 57% of men and 40% of women in Canada are considered overweight or obese1 [1]. Overweight is defined by a body mass index (BMI) of [25, and obesity [30, and it is linked to several diseases, including hypertension, diabetes mellitus, hyperlipidemia, coronary artery disease, obstructive sleep apnea, and cancers of the breast, uterus, prostate and colon [1, 2]. The BMI guidelines assume that body mass is closely associated with body fatness [3]. However, some overweight individuals are not necessarily overfat (e.g., bodybuilders), while others having a normal BMI have a high percentage of their body weight as fat [3]. This suggests that we need a more reliable indicator of obesity for actual body fat measurement. BMI is one of several ways to determine body fat or monitor obesity. Other techniques include skinfold thickness, circumference, bioelectrical impedance, dual energy X-ray absorptiometry, underwater weighing, computed tomography, and near-infrared interactance. The vast majority of these measurements are indirect and based on 1
Body mass index (BMI), by sex, household population aged 18 and over excluding pregnant women, Canada, provinces, territories, health regions and peer groups, 2003. http://www.statcan.ca/english/ freepub/82–221-XIE/00604/tables/html/1228_03.htm.
123
98
assumptions and models. Magnetic resonance imaging (MRI) has been used as a more reliable technique, but it is expensive and inaccessible. As a result, we have developed a new technique to directly scan and evaluate subcutaneous body fat content in humans using Fourier Transform near infrared (FT-NIR) spectroscopy. The FT-NIR response can be compared to reference materials with known fat content and composition or used in an empirical equation with height, weight, gender and age as covariates. The use of near infrared spectroscopy to determine fat content in humans was first reported in 1984 [4]. Ruchti et al. [5] reported the use of bovine fat as a reference material in their NIR determinations of the human fat content. We have shown that there are significant differences between bovine and human fat in composition, and since the FT-NIR signal is based on the fatty acid profile of the fat, a more appropriate reference material was developed that more closely resembled the human fatty acid composition [6]. Recent improvements in the resolution of FT-NIR spectroscopy coupled with chemometric analysis has allowed for the complete fatty acid profiling of fats and oils including trans fatty acids [7–9].
Methods Subjects Data were obtained from five different locations in Ontario: Mississauga City Hall, Mississauga South Common Mall, Mississauga River Grove Community Centre, Mississauga Meadowvale Community Centre, and Fitness Corner, Port Elgin. A total of 353 subjects (199 females and 154 males) were scanned, and their height, weight, and age recorded (as provided by the subjects). All subjects were required to read the information sheet describing the procedure and then sign a consent form.
Lipids (2008) 43:97–103
light was held against the back of the subjects’ upper ear to avoid eye contact; see demonstration of technique in [10]. On average five measurements of five scans each per subject were taken (for a total scan time of 25 s) and the resulting absorption spectra were averaged. The averaged spectrum was then integrated focusing on the frequencies associated with the fat peaks to obtain the FT-NIR response. This response proved to be related to subcutaneous fat of the subjects, and was matched against the FT-NIR response of the reference material with a known fat content [6].
FT-NIR Reference Material Preliminary results indicated that bovine fat used by Ruchti et al. [5] did not resemble human body fat composition and hence could not be used as a FT-NIR reference. Figure 1 shows the second derivative spectra of human, reference material and bovine fat; the latter showed a more pronounced peak for saturated fatty acids and less unsaturated fatty acids compared to human fat. In addition, a clear chemical shift of about 8 cm–1 wavenumbers was evident in the bovine fat at 5,777 cm–1 compared to human fat spectrum at 5,785 cm–1. This chemical shift and the profile differences have a significant effect on the quantitative determination in near infrared spectroscopy. A more suitable reference material was prepared consisting of water, fat, and protein, which more closely resembled human body fat profile (Fig. 1). To generate a calibration curve, the FT-NIR response of increasing amounts of fat in the reference material was measured (Fig. 2). These results were used to obtain the calibration curve (Fig. 3). The underlying protein signal from the cartilage was established by purchasing pig’s ears and measuring the FT-NIR
FT-NIR Methods and Procedures A Matrix-F FT-NIR Spectrometer equipped with a standard fibre optic probe from Bruker Optics in combination with OPUS (Optics Users Software) software was used to obtain the spectra. Several body parts were examined to obtain reliable and intense spectra. Most abdominal and arm measurements proved unusable because of high dispersion and low signal to noise ratio. It was found that the upper ear yielded the best spectral characteristics for this study with the least amount of background noise and water interference, and strong reflectance signal. The resulting absorption spectra were a composite of different components in the sample. The fibre optic probe carrying the near infrared
123
Fig. 1 Second derivative spectra for bovine, human, and the reference material: dashed line, bovine; dotted line, reference 24%; solid line, human 23.5%
Lipids (2008) 43:97–103
99
Magnetic Resonance Imaging Methods and Procedures Methods and procedures for the collection of MRI data were published [13].
Results
Fig. 2 FT-NIR response in the fat region of the reference materials containing 0–51% fat: solid line 0%; dotted line 5%; dashed line 10%; solid line 24%; dotted line 33%; dashed line 41%; solid line 51%
spectra with the skin plus fat layer, and after removal of the outer layers, recording the signal of the cartilage.
FT-NIR Empirical Equation The fat content measurement results using the reference material with known fat content were verified by developing an empirical equation. This equation was based on the existing body surface area equations [11, 12] and the ratio of subcutaneous to total fat content from MRI studies taking gender and age into consideration, which were combined with the FT-NIR response. It was assumed that the total body fat volume could be determined by measuring the body surface area multiplied by the thickness of subcutaneous fat layer. The values of the FT-NIR response were substituted for the thickness of subcutaneous fat layer in these equations.
Representative values for both female and male subjects are shown in Table 1. Their body fat content was measured by FT-NIR spectroscopy and determined using the calibration curve (Fig. 3). A number of subjects were compared according to height, weight, age, BMI and fitness level to observe differences in their fat content (Table 2). One pair is shown in Fig. 4, which shows the second derivative spectrum for male subjects 1007 and 1010. Subject 1007 was involved in bodybuilding activities, while subject 1010 is an average male. Both subjects were of a similar weight and height with a similar BMI. The FT-NIR spectrum for subject 1010 in Fig. 4 shows more intense fat peaks than that for subject 1007, indicating a higher body fat content. The data in Table 2 demonstrate that BMI is an unreliable measure of body fat content, and confirms the findings by other that fat percentage measurements are more informative than BMI [3]. An empirical equation was developed based on the assumption that subcutaneous fat thickness, substituted by FT-NIR response and multiplied by body surface area generates subcutaneous fat volume. The surface area was determined using existing body surface area equations [11, 12]. The ratio of subcutaneous to total adipose tissue was obtained from MRI data that was necessary to convert the FT-NIR response (subcutaneous) to total body fat content. Figure 5 shows the MRI data for subcutaneous/ total adipose tissue ratio with respect to age and gender. As can be seen, at a younger age, almost 95% of the adipose tissue for males and 97% for females is stored subcutaneously. A separate empirical equation for males and females was developed using the body surface area equations and MRI ratio of subcutaneous to total adipose tissue, and incorporating a factor for the FT-NIR response. These equations are shown below: Female
Male
Fig. 3 Calibration curve for fat content determination
TBF ¼
TBF ¼
64:719N W0:51456 H0:42246 ð0:001A þ 0:989ÞW
64:719N W0:51456 H0:42246 ð0:003A þ 0:997ÞW
TBF total body fat; N NIR response; W weight in kg; H height in cm; A age in years. The scanned data for 353 subjects were re-analyzed for body fat content using the FT-NIR empirical equations, and
123
100
Lipids (2008) 43:97–103
Table 1 Body fat content of representative female or male subjects using FT-NIR spectroscopy
Mean and standard deviations for all females were: age 43 ± 14 years; height 163 ± 7 cm; weight 64 ± 10 kg; FT-NIR response 0.39 ± 0.11; BMI 24 ± 4; body fat content by FT-NIR 29 ± 8%. Mean and standard deviations for all males were: age 39 ± 16 years; height 178 ± 7 cm; weight 82 ± 13 kg; FT-NIR response 0.22 ± 0.09; BMI 26 ± 3; body fat content by FT-NIR 16 ± 7%. The percent (%) body fat content for each of the subjects was calculated using the equations and the reference material
ID No.
Gender
Age
Height (cm)
Weight (kg)
FT-NIR response
BMI
FT-NIR Fat % equation
FT-NIR Fat % reference
1002
F
24
165
61.4
0.46
23
36
33
1006
F
21
163
67.3
0.38
25
28
27
1009
F
27
163
72.0
0.26
27
19
18
1011
F
47
156
60.5
0.35
25
28
25
1020
F
43
165
68.0
0.71
25
54
52
1022
F
56
168
85.9
0.45
30
31
33
1025 1027
F F
48 39
173 165
68.2 72.7
0.50 0.38
23 27
39 28
36 27
1033
F
46
163
62.0
0.64
23
51
47
1036
F
44
155
59.1
0.24
25
19
17
1038
F
37
170
61.4
0.50
21
40
36
1039
F
51
155
72.7
0.41
30
30
30
1001
M
49
174
79.1
0.26
26
19
19
1005
M
21
183
79.5
0.14
24
10
9
1007
M
25
188
93.0
0.07
26
5
4
1010
M
23
191
95.0
0.28
26
19
20
1014
M
22
185
90.0
0.23
26
16
16
1018
M
24
170
63.6
0.20
22
16
14
1019
M
24
178
95.4
0.31
30
20
22
1023
M
40
178
75.0
0.29
24
22
21
1029
M
42
198
78.8
0.31
20
24
22
1031 1034
M M
31 21
175 185
77.3 88.6
0.23 0.29
25 26
17 20
16 21
1042
M
60
173
78.0
0.20
26
15
14
comparing the results to those using the FT-NIR reference material. Setting the intercept at 0, a significant linear correlation was found (R2 = 0.94) tested using a two-tailed paired t test (P \ 0.0001), as shown in Fig. 6 (all data). Table 1 includes the determination of percent (%) body fat using the reference material and the empirical equation for each of the representative male and female subjects (the last two columns). The results using the empirical equation were
dependent upon subjects accurately reporting their height and weight, which may explain some minor discrepancies between the two data sets. These equations are very sensitive to age, weight and the NIR response. For example a change of 0.02 in the NIR response would result in 1% drop in body fat %, and in some minor cases up to 2% drop in both female and male subjects. The same is true for the age coefficient. Switching the empirical equation for males and females
Table 2 Effect of physical activity on subject’s body fat content ID No.
Gender
Age
Height (cm)
Weight (kg)
Activity level
1007
M
25
188
93
High—body building
0.07
26
4
1010
M
23
191
95
Average
0.28
26
20
1005
M
21
183
80
High—body building
0.14
24
10
1131
M
18
183
80
Active
0.24
24
17
1156
M
22
175
82
High—fitness training
0.12
27
8
1132
M
23
173
80
Active
0.23
27
16
1201 1002
F F
26 24
165 165
61 61
Active Average
0.29 0.46
22 22
21 34
1211
F
25
157
55
Active
0.26
22
19
1015
F
24
158
55
Average
0.43
22
32
123
FT-NIR fat response
BMI
FT-NIR Fat %
Lipids (2008) 43:97–103
Fig. 4 Second derivative comparisons for two subjects with the same BMI but different body fat content: dotted line, volunteer 1007; Solid line, volunteer 1010
Fig. 5 Subcutaneous to total fat ratio with respect to age for males and females as determined by MIR: SAT, Subcutaneous Adipose Tissue; TAT, Total Adipose Tissue; MRI, magnetic resonance imaging
result in a fat % difference of up to 2%. This is due to the difference in the slope of subcutaneous to total ratio in males and females (see Fig. 5). Figure 6 also shows that the majority of male subjects are located at the lower end of the scale, whereas female subjects are located at the higher end of the scale confirming that, on average, females have more body fat than males [3]. A comparison between the FT-NIR data for 353 subjects in Ontario and MRI data for 420 subjects from the St. Luke’s-Roosevelt Hospital in New York was subsequently performed. The FT-NIR and MRI data for 39 pairs of subjects were matched according to gender, age, height and weight. Table 3 shows representative comparisons for 10 female pairs and 10 male pairs. The R2 value for all 39 pairs was 0.96 with a two-tailed P value of P = 0.0468. The last two columns in Table 3 show the % fat content measured by the FT-NIR and MRI for the paired subjects. The only discrepancies would be similar to the example
101
Fig. 6 Comparison of two FT-NIR methods for determining fat content for males and females
provided in which a body builder is compared to an average person with similar weight and height (Table 2). Although the two subject groups were measured using two different techniques (FT-NIR and MRI) the results show a remarkable resemblance in body fat content. The comparison of these two groups was to some extent justified since both groups were from North America with similar life styles and diets. The FT-NIR technique was recently applied to also monitor body fat content of sleep apnea patients. The patient was diagnosed with obstructive sleep apnea having an Apnea Hypopnea Index (AHI) of 15.2 (event/h); below 10 is considered normal. The person’s weight was 81.8 kg and height 1.74 m for a BMI of 27.0. The patient then underwent a weight loss program losing 6.8 kg, which corresponded to a 9% drop in body fat content as measured by FT-NIR (25–16%), or a decrease in BMI from 27.0– 24.8. A subsequent obstructive sleep apnea test showed a marked improvement in the AHI down from 15.2 to 2.5 (event/h).
Discussion The body fat content of 353 subjects was measured by FT-NIR spectroscopy and determined using the calibration curve. The results were confirmed by calculating the fat volume using an empirical equation that was developed based on substituting the FT-NIR response for the subcutaneous fat thickness and the body surface area. A comparison between FT-NIR and MRI data for 39 pairs of subjects were matched according to gender, age, height and weight (R2 = 0.96, P = 0.0468). Both groups were from North America with similar life styles and diets.
123
102
Lipids (2008) 43:97–103
Table 3 Comparison of FT-NIR and MRI results for selected female or male subjects Gender ID No.
Age Height Weight BMI Fat Fat % (cm) (kg) % by by MRI FT-NIR
F1
NIR 1141
17
157
56
23
35
–
MRI 153
20
158
58
23
–
35
NIR 1202
18
165
58
21
20
–
MRI 190
23
162
51
19
–
20
NIR 1285
25
168
57
20
25
–
MRI 146
24
163
59
22
–
23
NIR 1149
22
170
64
22
27
–
MRI 166
24
170
65
22
–
28
NIR 1370
27
170
55
19
28
–
MRI 105
27
170
55
19
–
28
NIR 1060
27
160
68
27
30
–
F2 F3 F4 F5 F6
MRI 1184
29
158
65
26
–
29
F7
NIR 1191
30
156
59
24
30
–
F8
MRI 233 NIR 1204
30 36
157 163
56 61
23 23
– 28
31 –
MRI 313
36
160
58
23
–
29
NIR 1078
42
168
74
26
37
–
MRI 0329
42
167
73
26
–
37
NIR 1253
66
168
61
22
31
–
MRI 267
70
164
61
23
–
32
NIR 1323
20
180
68
21
17
–
MRI 202
20
183
73
22
–
17
NIR 1375
20
180
70
22
8
–
MRI 336
22
181
72
22
–
8
NIR 1010
23
191
95
26
21
–
MRI 17816 25
185
96
28
–
22
NIR 1018
24
170
64
22
14
–
MRI 0011
F9 F10 M1 M2 M3 M4
25
173
65
22
–
13
M5
NIR 1017 25 MRI 17486 25
183 187
68 70
20 20
10 –
– 11
M6
NIR 1195
25
180
80
25
11
–
MRI 32
26
178
78
25
–
12
NIR 1031
31
175
77
25
17
–
MRI 129
31
176
74
24
–
18
NIR 1059
37
180
86
27
18
–
MRI 0111
38
181
88
27
–
19
NIR 1123
40
178
86
27
19
–
MRI 2
38
181
88
27
–
19
NIR 1001
49
174
79
26
19
–
MRI 415
49
174
78
26
–
20
M7 M8 M9 M10
The newly developed method based on FT-NIR shows the potential for a quick, accurate and relatively inexpensive determination of human body fat content. Previous NIR studies were not successful since dispersive NIR technology
123
and inappropriate reference materials were used [4, 5]. The discovery of using the back of the upper ear providing a suitable reflectance surface for the NIR light, as well as, the finding that the fat layer was representative of the subcutaneous fat in the human subjects proved to be important steps in this research. The reflectance of NIR light could not be obtained by measuring other body parts due to dispersion of the NIR light. The MRI data established that a high percentage of body fat was stored subcutaneously (Fig. 5), which was used to correlate the subcutaneous fat layer on the back of the upper ear to the whole body fat content. The significant correlation of the FT-NIR to the MRI data suggests that the upper ear was representative of the whole body fat content when measured using FT-NIR spectroscopy. Although, the FT-NIR and MRI tests were performed on different subjects at different locations and times, the correlation between the results of the two techniques is notable and the similarities are gender neutral with no obvious differences for scanning male or female subjects. The similarity of life styles and diets between the two study groups allowed for a comparison of the two techniques. The ultimate test of acceptance for the FT-NIR technique would be a direct validation using both techniques and several groups of male and female subjects, old and young, normal and obese. These results demonstrate the potential of using FT-NIR to measure the body fat content of human subjects rapidly and accurately providing data comparable to the more costly and less accessible MRI technique. The FT-NIR technique is a low cost, portable and safe method that could prove valuable to determine the fat content of individuals being treated for obesity and sleep apnea and to more accurately monitor weight loss. The FT-NIR method has already been used for fatty acid profiling of fats and oils and there is the potential of doing the same for humans. Acknowledgments The authors acknowledge partial financial contributions by the Industrial Research Assistance Program of National Research Council in Canada, and David Hawks a technology advisor from that center. The authors also acknowledge the technical assistance of Anthony Kamalian, Michael Younikian, and Carolyn Winsborough in the scanning sessions.
References 1. Douketis JD, Paradis G, Keller H, Martinea C (2005) Canadian guidelines for body weight classification in adults: application in clinical practice to screen for overweight and obesity and to assess disease risk. Can Med Assoc J 172:995–998 2. Douketis JD, Feightner JW, Attia J, Feldman WF (1999) Examination Canadian task force on the periodic health. Periodic health examination, 1999 update: 1. Detection, prevention and treatment of obesity. Can Med Assoc J 160:513–525
Lipids (2008) 43:97–103 3. Gallagher D, Heymsfield SB, Heo M, Jebb SA, Murgatroyd PR, Yakamoto Y (2000) Healthy percentage body fat ranges: an approach for developing guidelines based on body mass index. Am J Clin Nutr 72:694–701 4. Conway JM, Norris KH, Boldwell CE (1984) A new approach for the estimation of body composition: infrared interactance. Am J Clin Nutr 40:1123–1130 5. Ruchti TL, Malin SF, Hazen KH, Makarewicz MR, Acosta GM (2002) Classification and characterization of tissue through features related to adipose tissue. Canadian Patent Application 2 397 611 (Jul 26, 2001) 6. Azizian H, Winsborough S, Younikian M, Winsborough C (2003) Method of in-vivo measurement of fat content of a body and apparatus thereof. Canadian Patent No. 2,404,891 (Issued November 18, 2003) 7. Azizian H, Kramer JKG (2005) A rapid method for the quantification of fatty acids in fats and oils with emphasis on trans fatty acids using Fourier transform near infrared spectroscopy (FTNIR). Lipids 40:855–867 8. Azizian H, Kramer JKG, Winsborough SL (2007) Factors influencing the FA determination in fats and oils using Fourier
103
9.
10. 11.
12.
13.
transform near infrared spectroscopy. Eur J Lipid Sci Technol 109:960–968 Azizian H, Kramer JKG, Kamalian AR, Hernandez M, Mossoba MM, Winsborough SL (2004) Quantification of trans fatty acids in food products by GC, ATR-FTIR, and FT-NIR methods. Lipid Technol 16:229–231 Azizian H, Kramer JKG (2005) A non-invasive analytical tool for many applications. Inform 16:656–658 Du Bois D, Du Bois EF (1916) A formula to estimate the approximate surface area if height and weight be known. Arch Intern Med 17:863–871 Gehan EA, George SL (1970) Estimation of human body surface area from height and weight. Cancer Chemother Rep (Part 1) 54:225–235 Engelson ES, Kofler DP, Tan YX, Agin D, Wang J, Pierson RN Jr, Heymsfield SB (1999) Fat distribution in HIV-infected patients reporting truncal enlargement quantified by whole-body magnetic resonance imaging. Am J Clin Nutr 69:1162–1169
123