Introduction

Techniques for measuring the volume and distribution of body fat have been developed primarily to aid the management of obesity related diseases and their comorbidities^1,2. Whilst the prevalence of these is increasing, so too are eating disorders and extreme exercise regimes associated with body dysmorphic disorders³ for which we believe the fat measurement techniques have not been validated. Total body fat is commonly estimated using a multi-compartment model from body density, where the body volume is determined by air-displacement plethysmography (ADP)⁴. The aim of this work was to perform whole-body fat mass determination using MRI and compare this with total body fat mass estimates from ADP. This was done on human subjects with body mass indexes (BMIs) covering the range of non-pathological BMIs, to estimate the agreement of total body fat mass determination between the two methods and the effect of body composition on that agreement. In a previous comparison between the two techniques⁵ a higher range of BMIs was studied (29.5±3.8kg/m²) at a different field strength.

Methods

MRI measurements were made on locally fabricated fat-in-gelatine test objects and on ten volunteers on a GE Discovery 750w 3T scanner (GE Healthcare, UK). Twelve test objects, with fat percentages ranging from 0% to 100%, were made from lard, gelatine and surfactant and housed in Universal Containers. The ten volunteers, with a range of BMI:20.4-29.7kg/m²; average BMI:23.8±3.7kg/m², comprised of five males (average age:33±12, average BMI:24.6±4.1kg/m²) and five females (average age:27±8, average BMI:23.0±3.7kg/m²), were imaged using a 21 element head and neck coil, 16 element anterior array body surface coil, 40 element spine array coil and 36 element peripheral vascular coil. All measurements were performed using two different three-dimensional gradient echo Dixon based protocols⁶, LAVA-flex (an extended two-point Dixon method) and IDEAL IQ (a three-point Dixon method), with the following parameters: LAVA-flex - TE/TR=2.0/4.9ms, flip angle=12°, receiver bandwidth=558Hz/pixel, FOV=48cm, slice thickness=8mm, matrix size=300x224; IDEAL IQ - TE/TR=3.2/8.4ms, flip angle=4°, receiver bandwidth=1116 Hz/pixel, FOV=44cm, slice thickness=8mm, matrix size=192x188. Percentage fat images were calculated from ‘fat-only’ (f) and ‘water-only’ (w) images (Figure 1) on a voxel-by-voxel basis using the following formula: $$$\%fat=\frac{100*f}{f+w}$$$. Consistent with previous work⁷, voxels calculated to contain a percentage fat above half of the maximum percentage fat in each slice were considered to be fat. The total fat mass was then calculated as the total number of fat voxels multiplied by the voxel volume and an assumed density of fat (0.9007kg/m³). Subjects underwent ADP measurements on the same day as the MRI measurements. ADP was performed in the BODPOD^® Body Composition System (COSMED USA), and the whole-body fat mass estimated using a multi-compartment model⁴.

Results and Discussion

The percentage fat calculations made on the test objects showed high correlations with the expected fat percentage for both protocols (Figure 2): LAVA-flex (r=0.99, p<0.001); IDEAL IQ (r=1.00, p<0.001). The LAVA-flex protocol gave a higher fat percentage at all fat concentrations except for 100% fat compared with the IDEAL IQ protocol, where the measurement was approximately 11% lower. The whole-body MRI fat mass calculations showed high correlations with the ADP estimates (Figure 3) for both sequences (r=0.99, p<0.001). However, consistent with previous work⁵, calculating a fat mass using a single assumed fat density throughout the body yielded, a larger value for total body fat mass compared with the ADP estimate of body fat mass (Figures 4 and 5). The mean difference between the LAVA-flex protocol and the ADP estimation (4±3kg) was larger than the mean difference between the IDEAL IQ protocol and the ADP estimation (2±3kg), corresponding to approximately 5% and 2.5% of total body mass respectively. These findings are consistent with those in the test objects where the MRI calculated fat mass using the LAVA-flex protocol was systematically larger in value compared to the IDEAL IQ protocol calculated fat mass. This difference was probably due to the inclusion of more ‘fat’ voxels as a result of the lower maximum fat percentage measured by the LAVA-flex protocol. The difference between the fat mass calculated from either MRI protocol and the ADP fat mass estimate was not affected by the subject’s BMI (runs test, p>0.05).

Conclusion

This work has shown that there is good agreement between whole-body MRI using Dixon sequences and ADP in estimating whole-body fat mass over a wide range of non-pathological BMIs. It has further shown that mean differences are independent of BMI.

Acknowledgements

The authors acknowledge the support of Eddie Ng’Andwe, Alison Campbell and the staff of the MRI Department and Human Metabolic Research Unit in making the measurements. We also wish to thank Simon Dezonie from GE Healthcare for his help with the initial protocol set-up.