The BELA signal - the loss changing rate as a function of frequency (m
loss) - did not correlate significantly with the amount of subcutaneous fat and, as had been predicted from the phantom study [18], tended to correlate with the amount of visceral fat. Comparison of the correlations found in this study to those found for a much simpler method waist circumference (WC) [22, 23] do not suggest much advantage for the BELA measurement over WC. However, both the sensitivity and specificity of WC tend to weaken in individuals over 40 years of age [23], likely because of the accumulation of SF. The low sensitivity of BELA for SF could make it more specific in this age group.
The comparison of the BELA correlations to those found for abdominal BIA and ultrasonography (US) do not show much advantage either. However, large scale studies in abdominal BIA [9, 10, 12] and US [13] do not report receiver operating characteristics (ROC) graphs, which we think are important when the performance of the method is estimated. Furthermore, one may question the performance of the bioelectrical impedance measurement in the abdominal BIA [24], because the final analysis commonly includes several additional parameters like sex, age [11] and body shape [8, 10, 11], or WC [12]. The nature of the eddy currents may justify including WC in the BELA analysis, but there is no need to use the other parameters often combined with BIA. Thus, in some groups of individuals, BELA will potentially perform better than BIA. BELA would also be more convenient in medical check-ups and repeated individual use than US, because BELA measurements do not need medical personnel to perform the measurement or interpret the results.
The loss changing rate m
loss is a relative quantity. It does not include information on the bulk losses caused by the subject but rather the rate of loss as a function of frequency. Based on this, we assumed that even if larger subjects introduce more losses than smaller ones, m
loss would not be sensitive to the body size. However, a strong correlation to WC at all levels suggests that m
loss is sensitive to body geometrics, which has to be targeted in further studies to make sure that the sensitivity to changes in fat content is sufficiently larger than the sensitivity to changes in WC.
The BELA setup used in this study was similar to that used in [18], except the doubled field strength (1V across the coil) and the electrostatic shield made to cover the coil's inner surface. By doubling the field strength, the signal from human subjects corresponded to what was observed with phantoms in [18]. This change was necessary, because at low frequency, the losses would have been too small. The electrostatic shield was included because, in human subjects but not in phantoms, losses were observed to be somewhat higher without shielding. The shield's purpose is to reduce capacitive coupling with the subject inside the coil. Therefore, only the inner surface of the coil was shielded. Shielding the outer surface would only reduce the coil's quality factor unnecessarily.
The losses were measured at 103 kHz and 185 kHz frequencies. The frequency selection is not critical and therefore these frequencies were not tuned precisely to some more intuitive frequencies like 100 kHz and 200 kHz by using trimmer capacitors but discrete components. It is important to have two separate frequencies low enough so that the wave length is clearly above the subject's diameter. Moreover, frequencies from 100 kHz to 200 kHz are low enough to give signal also from the interior of the subject. This was tested with air core and filled phantoms in [18] - a similar test than in [25].
One of the volunteers had an exceptionally large amount of VF, which the BELA measurement could not differentiate. The axial MR image, shown in figure 4A, indicated that in this volunteer, the VF and SF pools came into close contact with each other due to very thin or near non-existent abdominal muscles, thus significantly reducing the conductive volume (of muscle) between SF and VF pools. Since the BELA measurement needs an electrically conductive volume between the SF and VF pools, the measurement could not differentiate between the two fat pools, leading to underestimation of the VF and delaying possible treatment.
Individuals who do not exercise and maintain normal BMI through diet only may have a considerable proportion of VF. This "slim but fat inside" or thin-on-the-outside fat-on-the-inside (TOFI) body constitution carries a metabolic risk [26] and needs to be recognized. For BELA to perform well in a large range of body compositions, additional parameters are needed to give more information on the total conductivity of the measured area. Combining bioelectrical impedance measured at the waist level with BELA may allow a more reliable assessment of VF across a large range of body compositions. This combination of measurements should give enough information on subject's fat distribution so that TOFIs and cases similar to the one shown in figure 4A can be correctly classified. Practically, in these subjects, bioelectrical impedance should show exceptionally large impedance while the rate of loss remains small.
Both the accuracy and resolution of the BELA instrumentation need to be improved before performing evaluations in larger and more heterogeneous test groups. The measured voltage changes across the coil map to changes in the coil resistance. This change of resistance is very small, below mΩ at low frequency [18], which makes the measurement challenging. To improve the accuracy, we have developed a new pre-amplifier especially with the BELA measurement in mind [27]. The usage of this amplifier allows reducing the signal processing and provides higher gains in an analog front-end design. In this study, the accuracy and repeatability was limited by 10-bit analog-to-digital converter, which had an absolute accuracy of ±2 LSB. With 1.1V voltage reference, this is ±2.15 mV which corresponds the average standard deviations of the measured voltage changes through the study, 1.71 mV and 1.89 mV for low and high frequencies, respectively. Ten times better accuracy should be achievable which would suffice for clinical measurements. In addition, the average relative difference between the m
loss at different levels of navel was only ~ 3.3 %, and the effective slice thickness of the initial version of the BELA setup was ~ 5 cm. To achieve a thinner slice, and thus better spatial resolution, the form of the magnetic field has to be changed. Work is now under progress towards a new instrumentation that will allow more extensive clinical testing.