Magnetic resonance imaging (MRI) can be used to estimate adipose tissue and muscle volumes ^1,2. However, current MRI methods to estimate body tissue compartment volumes rely on manual segmentation, which is laborious, expensive, not widely available outside specialized centers, and not standardized. Thus, currently implemented MRI-based tissue volume assessment techniques are impractical for many research studies, and not feasible for clinical care. Rapid, inexpensive, standardized methods are needed to reduce analysis time and cost, and to enable the use of accurate and precise imaging biomarkers for research and, perhaps in the future, clinical practice. To address the problems inherent in manual analysis methods, a novel, semi-automated image analysis method has been developed ^3-6. Image acquisition takes about six minutes and requires overlapping stacks of MRI fat-water separated images, now widely available as product sequences. Thus, the aim of this study was to assess the repeatability and accuracy of a semi-automated analysis method to estimate abdominal adipose tissue and thigh muscle volumes.

In this single-site, prospective, cross-sectional, observational study of 20 adults, an axial, 3D, dual spoiled-gradient-echo, fat-water separation MRI sequence was used to estimate adipose and thigh muscle tissue volumes on a 3T scanner (GE Signa EXCITE HDxt, GE Medical Systems, Milwaukee, WI). Two MRI exams were performed. In the first MRI exam, the sequence was acquired twice (i.e., repeats 1 and 2). Then, the subject was taken off the table, placed back on the table, and the second MRI exam was performed in which that sequence was acquired again (i.e., repeat 3). Hence, this sequence was acquired three times for each subject.

Source images were reconstructed offline to generate water, and calibrated fat images. Each composite image stack (repeats 1, 2, and 3) was segmented into tissue compartments using a novel, semi-automated, 3D, non-rigid, multi-atlas segmentation method 3-6. Sixteen labeled atlases were registered to the image volumes. A voting scheme was then used to combine the 16 labels into a 3D segmentation of each compartment. A subset of 20 images obtained in repeat 1 for each subject was selected for assessment of volume estimation accuracy. As a reference standard, these 20 images were also segmented manually (sliceOmatic software package; Tomovision, Ontario, Canada).

Intra- and inter-examination repeatability was assessed with Bland-Altman analysis, intra-class correlation (ICC), and coefficient of variation (CV) for paired data. Accuracy was assessed by linear regression using the manually-segmented tissue volume measurements as reference.

Cohort characteristics are summarized in Table 1; subjects spanned a broad range of body habitus. Examples of semi-automated adipose and muscle tissue segmentations are shown in Figure 1.

Intra- and inter-examination volume estimation repeatabilities based on all images in the composite image stacks are summarized in Tables 2 and 3, respectively. Tissue volume repeatability was excellent, with intra- and inter-examination ICCs ranging from 0.996 to 0.998, and CVs ranging from 1.5 to 3.6%.

Tissue volume accuracy indices for the semi-automated method, using manually-determined measurements as reference for the 20 selected images, are summarized in Table 4. Abdominal adipose and thigh muscle estimated volumes were close to the corresponding manually-determined reference measurements. 95% confidence intervals (CIs) for all regression slopes included 1, and 95% CIs for all intercepts included 0. Regression bias was small for all measures, in the range 28.8 to 100.5 cm³.

One limitation of this study was that only a single 3T MR scanner from a single manufacturer was used. Another limitation is that we assessed accuracy in only a subset of non-randomly selected images, but accuracy of this method has been demonstrated previously ^4,7. We did not address intra- or inter-operator repeatability using the semi-automated analysis method, but the tested method requires minimal operator supervision, and intra- and inter-reader agreement was previously shown to be high ⁷.

In summary, a novel, semi-automated tissue segmentation and volume estimation method permits comprehensive body compartment analysis and provides high repeatability and accuracy. We are not aware of other segmentation methods with comparable levels of repeatability. Current and future clinical and drug development studies may benefit from this methodology, as may clinical settings where monitoring change in these measures is desired. Future studies might address ways to further develop, qualify, and validate semi-automated tissue volumes as possible biomarkers of clinical endpoints.