Extraocular Muscle Scan Analysis
Orbital image analysis techniques in Miller et al (2013) are significantly improved over those we originally introduced (Miller 1989), versions of which remain in general use. The old techniques give less accurate volume estimates, do not yield true muscle crossections without subsequent correction, and do not address errors and biases that can affect MRI quantification.

[1] Scan planes must be roughly perpendicular to the muscle's long axis.

Determining size in tomographic images begins by segmenting the structure of interest, that is, delineating its outline. High contrast between object and surround, and small pixel size are essential for accurate segmentation. For extraocular muscles, high contrast is achieved by viewing against contrasting orbital fat (eg, in quasi-coronal sections for the rectus muscles), rather than against poorly contrasting globe and optic nerve (eg, as in Lee et al 2007). Tightly framing the region of interest (eg, one or both eyes) in the scanner’s field of view (FOV) minimizes pixel size for given scanner settings. Volume averaging (signal summation across an elemental volume of tissue, or voxel), tends to blur and displace object borders. Thinner slices have less volume averaging, but also weaker signals. With elongated structures, such as extraocular muscles, volume averaging artifacts can easily be minimized by orienting the scan plane perpendicular to the long axis. Thus, quasi-crossectional scan planes should always be used (Demer & Miller 1999). Nevertheless, one still sees extraocular muscle studies using longitudinal sections, under the erroneous assumption that they provide "direct", complete pictures (eg, Lee et al 2007), whereas actually, because of volume averaging apparent muscle sizes are strongly affected by thickness and position of slice planes, and unless the slice precisely bisects the muscle along its entire path, muscle shape is distorted as well. The foregoing considerations were already reflected in our 1989 procedure.

[2] Multiple independent segmentations should be obtained.

MRI EOM SegmentationWith MRI used to make more subtle comparisons, it becomes critical to control operator bias. Even with modern clinical scanners and quasi-crossectional slice orientation, muscle contours must be delineated manually, and are therefore subject to errors of judgment. Nerves, blood vessels, and dense connective tissue bands can distort the apparent outline of a muscle, but can usually be excluded on the basis of known anatomy (figure above). Except for relatively gross comparisons, it is essential to obtain multiple independent judgments of these contours.

[3] 3D Reconstruction Yields Accurate Volumes & Cross-sections.

MRI Analysis SummaryIndividual quasi-coronal scans are problematic. In comparisons of scan planes fixed relative to an orbital reference, a muscle may appear to have hypertrophied, eg, simply because a larger crossection moved into the chosen plane. If one instead compared the single scans containing the maximum crossections, orbital positions of the compared slices would generally be different, creating a confusing or contentious presentation (were true maxima chosen?). For most purposes, then, a full range of slices is necessary to make a fair assessment of muscle size. Even so, if such a series is analyzed as "stacked blocks", we originally proposed, volumes will have difficult-to-correct errors, and crossectional areas will require projection-angle corrections (see schematic on right)

Consequently, unless the effects one is measuring are large enough to survive volume averaging artifacts, projection distortions, volume quantification errors, and slice selection biases, the preferred method for quantifying elongated structures such as extraocular muscles is 3D reconstruction based on multiple quasi-crossectional slices. Techniques for producing a ring model from muscle contours, fitting the ring model with a smooth surface, and then reslicing the surface perpendicular to the muscle's long axis, are straightforward (figure below), and can be performed with many image manipulation packages (we use Amira).

[4] EOM Imaging Should Extend to the Origin.

The origin is the preferred orbital reference. Posterior muscle segments show large changes in crossection with contraction.

Origin Drawing
Origin MRI
We use 2 measures of muscle size. The first is essentially the total volume of the muscle, apart from the flat portion that approaches and wraps around the globe (this segment is difficult to visualize, but also contains little muscle volume). Where it is not possible to accurately segment a muscle all the way to the point of tangency, it is essential to compare the same extent of muscle in various samples. This requires a reliable orbital reference. When we introduced EOM path and size measurement (Miller 1989) we used the globe-optic nerve junction as a longitudinal position reference because, although it did not yield an orbit-relative measurement, it could be reliably visualized with the scanning technology of the time. Generally available clinical scanners can now visualize the orbital apex sufficiently well to use muscle origins as head-fixed referents. For studies involving extraocular muscle contraction, which is known to be reflected mostly in posterior crossection increases (Miller 1989), it is essential for measurements to extend to the origin. Imaging to the muscle origins should therefore become general practice. As these figures indicate, muscle origins can be visualized with about 2 mm longitudinal accuracy.