Magnetic resonance neurography (MRN) is an imaging method by which nerves can be selectively highlighted. Using commercial software, the authors explored a variety of approaches to develop a three-dimensional volume-rendered MRN image of the entire brachial plexus and used it to evaluate the accuracy of infraclavicular block approaches.
With institutional review board approval, MRN of the brachial plexus was performed in 10 volunteer subjects. MRN imaging was performed on a GE 1.5-tesla magnetic resonance scanner (General Electric Healthcare Technologies, Waukesha, WI) using a phased array torso coil. Coronal STIR and T1 oblique sagittal sequences of the brachial plexus were obtained. Multiple software programs were explored for enhanced display and manipulation of the composite magnetic resonance images. The authors developed a frontal slab composite approach that allows single-frame reconstruction of a three-dimensional volume-rendered image of the entire brachial plexus. Automatic segmentation was supplemented by manual segmentation in nearly all cases. For each of three infraclavicular approaches (posteriorly directed needle below midclavicle, infracoracoid, or caudomedial to coracoid), the targeting error was measured as the distance from the MRN plexus midpoint to the approach-targeted site.
Composite frontal slabs (coronal views), which are single-frame three-dimensional volume renderings from image-enhanced two-dimensional frontal view projections of the underlying coronal slices, were created. The targeting errors (mean +/- SD) for the approaches-midclavicle, infracoracoid, caudomedial to coracoid-were 0.43 +/- 0.67, 0.99 +/- 1.22, and 0.65 +/- 1.14 cm, respectively.
Image-processed three-dimensional volume-rendered MNR scans, which allow visualization of the entire brachial plexus within a single composite image, have educational value in illustrating the complexity and individual variation of the plexus. Suggestions for improved guidance during infraclavicular block procedures are presented.
MAGNETIC resonance neurography (MRN) is an imaging modality that can reliably and selectively image peripheral nerves.1,,2 In MRN, the use of pulsed magnetic gradients, oriented parallel and perpendicular to the nerve orientation, exploits the difference in T1 and T2 relaxation differences and/or diffusional anisotropy of peripheral nerve tissue versus muscle and enhances the intensity of the nerve signal relative to the surrounding tissue.
A peripheral nerve has loose fatty connective tissue between and around its fascicles. The nerve fibers are surrounded by a compact layer of collagenous perineurium. The axons, Schwann cells, and endoneurium are bundled together into fascicles, each of which is encompassed by a collagenous perineurium. An outer connective tissue sheath, the epineurium, envelops the nerve and the fascicles. Fascicles in cross-sectional views appear in high-tesla MRN as dot-like honeycomb structures. Magnetic resonance (MR) imaging difficulties arise because the longitudinal course of the plexus at varying depths from the skin surface leads to variable signal-to-noise (SNR), which leads to bulk susceptibility artifact and inhomogeneity of fat saturation.
To improve image contrast in nerve imaging, the interfascicular fat signal must be suppressed. This can be accomplished with long TR sequences (T2-weighted fast spin echo with fat suppression) and via fat-suppressive STIR sequences.2,,3 In a STIR sequence (short T1 inversion recovery), fat suppression is attainable at all field strengths due to the short T1 relaxation properties of lipids. In this sequence, the longitudinal magnetization is first inverted by a 180° pulse, which causes all tissues to have negative values. The longitudinal magnetization of fat becomes progressively less negative at a faster rate than other tissues. Because fat has the shortest T1 (250–350 ms) of most substances in the body, it passes through its null point (the point of zero magnetization) first before other tissues do. At the null point, fat suppression is effected, while the remaining tissues still have a negative longitudinal magnetization. These negative values are then inverted and appear bright in the final image in direct proportion to their T1 values. This technique allows the creation of an MRN image, where the peripheral nerve tract remains the most prominent feature in the generated image. This has allowed MRN selective imaging of the brachial plexus for diagnostic purposes to become an established radiologic procedure.5,,6,,7
Our goal in this study was to develop a method by which to transform the sequence of discontinuous two-dimensional (2-D) MRN images, each of which exhibits a component segment of the brachial plexus, into a single-frame composite three-dimensional (3-D) volume-rendered MRN image of the entirety of the plexus. For multiple subjects, we then used the composite 3-D MRN volume-rendered image as a method by which to determine the accuracy of needle placement recommendations for three infraclavicular regional block approaches.
Materials and Methods
With approval of the institutional review board (University of Southern California, Los Angeles, California) and written informed consent, MRN of the right brachial plexus was performed in 10 adult volunteer subjects in the supine position, with arms adducted and head slightly rotated to the left. Subjects (3 men, 7 women) were aged (mean ± SD) 47 ± 10 yr, weighed 69 ± 11 kg, and were 167.5 ± 9.6 cm tall. All studies were performed on a GE 1.5-tesla MR scanner (General Electric Healthcare Technologies, Waukesha, WI) with 9.0 software and HighSpeed gradient platform using a four-channel phased array torso coil.
Coronal and oblique sagittal STIR sequences of the brachial plexus were obtained using 4- to 6-mm slices (no interslice gap) as follows:
Coronal STIR sequence parameters: TR 4,500 ms, effective TE 40 ms, TI 150 ms. Echo train length 16 with 4 excitations. Resolution parameters: field of view 22 cm, slice thickness 4 mm, skip 0 mm (no interslice gap), 256 frequency and 224 phase encoding steps. Frequency direction and flow compensation was directed in the superior–inferior direction.
LOblique sagittal STIR with slice orientation perpendicular to the long axis of the brachial plexus: TR 4,500 ms, effective TE 40 ms, TI 150 ms. Echo train length 16 with 4 excitations. Resolution parameters: field of view 14 cm, 4- to 5-mm slices, skip 0 (no interslice gap), 256 frequency and 192 phase encoding steps. Frequency gradient was oriented in the superior–inferior direction with flow compensation in the slice direction.
Oblique sagittal T1 sequence parameters: TR 600 ms, effective TE 12 ms. Echo train length 4 with 3 excitations. Resolution parameters: field of view 14 cm, 4- to 5-mm slices, skip 0 (no interslice gap), 256 frequency and 192 phase encoding steps. Frequency direction was directed anteroposteriorly with flow compensation in the slice direction.
The proximity of arteries, veins, and the lungs results in significant breathing and flow artifacts that must be minimized. Measures taken to reduce artifacts included the use of “presaturation” pulses, applied before a radiofrequency excitation pulse; the use of saturation bands, which helps to suppress blood flow artifacts and movement ghosts; flow compensation, to saturate inflowing blood; and use of a selective phase direction, to lessen motion and flow artifacts carried through the images in the phase direction. It should be noted that flow compensation pulses and saturation bands do not entirely remove artifact from flowing blood.
Multiple software programs were explored for possible display and manipulation of the MR images. The brachial plexus component of each composite slab was created from an average of 10 (range, 5–12) sequential coronal slices from an original total of 20 slices. The slices were selected by viewing the entire data set in the DicomWorks viewer program, version 1.3.5. DICOM is short for Digital Imaging and Communications in Medicine and is a standard in the field of medical informatics for exchanging digital information between medical imaging equipment (such as radiologic imaging) and other systems. Image processing programs used were PhotoShop version 7.0 and ImageReady version 7.0 (Adobe Systems Inc., San Jose, CA). Automatic segmentation, which depends on distinguishing different gray levels, was inadequate to separate the subtle differences of the nerves from the surrounding tissue and required supplemental manual segmentation, complete or partial, in nearly all cases.
We developed a composite frontal slab approach, which consists of enhanced 2-D frontal view projections of a limited number of coronal slices onto a single composite 3-D volume-rendered slice. The selected 2-D slices were those that included a brachial plexus component, whereas all non–plexus-related slices were excluded, other than those required to complete the background skeletal composite. Had we excluded all images lacking a brachial plexus component, we might have excluded images of bony structures relevant to nerve blocks and made it more difficult for the viewer to appreciate the block-related plexus anatomy. Hence, a minimal number of slices were added to the composite that would provide a reference skeleton, both in terms of background (spine) and foreground (clavicle, coracoid, humerus). To enhance slab image realism, the background image was created by merging two nonconsecutive representative slices from the data set that would reflect the directional course of the right-sided brachial plexus, i.e. , from the upper medial to the lower lateral. Hence, a background image at the medial cephalad region, at the level of the vertebral spine, was interconnected with a more shallow background image, which reflected the more anterior and lateral portion of the brachial plexus.
These images are not free of conventional MR imaging difficulties. A “partial voluming” structure discontinuity can be produced with the thicker 2-D slices (4–6 mm), when the studied structure (e.g. , thin nerve in the axilla) is comparable in size to the voxel (a voxel is defined as a pixel times slice thickness). When the structure straddles two adjacent voxels, it may not be possible to separate the signals from different tissue compartments, and this can result in imperfect continuity and diminished brightness in the resulting image. However, there has been no insertion of any extraneous images to patch up seeming discontinuities, as might occur at nerve branch points.
The frontal slab composite–MRN image segmentation approach shares a common “projection” aspect with the maximum intensity projection technique used in angiography. Maximum intensity projection is an automated technique, commonly used in angiography, where the brightest voxel in a 2-D slice (usually that corresponding to the vessel) is projected onto a composite image. Unlike this automated technique, frontal slab composite–MRN involves a qualitative identification of the nerve component (independent of the maximal brightness) and a segmentation of the nerve image, which is then projected onto a composite image.
Each of the identified slices was converted into a TIFF file in DICOM Works. These TIF files were then brought in as sequential layers in the form of a single Adobe Photoshop image, while maintaining the original order of slices. For improved spatial resolution and edge detection, the image display matrix was increased from 256 × 256 pixels to 1,024 × 1,024 pixels using Photoshop's bicubic resampling method.
Using the selection and mask tools in Photoshop, the areas of interest (bones, nerves, vessels) were cropped appropriately on each layer of the Photoshop image. Blending between the layers was done to create smooth transitions. The structures of interest were enhanced (using the dodge and burn tool) or colorized with semitransparent color to bring out the details and emphasize the structures that were atop underlying ones. Specifically, bony structures used as reference landmarks (coracoid process, clavicle) were colorized. The areas of interest (nerve, vessels, bone) were manually segmented using Photoshop's selection tools and filled with different semitransparent colors. The transparency of the color was typically set to 30% opacity—enough to tint the area with color but still allow the underlying structures to show through. Each of these colorized slices in the data set was brought in as a sequential layer in one composite Photoshop image (keeping the original order of slices).
Brachial Plexus Needle Targeting
For each subject, a composite 3-D volume-rendered MRN image of the brachial plexus was obtained. Using the composite MRN plexus as the reference structure, we assumed that ideal needle placement, for any infraclavicular approach that relies on a posteriorly directed needle, was for the needle to be directed toward the plexus midpoint (defined as the midpoint of the MRN plexus along its vertical extent). We compared three anatomical landmark–based infraclavicular approaches that rely on a posteriorly directed needle and determined their targeting accuracy relative to the frontal midplexus location specified by the composite MRN image. Specifically, we examined two coracoid approaches8,,9 (2 cm medial and 2 cm caudad to coracoid, or 2–3 cm caudad to coracoid) and one midclavicular (using a needle directed posteriorly immediately below the midclavicle) and ascertained the specific point in the plexus targeted by a posteriorly directed needle. Hence, the targeting error (mean ± SD) for any one approach was measured as the distance from the true MRN midplexus point to the actual site targeted by the specific approach-guided needle.
Ten right-sided image-enhanced brachial plexus MRNs were obtained. In , we present the single-frame frontal slab composite MRNs of the brachial plexus for six of those individuals.
The nerve roots, derived from the ventral primary rami that arise at the dorsal root ganglia, are generally well delineated. In a few individuals, even the dorsal primary rami that supply the posterior neck muscles and skin can be seen. To some degree, the subsequent branching structure tends to follow the classic pattern succession of trunks, divisions, cords, and branches, but this is subject to significant individual variation.
Certain bony structures—the coracoid process, medial acromion, clavicle, and humeral head—were colorized and left in place. In these composite frontal view MR neurographs, the subclavian veins are readily evident, with descending pectoral venous branches. In some individuals, the internal and jugular veins are well delineated. The subclavian artery is suppressed because of the use of artifact-suppressive saturation bands.
For coracoid-based approaches, Wilson et al. studied 20 male and 20 female patients in the supine position, with arms adducted. With the needle placed initially 2 cm medial and 2 cm caudad to the coracoid process, they advanced the stimulator-directed needle posteriorly to a depth of 4.2 ± 1.49 cm (2.25–7.75 cm) in men and 4.01 ± 1.29 cm (2.25–6.5 cm) in women, with a 94.8% success rate. The depth in question related to the image-determined posterior distance to the anterior aspect of the axillary artery. In their study measurements, the anterior midpoint of the coracoid process was used as the zero reference point. In this limited study of 10 subjects, we pursued the mediocaudad recommendations of Wilson et al. to determine the part of the brachial plexus that would be targeted by a posteriorly directed needle. The targeting error, defined as the vertical distance between the MRN plexus midpoint (y in ) and the Wilson targeted point, was positive if the Wilson approach point was above the MRN midpoint and negative if below it. As a starting point, we used a point on the coracoid process 0.5 cm from its medial edge. In 6 of the 10 subjects, the needle tip led directly toward the midpoint of the brachial plexus. In 2 subjects, the needle tip was directed to the lateral superior edge of the plexus, at a point close to the musculocutaneous nerve takeoff (2 subjects), and in another subject, it was distal to the musculocutaneous nerve takeoff. For these three cases, a recommendation of 2.5–3 cm medial to the coracoid process would have resulted in better positioning. In one subject (), the needle path completely missed the brachial plexus and ended up a full 2 cm above the lateral plexus, 3.4 cm away from the plexus midpoint. This individual was tall and had the longest clavicle (26.5 cm) in the group. Based on these preliminary observations and subject to confirmation by a larger MR study, we conclude that the needle skin placement recommendation of Wilson et al. targets well the brachial plexus in average-sized individuals. The average distance of the entry point above the ideal midplexus target was 0.65 ± 0.68 cm (range, −0.76 to 3.38 cm). The recommendation scheme may require, for broad-shouldered individuals, a more medial and more caudad skin entry point. In column II of , we show an illustrative oblique sagittal slice from each of two subjects in our study, which shows the position of the brachial plexus in relation to the coracoid process, as per the Wilson approach.
An infracoracoid process approach was advocated by Kapral et al. , namely, that after needle contact with the coracoid process, the needle should be reinserted 2–3 cm immediately caudad to the coracoid process. The targeting error was defined in the same manner as that used with the Wilson approach. In , z denotes the MRN plexus infracoracoid midpoint. Consistent with the more caudad course of the plexus in this region, in 5 of the 10 cases, a 3-cm caudad approach led to a midplexus position, and in 3 cases, a 2.5-cm caudad entry point led to the plexus midpoint. In one individual (D), the 3-cm caudal recommendation targeted the superior edge of the lateral plexus. In the broad-shouldered individual mentioned above, the 3-cm recommendation led to a position 1 cm cephalad to the lateral plexus branches. In summary, the Kapral approach leads to overall results similar to those indicated by Wilson. The targeting error, using the 3-cm caudad approach recommendation, was 0.99 cm ± 1.22 cm (range, −0.34 to 3.84 cm). Column III of shows, from two subjects, the representative oblique sagittal cross-sectional view of the coracoid-to-plexus relation in this more lateral region.
In the midclavicle (marked with a yellow vertical bar), the frontal image width of the plexus is at its narrowest (approximately 3 cm in width), which suggests the use of the midclavicle as a site suitable for a single injection technique. For infraclavicular blockade, Kilka et al. suggested that initial needle placement be made posteriorly just under the midclavicle. Kilka defined the midclavicle as half the distance between the ventral acromion (VA) and the middle of the jugular (suprasternal) notch (JN), i.e. , midclavicle =½(VA − JN). Alternatively, it is possible to define the midclavicle as the midpoint between the acromioclavicular (AC) joint and the sternoclavicular (SC) joint, i.e. , as midclavicle =½(AC − SC). For our 10 subjects, the difference (mean ± SD) between these two methods of determining the midclavicle site is −0.241 ± 0.35 cm. In 8 of 10 subjects, the Kilka acromion-to-jugular notch approach resulted in a midclavicle location slightly more medial than the method based on use of the two ends of the clavicle.
To obtain a targeting error estimate via the Kilka method, with use of the ½(VA − JN) calculation, we located the MRN plexus midpoint (x in ) and the corresponding Kilka recommended midpoint and vertically projected these points to the clavicle immediately above; the targeting error was then defined as the distance along the clavicle between the Kilka and MRN projections. The targeting error was positive if the MRN clavicle midpoint was medial, i.e. , to the right of the Kilka clavicle midpoint, and negative if it was to the left. The targeting results are shown in . In 7 of the 10 cases, this recommendation led to satisfactory needle targeting within the plexus. In 2 subjects, the Kilka target was at the superior edge of the lateral plexus, and in the 1 other subject, it was approximately 0.5 cm cephalad to the lateral trunk. The mean targeting error was 0.43 cm ± 0.67 (range, −0.54 to 1.44). Hence, the Kilka approach is associated with the smallest mean targeting error of the three infraclavicular approaches using a posteriorly directed needle. The results using the ½(AC − SC) midclavicle calculation, as shown in , were comparable.
We calculated the angle (in degrees) in the frontal slab composite coronal plane at which the plexus courses laterally beyond the midclavicle. Relative to a horizontal line drawn through the true midplexus site x, we calculated the angle between points x (midplexus at midclavicle) and z (midplexus at infracoracoid). As shown in , the mean angle is 27.8°± 6.2° (range, 18°–35°).
In column I of , we have from two subjects an oblique sagittal cross-section of the brachial plexus at the midclavicle position.
In a study of 10 volunteers, Klaastad et al. also noted that the midpoint of the clavicle was lateral to the crossing of the subclavian artery by 17 mm (19–23 mm). Only in one volunteer was the distance less than 10 mm. Hence, Klaastad et al. concluded that the clavicular midpoint and the subclavian artery crossing are not clinically interchangeable reference points. Our findings are consistent with this conclusion but indicate a lesser degree of laterality of the clavicular midpoint than the Klaastad study. In , for a needle positioned perpendicular to the skin, our results indicate that the averaged ideal anatomical reference point, so as to achieve localization of the midplexus immediately below the clavicle, is a point approximately 4–5 mm medial to the midclavicle (measured along the clavicle).
In the same study, Klaastad suggested modifications to the infraclavicular block approach of Raj et al. , which is essentially a midclavicular approach. As described by Klaastad and Borgeat et al. , the modified Raj approach requires that the arm be abducted to 45–90° and elevated to 30°; the needle should be inserted 1 cm below the midclavicle [½(VA − JN)] and angled toward the emergence of the axillary artery in the fossa axillaris. By comparison in our study, all subjects had the arms adducted. We question whether it is appropriate to make a targeting error comparison with this positional difference, because the effect of arm abduction is to alter the relative locations of key landmarks. Nonetheless, the anatomical findings and recommendations of Klaastad and Borgeat are largely consistent with our own imaging results.
We have developed an approach that we have called the frontal slab composite MRN approach, which exploits the weak selective highlighting of nerves in MR STIR and T1 spin echo sequences. Through the use of a composite of several slices (as opposed to successive visualization of single MR slices) and through the frontal projection of the underlying slices onto a single 2-D slice, an image segmentation of the brachial plexus is achieved in the form of a composite 3-D volume-rendered image, as shown in , wherein the entirety of the brachial plexus can be seen in an instant.
Representative oblique sagittal cross-sectional views of the plexus for the three infraclavicular block approaches are shown in the two rows of .
Image-processed frontal slab composite 3-D rendered MRNs of the brachial plexus have educational value in illustrating the significant individual variation, a feature that is difficult to quantify. There are at least 29 documented anatomical variants in the organization or course of brachial plexus components. These variants diverge from the conventional textbook presentation of the plexus interconnections. For example, for the median nerve, the site of union between its branches from the lateral and medial cords is quite variable and has been found as far down as the elbow. Similarly, the musculocutaneous nerve arises from the lateral cord (90.5% of 75 arms), from the lateral and posterior cord (4%), from the median nerve (2%), as two separate bundles from the medial lateral cords (1.4%), or from the posterior cord (1.4%). Instead of piercing the coracobrachialis muscle, the musculocutaneous nerve adheres to the median nerve for some distance down the arm and then passes between the biceps and coracobrachialis muscles (approximately 22% of cases). These and other variations are detailed in the referenced compendium.
The obtained MRNs reproduce the natural skin plane in the region of the plexus that would directly face the anesthesiologist performing the nerve block and allow for the envisioning of the nerve structures beneath the skin in the anesthesiologist's line of sight.
Without access to imaging technology, a clinician must rely on specific palpable anatomical landmarks to achieve needle access to the plexus. In this regard, high-frequency ultrasound has gained momentum as an imaging technology that can image peripheral nerves, but the modality is practically limited to nerves within 2 cm of the skin surface. Coarser quality ultrasound nerve images are obtained at greater depths, such that the degraded echotexture makes them difficult to distinguish from tendons. Ultrasound does not provide an overview of the entirety of the plexus, but rather it generates a focused view of a segment of the brachial plexus.
If it is possible to achieve visualization of the entirety of the brachial plexus, as with an MRN brachial plexus image, one has in principle an infinite number of potential skin entry points to access the brachial plexus with a directed regional block needle. With conventional noninteractive MR imaging devices, access to the patient is still an issue. Nor is quick MR image visualization yet a practical clinical reality owing to greater cost and long scanning times (approximately 30 min in this study) in comparison to ultrasonography, which uses a portable, handheld, and readily activated device. Nonetheless, the MR image postprocessing discussed herein, which is achievable with inexpensive commercial software, clearly enhances the results of presently attainable MR neurographs. The improved MR images are of educational value and demonstrate the considerable individual variation. Whether this image postprocessing enhancement procedure can be programmed for possible automated use in an interactive MR suite requires further investigation.