Ex Vivo Co-Registration

Magnetic Resonance Imaging

MRI scans were acquired after the CEA samples were embedded in 3% agarose gel in 15-mL culture tubes. These tubes were placed in a custom-built holder designed specifically for acquiring tissue MRI scans. Specimens were imaged using a 6-cm phased array 4-channel carotid coil (Pathway Med Tech, Redmond, WA) on an Excite 3.0 Tesla MRI scanner (GE Healthcare, Wauwatosa, WI). Serial axial proton-density weighted (PDW), T1-weighted, and T2-weighted images were acquired (2-mm slice thickness, matrix 512 × 512, field of view 100 × 100 mm) using a fast-spin echo sequence providing 8–31 slices depending on tissue size with an in-plane resolution of ∼0.195 mm. Correction algorithms adjusted for magnetic field strength gradients across the sample image.

Real-time Three-Dimensional US

After MRI, each of the 13 CEA tissue samples embedded in agarose was extruded intact with the agarose column from its culture tube. The column was then transferred to a plastic box (10.5 × 12.5 × 4.0 cm), and additional 3% molten agarose was added to form an agarose bed prepared as described under the sample preparation section. This procedure was performed to provide adequate surface contact for the US probe. Samples were imaged with a LOGIQ E9 US system (GE Healthcare) using a “free-hand 3-dimensional scanning with EM field sensors” approach. This approach used a mid-range DC magnetic transmitter to generate a weak magnetic field. Position sensors attached to a two-dimensional vascular US probe sensed this field and recorded the probe position through an in-built circuit system, which functioned similar to a global position system (GPS), thereby allowing the computer to know the position and orientation of the transducer and allow it to reconstruct three-dimensional volumes.

Automated Co-Registration

The PDW sequences of the MRI of a given CEA sample were uploaded from a CD-ROM to the US system, and manual marking of two to three features was performed between the MRI and a real-time B-mode US scan. The PDW sequence was selected over T1- and T2-weighted sequences because the PDW sequences were found to have the best image quality (reduced noise, reduced motion artifacts, reduced blurring, and sharp edges). Common features used for initial marking included the plastic marker tip, calcifications, and, if present, the “bifurcation.” After two to three common points between the MRI and the real-time US scans were marked, the US instrument ran an algorithm to create a transformation matrix, which mapped the two-dimensional probe position to a multiplanar reconstruction of MRI scans. The real-time US scan was now registered to the MRI scans, such that movement of the US probe resulted in movement through reconstructed MRI scan planes, displayed next to the US scan.

Validation of US Co-Registration by Manual Methods

All image data were exported in Digital Imaging and Communications in Medicine (DICOM) format to CD-ROMs for offline analysis. Image sets were loaded onto a free DICOM viewer. Each image in a set had three components: (1) a US image; (2) a US frame number that identified the order of the image in the series; and (3) an MRI scan co-registered such that several US frames in the sweep would have the same MRI scan (ie, more US frames than MRI slices) ( Figure 1 ).

Transverse US scan of a CEA tissue specimen ( left ) co-registered with its corresponding axial MRI slice ( right ). The specimen appears U-shaped as a result of being sliced longitudinally during surgical resection. The calcifications at both ends of the specimen appear as bright hyperechoic regions on US and dark hypointense regions on MRI. The circular ring enclosing the tissue specimen on US represents the interface between the surrounding agarose bed and the original agarose cylinder formed when the tissue was initially embedded in a 15-mL cylindric tube for MRI.

Initially, only the US image was blinded out of the three components, and two readers jointly identified two to three features on the MRI scan per CEA tissue sample. The US frame number for a feature and a description of the feature were noted. The total number of MRI scans per tissue sample was also noted.

After all features were identified, the readers independently loaded the image sets in a pre-randomized order unique to each reader, and now they were blinded to the US frame number and MRI scan and unblinded to the US image. Each reader identified the previously described MRI features on the US images. After a feature was satisfactorily identified on a US image, the US frame number was unblinded and recorded as the manually registered frame for that feature.

In Vivo Co-Registration

Reproducibility of C-IMT: US-US Co-Registration

We then examined application of the co-registration technology for improving the reproducibility of carotid intima-media thickness (C-IMT) measurements. Healthy individuals ( n = 10) without coronary heart disease, stroke, peripheral vascular disease, hypertension, or current smoking were invited to participate. All volunteers underwent bilateral US scans of the common carotid artery (CCA) for C-IMT measurements on a LOGIQ E9 US machine. Scans were acquired using traditional methodology and the GPS-like technology. Repeat scans using both methods were performed 2 days later. Neither the traditional method nor the GPS method was electrocardiography (ECG) gated.

The traditional method involved the use of Meijer’s arc for standardizing the probe angle. Volunteers were positioned supine with the neck slightly hyperextended and rotated ∼30 degrees from midline contralateral to the side imaged. The angle on the Meijer’s arc was recorded on the initial visit, and the same angle was used for imaging during the return visit.

For the GPS method, a three-dimensional US sweep of the CCA was performed on the initial visit to create a three-dimensional US map, which was stored. For measurement of C-IMT, both during the initial and return visits, real-time US scans were obtained and co-registered to the three-dimensional map with the GPS-like technology ( Figure 2 ). Semiautomated feature registration using co-registration of corresponding planes followed by point-to-point co-registration (ie, plane-point method) was performed between the real-time US and the three-dimensional map. The plane containing the carotid bifurcation became the internal registration plane for each subject. The carotid bifurcation served as the registration region.

Carotid US scan ( left ) co-registered to three-dimensional map ( right ). Green crosses ( circled ) seen on both the left and right indicate that the scan plane was aligned with chosen features on the three-dimensional map. The first cross, labeled with a 1, is located in the far wall of the internal carotid artery just superior to the bifurcation. The second marker, labeled with a 2, is located in the far wall of the CCA ∼1 cm inferior to the beginning of the bulb.

Once the plane-point co-registration was completed during the initial visit, point markers (ie, GPS markers) were placed on the co-registered real-time US scan and saved along with the three-dimensional US map. These markers, which appeared as green crosses when the scan plane was aligned with the markers and as red or blue boxes when the scan plane was directed away from the markers, were used to identify the scan plane/image sequence where the C-IMT measurements will be made. After the plane-point co-registration was completed on the return visit, the GPS markers were loaded from memory, enabling the operator to return to the location of the scan plane where the C-IMT measurements will be made.

All scans were de-identified and exported in DICOM format for offline processing on a workstation equipped to measure C-IMT (Carotid Analyzer, Medical Imaging Applications LLC, Coralville, IA). The C-IMT measurements took place approximately 2 months after image acquisition. For each side, the mean C-IMT value of the far wall of the distal CCA over a 1-cm length was measured, and the combined average of the right and left sides was used for our analysis. Two trained readers blinded to method and to visit order performed the measurements. During the analysis, the readers noted that the quality of the image sequences varied, and they performed a post hoc, blinded, subjective image quality assessment.

US-MRI Co-Registration in Patients with Known Carotid Atherosclerosis

To test US-MRI co-registration in a clinical setting, we invited a subset of individuals ( n = 3, 6 carotid arteries) participating in an ongoing trial ( ClinicalTrials.gov NCT00860184 ) for which Baylor College of Medicine is one center to participate in an ancillary study. The trial enrolls and follows asymptomatic subjects with 50% to 79% carotid artery stenosis in whom a carotid MRI is performed at baseline. The ancillary study that added three-dimensional US imaging was approved by the trial principal investigator.

MRI was first obtained using sequences specified by the trial protocol. The PDW MRI series obtained over an 8-cm length of carotid artery centered on the bifurcation of the carotid artery (2-mm slice thickness, 40 slices total) was then exported in DICOM format and uploaded to a LOGIQ E9 US system. For the US scan, we positioned the subjects similarly to that used for MRI to achieve in-plane registration.

The co-registration method between real-time scans and uploaded MRI scans was similar to that used for ex vivo co-registration between US and MRI. Briefly, real-time US scans were obtained and co-registered to the uploaded MRI scan using semiautomated feature registration of corresponding planes followed by point-to-point registration ( Figures 3 and 4 ). Validation by manual methods was conducted as described for CEA specimens; we used the disappearance of the flow divider as a common feature to evaluate registration between the imaging modalities.

Real-time long-axis US of the right carotid artery ( left ) co-registered with the corresponding MRI scan ( right ) in a patient with minimal carotid atherosclerosis in the vessel imaged. In both images, the common carotid artery is seen starting on the right, expanding to the carotid bulb, and tapering down to the internal carotid artery on the left. ICA, Internal carotid artery.

A–C , Real-time carotid US ( left ) has been co-registered with the corresponding PDW MRI series ( right ). A , Transverse views of the carotid arteries just after the carotid bifurcation with the external carotid artery to the left of the internal carotid artery. The vessel wall boundaries can be seen clearly on MRI. Long-axis views of the same artery can be seen (B) with the internal carotid artery on the left and the CCA on the right. Multiple regions of calcified plaque appear as bright hyperechoic regions on real-time US. Plaques may not appear on the corresponding PDW MRI scans, because calcifications appear as dark hypointense regions on MRI. Corresponding regions of plaque ( arrows ) have been marked in the carotid bulb. C , Color Doppler has been overlaid on the real-time US.

Statistical Methods

The MRI slices per sample, US frames per sample, and US frames per co-registered MRI slice were described using Stata 11 (Stata Corp, College Station, TX). Absolute agreement between US co-registered frames and manual co-registration were determined by calculating intraclass correlation coefficients (ICCs) using a two-way random effect analysis of variance (ICC [2, 1] method) model to test strength of absolute agreement. ICCs were also calculated using a multilevel hierarchic model with mixed effects, which adjusted for variations within each CEA specimen. Bland–Altman plots were generated to examine for biases inherent to registration methods used and to evaluate 95% limits of agreement between mutual MRI-registered and independent US-registered frames.

For the C-IMT measurements performed for US-US co-registration, the ICC [2, 1] method was used to determine inter-visit repeatability and inter- and intra-reader reproducibilities. Bland–Altman plots were also generated to examine for biases inherent between methods. For the US-MRI co-registration conducted in patients with known carotid atherosclerosis, the difference in frames for common features observed on US and MRI was noted and described.