Quantification of carotid plaque composition with a multi-contrast atherosclerosis characterization (MATCH) MRI sequence

Background and purpose Carotid atherosclerotic plaques with a large lipid-rich necrotic core (LRNC), intraplaque hemorrhage (IPH), and a thin or ruptured fibrous cap are associated with increased stroke risk. Multi-sequence MRI can be used to quantify carotid atherosclerotic plaque composition. Yet, its clinical implementation is hampered by long scan times and image misregistration. Multi-contrast atherosclerosis characterization (MATCH) overcomes these limitations. This study aims to compare the quantification of plaque composition with MATCH and multi-sequence MRI. Methods MATCH and multi-sequence MRI were used to image 54 carotid arteries of 27 symptomatic patients with ≥2 mm carotid plaque on a 3.0 T MRI scanner. The following sequence parameters for MATCH were used: repetition time/echo time (TR/TE), 10.1/4.35 ms; field of view, 160 mm × 160 mm × 2 mm; matrix size, 256 × 256; acquired in-plane resolution, 0.63 mm2× 0.63 mm2; number of slices, 18; and flip angles, 8°, 5°, and 10°. Multi-sequence MRI (black-blood pre- and post-contrast T1-weighted, time of flight, and magnetization prepared rapid acquisition gradient echo; acquired in-plane resolution: 0.63 mm2 × 0.63 mm2) was acquired according to consensus recommendations, and image quality was scored (5-point scale). The interobserver agreement in plaque composition quantification was assessed by the intraclass correlation coefficient (ICC). The sensitivity and specificity of MATCH in identifying plaque composition were calculated using multi-sequence MRI as a reference standard. Results A significantly lower image quality of MATCH compared to that of multi-sequence MRI was observed (p < 0.05). The scan time for MATCH was shorter (7 vs. 40 min). Interobserver agreement in quantifying plaque composition on MATCH images was good to excellent (ICC ≥ 0.77) except for the total volume of calcifications and fibrous tissue that showed moderate agreement (ICC ≥ 0.61). The sensitivity and specificity of detecting plaque components on MATCH were ≥89% and ≥91% for IPH, ≥81% and 85% for LRNC, and ≥71% and ≥32% for calcifications, respectively. Overall, good-to-excellent agreement (ICC ≥ 0.76) of quantifying plaque components on MATCH with multi-sequence MRI as the reference standard was observed except for calcifications (ICC = 0.37–0.38) and fibrous tissue (ICC = 0.59–0.70). Discussion and conclusion MATCH images can be used to quantify plaque components such as LRNC and IPH but not for calcifications. Although MATCH images showed a lower mean image quality score, short scan time and inherent co-registration are significant advantages.

Background and purpose: Carotid atherosclerotic plaques with a large lipid-rich necrotic core (LRNC), intraplaque hemorrhage (IPH), and a thin or ruptured fibrous cap are associated with increased stroke risk.Multi-sequence MRI can be used to quantify carotid atherosclerotic plaque composition.Yet, its clinical implementation is hampered by long scan times and image misregistration.Multi-contrast atherosclerosis characterization (MATCH) overcomes these limitations.This study aims to compare the quantification of plaque composition with MATCH and multi-sequence MRI.Methods: MATCH and multi-sequence MRI were used to image 54 carotid arteries of 27 symptomatic patients with ≥2 mm carotid plaque on a 3.0 T MRI scanner.The following sequence parameters for MATCH were used: repetition time/echo time (TR/TE), 10.1/4.35 ms; field of view, 160 mm × 160 mm × 2 mm; matrix size, 256 × 256; acquired in-plane resolution, 0.63 mm 2 × 0.63 mm 2 ; number of slices, 18; and flip angles, 8°, 5°, and 10°.Multi-sequence MRI (black-blood preand post-contrast T1-weighted, time of flight, and magnetization prepared rapid acquisition gradient echo; acquired in-plane resolution: 0.63 mm 2 × 0.63 mm 2 ) was acquired according to consensus recommendations, and image quality was scored (5-point scale).The interobserver agreement in plaque composition quantification was assessed by the intraclass correlation coefficient (ICC).The sensitivity and specificity of MATCH in identifying plaque composition were calculated using multi-sequence MRI as a reference standard.

Introduction
Stroke is the second global cause of both disability and mortality (1).Approximately 85% of all strokes are classified as ischemic strokes (2), and approximately 20% of ischemic strokes are linked to carotid atherosclerosis (3).At present, the degree of carotid stenosis is essential in determining the type of (secondary) stroke prevention treatment.Ischemic stroke related to carotid stenosis is mainly caused by embolization occurring after plaque rupture rather than by low perfusion resulting from the stenotic plaque (4)(5)(6).A vulnerable plaque is a plaque that is more susceptible to rupture and distal embolization.The morphological features of a vulnerable plaque include the presence of a large lipid-rich necrotic core (LRNC), intraplaque hemorrhage (IPH), and a thin or ruptured fibrous cap (7)(8)(9).It has been demonstrated that these features are strongly associated with an increased risk of (recurrent) stroke (10-12).
In the last two decades, magnetic resonance imaging (MRI) has emerged as the preferred imaging modality for visualizing and assessing plaque composition (9,13,14).By employing a combination of different MRI pulse sequences, especially pre-and post-contrast T1-weighted (T1w) turbo spin echo (TSE), magnetization prepared rapid acquisition gradient echo (MPRAGE), and time of flight (TOF) MRI, various plaque components can be distinguished (15).Recently, expert consensus recommendations on using the abovementioned multi-sequence carotid vessel wall imaging protocol have been published (15,16).However, multi-sequence carotid MRI still has some limitations.Limited slice resolution associated with 2D sequences, long acquisition time, and image misregistration due to patient motion between scans may hamper clinical implementation.
A few years ago, a new pulse sequence, i.e., multi-contrast atherosclerosis characterization (MATCH), was introduced (17).MATCH employs a 3D segmented spoiled gradient echo (GRE) readout to acquire data with three different contrast weightings, including hyper-T1w, gray-blood, and T2-weighted (T2w), following a non-selective inversion pulse and various inversion recovery times.On the hyper-T1w and T2w images, luminal blood signals are suppressed using flow-sensitive dephasing (FSD) magnetization preparation (18).IPH and calcifications can be delineated on the black-blood hyper-T1w and gray-blood images, respectively.The T2w images provide information on overall plaque morphology and the presence of LRNC (17).In addition to the unique tissue contrast weightings available in MATCH, all three image sets are simultaneously acquired in a 5-min scan and therefore are inherently co-registered and more immune to patient intolerance than conventional multi-sequence protocols.In a study conducted on a small group of six patients, good agreement between MATCH and multi-sequence MRI was demonstrated in detecting plaque components (17).However, this study had a few limitations, including a small sample size and no quantitative analysis of the plaque components.Later, in a study conducted on a large group of 46 patients, MATCH and multi-sequence MRI showed similar performance in detecting plaque components (19).An essential limitation of both studies was the absence of a dedicated sequence for detecting IPH or LRNC in the multisequence protocol.We used a dedicated sequence for the identification of IPH (MPRAGE) and LRNC (pre-and postcontrast T1w TSE images) as part of the multi-sequence protocol in the current study, as recommended in the consensus paper (16).
Therefore, the current study aims to compare the diagnostic performance of MATCH with multi-sequence MRI, including a dedicated sequence for IPH and LRNC to quantify carotid plaque composition.
institution.These studies were registered at ClinicalTrial.gov (NCT03291093, NCT02640313, and NCT04569006).The common objective of these previous trials was to evaluate novel PET tracers and novel MRI sequences.Our institutional review board approved all three studies.A signed informed consent form was obtained from each patient.The inclusion criterion was the presence of a carotid plaque of at least 2 mm in the ipsilateral carotid artery on either duplex ultrasonography or computed tomography angiography (CTA).Both symptomatic patients, who had experienced a TIA, stroke, or retinal ischemia, and asymptomatic patients were eligible for inclusion.Only patients who underwent MATCH and multisequence MRI were eligible for inclusion in our analysis.

MRI protocol
All examinations were performed on a 3.0 T hybrid integrated PET-MRI scanner (Biograph mMR, Siemens Healthineers).MATCH and multi-sequence MRI were combined in one exam.A four-channel special-purpose coil (Siemens Healthineers) was used to image the carotid bifurcation, allowing submillimeter-resolution imaging.The scan parameters for MATCH were as follows: repetition time/ echo time (TR/TE), 10.10/4.35ms; field of view (FOV), 160 mm × 160 mm; acquired matrix size, 256 × 256; in-plane acquired resolution, 0.63 mm 2 × 0.63 mm 2 ; slice thickness: 2 mm; flip angles, 8°, 5°, and 10°; number of slices, 18; and bandwidth, 130 Hz/pixel.The multi-sequence MRI protocol acquired 14 adjoining transverse 2 mm slices covering the entire plaque.The parameters of multi-sequence MRI and MATCH are listed in Table 1.A 0.1 mmol/kg dose of gadolinium-based contrast medium (Gadovist, Bayer AG) was used for post-contrast T1w imaging with a delay of 6 min postinjection.

Image analysis
All image datasets were anonymized and processed using the dedicated software (VesselMass, Department of Radiology, Leiden University Medical Center).Two trained observers (MK and KN) with 1-3 years of experience reviewed the MR images independent of the clinical data and each other.The multisequence MRI and MATCH images were evaluated blindly from the delineations and the scores of the other MRI method with a time interval of at least 1 month.
The observers manually co-registered the images of the different contrast weightings of the multi-sequence protocol inplane (x-direction and y-direction) and out-of-plane (z-direction) using the same dedicated software package (VesselMass).The time required for manual co-registration was included in the overall image analysis time.In the case of MATCH images, co-registration was inherent and did not require additional manual adjustments.Multi-sequence and MATCH images were assessed based on the previously validated criteria (17,20).( 1) The inner and outer vessel wall of the plaque was delineated on the pre-contrast T1w TSE images (multi-sequence protocol) or T2w images (MATCH protocol).( 2) The LRNC was defined as a region within the plaque that exhibits no contrast enhancement on the post-contrast T1w images (multi-sequence protocol) or a hypo-intense region within the plaque on T2w images and isointense on hyper-T1w images (MATCH protocol).(3) IPH was characterized as a hyper-intense signal in the bulk of the plaque compared to surrounding muscle tissue.For the multisequence protocol and MATCH protocol, the MPRAGE images  and hyper-T1-weighted images were used, respectively.( 4) For the multi-sequence protocol, calcifications were identified as areas with a hypo-intense signal relative to the sternocleidomastoid muscle on at least two different MRI weightings.Juxtaluminal calcifications can be obscured on the dark blood MRI weightings; thus, the TOF images are used to identify juxtaluminal calcifications.For the MATCH protocol, calcifications were identified as a hypointense signal on the gray-blood images.The quantification of the total fibrous tissue involved subtracting the combined volume of the LRNC (including IPH) and calcifications from the overall volume of the vessel wall (21).The normalized wall index (NWI) was defined as wall area/(lumen area + wall area), and the wall area was defined as the area between the lumen and outer wall (22).The percent wall volume (PWV) is calculated by dividing the volume of the vessel wall by the total volume of the carotid artery segment and then multiplying this number by 100 to express it as a percentage (23).The scan time (acquisition time and planning) and the image analysis time (co-registration time and time to delineate the vessel wall and plaque components) were recorded for MATCH and multi-sequence MRI.
A region of interest (ROI) was drawn in the sternocleidomastoid muscle at the level of the carotid bifurcation.The signal-to-noise ratio (SNR) was calculated as the ratio between the mean signal intensity and the standard deviation (SD) of this ROI.The contrast-to-noise ratio (CNR) of IPH was calculated for MPRAGE and MATCH as the difference in mean signal intensity of IPH and muscle tissue divided by the noise of muscle tissue (SD) (24).The image quality was scored on a 5-point scale on a slice-by-slice basis: 1 for low image quality and 5 for excellent image quality (20).

Histological analysis
In one patient who underwent carotid endarterectomy as part of his/her routine clinical care 1 day after the MRI examination, the MRI findings were compared to histology.The specimen was collected directly after carotid endarterectomy.The specimen was cut into ∼3 mm slices, coded, and alternately frozen and stored at −80°C.The sample was fixated in a 4% paraformaldehyde phosphate buffered saline (PBS) solution for 18-48 h, decalcified using ethylenediaminetetraacetic acid (EDTA) for 4 h, and then embedded in paraffin.Cross-sectional sections of 4 mm were stained using hematoxylin and eosin (H&E).An experienced vascular pathologist macroscopically assessed the carotid plaque composition.All procedures conducted during the research adhered to the guidelines outlined in the Dutch Code of Conduct for Observational Research with Personal Data (2004) and Tissue.

Statistical analysis
Two-way mixed-effects model of intraclass correlation coefficients (ICCs) and Cohen's kappa test (κ) were used to assess the interobserver agreement for the quantification and identification of carotid plaque components on multi-sequence and MATCH images.The ICC is a numerical value ranging from 0 to 1. Values below 0.5 suggest poor agreement, values between 0.5 and 0.75 indicate moderate agreement, values between 0.75 and 0.9 represent good agreement, and any value above 0.9 indicates excellent agreement (25).Kappa values also range from 0 to 1, where values from 0 to 0.2 indicate slight; 0.21-0.4,fair; 0.41-0.60,moderate; 0.61-0.8,substantial; and 0.81, upward excellent agreement (26).The comparison between MATCH and multi-sequence protocol was as follows: hyper-T1w vs. MPRAGE, gray-blood vs. TOF, and T2w MATCH vs. post-contrast T1w TSE.Sensitivities and specificities of MATCH in identifying plaque components on an artery basis were calculated using the multi-sequence protocol as a reference standard.A paired t-test or Wilcoxon signed ranked test was used to evaluate the differences in vessel wall volume, NWI, and volumes of the various plaque components between the MATCH and multisequence protocol, as appropriate.In addition, the differences between the IPH, LRNC, calcifications, and NWI measurements on MATCH and multi-sequence images were plotted against the mean difference as Bland-Altman plots.Limits of agreement were calculated as mean difference ± 1.96 × standard deviation of difference.Statistical analyses were performed using IBM SPSS Statistics for Windows, Version 24.0 (IBM Corporation).Normally distributed continuous variables were presented as mean ± standard error.Otherwise, the median and interquartile range (IQR) was presented.P < 0.05 indicated statistical significance.

Results
A total of 27 patients (54 carotids) underwent MATCH and multi-sequence carotid MRI, of which 21 were male.A flowchart of the patient inclusion is shown in Figure 1.One artery was excluded due to total occlusion.Two patients did not undergo contrast injection because of a low glomerular filtration rate (GFR) (<30 ml/min).The patients had a mean age of 70 ± 6.8 years.Patient characteristics are presented in Table 2.The mean MATCH and multi-sequence scan time was 7 min and 25 s and 39 min and 31 s, respectively.The mean time to delineate the vessel wall and the plaque components, including co-registration, for one artery on MATCH was shorter than that for the multisequence images (7:42 min ± 2:30 min and 13:24 ± 1:55 min, respectively; p < 0.05).

The performance of MATCH for identifying and quantifying carotid plaque features
The results of true-positive/negative and false-positive/negative identification of plaque components on MATCH images compared with multi-sequence protocol as the reference standard for both readers are listed in Table 5.For readers 1 and 2, the sensitivity and specificity in detecting IPH on hyper-T1w images and LRNC on T2w images of the MATCH protocol were ≥88.9% and 90.9% and ≥81.8% and ≥85.4%, respectively.However, lower sensitivity and specificity of 71.4% and 32.0%, respectively, were observed for scoring calcifications on the MATCH protocol gray-blood images.Figure 3 shows an example of good agreement between MATCH and multi-sequence images for identifying plaque composition.Figure 4 also shows the presence of IPH, LRNC, and calcifications on MATCH and multi-sequence images with histology as a reference in a carotid endarterectomy (CEA) specimen in a patient that underwent CEA 1 day after the MRI examination as part of routine clinical care.

Discussion
In the present study, we validated MATCH with a multisequence carotid MRI protocol that was recommended in a white paper (16).Our results showed a substantial-to-excellent interobserver agreement of detecting and quantifying all plaque    The Bland-Altman plots of the difference (mm 3 ) between two readers and the mean value for the total volume of the lipid-rich necrotic core, intraplaque hemorrhage, calcifications, and the normalized wall index for MATCH and the multi-sequence protocol.The continuous lines display the mean bias, and the dashed lines display two times the SD.features with MATCH except for calcification which showed a fairto-moderate agreement.In addition, the sensitivity and specificity of identifying IPH and LRNC on the MATCH images using multi-sequence MRI as the reference standard were high.Moreover, the quantification of vulnerable carotid plaque components such as IPH and LRNC using MATCH were in agreement with the quantification using the multi-sequence protocol, whereas a moderate and poor agreement was seen for the total volume of fibrous tissue and calcifications, respectively.These findings are important since the scan time and the time needed for image analysis were substantially less for MATCH than for the conventional multi-sequence protocol.Dai et al. (19) demonstrated that MATCH showed a comparable, if not superior, performance compared to a conventional protocol in identifying and quantifying major carotid plaque components.This previous study used the T1w TSE, T2w TSE, and TOF images as a reference.Therefore, LRNC was determined using T2w TSE images, which is less accurate than that of the contrast-enhanced T1w TSE images, which have been used in the present study (27).In addition, in the current work, we also measured additional practical outcome variables, such as the total scan time, which includes acquisition time and planning.Similarly, we compared how long it takes to identify and quantify carotid plaque features and the composition of one carotid plaque.In comparison, both MATCH and MPRAGE have similar scan times and can be used to identify IPH, an important risk factor for predicting stroke (11,12).The advantage of MATCH is that without a time penalty, it can also provide multiple co-registered images, which can be used to identify additional plaque features, such as LRNC volume, that are not possible by using only MPRAGE.In addition, the MPRAGE sequence contains a non-selective inversion pre-pulse.This non-selective inversion pre-pulse is limited to the field of view of the body coil.Therefore, in case we observe that the luminal blood is not entirely suppressed, the patient's position can be slightly off-centered toward the foot direction.In the MATCH protocol, FSD pre-pulses are applied to suppress the signal originating from the luminal blood (17, 18).The FSD prepared is flow-dependent but does not rely on the inflow of blood (18).We observed a poor agreement on the quantification of calcifications.The visualization of small calcifications on MR images can be challenging due to their dark appearance.Therefore, partial volume effects can obscure small regions of calcifications on MRI.In line with this, a higher measurement error for calcifications compared to that for the LRNC and wall volume was reported in a previous study on quantitative assessment of carotid atherosclerosis (28).In the multi-sequence MRI protocol, regions that were not identified as LRNC and that appeared dark on at least two different weightings were identified as calcifications.In the MATCH protocol, calcifications were identified on the gray-blood image.This may explain the poor agreement on the quantification of calcification.Also, this will lead to a lower agreement in the quantification of fibrous tissue since the fibrous tissue is identified as the remaining tissue after delineating the LRNC, IPH, and calcifications.Previous studies demonstrated that IPH and LRNC were important parameters for the risk prediction of stroke (11,29).Importantly, these parameters can be quantified with a good-to-excellent agreement.The association with stroke risk was not reported for calcifications and fibrous tissue as identified on MRI.
Despite the lower image quality observed in the hyper-T1w and T2w MATCH images, good agreement was still observed for the quantification of IPH and the LRNC, respectively.In addition, although the CNR between IPH and muscle tissue was lower for the hyper-T1w MATCH images compared to that for the MPRAGE images, the CNR remained relatively high, allowing for a clear appearance of IPH.These findings suggest that despite the lower image quality of the MATCH images, the vulnerable plaque components (IPH and LRNC) remained discernible.
MRI-based tissue quantification is accurate and reproducible (30,31).The requirement of acquiring multiple weightings to obtain accurate measurements can lead to longer scan times.Moreover, the acquisition of different weightings increases the risk of image misregistration, resulting in the misalignment of anatomical structures and potential inaccuracies in tissue quantification.Despite these limitations, ongoing research and technological advancements aim to optimize MRI protocols to enhance the efficiency, accuracy, and reproducibility of MRIbased tissue quantification.Apart from MATCH, other novel multi-contrast MR sequences, such as simultaneous noncontrast angiography and IPH (SNAP) (32) and bright-blood and black-blood phase-sensitive inversion recovery sequence (BOOST) (33), have recently been developed.These sequences also offer the advantage of acquiring multi-contrast images using a single sequence, thereby greatly reducing scan time and providing inherent image co-registration.However, it is important to note that these novel sequences still require additional validation to ensure their reliability and accuracy in clinical practice.In addition to the advancements in MR sequences, the integration of artificial intelligence (AI) holds tremendous potential in further enhancing MR image quality and shortening analysis times.AI techniques can aid in automating image analysis tasks, enabling more efficient and streamlined processing of MR images.A limitation of the MATCH sequence is that no criteria are available for determining the fibrous cap status on MATCH images.A thin or ruptured fibrous cap on MRI is a well-known risk factor for recurrent stroke (29,34), and it can be easily scored on post-contrast T1w as part of the multi-sequence protocol (27,35).Thus, if time allows, it is still beneficial to add a post-contrast dark blood T1w MRI sequence to score the fibrous cap status.This sequence, which takes 3-5 min, will Bland-Altman plots of difference (mm 3 ) between two protocols (multi-sequence and MATCH) vs. the mean value for the total volume of lipid-rich necrotic core intraplaque hemorrhage, calcifications, and normalized wall index with a mean bias (continuous lines) and two times the SD (dashed lines).require contrast injection and needs to be acquired approximately 6 min after contrast injection.This 6 min could be used to acquire a contrast-enhanced magnetic resonance angiography (MRA) to quantify the degree of stenosis.In the present study, the MATCH images were only acquired before contrast injection.Future studies could investigate whether post-contrast MATCH images could be of added value.One of the limitations of our study is that most patients had only an intermediate or mild degree of stenosis (<70%).Symptomatic patients with high-grade carotid stenosis (70%-99%) benefit from carotid endarterectomy (36).Thus, comparing MATCH with histology as a gold standard can be considered in the future.Symptomatic patients with mild-to-moderate stenosis (<70%) are still at increased risk of recurrent ischemic events, and the usefulness of carotid endarterectomy has not yet been determined (37).Therefore, studying the risk factors of plaque components such as IPH and LRNC using a short carotid MRI protocol and inherently co-registered images is beneficial.In addition, while we made efforts to minimize bias by conducting the image evaluation with a time interval of at least 1 month, it is important to acknowledge that blinding readers to whether they delineated the plaque components on MATCH or multi-sequence MR images is inherently impossible since the observers can recognize the weightings.Last, although the sensitivity and specificity for the identification of the plaque components were similar between the two readers, reader 2 had a larger number of false-positive findings.These results suggest that although both readers were well-trained, there is still a difference in their performance.
In conclusion, MATCH can be used for identifying and quantifying carotid plaque composition except for calcifications and fibrous cap status.Inherently co-registered images and short scan and analysis times are major advantages of MATCH.

FIGURE 3
FIGURE 3An example of a patient that exhibits an atherosclerotic plaque in the left carotid artery.The four multi-sequence MR images have been manually coregistered and are displayed in the upper row.In addition, the MATCH images with inherent co-registration are shown in the bottom row.IPH appears hyper-intense (white arrow) on hyper-T1w MATCH and MPRAGE images.Contours are shown on T2w MATCH and MPRAGE: green for the outer vessel wall, red for the lumen, yellow for LRNC, and blue for IPH.A good agreement is shown between the contours that are delineated on the MATCH vs. the multi-sequence MR images, although slight deviations can be observed.

FIGURE 4
FIGURE 4Example of histological comparison of MATCH and multi-sequence protocol at the left common carotid artery.IPH appears hyper-intense (white arrowhead) on hyper-T1w MATCH and MPRAGE images.A dark calcified nodule on the gray-blood image and TOF (white arrows).Histological specimen with hematoxylin-eosin staining confirms the presence of IPH, LRNC (black arrowhead), and calcifications (black arrows).Contours are shown on T2w MATCH and T1w post-contrast: green for the outer vessel wall, red for the lumen, yellow for the lipid-rich necrotic core, brown for calcifications, and blue for IPH.

TABLE 1
MATCH and multi-sequence carotid MRI protocol.
The data are presented as mean ± SD, median, IQR, or absolute numbers of patients (%).NASCET, North American symptomatic carotid endarterectomy trial; CVD, cardiovascular disease; PAD, peripheral arterial disease; BMI, body mass index.

TABLE 3
Image quality comparison between MATCH and multi-sequence protocol.

TABLE 4
Interobserver agreement for multi-sequence and MATCH images.

TABLE 5
Concordance between multi-sequence and MATCH images.Four carotids were excluded (contraindication of contrast injection).