In-vivo carotid T₂ mapping can accurately quantify plaque lipid content to discriminate between symptomatic and asymptomatic patients: histological validation, scan-rescan reproducibility and clinical study

Background

In-vivo carotid CMR is able to identify features of plaque vulnerability such as lipid core size. However, the current standard (multicontrast CMR) requires contrast media, extensive post-processing and subjective interpretation. Recently we proposed to use quantitative T₂ mapping to distinguish plaque lipid from surrounding fibrous tissue and measure lipid core size. This study aimed to (1) validate plaque lipid quantification by T₂ mapping against histology and (2) investigate if it could discriminate between symptomatic and asymptomatic patients.

Methods

CMR

40 patients scheduled for carotid endarterectomy (50-99% stenosis on Ultrasound), either symptomatic or asymptomatic, were imaged at 3T (Siemens Verio) max 24 h before surgery (IRB approved, written consent obtained). We used our novel black-blood Multiecho Spin-Echo sequence for T₂ mapping (DANTE-MESE) to acquire 5 slices in 4 min (TR = 2 s, TE = 9-18-…-126 ms, partial Fourier = 5/8, FOV = 128 × 128 mm, matrix size = 384 × 384, slice thickness = 2 mm, slice gap = 2 mm). T₂ maps were generated using nonlinear fitting; lumen and external carotid boundary were contoured semi-automatically; lipid core was segmented from other plaque components by T₂ thresholding; and the optimal segmentation (i.e. highest correlation with histology) was automatically calculated using leave-one-out cross-validation (Figure 1).

Figure 1

Acquisition and analysis pipeline for plaque lipid quantification. I) Time-of-Flight (TOF) bright-blood images were acquired to localize carotid bifurcation and stenosis. II) Multiple T₂ map slices were acquired using our novel DANTE-MESE sequence that combines black-blood preparation based on non-selective DANTE pulse trains with Multiecho Spin-Echo. T₂ maps were generated using nonlinear fitting; lumen and external carotid boundary were contoured semi-automatically; on T₂ maps plaque calcification is shown in black, lipid in blue, normal vessel wall and fibrous tissue in light blue/green, and intraplaque haemorrhage in yellow/red. III) Plaque lipid area was automatically segmented on carotid T₂ maps using leave-one-out cross-validation. IV) Plaque lipid area was manually segmented (blinded to T₂ maps) on matching histological sections stained with H&E and Masson Trichrome, with Oil-red-O used to confirm lipid distribution.

Histology

Plaques were collected at the time of carotid endarterectomy, processed and cut at 1 mm intervals using carotid bifurcation and T₁w images as references to match T₂ map locations. Plaque lipid area was manually segmented on sections stained with H&E and Masson Trichrome, with Oil-red-O used to confirm lipid distribution.

Results

26 patient scans (median age 69) had good quality and 60 slices with matching locations on plaque histology. Lipid without haemorrhage had shorter T₂ than normal vessel wall and fibrous tissue, whereas haemorrhage infiltrating the lipid core increased T₂. Thresholding T₂ < 42 ms and T₂ > 90 ms resulted in the globally optimal lipid core segmentation (highest R = 0.85 with histology, Figure 2A). We found significantly more lipid in symptomatic than in asymptomatic plaques (Figure 2B); ROC analysis showed a fair/good ability to discriminate between the 2 groups; and the optimal cut-off value ~25% agreed with histological studies that found higher risk of rupture associated with lipid core >25% of plaque area (Figure 2C). Scan-rescan reproducibility in 9 patients was excellent: ICC = 0.89 (95% CI 0.59-0.98) and CoV = 8.9%.

Figure 2

Histological validation and clinical results. A) The highest correlation (R = 0.85, P < 0.001) between plaque lipid area (%) measured on T₂ maps and on histology was achieved by the thresholding combination T₂ < 42 ms and T₂ > 90 ms (using leave-one-out cross-validation). B) Symptomatic plaques contained significant more lipid than asymptomatic plaques (31.5 ± 3.7% vs. 15.8 ± 3.1%, P = 0.005) despite similar degree of luminal stenosis (15 symptomatic patients, median age 73, stenosis on Ultrasound = 80 ± 9% vs. 11 asymptomatic patients, median age 60, stenosis = 83 ± 9%). C) ROC curve analysis showed that T₂ mapping had a fair/good ability to discriminate between symptomatic and asymptomatic plaques (AUC = 0.79, P = 0.012). The optimal cut-off value for lipid area was ~25% (sensitivity = 67%, specificity = 91%).

Conclusions

We have demonstrated that carotid T₂ mapping can be used to quantify plaque lipid content with good accuracy and reproducibility, and to classify plaques as symptomatic or asymptomatic based on their lipid core size. This technique can potentially be used to identify patients at risk of plaque rupture; informing decisions of stents vs. surgery; stratify for more intensive lipid treatment; and monitor response to treatment in clinical trials.