MRI ensemble model of plaque and perivascular adipose tissue as PET-equivalent for identifying carotid atherosclerotic inflammation

Abstract

Background: Severe cerebrovascular events are associated with carotid atherosclerotic plaque progression and rupture that is mediated by inflammation. ¹⁸F-fluorodeoxyglucose ([¹⁸F]FDG) PET is important for assessing the inflammation of carotid atherosclerotic plaque, but it suffers from the limitations of radiation exposure. Additionally, inflammation of perivascular adipose tissue (PVAT) has been found to promote atherosclerosis progression through paracrine signaling mechanisms. The study aimed to develop an ensemble model based on carotid plaque and PVAT MRI radiomics for identifying highly inflammatory plaques (HIPs).

Results: 159 asymptomatic carotid atherosclerosis patients (137 males; 65 ± 8 years old) with 209 plaques (104 HIPs) were consecutively enrolled. 47.95% (70/146) of cases and 53.97% (34/63) were defined as HIPs in the training and testing datasets, respectively. There was more lipid core, more intraplaque hemorrhage, and less calcification in the HIPs compared to the non-highly inflammatory plaques (NHIPs) in the training dataset (p = 0.002, 0.019, and 0.013, respectively). Notably, the incidence of indistinct PVAT (IPVAT) in HIPs was higher than that in NHIPs, both in the training (81.43% vs. 46.05%; p < 0.001) and the testing (88.24% vs. 58.62%; p = 0.007) datasets. The correlations between plaque MRI characteristics and [¹⁸F]FDG uptake differed between the NHIPs and HIPs. However, IPVAT consistently correlated with SUV_max (r = 0.35, 0.30; p < 0.001, p = 0.002; for NHIPs and HIPs, respectively). The ensemble model that integrates the radiomics of carotid plaque and PVAT outperformed all models in predicting HIP (area under the curve [AUC] = 0.92/0.91, training/testing dataset). The follow-up further validated the PET for predicting plaque progression with the same accuracy as the ensemble model (AUC: 0.85 vs. 0.79).

Conclusions: The ensemble model integrating the radiomics of carotid plaque and perivascular adipose tissue provides an equivalent tool to PET in the visualization of the evaluation of carotid atherosclerosis inflammation and progression.

Keywords: Carotid atherosclerosis inflammation; PET/MRI; Perivascular adipose tissue; Radiomics.

Fig. 1

Flowchart of this study based on the inclusion and exclusion criteria 179 patients underwent a high-resolution carotid artery [18F]FDG PET/MR scan after excluding 266 patients. A total of 159 patients with 209 carotid plaques were consecutively enrolled following the exclusion of 20 cases owing to PET/MRI restrictions. 35 patients with 41 carotid plaques were finally included to evaluate plaque progression at 12- month based on HR-VM MRI

Fig. 2

Radiomics workflow pipeline. PVAT indicates perivascular adipose tissue; RFs, radiomics features

Fig. 3

Relationship between [¹⁸F]FDG uptake and MRI features of carotid plaque. a The correlated link and heatmap between PET metabolic parameters and MRI characteristics in the NHIP group. b Boxplot of SUV_max and TBR_max about plaque with/without IPVAT in the NHIP group. The whiskers extend from the median to ± 1.5 × interquartile ranges. c Correlated link and heatmap between PET metabolic parameters and MRI characteristics in the HIP group. d Boxplot of SUV_max and TBR_max about plaque with/without IPVAT in the HIP group. The whiskers extend from the median to ± 1.5 × interquartile ranges. IPVAT indicates indistinct perivascular adipose tissue; IPH, intraplaque hemorrhage; IPS, irregular plaque surface; HIP, highly inflammatory plaque; NHIP, non-highly inflammatory plaque; SUV_max, maximum standard uptake value; TBR_max, maximum target background ratio

Fig. 4

Nomogram and performance of the ensemble model. a The nomogram of the ensemble model. b Calibration curve of the ensemble model in the training dataset. c ROC curves in the training dataset with 5-fold cross-validation. d ROC curves in the testing dataset. PVAT indicates perivascular adipose tissue; ROC, receiver operating characteristic

Fig. 5

The baseline and follow-up imaging of HIP and NHIP. a One 60-year-old woman with HIP is identified by PET/MRI at baseline. MRI shows plaque located from the bifurcation to the C1 of the right carotid artery with IPVAT. Inflammation is observed on the actual PET images and virtual PET images based on MRI radiomics. Carotid MRI shows a significant increase in wall volume at 12-month follow-up. b a 72-year-old man performs PET/MRI at baseline, which shows carotid plaque located from the bifurcation to the C1 of the left carotid artery and non-IPVAT. Actual PET images and virtual PET images exhibit low [¹⁸F]FDG uptake. Carotid MRI at 12-month shows no progression in wall volume but an increased degree of luminal stenosis. IPVAT indicates indistinct perivascular adipose tissue

Fig. 6

Follow-up validation of models for predicting HIP. a Heatmap of risk factors associated with HIP in carotid plaque progression and non-progression. b The ROC curve of HIP based on PET for predicting carotid plaque progression. c The ROC curve of nomogram_score from the ensemble model for predicting progression. ROC indicates receiver operating characteristic