Integrative analysis of hybrid PET/MR improves recovery prediction of left ventricular contractility after percutaneous revascularisation of coronary chronic total occlusions

Abstract Background The clinical indication to revascularization of coronary chronic total occlusions (CTO) is set on viability assessment through well-established imaging methods. Precise characterization of the myocardium allows a better interventional risk stratification before revascularization. Hybrid positron emission tomography/magnetic resonance (PET/MR) scanners allow the combination of image based biomarkers through simultaneous acquisition and may potentially improve the prediction of therapy response. This study aims to evaluate the accuracy in recovery prediction after CTO revascularization by combining information of myocardial viability and wall motion abnormalities (WMA) from PET/MR exams. Methods Viability was assessed with PET/MR imaging in 22 patients before percutaneous revascularization of a CTO. A cardiac MR imaging was performed at follow-up after 6 months. Segmental WMA was assessed on the AHA 17-segments model from cine images at baseline and follow-up with a 5-point scale. Recovery of WMA was predicted using six different models combining parameters from fluorodeoxyglucose (FDG) uptake, transmural extent of the scar with late gadolinium enhancement image (LGE), T1 mapping, and WMA at baseline. From 374 myocardial segments, we considered only 82 that were completely characterized and presented at least hypokinetic WMA at baseline. 46 of them were supplied by the treated CTO artery (CTO-subtended), wherefrom 16 segments recovered the contractility. A Random Forest classifier was applied for recovery prediction. Training and testing data were selected through stratified random sampling (50 iterations) with replacement from the two sets: CTO-subtended and non-CTO-subtended segments. To compensate for class imbalance and limited sample size, SMOTE oversampling was applied to the training data. Prediction assessment was based on balanced accuracy (bAcc), sensitivity, specificity, ROC AUC, and Precision-Recall AUC (PR AUC), while Wilcoxon test with a Bonferroni's correction was used for comparison between the proposed models and LGE+FDG, which has shown the best performance in previous studies [1,2]. Results The best model for the prediction of recovery after revascularisation was the combination of LGE+FDG+WMA+T1mapping (bAcc 0.62±0.11, sensitivity 0.72±0.18, specificity 0.70±0.19, ROC AUC 0.67±0.12, PR AUC 0.60±0.12), followed by LGE+FDG+WMA, LGE+FDG+T1mapping, LGE+FDG, LGE, and FDG, respectively. The first model statistically significantly (p-value<0.01) outperformed the reference model LGE+FDG (bAcc 0.54±0.11, sensitivity: 0.67±0.18, specificity 0.55±0.19, ROC AUC 0.56±0.13, PR AUC 0.47±0.10). Conclusion This study documents the benefits of a hybrid PET/MR multiparametric assessment beyond the standard analysis based on FDG uptake and LGE for the prediction of contractility recovery after revascularization of CTO of the coronary arteries. Funding Acknowledgement Type of funding sources: Public grant(s) – EU funding. Main funding source(s): This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 764458 PETMR protocolROC AUC comparsion