Predicting Plaque Vulnerability Using Machine Learning-Enabled Nanoagents Sensitized Molecular High-Resolution Magnetic Resonance Imaging Data

Molecular imaging based on paramagnetic nanoagents has emerged as an intriguing strategy to sensitize the local magnetic properties of pivotal pathological processes related to atherosclerotic plaque destabilization, opening up a potential possibility for noninvasively predicting plaque vulnerability. Unfortunately, current magnetic resonance (MR) imaging interpretation fails to provide objectively and precisely quantitative imaging descriptors, thus showing limited values in stratifying the plaque risk from MR images. To address this need, we originated a synergistic nanoagents (tFM-Nanoagents)-assisted machine learning (nano-AML) technology for directly reading out plaque vulnerability from molecular high-resolution vessel wall MR imaging (HR-VWI). The proposed diagnostic paradigm provided a holistic visualization of the distribution of foamy macrophage-defined plaques; by using a machine learning (ML) approach to decode data of tFM-Nanoagents sensitized HR-VWI, an imaging-derived risk score (nano-AML score) correlating with the pathology vulnerability index of plaques was generated and validated in a preclinical atherosclerotic model. Our data showed that the nano-AML score could effectively phenotype plaques into "vulnerable" and "stable" classes, with an area under the curve (AUC) of 0.871 in the training cohort and 0.870 in the validation cohort. We also demonstrated that the predictive performance of nano-AML score outperformed that of commercial contrast agent Gadovist (AUC of 0.560 in the training cohort and 0.538 in the validation cohort), suggesting its robust potency for serving as a reliable predictor for vulnerable plaques.