Predicting High-Risk Carotid Plaque: Integrating radioMics and biomEchanics

Stroke is a leading cause of mortality and morbidity globally, accounting for 12 million cases worldwide in 2019 and a global cost of $891 billion in 2017. One of the main causes of stroke is the rupture of atheromatous plaque in the carotid artery. For this reason, identifying and studying plaques is essential to preventing stroke and cerebrovascular events.

European guidelines recommend risk stratification strategies for carotid plaques based on the degree of stenosis they cause and the assessment, through radiological imaging, of their core characteristics. However, this stratification alone is still not enough. In recent years, scientific research has succeeded in identifying additional markers of instability in the morphological and compositional characteristics of plaques based on imagery.

Finally, radiomics, a new field of high-throughput extraction of quantitative features observable in radiological images, has demonstrated promising results in recognizing and stratifying unstable carotid plaques.

The overall aim of the study is to create a predictive model that aims to identify unstable plaques in asymptomatic patients at initial diagnosis. Currently, the most effective objective method for assessing plaque instability is histopathological analysis, but this is only available after surgery. Therefore, it is crucial to have a noninvasive predictive model for identifying the risk of cerebrovascular accidents in asymptomatic patients.

Specifically, the primary objective is to identify carotid plaque instability by extracting radiomic markers belonging to the "shape and size" (26 shape and size features), "first-order statistics" (19 statistical features), and "textural" (75 texture features) classes from each CTA image. A further objective will be the calculation of mechanical markers (von Mises stress, maximum principal stress, ultimate stress, and thickness) capable of identifying carotid plaque instability in patients with carotid stenosis and candidates for carotid artery stenosis surgical removal.

The study will involve patients attending the Foundation's Vascular Surgery Unit and candidates for carotid artery stenosis. Two specific phases are planned. The first phase will be retrospective and will involve the retrieval of clinical and imaging data to activate machine learning for the analysis of radiomics data in combination with clinical data. Preoperative CT images and perioperative clinical data will be analyzed for training and preliminary development of a machine learning model based on radiomics, clinical, and biomechanical data, calculated from structural and fluid dynamics numerical simulations. Patient-specific preoperative CT images will be segmented using 2D convolutional neural networks (CNNs). The resulting 3D models of vessels and plaques will be used for structural and fluid dynamic simulations and radiomics analysis.

In the second prospective phase, patients will be enrolled consecutively, according to the SOC's indication for surgery. After obtaining written informed consent, the inclusion/exclusion criteria will be evaluated and each patient will be assigned a progressive identification code. Preoperative clinical and imaging (CT) data will be collected. The prospectively enrolled patients will undergo the same procedure (carotid TEA), according to the protocols and standards proposed by the European Society of Vascular Surgery (ESVS) and adopted in common clinical practice by our SC.

At the end of the TEA procedure, in addition to collecting pre- and post-operative clinical data, histopathological and biomechanical analysis of the surgically removed plaque, which is usually not analyzed and disposed of, will be performed. The results will be used to prepare and validate the virtual predictive model. The surgeon will send the removed material, in alternating sequence, once for histology and once for mechanical testing. Subsequently, numerical strategies will be adopted to calculate the biomechanics of carotid arteries affected by the presence of plaque. In addition to these numerical approaches, a machine learning algorithm will be used to couple the previously identified variables (i.e., morphological, histological, radiomic, and biomechanical) to identify markers capable of identifying unstable plaques.

Finally, 3D printing will be used to produce anatomical and functional models of the segmented geometries of patients undergoing TEA. These prints will be inserted into a hydraulic circuit capable of reproducing physiopathological flow and pressure conditions. This strategy will be used to validate the results from the structural and fluid dynamic simulations in a fully controlled environment in terms of postoperative geometric conformation (postoperative diameters achieved by the anatomy) and flow distribution at the fluid dynamic level.