Peri-luminal COROnary CTa AI-driven radiOMICS to Identify Vulnerable Patients (CORO-CTAIOMICS)

CAD is a leading cause of mortality in Europe. cCTA is recommended to rule out obstructive CAD, but, in most patients, it shows non-obstructive CAD. The management of these patients is unclear due to lack of reproducible quantitative measurement, beyond stenosis severity, capable to assess the risk of disease progression towards developing MACEs. To improve identification and phenotypization of patients at high risk of disease progression, the investigators propose the application of artificial intelligence algorithms to cCTA images to automatically extract periluminal radiomics features to characterize the atherosclerotic process. By leveraging machine-learning empowered radiomics the investigators aim to improve patients' risk stratification in a robust, quantitative and reproducible fashion. By developing a novel quantitative AI based cCTA measure, the investigators expect to provide a risk score capable to identify patients who can benefit of a more aggressive medical treatment and management, thus improving outcome

Background and rationale In the last years, cCTA has become a pivotal diagnostic tool in the setting of suspected CAD. The latest ESC guidelines recommend cCTA as the first line diagnostic test for patients with symptoms suspected to originate from CAD, especially in intermediate pre-test probability (15 to 85%). patients. The AHA and Italian guidelines essentially suggest the same approach. The recommendations are based on the extremely high accuracy of cCTA in ruling-out obstructive CAD. As result the investigators assist to a tremendous increase in the number of patients undergoing CCTA in the daily clinical routine. Most of these patients (80%) result to not have obstructive CAD, but this result does not mean that all this patients are at low risk. The prognostic stratification of these patients is still an urgent unmet need. Among patients with absence of obstructive CAD the investigators found patients with different degree of atherosclerotic burden and patients with different kind of plaques (high or low risk plaques) and different degrees of coronary wall and pericoronary inflammation. The prognostic stratification of these patients is still an urgent unmet need. The cCTA images include a lot of information with a great potential informative content about the atherosclerotic burden and the vulnerability of the non-obstructive CAD of each single patients, but all these information are not currently exploited in the clinical routine to change the patients management for a lack of robust and reproducible tools to extract this data in quantitative way and o integrate them in a prognostic risk score. In the last years, it has been shown that coronary artery plaque characteristics (e.g. lesion length, volume, stenosis, attenuation, remodeling index, etc) (2) and pericoronary adipose tissue (3) attenuation carry a significant prognostic value. However, it is known that many of these biomarkers suffer from low reproducibility, mainly due to technical constrains in automatic or semiautomatic separation of plaques from surrounding adipose tissue. This may cause the incorrect inclusion of plaques into the segmentation of pericoronary adipose tissue and vice-versa, leading to unreliable results in the assessment of plaque burden and plaque attenuation (4). Furthermore, the qualitative or semiquantitative evaluation of the aforementioned plaque characteristics and pericoronary fat density might reflect only part of the information available. In this scenario, radiomics-based assessment may unveil information hidden to the human eye, leading to better risk stratification of cardiovascular risk (5, 6). This study adds two major improvements in prognostic risk stratification based on cCTA plaque and adipose tissue analysis. First, the investigators propose a method to segment both the plaques and the pericoronary adipose tissue easily and independently by semiautomatic or automatic identification of the edge between plaque and fat (which is critical). In fact, this tool will perform a segmentation of all the tissue included in a circular range outside the coronary lumen with a diameter based on the lumen diameter itself. This technical solution will greatly increase reproducibility of segmentation. Secondly, the investigators propose the use radiomics to analyze this circular pericoronary milieu, potentially individuating novel biomarkers -invisible to human eye- of CAD instability capable of providing important prognostic value. In detail, the novelty of this research lies in the fact that 1) pericoronary adipose tissue and plaques will be treated as a single milieu, thus significantly reducing the effort for accurate tissue segmentation and reducing reproducibility concerns; 2) radiomics will be applied to extract meaningful information invisible to human eye capable. Furthermore, this radiomics approach will be supported by machine learning (ML) models including regularized regression, genetic algorithms and deep learning, due to the capability of ML to directly manage and assess the huge amount of data extracted from the radiological images. Thus, the final outcome of this research will be an algorithm capable of predicting the risk of MACEs of the single patient by automatically analyzing peri-luminal coronary tissue radiomics data derived from cCTA performed in the routine clinical practice.