Artificial Intelligence for Preventing Heart Disease: Observational, Single Center, Prospective and Retrospective Study
Artificial Intelligence for Preventing Heart Disease: Observational, Single Center, Prospective and Retrospective Study
Coronary artery disease (CAD) is a leading cause of mortality in western countries. Coronary computed tomography angiography (cCTA) is the first-line imaging test in patients with suspected obstructive CAD. However, in most patients, cCTA shows non-obstructive CAD. The management of patients with non-obstructive CAD is unclear. This is due to the lack of cCTA-based methods capable to assess the risk of disease progression towards developing major adverse cardiovascular events (MACEs) based on the atherosclerosis characteristics of each patient. A solution for prognostication in these patients is particularly appealing since it could allow to identify patients who can benefit of a more aggressive medical treatment and management, thus improving outcome.
Proposed methods, which include qualitative evaluations such as the identification of adverse atherosclerotic plaque characteristics or quantitative evaluations such as the quantification of atherosclerotic plaque burden, may in some cases suffer of limited reproducibility between operators and software. Most importantly, each single biomarker is insufficient to accurately predict patient risk, hence potential synergic integration of cCTA and clinical biomarkers is the key to efficiently guide the personalization of patient's management. Furthermore, the few risk stratification methods that have been proposed are not designed to work on platforms capable of deploying the solution to other clinical settings, promoting prospective or external validation
Background and Rationale
In recent years, cCTA has become a crucial diagnostic tool for suspected CAD, recommended as the first-line test for patients with an intermediate pre-test probability of CAD by the European Society of Cardiology (ESC) guidelines and the American HeartAssociation (AHA) and Italian guidelines , because of the well-known high negative predictive value of cCTA in ruling out obstructive CAD. However, most patients have non-obstructive CAD.
While the management of patients with obstructive CAD is established, as it revolves around further diagnostic test for ischemia evaluation or upfront coronary artery revascularization, this is not the case for patients with non-obstructive CAD. However, these cohort of patients still has a significant risk of developing major adverse cardiovascular events (MACEs) that could be prevented by implementing adequate medical therapy.
To date, many approaches have been proposed to tackle this issue. However, these proposed solutions lack the ability to provide quantitative and reproducible results with a sufficiently strong predictive value, are often proposed as a stand-alone solution without the integration with multiple prognosticator imaging and clinical parameters, and are not delivered through platforms capable of providing external validation and easy integration in the clinical workflow. Among the proposed prognostic approaches, some are based on the qualitative evaluation of coronary artery plaque features, such as positive remodeling, low attenuation of the plaque, presence of spotty calcification, and "napkin ring" sign , which is subject to significant inter-reader variability.
Other approaches rely on quantitative methods for evaluating atherosclerotic burden based on the extent of coronary artery plaques and their characteristics, such as calcium density, number of lesions, regional distribution, plaque volume, non-calcified plaque volume etc.
However, these approaches may be hampered by low reproducibility, especially among different scanner vendors. Interestingly, a new research has also shown that, besides coronary artery vessel wall characteristics, pericoronary adipose tissue attenuation carries significant predictive value, as it reflects the state of coronary inflammation that plays a key role in the development and progression of coronary atherosclerosis.
All these CAD characteristics are often analyzed independently from one to another, reducing their potential synergistic prognostic value and creating redundant variables that have negligible effect on prognosis. We propose an AI-based analysis that can integrate all this data in order to select the most important determinant of CAD progression and to discard futile features, thus creating an agile and clinically valuable risk stratification model.Furthermore, we plan to create a novel imaging marker of CAD with unfavorable outcome, to be integrated in the AI-based model, which will be based on topological features of the coronary artery tree. In fact, data on the association between coronary artery topology (e.g., vessel-length, coronary artery volume index, cross-sectional area, curvature, and tortuositv) and prognosis is scarce. However, it is known that vessel tortuosity influences wall shear stress and leads to disruption of laminar flow, resulting in endothelial dysfunction and flow alterations that may lead to atherosclerosis, eventually causing adverse cardiac events . Thus, this novel biomarker may carry a significant prognostic role. Based on these premises, our research aims to develop a novel clinical-imaging AI-based model to identify and categorize patients at high risk of disease progression and provide a more personalized management approach to improve patient outcomes.
However, besides the primary objective of creating an AI-based model for CAD risk stratification, we aim to overcome some issues that currently hamper the widespread clinical application of AI in cardiovascular care. In fact, it is recognized that the integration of AI-based applications into the clinical workflow, which will increase usability and decrease costs, is currently lacking.
We aim to tackle these issues with the help of the industrial partners involved in this project that will build a platform capable of delivering the software solution to provide external validation of the algorithm.
This platform will be characterized by state-of-the-art security measures, interoperability with current clinical software, and easy-to-use interface.
Inclusion Criteria:
Exclusion Criteria: