Identifying Vulnerable CoronAry PLaqUes With Artificial IntElligence-assisted CT Angiography

Development and Validation of Multi-scale Deep Neural Network-Based CT Intelligent Diagnosis System for Coronary Vulnerable Plaques: A Chinese Multicenter Study

The goal of this observational study is to develop an automatic whole-process AI model to detect, quantify, and characterize plaques using coronary CT angiography in coronary artery disease patients. The main questions it aims to answer are:

1. Whether the AI model enables to detect and quantify coronary plaques compared with intravascular ultrasound or expert readers;
2. Whether the AI model enables to identify vulnerable plaques using intravascular ultrasound or optical coherence tomography as the reference standard.
3. Whether the AI model enables to predict future adverse cardiac events in a large cohort of 10,000 patients with non-obstructive CAD.
4. Whether the AI model enables to influnece downstream clincial decision-making.

Coronary artery disease (CAD) remains the leading cause of death worldwide. Atherosclerotic plaques play a pivotal role in CAD-related patient mortality. Thus, the detection, quantification, and characterization of coronary plaques are clinically significant for early prevention and interventions for CAD.

Coronary CT angiography (CCTA) has emerged as a robust noninvasive tool for the evaluation of CAD. In clinical practice, the coronary plaque assessment is performed by a time-consuming manual process dependent on the clinician's experience and subjective visual interpretation. With the development of artificial intelligence, many automatic computer-aided methods have been proposed to post-process the CCTA images. However, previously proposed algorithms of plaque evaluation were not developed based on intravascular ultrasound (IVUS) or optical coherence tomography (OCT), which were regarded as the gold reference for plaque evaluation. Thus, we aimed to develop a deep learning model in a whole-process automatic and intelligent system on CCTA to detect, quantify, and characterize plaques using IVUS or OCT as reference standard. Then we will work on the validation in different clinical scenarios: (1) Validation of the accuracy of the new deep learning model; (2) Prognosis of the model in different populations with CAD; (3) Impact of the model on guiding clincial therapies.

The main questions it aims to answer are:

1. Whether the AI model enables to detect and quantify coronary plaques compared with intravascular ultrasound or expert readers;
2. Whether the AI model enables to identify vulnerable plaques using IVUS or OCT as the reference standard.
3. Whether the AI model enables to predict future adverse cardiac events in a large cohort of 10,000 patients with non-obstructive coronary artery disease (China CT-FFR study 2).
4. Whether the AI model enables to influnece downstream clincial decision-making in real-world clincial practice.

Inclusion Criteria:

• Intravascular imaging (including intravascular ultrasound or optical coherence tomography) was performed within 3 months after CCTA;
• No change in medications or clinical symptoms during CCTA and intravascular imaging examinations;
• Coronary artery diameter stenosis of 30% to 90% on invasive coronary imaging.

Exclusion Criteria:

• Image quality of CCTA or intravascular US was inadequate to analyze;
• Intravascular imaging was performed after percutaneous coronary intervention (PCI) or pre-dilation of the target lesions;
• Lesions could not be co-registered between CCTA and intravascular US;
• Missing CCTA or intravascular US data