The Machine Learning Models in Major Cardiovascular Adverse Events Prediction Based on Coronary Computed Tomography...

Area of Science:

• Cardiology
• Radiology
• Artificial Intelligence

Background:

• Coronary computed tomography angiography (CCTA) is a primary noninvasive test for high-risk coronary artery disease (CAD).
• Machine learning (ML) enhances CCTA's diagnostic accuracy for major adverse cardiovascular events (MACEs).
• Radiomics offers multidimensional features for identifying high-risk patients and improving CCTA performance, though its role in MACE prediction is debated.

Purpose of the Study:

• To evaluate the diagnostic value of ML models using CCTA-derived radiomic features for MACE prediction.
• To compare the performance of various ML algorithms and models.
• To provide clinical recommendations for MACE diagnosis, treatment, and prognosis.

Main Methods:

• A systematic search of 5 databases (PubMed, Web of Science, etc.) up to September 10, 2024, identified studies on ML models, CCTA, and MACE prediction.
• Risk of bias was assessed using the Prediction Model Risk of Bias Assessment Tool.
• Radiomics quality score (RQS) and TRIPOD guidelines ensured methodological quality and transparency.
• Meta-analysis was conducted using Meta-DiSc and StataMP, with subgroup analysis for heterogeneity.

Main Results:

• Ten studies were included, with 17 models in training sets and 26 in testing sets.
• Pooled area under the receiver operating characteristic (AUROC) curves were 0.7879 (training) and 0.7981 (testing).
• Logistic regression (LR) models achieved AUROCs of 0.7983 (training) and 0.8229 (testing); Random Forest (RF) models reached 0.8444 (training).

Conclusions:

• ML models using CCTA radiomic features outperform general models for MACE prediction.
• LR-based ML models show significant clinical potential, especially when integrated with clinical features.
• Further validation through extensive clinical trials is recommended for LR-based ML models.