Table 1.

Eligibility criteria for original studies screened for this systematic review.

Items	Inclusion criteria	Exclusion criteria
P (Populations)	General population	None
E (Exposure)	Studies involving the complete development of a machine learning model (MLM) for CA diagnosis, with no restrictions on the types of predictor variables;	(1)Studies that only performed differential factor analysis without developing a complete MLM; (2)Studies that only performed image segmentation without constructing an MLM for identifying CA.
C (Control)	The control group for our study comprised non-CA subjects, including healthy individuals or patients with other cardiac conditions in the differential diagnosis of CA.	None
O (Outcomes)	The outcome measures were those used to evaluate model performance, including the receiver operating characteristic curve (ROC), c-statistic, c-index, sensitivity, specificity, accuracy, recall, precision, confusion matrix, diagnostic four-cell table, F1 score, and calibration curve.	Research lacking any outcome measures to assess the model accuracy.
S (Study design)	(1)Case-control studies, cohort studies, cross-sectional studies, and randomized controlled trials; (2)A limited number of studies employed only internal validation techniques (e.g., cross-validation or bootstrap validation). Despite the lack of an independent external validation set, these studies were included given the importance of evaluating overfitting in ML; (3)A small number of studies may be based on different ML studies published in the same open database. These studies were also included; (4)Research published in English.	(1)Meta-analyses, reviews, guidelines, and expert opinions; (2)Studies with a sample size of fewer than 20 total cases. This criterion ensures compliance with the Events Per Variable (EPV) rule of thumb of >10 for multivariate MLM development to mitigate overfitting. As a minimum of two predictor variables is expected in such models, a lower limit of 20 cases was set.