# The diagnostic performance of machine learning-based FFRCT for coronary artery disease: A meta-analysis

## TL;DR

This study shows that machine learning-based FFRCT is highly accurate in diagnosing coronary artery disease compared to invasive tests.

## Contribution

The study provides a meta-analysis of ML-FFRCT's diagnostic performance for CAD, supporting its clinical use.

## Key findings

- Pooled sensitivity was 0.84 and specificity was 0.83 for ML-FFRCT in diagnosing CAD.
- The diagnostic odds ratio was 25.15, and the area under the curve was 0.90, indicating strong accuracy.
- No significant publication bias was found in the analysis.

## Abstract

This meta-analysis evaluates the diagnostic accuracy of machine learning-derived FFRCT (ML-FFRCT) for CAD, using invasive coronary angiography-derived fractional flow reserve (ICA-FFR) as the gold standard to provide evidence for clinical translation.

We systematically searched PubMed and Embase for relevant studies. Study quality was assessed using QUADAS-2 in RevMan 5.3. Diagnostic performance was evaluated by pooling sensitivity (SEN), specificity (SPE), positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and the area under the curve (AUC) using Stata 14.0. Meta-regression and subgroup analyses were conducted based on the publication year, country, study design, sample source, and sample size.

The pooled SEN was 0.84 (95% CI: 0.79–0.87) and SPE was 0.83 (95% CI: 0.77–0.88). The PLR and NLR were 4.95 (95% CI: 3.58–6.84) and 0.20 (95% CI: 0.15–0.26), respectively. The DOR was 25.15 (95% CI: 14.87–42.52) and the AUC was 0.90 (95% CI: 0.87–0.93), indicating high diagnostic accuracy. Deeks’ funnel plot revealed no significant publication bias.

ML-FFRCT demonstrates high SEN and SPE in diagnosing CAD. These findings support its potential as a promising noninvasive tool for CAD assessment.