Abstract Body (Do not enter title and authors here): Introduction: Fractional Flow Reserve (FFR) is a stenosis-specific index for coronary artery stenosis (CAS) detection and its effect on myocardial perfusion. Visualizing stenosis through an invasive coronary angiography (iFFR) lacks consistency and is highly subjective. Recently, non-invasive techniques have been developed using artificial intelligence (AI) and Machine learning (ML) to measure FFR (ML-FFR). We aimed to analyze the diagnostic performance of ML-FFR in comparison to iFFR in the detection of CAS.

Methods: A search was carried out in PubMed, IEEE Xplore digital library, and clinicaltrials.gov where studies evaluating the risk of CAS in patients undergoing ML-FFR and iFFR were considered in terms of FFR values from 2014 to 2024. OpenMetaAnalyst was used to evaluate the diagnostic performance in terms of pooled Sensitivity(SN), Specificity(SP), Negative Likelihood Ratio(NLR), Positive Likelihood Ratio(PLR), and Diagnostic Odds Ratio(DOR) at 95% confidence interval (CI). P<0.05 was considered significant.

Results: A total of 9 studies including 1358 vessel lesions were analyzed. The pooled SN, SP, NLR, and PLR were 81.1%(74.3%-86.4,p=0.014), 91.8%(84.9%-95.7%, p<0.001), 0.187 (0.151-0.232, p=0.575) and 9.898(5.492-17.839,p<0.001), respectively with >75% heterogeneity in SP and PLR. After excluding two studies, a “leave-out” analysis showed pooled SN, SP, NLR, and PLR as 79.9% (71.4%-86.3%, p=0.007), 94.4% (90.2%-96.9%, p=0.002), 0.187 ( 0.140-0.250, p=0.540) and 14.307 (8.094-25.289, p=0.004) with <70% heterogeneity. The pooled DOR was 53.42 (28.26-100.99, p=0.005) at random effect model.

Conclusion: ML-FFR showed significant diagnostic values signifying that it can be used as a potential non-invasive alternative to iFFR for detecting lesion severity and guiding clinical decision-making in patients with suspected CAD, especially for those who are considered high-risk or ineligible for invasive angiography.