How does an AI-based angiography system fare in PCI?

In the FLASH study, a recently developed artificial intelligence-based quantitative coronary angiography (AI-QCA) system is noninferior to optical coherence tomography (OCT)-guided percutaneous coronary intervention (PCI) in achieving optimal post-PCI minimal stent area (MSA) in patients with less complex coronary artery disease (CAD).

The post-PCI MSA was 6.3 mm² in the AI-QCA arm and 6.2 mm² in the OCT arm (difference, -0.16; p for noninferiority<0.001). The researchers noted that noninferiority was tested with a margin of 0.8 mm². [JACC Cardiovasc Interv 2025;18:187-197]

Similar patterns were seen on sensitivity analysis, particularly in the proximal (7.2 vs 7.1 mm² [AI-QCA vs OCT]; difference, -0.07; p for noninferiority<0.001) and distal half segments (6.5 vs 6.3 mm²; difference, -0.20; p for noninferiority<0.001), as well as the proximal (8 vs 7.4 mm²; difference, -0.59; p for noninferiority<0.001) and distal stented segments separated by a side branch >2 mm in diameter (6 vs 5.8 mm²; difference, -0.13; p for noninferiority=0.008).

OCT-defined secondary endpoints, safety

Save for a higher rate of stent malapposition in the AI-QCA vs OCT arm (13.6 percent vs 5.6 percent; p=0.007), there were no significant between-group differences in overall stent expansion (78.7 percent vs 79.2 percent; p=0.78), stent underexpansion (50.8 percent vs 54.6 percent; p=0.48), dissection (15.6 percent vs 12.8 percent; p=0.42), or untreated reference segment disease (15.1 percent vs 13.3 percent; p=0.61).

Immediate post-procedural safety outcomes and overall clinical event rate were uncommon and did not significantly differ between arms.

Bridging the gap

“Conventional angiography-guided PCI has long been the de facto strategy despite its lack of standardization and reliance on operators’ subjective visual estimation and experience, which can lead to inaccuracies and interobserver variability,” the researchers noted.

In the GUIDE-DES trial, 1-year outcomes were similar between protocolized PCI using onsite manual QCA and intravascular ultrasound-guided PCI. [JAMA Cardiol 2024;9:428-435] According to the researchers, findings from this trial underpin the potential of manual QCA-based stent and balloon size selection, along with routine high-pressure post-dilatation, to overcome the limitations of conventional angiography-guided PCI.

The current findings build on these results, with AI-QCA offering “a streamlined approach with a rapid, automated, and objective analysis of coronary angiograms in real time. This may improve procedural efficiency while reducing workflow interruptions tied to manual measurements.”

In this investigator-initiated trial, 400 patients (mean age 65.1 years, 81.8 percent men) were randomized 1:1 to AI-QCA-assisted or OCT-guided PCI with contemporary drug-eluting stents for significant coronary artery lesions from 13 participating centres in South Korea. AI-QCA analysis was done using the MPXA-2000 system, while OCT-guided PCI was performed according to a standardized protocol.

“FLASH introduces AI-QCA as a promising approach for guiding coronary intervention [and] is the first randomized study to show that AI-QCA technology can be effectively integrated into routine PCI practice, potentially bridging the gap between conventional angiography-guided PCI and state-of-the-art PCI guided by intracoronary imaging,” the investigators said.

They added that the system may be particularly beneficial in resource-limited settings or less complex CAD cases, for which the clinical benefits of intravascular imaging are not fully established.

The researchers called for larger studies exploring long-term outcomes to fully establish the role of AI-QCA-assisted PCI in daily interventional cardiology practice and trials conducted in other regions. “South Korea is a region where imaging-guided PCI is more frequently performed … To enhance generalizability of our findings, further validation of this system is necessary in non-Asian regions, particularly in areas where coronary imaging [is] underused.”