OCT image interpretation predicts recurrent heart attacks

04 Sep 2025

Radboud University Medical Center uses AI to assess coronary artery condition.

Optical coherence tomography (OCT) can provide detailed in vivo evaluation of coronary arteries, as a way to identify high-risk plaques.

Most studies of OCT for this purpose have relied on offline image analysis by specialist laboratories, however, slowing down the assessment process and moving it away from the clinician carrying out the OCT imaging.

A project at Radboud University Medical Center (Radboudumc) in the Netherlands has now demonstrated that AI can reliably take over this analysis, and assess arteries for weak spots based on OCT data.

"One of the challenges with this technique is that it is extremely difficult for physicians to interpret OCT images," commented Jos Thannhauser from Radboudumc.

"The technique is already used in clinical practice to guide angioplasty and to check whether a stent has been placed correctly, and even assessing just the stent placement is challenging. Only a handful of specialized labs can interpret these images, and even they cannot review everything."

Published in European Heart Journal the new study suggests that a combined technique can successfully identify patients at increased risk of adverse cardiovascular outcomes.

Towards clinical applications with AI

"It has been shown that OCT imaging reduces the risk of new infarctions and complications," said Thannhauser. "But in those cases, physicians only look at a very small part of an artery, the site of the infarction. Our study shows that this technique, combined with AI, has much greater potential to map entire vessels."

The project built upon its previous development of OCT-AID, an AI model for automated full-vessel segmentation of OCT images. OCT-AID is able to accurately quantify plaques and assess the thickness of the fibrous cap overlaying the arterial plaque, referred to as a thin-cap fibroatheroma (TCFA).

The new study assessed if the AI algorithm can identify TCFA when compared with core lab analyses, and whether AI-detected TCFA is associated with adverse clinical outcome in patients with myocardial infarction and non-culprit plaques.

Statistical analysis of data from 438 patients over a two-year trial showed that the agreement between AI-TCFA and assessment by a core laboratory was fair to moderate on a patient and lesion level, which the team believes it can improve upon. But AI-TCFA showed a significant association with adverse cardiovascular events, and evaluating image sequences from OCT pullback rather than just a target lesion showed that the technique offered a predictive indication of later patient outcomes.

"The study shows that AI detects vulnerable spots in the arterial wall just as well as specialized laboratories, and even predicts new infarctions or death within two years more accurately," said Rick Volleberg of Radboudumc. "If we know who has high-risk plaques and where they are located, we may in the future be able to tailor medication or even place preventive stents."