An AI algorithm that analyzes entire coronary arteries via OCT imaging more accurately predicts adverse events than expert analysis of target lesions.

Original Title: Artificial intelligence-based identification of thin-cap fibroatheromas and clinical outcomes: the PECTUS-AI study

Journal: European heart journal

DOI: 10.1093/eurheartj/ehaf595

AI Identifies High-Risk Plaques to Predict Outcomes

Overview

This study investigated an artificial intelligence algorithm, called OCT-AID, for its ability to predict future cardiovascular problems. The research was a secondary analysis involving 414 patients who had previously experienced a heart attack. These patients had undergone optical coherence tomography (OCT) imaging of their coronary arteries. The AI and a core laboratory of human experts independently analyzed these images to detect high-risk plaques known as thin-cap fibroatheromas (TCFA). The presence of AI-identified TCFA in a target lesion was significantly associated with adverse outcomes over two years (Hazard Ratio 1.99), whereas the association for expert-identified TCFA was not statistically significant. The study aimed to validate the AI’s prognostic value against the expert standard.

Novelty

While experts have linked TCFA to poor outcomes, manual analysis is slow and variable. This study’s novelty is in evaluating a fully automated AI for this task and comparing it to the expert standard. A key finding was that analyzing the entire imaged artery with AI provided much stronger risk prediction (Hazard Ratio 5.50) than analyzing only the target lesion. This comprehensive vessel-wide assessment is practically unfeasible for manual expert analysis, highlighting a distinct advantage of the automated approach. The system demonstrated superior discriminatory ability for outcomes when analyzing the full vessel compared to the core lab’s analysis of the target lesion (C-statistics 0.66 vs. 0.56).

My Perspective

I find the relationship between the AI and the expert analysis particularly noteworthy. The study reported only fair to moderate agreement between the two, which might initially seem like a weakness of the AI. However, the AI’s predictions were more strongly associated with actual patient outcomes. This suggests the AI may be identifying subtle, yet prognostically critical, features that are not consistently captured by human interpretation. The AI’s strength is not just in mimicking human experts but in its exhaustive, frame-by-frame analysis of the entire vessel, free from fatigue or bias. This capability may allow it to establish a more direct and reliable link between plaque characteristics and clinical risk, potentially redefining the “ground truth” from expert opinion to patient outcome.

Potential Clinical / Research Applications

In a clinical setting, this technology could be integrated into cardiac catheterization labs to provide real-time, objective risk stratification. An automated analysis could help physicians make immediate decisions about treating non-flow-limiting plaques that are identified as high-risk. The algorithm’s high negative predictive value (97.6% for full-segment analysis) is especially valuable, as it could reliably identify patients who are at very low risk of future events, thereby avoiding unnecessary procedures. For research, this tool offers a standardized and reproducible method for plaque quantification. This can reduce variability in clinical trials, making them more powerful and efficient. It also facilitates large-scale analysis of entire coronary arteries, which could lead to the discovery of new imaging biomarkers for cardiovascular disease.

Similar Posts

  • Ensuring Health Equity in the Medical AI Revolution

    Original Title: Keeping Health Equity at the Forefront of the Artificial Intelligence Revolution in Medicine and Health Journal: JAMA health forum DOI: 10.1001/jamahealthforum.2025.6477 Overview OverviewThe rapid deployment of artificial intelligence in healthcare offers potential for increased efficiency and improved health outcomes. However, significant concerns exist regarding its impact on health equity. Historically, technological innovations have often benefited advantaged populations first, a phenomenon known as the 'inverse equity hypothesis'. Evidence from studies across 89 low- and middle-income countries demonstrates that without deliberate strategies, new technologies widen existing health gaps. Digital health tools frequently sustain inequities related to socioeconomic status, race, and geographic location. For instance, individuals with lower socioeconomic status are…

  • AI-Powered Speech Analysis for Alzheimer’s Detection

    Original Title: Biomarkers Journal: Alzheimer's & dementia : the journal of the Alzheimer's Association DOI: 10.1002/alz70856_107467 Overview The study investigates the utility of spontaneous speech as a non-invasive biomarker for Alzheimer's disease by developing an automated analysis pipeline. Utilizing the ADReSS 2020 Challenge dataset, which comprises audio recordings from 108 training participants and 48 testing participants performing the Cookie Theft picture description task, the researchers explored the transition from raw audio to diagnostic classification. The methodology involved transcribing audio using commercial tools like OpenAI Whisper and AssemblyAI, followed by the generation of semantic vector embeddings using large language models. These embeddings were then used to train machine learning classifiers, including…

  • Deep Learning MRI Super-Resolution for Alzheimer’s Atrophy

    Original Title: Biomarkers Journal: Alzheimer's & dementia : the journal of the Alzheimer's Association DOI: 10.1002/alz70856_107471 Overview Alzheimer's disease involves grey matter loss in regions like the hippocampus. Accurate atrophy measurement is essential for monitoring progression. Deformation Based Morphometry (DBM) quantifies these changes but is limited by the 1 millimeter cubed resolution of standard Magnetic Resonance Imaging. This study evaluates whether deep learning-based super-resolution improves the detection of subtle brain changes. The researchers used a dataset of 497 individuals from the Alzheimer’s Disease Neuroimaging Initiative. They compared standard 1 millimeter resolution images against high-resolution 0.5 millimeter isotropic images generated via an autoencoder-based model. By correlating measurements with ADASCog13 cognitive scores,…

  • Automating Expert-Level Medical Reasoning Evaluation for AI

    Original Title: Automating expert-level medical reasoning evaluation of large language models Journal: NPJ digital medicine DOI: 10.1038/s41746-025-02208-7 Overview Large language models increasingly assist in clinical decision-making, yet their internal reasoning processes often remain opaque. Current evaluation methods frequently rely on multiple-choice question accuracy, which fails to capture whether a model reached a correct conclusion through sound medical logic or mere pattern matching. While human expert review provides a highly reliable assessment, it is time-consuming and difficult to scale. To address these limitations, researchers developed MedThink-Bench, a dataset of 500 complex medical questions across ten domains, including pathology and pharmacology. Each question is paired with expert-authored, step-by-step reasoning paths. Alongside this…

  • A study of 950 AI medical devices found that lack of clinical validation and public company status were linked to higher odds of early recalls.

    Original Title: Early Recalls and Clinical Validation Gaps in Artificial Intelligence-Enabled Medical Devices Journal: JAMA health forum DOI: 10.1001/jamahealthforum.2025.3172 AI Medical Device Recalls and Validation Gaps Overview Artificial intelligence-enabled medical devices (AIMDs) are increasingly common in clinical practice, yet many receive US Food and Drug Administration (FDA) clearance through an accelerated pathway that does not require prospective human testing. This raises concerns about their performance and safety after entering the market. This study investigated the frequency of recalls among AIMDs and examined whether recalls were associated with two key factors: the lack of premarket clinical validation and the type of manufacturer (publicly traded vs. privately held). Researchers analyzed 950 FDA-cleared…

  • An AI algorithm for coronary imaging standardizes high-risk plaque detection, improving risk prediction when assessing the entire vessel.

    Original Title: Artificial intelligence-based identification of thin-cap fibroatheroma: a new paradigm for risk stratification? Journal: European heart journal DOI: 10.1093/eurheartj/ehaf662 AI for Identifying Risky Heart Plaques Overview Atherosclerosis involves the buildup of plaques in arteries, but not all plaques are equally dangerous. Thin-cap fibroatheromas (TCFAs) are considered particularly high-risk and are associated with heart attacks. Identifying these TCFAs using intracoronary imaging like optical coherence tomography (OCT) is challenging, as manual interpretation by experts can be time-consuming and inconsistent. This editorial examines the PECTUS-AI study, which tested an artificial intelligence algorithm designed to automatically detect TCFAs from OCT images in 414 patients who had recently suffered a heart attack. The study…

Leave a Reply

Your email address will not be published. Required fields are marked *

CAPTCHA