Real-world AI applications in cardiology

Artificial intelligence (AI) is reshaping heart failure (HF) diagnosis, providing novel tools such as voice-based risk prediction and AI-electrocardiography (ECG) algorithms that consistently outperform humans in identifying HF with reduced ejection fraction (HFrEF) and atrial fibrillation (AF). Real-world applications such as automated CT analysis of coronary inflammation and generative AI tools may offer diagnostic insights.

“Medicine is entering a new era where machines don’t just assist; they reason, predict, and evolve,” said Prof. Thomas Felix Lüscher (Royal Brompton Hospital, UK) [1]. In an era of increasing data complexity, such as omics and imaging, he pointed to the obvious: that AI tools offer unparalleled scalability and pattern recognition capacity, which human clinicians alone cannot match.

Prof. Lüscher began with vocal biomarkers, a novel area where AI models trained on just 5 sentences spoken by patients with HF can now detect fluid retention and predict outcomes. “The voice becomes a mirror of cardiac function,” he explained. Studies suggest voice analysis can stratify risk for hospitalisation and death, potentially offering a passive, non-invasive monitoring tool for remote care [2].

Deep learning has radically advanced ECG interpretation. “An AI-enabled ECG can now tell you if a patient had AF, even if they’re in sinus rhythm at the time of recording,” Prof. Lüscher said. He described studies in which neural networks predicted the likelihood of HFrEF or HFpEF from a single 12-lead ECG with over 90% accuracy. “This could empower primary care and emergency settings, where echo isn’t always immediately available.”

He emphasised the real-world impact: in Taiwan, the implementation of AI-ECG alerts in emergency departments led to a 30% reduction in mortality for high-risk patients [3]. “It’s not just innovation; it’s measurable impact.”

Cardiac imaging is another domain of interest. “AI already outperforms human readers in assessing left ventricular function,” Prof. Lüscher noted, referring to real-world examples of automated ECG with superior reproducibility. In coronary CT, he cited AI algorithms that quantify perivascular fat attenuation as a marker of inflammation surrounding the coronary arteries [4]. “We now have imaging biomarkers that go beyond stenosis. Inflammation is a predictor of events—sometimes even more than plaque burden.”

Prof. Lüscher also demonstrated the potential of generative AI, specifically large language models such as GPT-4, in synthesising complex clinical presentations. In a head-to-head trial at his institution, a case of cardiac sarcoidosis was presented to both experienced consultants and GPT-4. “Remarkably, GPT-4 gave the correct diagnosis on the first attempt,” he reported. “It may soon become a decision-support tool in routine care.”

Beyond diagnostics, Prof. Lüscher addressed the promise of precision cardiology, an approach integrating clinical, genomic, metabolomic, and imaging data, analysed through machine learning. He cited the use of AI to stratify patients based on genetic variants linked to cardiomyopathies, as well as emerging applications in predicting drug response and adverse events.

But he also issued a note of caution. “AI models must undergo the same rigour as any new therapy: external validation, reproducibility, and ethical oversight are essential,” he said. He called for the development of regulatory frameworks to ensure transparency, safety, and patient trust.

In conclusion, Prof. Lüscher stated: “AI won’t replace cardiologists, but cardiologists who use AI may replace those who don’t. Our responsibility is to guide it wisely, integrate it safely, and keep the patient at the centre of every innovation.”

1. Lüscher TF. precision medicine and tailored treatment. Heart Failure 2025, 18 May, Belgrade, Serbia.
2. Amir O, et al. J Am Coll Cardiol HF. 2020;8(9):712–722.
3. Liu WC, et al. J Pers Med. 2021 Nov 4;11(11):1149.
4. Nishii T, et al. Acad Radiol. 2023 Nov;30(11):2505-2513.

Medical writing support was provided by Dr Rachel Giles.

 

Brought to you by Pfizer

PP-VYN-NLD-0443



Posted on 24 June 202526 June 2025