New review shows how artificial intelligence could make nuclear cardiology safer, faster and more accurate.

Artificial intelligence (AI) is moving from the lab into the imaging suite—especially for nuclear cardiology tests such as SPECT and PET scans that check blood flow to the heart. A 2023 review in Cardiology Clinics maps where AI is already helping and what still needs to happen before it’s everywhere in clinical care.  

What’s new

• Cleaner images at lower dose. AI techniques can denoise and reconstruct scan data, which could maintain image quality while using less radiation or shorter scan times—an obvious win for patient safety.  
• Fewer technical hiccups. Algorithms can fix common issues like misregistration between scan types and even simulate attenuation‑correction images, reducing repeat scans.  
• Sharper reads and risk forecasts. Machine‑learning models that combine clinical, stress, and imaging information often beat traditional “by‑hand” measurements for detecting obstructive coronary artery disease and for predicting future cardiac events. Deep‑learning systems can also read images directly to estimate risk—no manual feature picking required. (See the review’s “Key Points,” p. 1.)  
• More transparent AI. “Explainable AI” tools (such as heat maps that show what parts of an image drove a decision) are being tested to help clinicians trust and verify model outputs—an essential step toward everyday use.  

Why it matters

If these tools continue to validate well, patients could see shorter appointments, lower radiation exposure, and fewer unnecessary invasive tests, while clinicians gain a second set of eyes for tough calls. The review underscores that AI’s biggest clinical value may be risk stratification—using what’s already in the images to flag who needs closer follow‑up and who can safely avoid extra testing.  

Important caveats

• Trust but verify. The review stresses rigorous training–testing separation, external validation, and attention to “temporal shift” (when performance drifts as practice patterns change). In short: no shortcuts on evaluation.  
• Mind the bias. AI can inherit biases from the data used to train it. Encouragingly, nuclear cardiology teams are already publishing methods to detect and mitigate bias, but it remains a work in progress.  
• Implementation matters. Even great models can stumble if they don’t fit real‑world workflows. The authors highlight simplified models and explainability as practical bridges from research to routine care. (See “Clinics Care Points,” p. 10.)  

What patients can ask right now

• Will my scan use any AI tools? If so, how do they help with image quality or interpretation?  
• Does AI change my radiation dose or time in the scanner? Many reconstruction tools aim to safely reduce both.  
• How are results double‑checked? Ask how AI outputs are reviewed and how your care team handles edge cases.  

Bottom line

AI is not replacing heart doctors—it’s enhancing them. In nuclear cardiology, the technology is already improving image quality and helping to sort patients by risk. With careful validation and attention to bias and transparency, these tools could make cardiac imaging safer, faster, and more precise for millions of people.  

Source: Miller RJH. “Artificial Intelligence in Nuclear Cardiology.” Cardiology Clinics (2023).  

Editor’s note: This article is for information only and is not a substitute for professional medical advice.