Current Landscape of Artificial Intelligence (AI) in Nuclear Cardiology: Insights from Seminars in Nuclear Medicine Review
A recent white paper by a working group of the International Atomic Energy Agency (IAEA) provided a comprehensive overview of the technical foundations, deployment areas, clinical significance, and existing challenges of Artificial Intelligence (AI) in nuclear cardiology.
The authors recommend promoting standardised acquisition and reporting practices, establishing globally representative reference datasets, advancing multimodality imaging frameworks, and developing AI-proficient clinical and technical personnel.
The review highlighted that under these conditions, AI has the potential to significantly enhance the diagnostic and prognostic capabilities of nuclear cardiology, support equitable implementation, and maintain clinical accountability.
The paper was published in December 2025 in the Journal Seminars in Nuclear Medicine
The Foundation of AI and Its Role in Nuclear Cardiology
Modern machine learning (ML)-based artificial intelligence (AI) relies on mathematical models or algorithms that iteratively improve their classification or predictive performance by exposure to new data. ML extends traditional statistical modelling by accommodating a large number of input variables and by capturing complex, non-linear relationships in the data. Among ML techniques, artificial neural networks (ANNs), which feature processing mechanisms such as convolutions and transformers, have demonstrated remarkable success in analyzing and generating images, text, and video. These advancements fall under the umbrella of deep learning (DL). There is a hierarchical relationship between AI, ML, and DL, with key DL architectures relevant to medical imaging, including convolutional neural networks (CNNs), generative adversarial networks (GANs), and transformer-based language models.
Implementation of AI across various stages of the nuclear cardiology workflow
A summary of machine learning (ML) applications and their advantages across the different stages of the nuclear cardiology workflow is presented in Table 1.
Table1: ML Applications Across the Nuclear Cardiology Workflow
| Workflow Stage | ML Applications | Potential Advantages |
| Patient Selection | Integration of clinical and imaging data Identify patients most likely to benefit from advanced imaging and stress-only imaging | Increased diagnostic efficiency Improved appropriate use criteria Reduced radiation exposure Fewer unnecessary procedures |
| Image Acquisition | Image reconstruction Denoising Motion correction Attenuation correction | Improved diagnostic accuracy Enhanced spatial and temporal resolution Reduced radiation exposure Accurate and personalized images |
| Image Processing | Automatic calcium scoring Quantification of cardiac activity | Reduced inter-reader variability Shorter reporting time Reduced radiation exposure Extends calcium scoring to non-dedicated scans |
| Image Interpretation | Highlighting abnormalities Structured report generation Extraction of structured data from free-text reports | Improved diagnostic accuracy Reduced inter-reader variability Shorter reporting time Supports research and communication |
| Prognosis and Risk Assessment | Integration of demographic, clinical, laboratory, and imaging data Personalized prognostic assessment and risk stratification | Improved prognostic accuracy Reduced bias in patients with atypical symptoms Supports clinical decision-making Advances precision medicine |
Foundations of safe and legally compliant AI in nuclear cardiology
As AI enters the field of nuclear cardiology, legal and clinical safety are vital. AI tools adapt to new data and require testing and monitoring.
The following three core areas are essential.
- Testing Before Use: AI tools must undergo verification and validation (testing on real clinical data). Skipping this step exposes hospitals to legal liabilities.
- Ongoing Monitoring: AI performance can decline over time owing to model drift from clinical environmental changes. A UK hospital's sepsis AI tool failed after changes to the laboratory. Continuous performance checks are essential to avoid unsafe usage.
- Traceability and Explainability: Maintaining logs of AI use and clinical overrides is vital for legal defence. Addressing the "black box" problem through XAI improves trust and compliance with decisions.
Implications for Stakeholders
AI has the potential to enhance multimodal nuclear cardiology by optimising the extraction of value from each imaging modality and by integrating imaging with clinical, functional, and other patient-level data, thereby addressing the limitations of individual techniques. The effective clinical implementation of AI will require the involvement of nuclear cardiology professionals trained in AI to facilitate the appropriate use, interpretation, and adaptation of AI tools, thereby supporting collaboration between humans and AI rather than replacement.
Reference: Wiefels C, Juárez-Orozco LE, Craviolatti PS, et al. Artificial intelligence in nuclear cardiology: Technical perspectives, strategic directions, and recommendations from an IAEA expert working group. Semin Nucl Med. Published online December 29, 2025. doi:10.1053/j.semnuclmed.2025.11.011
