Title : From attenuation correction to actionable cardiovascular intelligence: fusion CV for AI-enhanced multimodal SPECT/PET imaging
Abstract:
Background: Low-dose CT attenuation-correction (CTAC) images acquired during SPECT and PET myocardial perfusion imaging are often treated primarily as technical inputs, even though they contain potentially actionable anatomic information. Coronary artery calcium (CAC), epicardial adipose tissue (EAT), perfusion findings, electrocardiographic data, and clinical context usually remain separated across imaging and reporting systems. This fragmentation can limit recognition of atherosclerotic burden and complicate interpretation of equivocal or attenuation-prone studies.
Program description: Fusion CV is a clinician-centered, AI-enabled decision-support framework designed to transform same-session CTAC into reviewable cardiovascular intelligence and integrate it with myocardial perfusion imaging. The initial workflow assesses CTAC image adequacy, segments and localizes coronary calcification, estimates CAC burden with clinically interpretable risk categories and confidence measures, and aligns coronary anatomy with perfusion territories and polar-map displays. Structured outputs and fused overlays are intended to return to the existing nuclear cardiology workflow for physician verification and reporting. Subsequent modules extend the platform to EAT quantification and multimodal models that combine perfusion, CAC/EAT, stress ECG, and clinical variables to estimate the likelihood of angiographically significant coronary disease.
Validation and clinical value: Development is organized as a staged translational program. Analytical validation will examine segmentation accuracy, CAC agreement, registration, repeatability, and failure detection. Clinical validation will evaluate diagnostic discrimination, reclassification, prognostic associations, and performance across scanners, tracers, body habitus, and patient subgroups. Prospective usability testing will assess interpretation time, reader agreement, downstream testing, preventive treatment recommendations, and workflow integration. Potential value includes fewer equivocal interpretations and repeat studies, improved differentiation of artifact from ischemia, more consistent recognition of previously unreported coronary atherosclerosis, and better use of the information already acquired during the imaging encounter.
Conclusion: Moving CTAC from attenuation correction alone to multimodal cardiovascular intelligence represents a pragmatic path for responsible AI adoption in nuclear cardiology. Fusion CV is designed as non-autonomous clinical decision support that augments, rather than replaces, physician interpretation. Its emphasis on same-session data, uncertainty display, interoperability, auditability, and staged validation is intended to support safe translation from retrospective development to prospective clinical deployment.
