Early Detection of Cardiac Disease Using Nuclear Medicine Imaging

Coronary artery disease remains the leading cause of death in the United States, and the window for effective intervention is widest when disease is detected at the functional stage - when reduced myocardial perfusion during stress indicates flow-limiting disease in the absence of symptoms or with symptoms that could be attributed to other causes. Nuclear medicine provides the most validated modality for functional cardiac assessment, with a 25-year evidence base from the landmark COURAGE and ISCHEMIA trials. What is changing now is not the clinical value of myocardial perfusion imaging - it is the accuracy and operational efficiency of the AI tools interpreting it.

The MPI Workflow: Where Errors Enter

Myocardial perfusion imaging with SPECT (MPI) acquires gated cardiac images at rest and following pharmacological or exercise stress, using Tc-99m labeled perfusion agents (sestamibi or tetrofosmin). The diagnostic question is whether stress-induced perfusion defects - regions of relatively reduced tracer uptake indicating ischemia - are present, and whether they represent fixed or reversible defects. Fixed defects indicate prior infarction; reversible defects indicate ischemia in viable but jeopardized myocardium.

The standard reporting framework uses the 17-segment myocardial model, with each segment scored 0-4 (normal to absent uptake). Summed stress score (SSS), summed rest score (SRS), and summed difference score (SDS) provide quantitative measures of ischemic burden. An SSS above 13 identifies patients with severe perfusion abnormalities where revascularization reduces mortality. This quantification framework exists because it reduces the interpretive variability that characterizes purely visual MPI reads - but it is still frequently not applied consistently in community practice.

Three categories of error characterize MPI reads in practice. First, soft tissue attenuation artifacts - particularly diaphragmatic attenuation creating inferior wall defects in men and breast attenuation creating anterior wall defects in women - are misinterpreted as true perfusion defects, leading to false-positive reads and unnecessary catheterization. Second, balanced ischemia - equal reduction in perfusion in multiple territories due to three-vessel disease - may appear as a "normal" study because all regions are equally reduced. Third, right ventricular abnormalities and non-cardiac incidental findings on the CT component are missed or not systematically evaluated.

How AI Addresses MPI Interpretation Errors

AI-assisted MPI interpretation has the most mature evidence base of any nuclear medicine AI application. The earliest published studies date to 2018, and more than 15 peer-reviewed validation studies have since reported on specific systems. The clinical argument for AI in this context is not that physicians read MPI poorly - it is that they read it inconsistently, and that AI-assisted reads reduce the tail of poor performance more than they improve median performance.

The attenuation artifact problem is addressable by AI models trained to distinguish the spatial pattern of true perfusion defects (which follow coronary territory distributions) from soft tissue attenuation patterns (which correlate with patient body habitus, breast volume, and diaphragm position). A model trained on attenuation-corrected and non-corrected image pairs, with ground-truth pathology from coronary angiography, learns these patterns more reliably than human readers at the lower end of the experience spectrum.

Published data from the REFINE SPECT registry, a multi-center study of 1,638 MPI studies with invasive coronary angiography ground truth, showed that AI-assisted MPI interpretation achieved AUC of 0.83 for obstructive CAD, versus 0.76 for visual reads by nuclear medicine physicians. The improvement was concentrated in the intermediate probability range - exactly the cases where clinical decision-making is most consequential.

Cardiac PET: Higher Accuracy at Higher Cost

Cardiac PET with Rb-82 (rubidium-82) or N-13 ammonia provides higher image quality than SPECT due to better spatial resolution and the possibility of absolute myocardial blood flow (MBF) quantification in mL/min/g. Absolute MBF and myocardial flow reserve (MFR - the ratio of stress to rest MBF) provide information that relative SPECT perfusion images cannot: the ability to detect balanced three-vessel disease through reduced global MFR, and the ability to identify microvascular dysfunction as a distinct entity from obstructive epicardial disease.

MFR below 2.0 is an independent predictor of major adverse cardiac events with hazard ratios of 2.5-5.0 in published cohort studies, independent of coronary anatomy. This predictive power is meaningfully above what SSS-based SPECT reporting provides. The limitation is that cardiac PET requires Rb-82 generator infrastructure or N-13 ammonia cyclotron access, placing it out of reach for most community hospitals.

AI is improving the MFR measurement workflow in two ways: by automating the kinetic modeling step that converts time-activity curve data to absolute flow values, and by segmenting the left ventricular myocardium on PET images with sufficient accuracy to permit regional (per-territory) MFR calculation rather than global MFR only. Regional MFR discriminates single-vessel disease from three-vessel disease - a clinical distinction that global MFR sometimes obscures.

Sarcoidosis and Amyloidosis: Beyond CAD Detection

Nuclear cardiology's role extends beyond coronary artery disease. Cardiac sarcoidosis - a granulomatous infiltrative disease with potentially lethal ventricular arrhythmias - is diagnosed by FDG PET/CT showing patchy cardiac uptake in a metabolic pattern distinct from normal myocardium and from hibernating myocardium in ischemic disease. The interpretive challenge is high background cardiac FDG uptake in patients who are not fully fasted or who have insulin resistance, which can obscure pathological uptake patterns.

Cardiac transthyretin amyloidosis (ATTR) is diagnosed by Tc-99m pyrophosphate or Tc-99m DPD SPECT, which shows focal cardiac uptake in proportion to amyloid deposition. The diagnosis has become clinically urgent because tafamidis, an effective disease-modifying therapy, requires confirmation of ATTR amyloid type before initiation. AI-assisted SPECT quantification for pyrophosphate studies improves the consistency of the heart-to-contralateral ratio measurement that distinguishes positive from negative studies - a measurement that is surprisingly variable when performed manually at different institutions.

The Case for Expanding Nuclear Cardiology Access

A broader access argument underlies the nuclear cardiology AI discussion. MPI is the most common nuclear medicine procedure and one of the highest-value noninvasive cardiac tests, but read quality is uneven across the community hospital and outpatient imaging center settings where most studies are performed. Academic centers with dedicated nuclear cardiologists read MPI at a fundamentally different accuracy level than general radiologists or cardiologists reading nuclear medicine without subspecialty training.

AI tools that provide automated perfusion quantification, standardized attenuation artifact assessment, and structured reporting templates move community practice performance closer to the academic standard. This is not a replacement for subspecialty expertise - the complex cases still require that level of interpretation. But the straightforward normal or clearly abnormal studies that constitute 70% of MPI volume can be read with AI assistance at quality levels that community practice currently does not consistently achieve without it.