Medical Technology · global
Estimating Ischemia From Coronary CT: Elucid Validation Shows 92.9% Accuracy, Falling to 84.1% Near the Threshold
FFR-IQ translates vessel morphology and plaque characteristics into an assessment of coronary artery function. Validation data from 347 patients support this use, but cases near the interpretation threshold and differences between regional datasets also define the limits of these figures.
Coronary imaging can reveal narrowing, but another question remains: does that narrowing restrict blood supply to the heart muscle? On November 3, Elucid announced results from the ELUCIDATE study. Its artificial intelligence software, FFR-IQ, estimates vessel function from coronary computed tomography angiography (CCTA), achieving vessel-level classification accuracy of 92.9% against invasive testing. However, interpretation remains more difficult closer to the clinical decision threshold.
The tool estimates fractional flow reserve (FFR), which is used to assess whether coronary artery lesions affect blood supply. According to the company, the model integrates the vessel lumen, vessel geometry, and features of calcified, noncalcified, and lipid-rich plaque to generate FFR-CT values along vessels from existing CT images. It provides an estimate of physiological function rather than a direct measurement of blood flow.
The FDA's publicly available submission summary states that this retrospective validation included 347 patients and 491 coronary arteries from nine sites in the United States and two outside the United States. CCTA and invasive FFR testing were performed no more than 60 days apart, and those processing the images were blinded to the invasive test results. Overall sensitivity was 92.2% and specificity was 93.1%, reflecting the ability to identify vessels with and without functional ischemia, respectively. The 92.9% accuracy figure describes vessel classification results and cannot be interpreted as the same degree of diagnostic certainty for every patient.
Performance near the threshold warrants the closest examination. The company reported accuracy of 84.1% in 145 vessels with invasive FFR values between 0.75 and 0.85, below the overall result. This range spans the clinical decision threshold of 0.80, where small differences in estimates can change the classification. It also shows how focusing solely on overall accuracy can obscure the most difficult cases to assess.
Data sources also affect how the figures should be interpreted. The FDA document lists sensitivity and specificity in the US group as 90.9% and 89.2%, respectively, both below the overall results. The document explains that the retrospective US data were more concentrated among patients whose condition was difficult to determine and who therefore underwent invasive testing. Some data from outside the United States were originally collected prospectively and included more vessels clearly without ischemia. The pooled results therefore cannot directly represent performance among patients presenting to each individual hospital.
The study also excluded patients with prior coronary revascularization or myocardial infarction, as well as cases with insufficient image quality, among other exclusions. Evidence for extending the results to these populations is limited. This validation primarily addresses how closely the software's assessments agree with invasive FFR. It has not yet demonstrated that introducing the software can reduce unnecessary invasive examinations or improve patients' long-term health outcomes.
FFR-IQ received 510(k) clearance from the US FDA on September 29; the results announced this time are the validation findings supporting that clearance. Its cleared use is to assist in assessing vessel function in clinically stable patients with suspected coronary artery disease, with results interpreted alongside medical history, symptoms, and other examinations. Validation supports extracting an additional layer of physiological information from CT images. How to translate that information into better diagnostic and treatment decisions remains a question for future clinical research.