What is the clinical evidence from this study?

Study design: Observational. Population: Coronary artery calcium (CAC) (n=527). Intervention: Artificial intelligence algorithm (NANOX AI) vs. Cardiologist interpretation. Primary outcome: Agreement between AI classification and cardiologist interpretation for CAC severity.

December 5, 2023Open Access

Using Artificial Intelligence to Semi-Quantitate Coronary Calcium as an ‘Incidentaloma’ on Non-Gated, Non-Contrast CT Scans, A Single-Center Descriptive Study in West Michigan.

CKConnor C. KerndtGrand Rapids Community College RCRajus ChopraMichigan State University PWPaul M. WeberProvidence Hospital

Key Result

An artificial intelligence algorithm demonstrated 88.76% agreement with cardiologist interpretation in detecting and semi-quantitating coronary artery calcium on non-gated, non-contrast chest CT scans.

Study Design

Type

Observational (n=527)

Blinding

None

Multicenter

Structured PICO

Does an AI algorithm accurately detect and semi-quantitate coronary artery calcium on non-gated, non-contrast CT scans obtained for non-cardiac indications?

Population

527 patients aged 30 years and older undergoing non-gated, non-contrast chest CT scans for non-cardiac indications, evaluated retrospectively for incidental coronary artery calcium using an AI algorithm.

Intervention

FDA-approved AI algorithm (NANOX AI) to measure and semi-quantitate coronary artery calcium (CAC)

Comparator

Cardiologist overread (visual estimation/semi-quantitation)

Outcome

Agreement between AI classification and cardiologist interpretation of CAC severity (normal-mild, moderate, severe)surrogate

AI can accurately detect and semi-quantitate coronary artery calcium on routine non-gated, non-contrast chest CTs, identifying a significant proportion of patients with previously undiagnosed ASCVD who may benefit from preventive therapies.

Limitations

Single-center study on a primarily Caucasian population, limiting generalizability
Non-ECG gated studies have a higher likelihood of missing calcium (false negatives) and cardiac motion artifact
Validating cardiologist was not blinded to the AI CAC results, introducing potential confirmation bias
Only one cardiologist was involved in the independent analysis
Lack of a randomized control population of healthy patients

Abstract

INTRODUCTION: Non-gated, non-contrast computed tomography (CT) scans are commonly ordered for a variety of non-cardiac indications, but do not routinely comment on the presence of coronary artery calcium (CAC)/atherosclerotic cardiovascular disease (ASCVD) which is known to correlate with increased cardiovascular risk. Artificial intelligence (AI) algorithms can help detect and quantify CAC/ASCVD which can lead to early treatment and improved outcomes. METHODS: Using an FDA-approved algorithm (NANOX AI) to measure coronary artery calcium (CAC) on non-gated, non-contrast CT chest, 536 serial scans were evaluated in this single-center retrospective study. Scans were categorized by Agatston scores as normal-mild (<100), moderate (100-399), or severe (≥400). AI results were validated by cardiologist's overread. Patient charts were retrospectively analyzed for clinical characteristics. RESULTS: Of the 527 patients included in this analysis, a total of 258 (48.96%) had moderate-severe disease; of these, 164 patients (63.57%, p< 0.001) had no previous diagnosis of CAD. Of those with moderate-severe disease 135 of 258 (52.33% p=0.006) were not on aspirin and 96 (37.21% p=0.093) were not on statin therapy. Cardiologist interpretation demonstrated 88.76% agreement with AI classification. DISCUSSION/CONCLUSION: Machine learning utilized in CT scans obtained for non-cardiac indications can detect and semi-quantitate CAC accurately. Artificial intelligence algorithms can accurately be applied to non-gated, non-contrast CT scans to identify CAC/ASCVD allowing for early medical intervention and improved clinical outcomes.

Perguntar à IA

Bookmark

View Full Paper