AI Tool Shows Consistent FAI Measurement Across CT Scanners

ARTIFICIAL intelligence (AI) software demonstrated good agreement with conventional workstation measurements of the pericoronary fat attenuation index (FAI), a non-invasive imaging biomarker of coronary inflammation, across five different computed tomography (CT) scanners.

The findings suggest that automated FAI measurement could offer a consistent approach to assessing coronary inflammation across different imaging platforms.

Evaluating AI-Based FAI Measurement

The retrospective study assessed coronary CT angiography data from 1,125 patients, including 584 males (51.9%), with a mean age of 61.5 years. The dataset covered 3,375 coronary arteries examined using five CT scanners from different manufacturers, models, and generations.

Researchers compared FAI measurements obtained using AI software with those calculated using conventional workstation software. They assessed the agreement between the two methods, their diagnostic performance, and whether differences between scanners affected measurement consistency.

Agreement Across Measurement Methods

At the patient level, AI-based FAI measurements showed a strong correlation with workstation measurements. Agreement analysis found a mean difference of 1.60 Hounsfield units (HU), with 94.84% of measurements falling within the established limits of agreement.

Using an FAI threshold of −70.1 HU, the AI method achieved an accuracy of 84.98% and an area under the curve of 0.851. These results indicated comparable diagnostic performance to conventional workstation-based measurement.

Researchers also assessed consistency across the five scanners. Correlation values ranged from 0.829 to 0.896, while only 3.70–6.06% of measurements were outliers beyond the limits of agreement.

Potential for Clinical Application

The findings support the feasibility of using AI to automate FAI measurement across different CT scanners. Researchers highlighted its potential to facilitate more efficient and standardised quantification of coronary inflammation, particularly in settings where imaging data are acquired using different equipment.

Researchers concluded that AI-based FAI measurement could provide a practical alternative to traditional workstation analysis and support cardiovascular risk assessment. However, they noted that multicentre, prospective studies and further evaluation across different software platforms and repeated measurements could strengthen the evidence for wider clinical application.

 Reference:

Qiu L et al. Artificial intelligence standardizes pericoronary fat attenuation index measurement: external multi-vendor CT validation. Br J Radiol. 2026; 99(1187): 2415–26.