AI-based CCTA plaque quantification vs IVUS: INVICTUS Registry abstract

The INVICTUS Registry abstract describes an automated artificial intelligence workflow for coronary CT angiography plaque quantification being assessed against intravascular ultrasound (IVUS), using co-registered coronary segments that include both diseased and non-diseased anatomy.

The comparison is framed around whether CT-derived automated measurements track IVUS-derived assessments of vessel and plaque morphology across the disease spectrum. In that context, the report focuses on quantitative concordance between modalities within matched segments.

According to the abstract, paired coronary CT angiography (CCTA) and IVUS analyses were available for 108 vessels from 85 patients, enabling head-to-head comparisons in co-registered segments. Image analyses were performed by independent core laboratories blinded to the other modality’s findings, with quantification of vessel external elastic membrane (EEM), lumen, and plaque volumes, along with plaque burden metrics, within matched anatomy. Overall, the dataset is presented as a paired, core-lab–read, co-registered segment set intended for modality-to-modality comparison.

Across whole co-registered segments, the abstract reports “whole-segment Pearson correlations” between AI‑QCT and IVUS for EEM volume (r=0.899), lumen volume (r=0.943), and plaque volume (r=0.833). It also reports correlations for length-normalized percent atheroma volume (PAV; r=0.851) and for a calcium index calculated for the whole segment (r=0.960). These values are presented as the primary whole-segment validation statistics in the abstract.

For plaque-component subsegments, the abstract reports correlations for vessel, lumen, and plaque volumes within non-calcified plaque segments (r=0.95, 0.97, and 0.83, respectively) and within low-attenuation plaque segments (r=0.90, 0.86, and 0.86, respectively). For lumen area agreement, it reports that minimum lumen area was smaller by AI-QCT than IVUS on average (mean difference 0.61 ± 1.18 mm2; 95% CI, -0.83 to -0.38), and that lumen area stenosis was similar between methods (mean difference 1.26 ± 24.17; 95% CI, -3.37 to 5.90). These results are described as component-level correlations alongside lumen area and stenosis agreement metrics.

In conclusion, the abstract characterizes the findings as showing high correlations and close agreement between AI-QCT and IVUS across whole segments and in segments containing non-calcified and low-attenuation plaques, and it highlights co-registration, independent core-lab reads, and blinding between modalities. As reported, the abstract provides numerical correlation and agreement summaries within a blinded, co-registered core-lab framework, while leaving key operational specifics unspecified.

Key Takeaways:

• Overall, the abstract reports that AI-QCT measurements showed strong correlation and close agreement with IVUS across whole segments, including non-calcified and low-attenuation plaque segments.
• In non-calcified and low-attenuation plaque subsegments, the abstract reports component-level correlations for vessel, lumen, and plaque volumes, and it reports that minimum lumen area was smaller by AI-QCT on average while lumen area stenosis was similar.
• The comparison is described as based on co-registered segments with independent, blinded core-laboratory analyses, and the abstract does not specify acquisition/quality criteria or operational limitations for the workflow.