Clinical evaluation of a motion correction software based on partial angle reconstruction in coronary CT angiography

Comprehensive Summary

Caballo et al. conducted a retrospective single-center study evaluating a deep learning (DL) motion correction algorithm based on partial angle reconstruction (PAR) in 62 consecutive patients with heart rates >70 bpm undergoing single-beat coronary CT angiography (CCTA). Motion-corrected and uncorrected images were assessed by two blinded cardiothoracic radiologists using a 5-point Likert scale at per-patient and per-vessel levels. Motion correction significantly improved image quality, increasing the proportion of patients with scores ≥3 from 64.5% to 80.6% (reader 1) and 69.4% to 88.7% (reader 2) (p < 0.02), while diagnostic interpretability (score ≥2) improved from 91.9% to 98.4% and 93.5% to 96.8%, respectively. Improvements were most pronounced in the right coronary artery (p < 0.001), with limited or non-significant effects in the left coronary system. Objective metrics showed no differences in noise, SNR, or CNR, but demonstrated reduced motion artifacts. Interobserver agreement was moderate-good (Gwet 0.67-0.78), and no consistent predictors of benefit were identified. The authors concluded that DL PAR-based motion correction significantly improves image quality and diagnostic interpretability in high-heart rate CCTA.

Outcomes and Implications

DL-based PAR motion correction improves image quality and diagnostic interpretability in high-heart rate CCTA without altering noise or contrast characteristics, with the greatest benefit in motion-prone coronary segments. However, the small, single-center cohort and lack of diagnostic accuracy or outcome data limit generalizability. External validation and studies linking these improvements to clinical decision-making are needed before routine adoption.