Abstract To evaluate a new deep learning (DL) motion correction (MC) software based on partial angle reconstruction (PAR) to reduce motion artifacts in patients with increased heart rate (HR) in coronary CT angiography (CCTA). This retrospective single-center study included consecutive patients with HR > 70 bpm who underwent single-beat wide-area-detector CCTA over a 6-month period. A DL PAR-based MC software was applied to each image, and corrected and uncorrected reconstructions were scored by two blinded independent cardiothoracic radiologists for coronary motion artifact severity. Scores were obtained on a per-vessel and on a per-patient level using a 5-point Likert scale (1 = non-interpretable, 2 = severe, 3 = moderate, 4 = mild, 5 = no artifacts). Scoring differences were analyzed with Chi-Squared test and interrater agreement with Gwet agreement coefficients. 62 patients (35 female) with (mean ± std.dev.) BMI 29.4 ± 6.8 kg/m2 and HR 81.9 ± 13.1 bpm were included. Without MC, the number of cases scored 3 or higher on a per-patient level were 40/62 (64.5%) and 43/62 (69.4%), respectively for reader 1 and 2. With MC, they improved to 50/62 (80.6%) and 55/62 (88.7%), respectively for reader 1 and 2. Improvements in scoring were significant for both readers ( p < 0.02). Per-vessel scores followed a similar trend, but showed significance for both readers only for the right coronary artery ( p < 0.001). The fraction of diagnostically-interpretable cases (score ≥ 2) were 91.9% (uncorrected) and 98.4% (motion-corrected) (reader 1), and 93.5% (uncorrected) and 96.8% (motion-corrected) (reader 2). Interrater agreement was between (0.67–0.78). The MC software significantly improved image quality by reducing coronary motion artifacts in CCTA patients with increased HR.
Caballo et al. (Mon,) studied this question.