Diagnostic Improvement in Endoleak Detection: The Role of Low-energy Virtual Monochromatic CT and Deep Learning Reconstruction.

To assess the diagnostic performance of low-energy virtual monochromatic CT imaging (VMI) combined with deep learning image reconstruction (DLIR) for detecting endoleaks. Seventy-one patients who underwent contrast-enhanced CT after endovascular aortic repair (EVAR) were retrospectively studied, with endoleaks being identified in 41 (58%) of them. CT raw data were reconstructed using three techniques: 70-keV VMI with conventional hybrid iterative reconstruction (HIR [ASiR-V50%]), and 40- and 70-keV VMI with DLIR (TrueFidelity-H). Three observers evaluated the presence or absence of endoleaks on a 5- point scale, taking into account the confidence level: score 1, definitely absent; score 2, probably absent; score 3, possibly present; score 4, probably present; score 5, definitely present. Scores of ≥ 3 were considered positive for endoleaks. Receiver-operating characteristic (ROC) analyses were performed to compare the area under the curve (AUC) values. ROC analyses demonstrated that the 40-keV VMI with DLIR achieved the highest AUC across observers (0.92-0.99), outperforming the 70-keV DLIR (0.91-0.97) and 70-keV HIR (0.88-0.96). The proportion of correctly identified endoleaks with high confidence (score ≥ 4) was significantly greater, or at least comparable, for 40-keV VMI with DLIR versus 70-keV VMI with HIR (Observer 1: 85% [35/41] vs. 73% [30/41], p = 0.02; Observer 2: 85% [35/41] vs. 78% [32/41], p = 0.20; Observer 3: 98% [40/41] vs. 90% [37/41], p = 0.10). The integration of low-energy VMI with DLIR significantly enhances the diagnostic confidence and accuracy in endoleak detection compared with conventional reconstruction techniques. These findings underscore the clinical utility of advanced reconstruction algorithms for optimizing post-EVAR surveillance.