Single Deep Learning Model Detects Endoleaks on CT Angiography with High Accuracy

A deep learning model detected endoleaks on post-EVAR CT with 91% sensitivity, 93% specificity per slice, and AUC 96% in 120 test scans, offering a fast single-model approach.

• Single object detection model processed CT slices with 99% negative predictive value per slice.
• Per-series sensitivity was 93% and specificity 95%, comparable to per-slice metrics.
• Study limited to 249 CT scans from a retrospective cohort; authors recommend dataset expansion for future work.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.