Journal of vascular and interventional radiology : JVIR · 2026
Automated Detection of Endoleaks in CT Angiography Using a Single Deep Learning Model.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Marcelin Clément, Bedague Théo, Maes Jean David, Maaloum Rim, Minvielle Ludovic, Valenzuela Gladys, Le Bras Yann, Fourage Eva
Аннотация
AIM: Detecting endoleaks on post-EVAR CT scans, especially in the presence of artifacts caused by embolic agents like Onyx, is a challenging but critical task. In this study, we propose a deep learning-based method for endoleak detection using a single object detection model.
MATERIALS AND METHODS: A retrospective dataset comprising 249 CT scans was used, divided into training (129 scans) and test (120 scans) datasets. Ground truth annotations were obtained through 3D manual segmentation, and a preprocessing pipeline was applied. The model was trained on slices extracted from the training dataset and validated using standard metrics, including sensitivity, specificity, and AUC.
RESULTS: The model achieved high performance at both the per-slice and per-series levels. Per-slice detection yielded a ROC AUC of 96%, with a sensitivity of 91%, specificity of 93%, PPV of 55%, and NPV of 99%. At the per-series level, sensitivity and specificity were 93% and 95%, respectively. Notably, our method introduces a per-slice detection metric, which was not reported in previous studies, and uses a single model, making it computationally efficient and easy to train.
CONCLUSION: This study demonstrates the feasibility of using deep learning for endoleak detection, offering a robust and efficient tool for clinical practice. Future work could focus on expanding the dataset and integrating additional predictive capabilities, such as endoleak classification and localization.
MATERIALS AND METHODS: A retrospective dataset comprising 249 CT scans was used, divided into training (129 scans) and test (120 scans) datasets. Ground truth annotations were obtained through 3D manual segmentation, and a preprocessing pipeline was applied. The model was trained on slices extracted from the training dataset and validated using standard metrics, including sensitivity, specificity, and AUC.
RESULTS: The model achieved high performance at both the per-slice and per-series levels. Per-slice detection yielded a ROC AUC of 96%, with a sensitivity of 91%, specificity of 93%, PPV of 55%, and NPV of 99%. At the per-series level, sensitivity and specificity were 93% and 95%, respectively. Notably, our method introduces a per-slice detection metric, which was not reported in previous studies, and uses a single model, making it computationally efficient and easy to train.
CONCLUSION: This study demonstrates the feasibility of using deep learning for endoleak detection, offering a robust and efficient tool for clinical practice. Future work could focus on expanding the dataset and integrating additional predictive capabilities, such as endoleak classification and localization.