Machine learning for endoleak detection after endovascular aortic repair

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Surgery

Background:

• Endovascular aortic repair (EVAR) requires follow-up CT angiography (CTA) for endoleak detection.
• Manual CTA review is time-consuming, prone to errors, and relies heavily on radiologist expertise.

Purpose of the Study:

• To evaluate a deep neural network (DNN) for automated endoleak detection on post-EVAR CTA.
• To assess the performance of a data-efficient training approach for the DNN model.

Main Methods:

• A custom augmentation method, Endoleak Augmentor, was used for robust DNN training.
• The DNN model generated segmentation maps for endoleak identification.
• Model performance was evaluated against a reference subspecialist and blinded general radiologists.

Main Results:

• The DNN achieved high diagnostic performance (95% accuracy, 90% precision, 100% recall, AUC=0.99) compared to the reference subspecialist.
• Performance varied among general radiologists (70-90% accuracy).
• The DNN demonstrated superior performance over general radiologists and comparable performance to subspecialists.

Conclusions:

• A machine learning-based approach offers a reliable and efficient method for endoleak detection in post-EVAR patients.
• This AI tool has the potential to assist radiologists, improving diagnostic accuracy and workflow efficiency.
• The developed model shows promise in augmenting clinical decision-making for EVAR surveillance.