Applying Artificial Intelligence to Predict Complications After Endovascular Aneurysm Repair

Area of Science:

• Vascular Surgery
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Post-operative complications following Endovascular Aneurysm Repair (EVAR) can be life-threatening.
• Current patient surveillance methods are challenging, with decreasing patient follow-up rates, especially after the first year.
• Predictive models are needed to identify patients requiring more intensive post-operative surveillance.

Purpose of the Study:

• To develop an artificial intelligence (AI) model to predict individual patient complication probability after EVAR.
• To identify patients who would benefit from more intensive post-operative surveillance.
• To optimize surveillance strategies for EVAR patients.

Main Methods:

• A deep convolutional neural network (VascAI©) was developed using pre-operative CT angiography (CTA) 3D reconstruction images from 273 patients who underwent EVAR.
• The model utilized 3D CT images to predict the risk of post-operative complications.
• Data down-sampling and augmentation techniques were employed to address data imbalance and enhance model performance.

Main Results:

• The AI model achieved 100% sensitivity in identifying patients who developed post-operative complications after EVAR.
• The model correctly identified all patients who experienced complications.
• The model demonstrated a high false positive rate (44%), indicating potential for reducing surveillance frequency in a significant portion of patients.

Conclusions:

• AI models can accurately predict the risk of post-operative complications following EVAR.
• The developed AI model requires only pre-operative AAA CTA images as input, eliminating the need for expert-annotated data.
• This AI tool can assist in identifying high-risk patients for EVAR complications, guiding surveillance compliance.