Imaging-Based Artificial Intelligence in Vascular and Interventional Radiology: A Narrative Review

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• AI excels in diagnostic radiology but has limited VIR application due to data variability.
• Vascular interventional radiology (VIR) procedures often involve complex imaging analysis.

Purpose of the Study:

• To review AI applications across preprocedural, intraprocedural, and post-procedural VIR stages.
• To assess AI's impact on patient care using various imaging modalities.

Main Methods:

• Comprehensive literature search of PubMed, Embase, and Web of Science.
• Categorization of AI studies by imaging modality (CT, MRI, fluoroscopy/DSA, ultrasound, X-ray, multimodal) and task type (segmentation, detection, prediction).

Main Results:

• Deep learning achieved high accuracy (0.82-0.962) in anatomical segmentation (e.g., aneurysms).
• AI detection tasks showed high performance (e.g., 95% accuracy for endoleak detection, 0.97 AUC for stenosis).
• Prediction models (AUC > 0.90) outperformed clinical assessments for outcomes after EVAR, TEVAR, TACE, TARE.

Conclusions:

• AI demonstrates transformative potential in VIR, particularly with radiomics and machine learning for treatment response.
• Challenges include data limitations, bias, and interpretability.
• Clinician-AI expert collaboration is vital for safe and effective deployment in interventional radiology.