Artificial Intelligence CT Texture Radiomics for Outcome Prediction After EVAR: A Narrative Review.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Endovascular aneurysm repair (EVAR) necessitates lifelong surveillance due to potential complications like endoleaks and sac expansion.
• Conventional follow-up methods may miss early microstructural changes, delaying detection of clinical deterioration.

Purpose of the Study:

• To review current evidence on texture-based radiomics and AI for predicting outcomes after EVAR.
• To assess the utility of these techniques in post-EVAR surveillance and complication detection.

Main Methods:

• Narrative review of original studies using radiomics and AI on CT/CTA for EVAR outcome prediction.
• Qualitative synthesis of studies evaluating radiomic features and AI models for endoleak, sac behavior, and adverse event prediction.

Main Results:

• Radiomic features capturing texture heterogeneity (e.g., entropy, spatial complexity) were extracted from various anatomical regions.
• AI models (machine and deep learning) demonstrated good to excellent performance (AUC 0.78-0.95) in predicting EVAR complications.
• Texture-based radiomics outperformed morphology-only assessments and complemented deep learning, even on non-contrast CT.

Conclusions:

• CT texture radiomics combined with AI is a promising approach for post-EVAR surveillance.
• This approach may enable earlier detection of complications by analyzing tissue heterogeneity beyond conventional morphology.
• Limited evidence, methodological heterogeneity, and reproducibility issues currently hinder widespread clinical translation.