AI-Based Automated Extraction of Manufacturer-Specific Stent-Graft Codes for EVAR Planning
Abstract
Purpose
The objective of this study was to assess whether an Artificial Intelligence (AI) model can automate the retrieval of Endovascular Aneurysm Repair (EVAR) related device codes used to identify specific stent-grafts from preoperative computed tomography angiography (CTA) imaging studies, with the goal of standardizing device selection across two different manufacturers
Methods
Twenty preoperative CTA scans from patients with infrarenal abdominal aortic aneurysm, were analyzed to asses an AI tool that automatically segmented the aorto-iliac proximal and distal sealing zones, and extracted key anatomical measurements (i.e., vessel diameters and lengths). The AI algorithm used these parameters to evaluate all feasible main body and iliac limb stent-graft component combinations and generate complete identification codes according to the instructions of two manufacturers. AI-generated codes were then compared with selections of two experienced vascular surgeons. Agreement was categorized as Exact Match, Near Match (minor differences due to measurements near dimensional thresholds, e.g. 1 mm variation in diameter or length resulting in adjacent codes), or Mismatch. Fisher's Exact test was used to compare AI-Surgeon code agreement to Inter-Surgeon Code agreement.
Results
High concordance was observed between the AI-generated stent-graft codes and those selected by the surgeon for each manufacturer. For Manufacturer 1, 65% of AI selections were exact matches, and 30% were near matches, while 5% were mismatches. For Manufacturer 2, 83% of AI selections were exact matches, and 15% were near matches, with only 2% being mismatches. No significant differences were found between AI-surgeon and surgeon-surgeon agreement distributions for either manufacturer (p = 0.67 and p = 0.38).
Conclusion
AI-based automated extraction of EVAR stent-graft codes demonstrates expert-level agreement with vascular surgeons across the two manufacturers. The AI tool automates converting patient-specific anatomy into manufacturer-specific device codes, eliminating tedious EVAR planning tasks, while maintaining standardisation and compliance with manufacturer device restrictions.