287 - AI-Based Cone-Beam CT Reconstruction and Embolization Planning for Prostatic and Genicular Artery Embolizations

Purpose: To assess the clinical impact of artificial intelligence (AI) based cone-beam CT (CBCT) reconstruction and embolization planning software during prostatic artery embolization (PAE) and genicular artery embolization (GAE).

Materials and Methods: This retrospective single-center study included two analyses.

First, image quality assessment: 134 CBCT acquisitions (PAE n=85, GAE n=49) from 97 procedures were selected among 100 consecutive cases (July 2024 - February 2025), excluding three non-conforming acquisitions. Each CBCT was reconstructed using both conventional and AI-based algorithms (CleaRecon DL, GE HealthCare) and reviewed side-by-side by two independent reviewers, blinded to reconstruction type. Evaluation included streak artifacts, vessel visualization (5-point Likert scale), and overall reconstruction preference.

Secondly, procedural planning evaluation: A randomized subset of 90 CBCT datasets (PAE n=50, GAE n=40) was analyzed to assess the added value of AI integration in procedural planning. Each dataset was processed twice using Embo ASSIST from the conventional reconstruction and using Embo ASSIST AI (GE HealthCare) from the AI-based reconstruction. The same two reviewers sequentially evaluated vessel segmentation accuracy, and the extend of target artery depiction . Segmentation performance was scored on a 4-point scale (0=None, 1=Proximal, 2=Distal, 3=Very Distal).

Results: In the image quality analysis, AI-based reconstruction was preferred over conventional in 100% of cases. It reduced streak artifacts in all cases and improved vessel visualization in 97% of cases (Significantly Better 59.3%, Better 37.7%, Equivalent 2.6% and Worse 0.4%).

In the procedural analysis, segmentation scores improved with AI-based planning. For the PAE subgroup, the average segmentation score increased from 1.9 ±1.0 to 2.7 ±0.5 (p < 0.001). For the GAE subgroup, the average segmentation score increased from 2.1 ± 1.0 to 2.4±0.7 (p < 0.001), as determined by the Wilcoxon signed-rank test.

Conclusion: AI-based CBCT reconstruction and embolization planning significantly improves image quality, vessel visualization, and segmentation performance during prostatic and genicular artery embolization. These enhancements lead to better visualization of vascular anatomy and more accurate identification of target arteries, which may increase procedural efficiency and operator confidence.