Real-Time Artificial Intelligence-Based Detection of Flow-Directed Microcatheter Tip Markers

Flow-directed microcatheters (FDMs) offer high flexibility and trackability for accessing distal arterial feeders. However, the radiopaque tip marker is small, making it difficult to identify on fluoroscopy. This study evaluated an intraoperative artificial intelligence (AI)-based system for real-time detection of FDM tip markers.

We retrospectively analyzed 10 consecutive cases of middle meningeal artery embolization for chronic subdural hematoma using the AI-based system. The detection rate was evaluated on a frame-by-frame basis during microcatheter placement. Exploratory subgroup analyses were performed based on the catheter diameter (1.5 Fr vs. 1.3 Fr) and whether the microcatheter tip or guidewire tip advanced ahead during navigation.

Twenty-five FDM placement scenes were analyzed. Mean microcatheter navigation time was 2.1 min. The precision, recall, and detection rate of the system were 95%, 51%, and 50%, respectively; the detection rate was calculated as the proportion of true-positive frames among all analyzed frames. Although this was an exploratory subgroup analysis, the detection rate appeared higher for the 1.5-Fr microcatheters than that for the 1.3-Fr microcatheters (78% vs. 46%; p = 0.006). Furthermore, the detection rate was significantly higher when the microcatheter tip advanced ahead of the guidewire tip than vice versa (65% vs. 43%; p = 0.002).

Although the detection rate was 50%, the AI system may help operators recognize the tip position when it becomes difficult to identify on fluoroscopy. Larger multicenter studies are required to validate these findings.