Incorporating AI-Driven Vision Systems to Quantify Learning Curve in EVD Placement

Abstract There is an application for artificial intelligence (AI) to augment medical education. The aim of this study was to incorporate AI-powered cameras to quantify the learning curve and performance metrics associated with external ventricular drain (EVD) placement. Fourteen participants, comprising medical students and neurosurgical residents, were recorded performing an EVD on a trainer head. Five panoramic cameras were installed within the simulation suite. The model employed convolutional neural networks to track anatomical landmarks and assess task completion. Quantification of the learning curve was achieved by aggregating scores across three phases: preparation, insertion, and closing. Additional metrics included fluidity, a proxy for surgical finesse. The model successfully itemized parameters that characterize EVD placement. The study demonstrated a clear learning curve in EVD placement. The overall scores were 64.4/126 (51.1%), 99.6/126 (79%), and 113/126 (89.7%) for the students, junior residents, and senior residents (p < 0.0001). Significant improvements were observed in the preparation, insertion, and closing phases. The mean scores for preparation were 16.4/37 (44.3%), 25.6/37 (69.2), and 30.5/37 (82.4) for the students, junior residents, and senior residents (p < 0.0001). The mean scores for insertion were 26.2/44 (59.5%), 37.8/44 (85.9%), and 38.5/44 (87.5%) for the students, junior residents, and senior residents (p = 0.026). The mean scores during closing were 13/25 (52%), 22.2/25 (88.8%), and 25/25 (100%) for the students, junior residents, and senior residents (p = 0.0034). Fluidity improved significantly with training level (p = 0.0006). Our platform effectively quantified the learning curve associated with EVD placement, underscoring the importance of objective feedback and AI's potential to facilitate skill acquisition.