SO24. Automated Vessel Edge Detection in Microsurgery Using an AI-Driven Convolutional Neural Network

PURPOSE: Precise visualization of vessel edges is crucial in microsurgical procedures like lymphaticovenous anastomosis (LVA), directly affecting surgical precision and patient outcomes. Integrating artificial intelligence (AI) into microsurgery offers significant potential for enhancing these outcomes and serves as a stepping stone toward AI-assisted and robotic microsurgery. This study aims to develop and evaluate an AI-driven convolutional neural network (CNN) model for automated vessel edge detection in microsurgical procedures. METHODS: We developed a custom CNN model using TensorFlow and Keras for vessel edge detection in microsurgery. A dataset of 326 high-resolution images from LVA procedure videos was divided into training, validation, and testing sets. Data augmentation techniques were applied to increase diversity. The model was optimized using the Adam algorithm with a custom loss function. RESULTS: The CNN model achieved 99% accuracy on training and test sets with low loss metrics. High Dice coefficient and Jaccard index values confirmed accurate edge detection. The model consistently identified vessel edges under LVA-specific conditions and demonstrated robustness to varying vessel sizes and tissue characteristics. CONCLUSION: We present the first AI-driven CNN model for automated vessel edge detection in microsurgical procedures. This model enhances visualization and may facilitate future AI integration into surgical microscopes and robotic systems, advancing the field of AI-assisted and robotic microsurgery and potentially improving surgical precision and patient outcomes.