Answers are grounded in this paper and cite their sources.
Objective: This study aims to develop an artificial intelligence (AI) system based on the U-Net deep learning architecture to automatically identify and stage lymphatic vessels in indocyanine green (ICG) lymphography images, thereby enhancing recognition efficiency and objectivity and assisting in the planning of lymphaticovenular anastomosis (LVA) surgery. Methods: A total of 560 ICG lymphography images from 140 patients with lower-limb lymphedema were collected. After standard preprocessing, the images were randomly split into a training set and a test set in a 7:3 ratio. The U-Net model was used for training, with hyperparameters optimized using a five-fold cross-validation approach. The Adam optimizer was employed for model training. Evaluation metrics included precision, recall, F1 score, and inference time. Results: The model demonstrated excellent performance in identifying lymphatic vessels across lymphedema stages I–IV. The precision scores for each stage were as follows: stage I: 95.29%, stage II: 92.58%, stage III: 93.17%, and stage IV: 90.08%. Recall rates for all stages were above 89%, and all F1 scores exceeded 90%, indicating strong overall accuracy and balance between precision and recall. The average inference time per image was just 56 milliseconds, highlighting the model’s suitability for near-real-time applications. Additionally, the confusion matrix revealed a high discriminative capability across the different lymphedema stages. Conclusion: The U-Net-based AI model demonstrated high accuracy and efficiency in the automatic identification of lymphatic vessels in ICG images, highlighting its strong potential for clinical applications. This system may significantly support LVA surgical planning, enable real-time intraoperative navigation, and serve as a valuable tool for physician training.
No takes yet. Share an insight, caveat, or question.
Aoshang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: