Deep learning based prediction of burn wound healing after plastic surgery(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2026年第2期
- Page:
- 255-260
- Research Field:
- 医学人工智能
- Publishing date:
Info
- Title:
- Deep learning based prediction of burn wound healing after plastic surgery
- Author(s):
- deep learning attention mechanism burn plastic surgery healing status
- Department of Burn and Plastic Surgery, General Hospital of Eastern Theater Command, Nanjing 210016, China
- Keywords:
- Keywords: deep learning attention mechanism burn plastic surgery healing status
- PACS:
- R318；R619.5
- DOI:
- DOI:10.3969/j.issn.1005-202X.2026.02.016
- Abstract:
- Objective To predict the healing status in burn patients after plastic surgery using deep learning models, focusing on scar formation, infection occurrence, and classifications of complete and partial healing, and assist clinicians in more accurately assessing postoperative recovery by extracting and analyzing features from burn image data. Methods The pre-trained visual geometry group (VGG) model was used to extract features from burn wound images, and an attention mechanism was introduced to enhance the models ability to focus on key regions. The extracted features were further input into various traditional machine learning classifiers for model training and testing. Results The classification model combining the VGG model with an attention mechanism achieved favorable performance in predicting burn wound healing status. Among all the models, the VGG model incorporating the attention mechanism and XGBoost classifier exhibited the optimal performance, with an accuracy of 0.853, an average macro F1 score of 0.844, and an average macro AUC value of 0.921, verifying its superior performance in distinguishing different healing statuses. Conclusion The model employing VGG feature extraction and incorporating an attention mechanism demonstrates significant clinical potential in predicting burn wound healing. The application of the attention mechanism enhances the models focus on key image features, which further improves classification accuracy and result consistency.
Last Update: 2026-01-27