Automated Classification of Second- and Third-Degree Burn Images Using Convolutional Neural Networks.

backgroundBurn severity assessment is clinically relevant and often requires timely decision-making. Visual classification of second- and third-degree burns remains subjective and prone to observer variability.

methodsThis study aimed to develop and evaluate a deep learning model for the automatic classification of second- and third-degree burn images. A dataset of clinical burn images from a private wound care clinic was used to train a convolutional neural network. Hyperparameter optimization and color channel sensitivity analysis were performed to identify the optimal model configuration. Model performance was evaluated using standard classification metrics on training, validation, and independent test datasets, and results were compared with transfer learning approaches.

resultsThe best performance was achieved using a compact model trained exclusively with the green color channel, obtaining an accuracy of 0.94, precision of 0.96, recall of 0.92, and an F1-score of 0.94 on the independent test set. This model outperformed more complex transfer learning approaches while reducing computational complexity.

conclusionsThese findings indicate that using only the green color channel enables efficient and accurate burn classification. The proposed model was also integrated into a graphical user interface, supporting its potential application in clinical and educational environments.