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

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Dermatology and wound care

Background:

• Accurate burn severity assessment is critical for timely clinical decisions.
• Current visual classification of burn degrees (second- and third-degree) is subjective and inconsistent.
• Observer variability in burn assessment can impact treatment outcomes.

Purpose of the Study:

• To develop and validate a deep learning model for automated burn classification.
• To investigate the efficacy of different color channels for burn image analysis.
• To compare the performance of the developed model against established transfer learning techniques.

Main Methods:

• A convolutional neural network was trained on a dataset of clinical burn images.
• Hyperparameter optimization and color channel sensitivity analysis were conducted.
• Model performance was assessed using accuracy, precision, recall, and F1-score on independent test sets.

Main Results:

• A compact deep learning model utilizing only the green color channel achieved high performance (accuracy: 0.94, F1-score: 0.94).
• The green channel model surpassed the performance of more complex transfer learning models.
• This approach demonstrated reduced computational complexity compared to other methods.

Conclusions:

• The green color channel is sufficient for efficient and accurate burn classification.
• The developed deep learning model shows potential for clinical and educational applications.
• Integration into a graphical user interface facilitates practical implementation in healthcare settings.