Application of multiple deep learning models for automatic burn wound assessment.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Computational Biology

Background:

• Accurate estimation of burn percentage total body surface area (%TBSA) is critical for effective burn injury management.
• Manual estimation of %TBSA is subjective and prone to significant discrepancies among clinicians.
• Identifying deep burn regions is essential for appropriate treatment planning.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) system for accurate %TBSA estimation in burn injuries.
• To enable the segmentation of deep burn regions from the entire wound area.
• To improve the consistency and accuracy of burn assessment compared to traditional methods.

Main Methods:

• Utilized deep learning models including U-Net, PSPNet, DeeplabV3+, and Mask R-CNN with ResNet101 encoder.
• Implemented boundary-based and region-based labeling strategies for training datasets.
• Calculated %TBSA burned by segmenting total burn wound area relative to palm size and identified deep burn areas.

Main Results:

• DeeplabV3+ demonstrated superior performance across three segmentation tasks: total burn wound, palm, and deep burn.
• Achieved high precision and recall values for all segmentation tasks, indicating robust model performance.
• Successfully trained and tested DL models on a substantial dataset of burn wound and palm images.

Conclusions:

• Deep learning models can automatically diagnose %TBSA burned, fluid resuscitation volume, and deep burn area percentage with pixel-level accuracy.
• The developed AI system offers consistent, accurate, and rapid assessments of burn wounds.
• This technology has the potential to significantly enhance clinical decision-making in burn management.