A Framework for Automatic Burn Image Segmentation and Burn Depth Diagnosis Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Burns are common traumatic injuries with significant morbidity and mortality.
• Accurate burn depth and area diagnosis are critical for effective treatment but challenging due to wound complexity.
• Current diagnostic methods for burns lack accuracy and standardization, leading to potential errors.

Purpose of the Study:

• To develop and evaluate a deep learning framework for automated burn area segmentation and burn depth diagnosis.
• To improve the accuracy and standardization of burn wound assessment.
• To reduce human error in burn diagnosis and facilitate timely, appropriate treatment.

Main Methods:

• A comprehensive burn dataset with detailed burn area segmentation and depth labeling was created.
• An end-to-end deep learning framework was designed for burn image analysis.
• The framework was utilized for segmenting burn areas and classifying multiple burn depths.

Main Results:

• The deep learning network achieved an Intersection over Union (IOU) of 0.8467 for burn vs. non-burn area segmentation.
• The framework successfully segmented multiple burn depth areas, with an average IOU of 0.5144.
• The system demonstrated potential for calculating the percentage of total body surface area (%TBSA) affected by burns.

Conclusions:

• Deep learning offers a promising approach to automate and standardize burn diagnosis.
• The developed framework shows efficacy in segmenting burn areas and assessing burn depth.
• This technology can potentially improve patient outcomes by enabling more accurate and rapid burn assessments.