Comparing Artificial Intelligence Guided Image Assessment to Current Methods of Burn Assessment.

Area of Science:

• Medical imaging
• Artificial intelligence in healthcare
• Burn management

Background:

• Accurate burn depth and size identification is critical for effective treatment.
• Clinical assessment remains the standard but has limited accuracy (67%) for partial-thickness burns.
• Existing aids like laser Doppler imaging (LDI) have limitations.

Purpose of the Study:

• To develop an AI-powered system for predicting burn severity and wound margins.
• To create a triaging tool for thermal injury management using mobile-device-captured images.
• To compare the AI system's performance against clinical assessment and LDI.

Main Methods:

• A convolutional neural network (CNN) based on Modified EfficientNet architecture was developed.
• A novel boundary attention mapping (BAM) algorithm was integrated for burn boundary recognition.
• The CNN-BAM system was validated using 144 patient charts, comparing its output to LDI assessments.

Main Results:

• The CNN achieved 85% accuracy in a 4-level burn severity classification.
• CNN-BAM accurately segmented burn areas with 91.6% accuracy, 78.2% sensitivity, and 93.4% specificity compared to LDI.
• Burn severity predictions from CNN-BAM showed a 66% agreement with LDI-extrapolated healing potential.

Conclusions:

• The CNN-BAM algorithm demonstrates high accuracy in burn depth and area detection, comparable to LDI.
• This AI system offers a more economical and accessible solution for burn assessment when integrated into mobile devices.
• The developed AI system shows significant potential as a triaging tool in thermal injury management.