Tissue classification and segmentation of pressure injuries using convolutional neural networks

Area of Science:

• Biomedical image analysis
• Medical diagnostics
• Computational pathology

Background:

• Pressure injuries are localized skin damages requiring frequent diagnosis and treatment.
• Accurate segmentation and tissue identification are crucial for effective pressure injury management.
• Existing methods may lack the precision needed for complex wound structures.

Purpose of the Study:

• To present a novel approach for automatic tissue classification in pressure injuries.
• To develop a reliable system for segmenting and identifying granulation, slough, and necrotic tissues.
• To improve treatment outcomes through enhanced diagnostic accuracy.

Main Methods:

• A Convolutional Neural Network (CNN) was developed for optimized tissue segmentation.
• Preprocessing involved flash light removal and creation of 5x5 sub-images for CNN input.
• The CNN classified sub-images into granulation, slough, or necrotic tissue categories.

Main Results:

• The system achieved an overall average classification accuracy of 92.01%.
• Average total weighted Dice Similarity Coefficient was 91.38%.
• Precision rates were 97.31% for granulation, 96.59% for necrotic, and 77.90% for slough tissue.

Conclusions:

• The developed system demonstrates feasibility in recognizing complex structures within biomedical images.
• This automated approach offers a promising tool for pressure injury assessment.
• The findings support the potential of CNNs in advancing wound care diagnostics.