Integrated image and location analysis for wound classification: a deep learning approach.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Wound care diagnostics

Background:

• The global prevalence of acute and chronic wounds necessitates improved diagnostic and treatment strategies.
• Current wound classification methods often lack precision, impacting patient care.
• Accurate wound categorization is crucial for effective clinical decision-making.

Purpose of the Study:

• To develop and evaluate an innovative multi-modal deep convolutional neural network for classifying diabetic, pressure, surgical, and venous ulcers.
• To enhance wound image classification by integrating wound images with corresponding body location data.
• To improve upon traditional wound classification techniques through a novel architecture and body map system.

Main Methods:

• Development of a multi-modal network integrating VGG16, ResNet152, and EfficientNet models.
• Incorporation of spatial and channel-wise Squeeze-and-Excitation modules, Axial Attention, and an Adaptive Gated Multi-Layer Perceptron.
• Utilizing a body map system for accurate wound location tagging alongside image data for training and evaluation on two distinct datasets.

Main Results:

• The proposed multi-modal network achieved high classification accuracies: 74.79-100% (ROI without location), 73.98-100% (ROI with location), and 78.10-100% (whole image).
• Performance significantly surpassed traditional wound image classification methods.
• Demonstrated superior accuracy compared to previously reported metrics in the literature.

Conclusions:

• The developed multi-modal network shows significant potential as an effective decision-support tool for wound image classification.
• The integration of wound images and location data enhances classification precision.
• This approach offers a promising advancement for clinical wound assessment and management.