Convolution Neural Network with Coordinate Attention for Real-Time Wound Segmentation and Automatic Wound Assessment

Area of Science:

• Medical imaging
• Computer vision
• Artificial intelligence in healthcare

Background:

• Emergency wound assessment is critical but often delayed by limited resources and inaccurate manual methods.
• Manual wound measurement is prone to errors and increases infection risk.
• An automated system is needed for real-time, accurate wound segmentation and size estimation.

Purpose of the Study:

• To develop an Automatic Wound Segmentation Assessment (AWSA) framework.
• To enable real-time wound segmentation and automatic wound region estimation.
• To improve the speed and accuracy of wound assessment in emergency settings.

Main Methods:

• Utilized a short-term dense concatenate classification network (STDC-Net) as the backbone for a balance between accuracy and speed.
• Integrated a coordinated attention mechanism to enhance segmentation performance.
• Developed a functional relationship model for wound area measurement based on image pixels and shooting height.

Main Results:

• The AWSA framework demonstrated superior performance over state-of-the-art methods in segmentation accuracy metrics (mAP, mIoU, recall, dice score) and AUC (approx. 99.5%).
• Achieved high real-time performance with FPS values of 100.08 and 102.11, significantly faster than existing methods.
• Accurately estimated wound area with a low relative error of approximately 3.1%.

Conclusions:

• The developed AWSA framework, using STDC-Net, successfully balances segmentation accuracy and prediction speed.
• The system provides a significant advancement in real-time wound assessment for emergency medical care.
• This automated approach offers a more efficient and precise alternative to manual wound measurement techniques.