Automatic foot ulcer segmentation using conditional generative adversarial network (AFSegGAN): A wound management...

Area of Science:

• Medical Technology
• Artificial Intelligence
• Image Analysis

Background:

• Chronic wounds pose significant health and economic burdens, especially in aging and diabetic populations.
• Accurate quantitative wound assessment is crucial for effective clinical management, but traditional methods are subjective and error-prone.
• Digitalization and deep learning offer promising alternatives for objective wound analysis.

Purpose of the Study:

• To develop an automated wound management system for accurate wound segmentation and morphological parameter estimation.
• To leverage deep learning, specifically conditional generative adversarial networks (cGANs), for improved wound image analysis.
• To enhance the precision and efficiency of chronic wound care through advanced technology.

Main Methods:

• Development of AFSegGAN, a novel conditional generative adversarial network (cGAN) model.
• Validation of the model on the MICCAI 2021-foot ulcer segmentation dataset.
• Implementation of adversarial loss and patch-level comparison to optimize GAN training and segmentation performance.

Main Results:

• AFSegGAN achieved superior performance compared to state-of-the-art methods.
• The model demonstrated a high Dice score of 93.11% and an IoU of 99.07% in wound segmentation.
• The system effectively estimates wound morphological parameters, indicating high accuracy.

Conclusions:

• The proposed wound management system automates critical aspects of wound assessment, reducing the burden on healthcare workers.
• AFSegGAN offers a reliable tool for objective wound segmentation and parameter estimation, facilitating better clinical decision-making.
• This technology supports remote healthcare initiatives and improves the overall standard of chronic wound care.