Fully Automatic Diabetic Wound Segmentation Using Lightweight Deep Convolutional Neural Networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Diabetic foot ulcers (DFUs) are a significant global health issue, often leading to severe complications like amputation.
• Early detection and monitoring of DFUs are critical for effective treatment and improved patient outcomes.
• Digital imaging and mobile devices are increasingly used for remote DFU assessment, necessitating automated wound analysis.

Purpose of the Study:

• To develop a computationally efficient deep learning model for automated diabetic foot wound segmentation.
• To improve the accuracy and objectivity of wound area measurement for tracking healing progression.
• To create a model suitable for deployment on resource-constrained devices in clinical settings.

Main Methods:

• Proposed a lightweight convolutional neural network (CNN) augmenting U-Net with ghost features and Convolutional Block Attention Modules (CBAM).
• Evaluated the model on 3450 annotated diabetic foot wound images, comparing it with state-of-the-art architectures.
• Implemented a two-step pipeline including prior foot segmentation for region of interest (ROI) detection.

Main Results:

• The proposed CNN achieved high performance metrics: 85.13% precision, 91.84% recall, 86.95% Dice coefficient, and 77.23% IoU with ROI detection.
• Demonstrated competitive segmentation accuracy compared to high-capacity models.
• Showcased significantly reduced computational complexity, suitable for real-time deployment.

Conclusions:

• The lightweight CNN offers an efficient and accurate solution for diabetic foot wound segmentation.
• The model's performance and low computational requirements make it ideal for remote, low-resource clinical applications.
• This approach supports objective wound monitoring, potentially reducing amputation rates and improving patient management.