Lower limb and feet wound image dataset.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Existing medical imaging resources lack balanced datasets for lower-limb conditions.
• There's a critical need for diverse datasets encompassing both healthy and pathological foot images.

Purpose of the Study:

• To develop a comprehensive, balanced wound-related image repository.
• To address gaps in medical imaging resources for lower-limb conditions.
• To facilitate machine learning research in wound detection and analysis.

Main Methods:

• Collected 5443 images from clinical wound cases and healthy volunteers.
• Expertly annotated 2686 wound images across eight types with segmentation masks.
• Acquired 2757 healthy foot images using consumer smartphones from diverse participants.
• Standardized images and applied augmentation for wound datasets.

Main Results:

• Created a unified dataset with 2686 annotated wound images and 2757 healthy foot images.
• The dataset is balanced, mitigating class imbalance issues.
• Enables development of machine learning models for distinguishing normal vs. pathological foot conditions.

Conclusions:

• The developed repository provides a robust foundation for machine learning in wound care.
• Facilitates scalable and generalizable deep learning research for wound detection and monitoring.
• Supports advanced tasks like wound segmentation, severity assessment, and clinical decision support.