Effectiveness of Semi-Supervised Active Learning in Automated Wound Image Segmentation

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Computational pathology

Background:

• Chronic ulcer management relies on accurate wound area measurement.
• Image analysis is a key instrumental method for wound assessment.
• Current methods can be time-consuming and require specialized equipment.

Purpose of the Study:

• To develop an AI-driven convolutional neural network (CNN) model for efficient wound segmentation.
• To automate clinical wound severity assessment using smartphone-acquired images.
• To establish a foundation for advanced ulcer characteristic analysis.

Main Methods:

• Utilized active semi-supervised learning for CNN training.
• Developed a custom app for image acquisition and a database for AI training.
• Tested various CNN architectures and validated with public datasets.
• Created an annotated dataset of 1564 ulcer images from 474 patients.

Main Results:

• Achieved a Dice Similarity Coefficient (DSC) greater than 0.95 after <50 epochs.
• Demonstrated robustness across diverse lighting conditions and image expositions.
• Validated the model's efficiency and reliability against public datasets.
• The AI model automates wound segmentation and assessment.

Conclusions:

• The proposed active semi-supervised learning strategy offers an efficient wound segmentation method.
• The AI model can significantly reduce clinician working times for wound measurements.
• The robust pipeline is suitable for clinical practice as a decision support system.