VGG19 demonstrates the highest accuracy rate in a nine-class wound classification task among various deep learning...

Area of Science:

• Medical Artificial Intelligence
• Computer Vision in Healthcare
• Digital Diagnostics

Background:

• Current deep learning models show limited accuracy in multi-class wound classification.
• There is a growing need for efficient wound management solutions to reduce the burden on healthcare professionals.
• Assisting non-specialists in wound diagnosis and management is crucial.

Purpose of the Study:

• To create a dependable and precise 9-class wound classification system.
• To support wound care specialists, and potentially patients and general practitioners, in wound management.
• To enhance diagnostic capabilities through artificial intelligence.

Main Methods:

• Trained and tested six deep learning networks (VGG16, VGG19, EfficientNet-B0, EfficientNet-B5, RepVGG-A0, RepVGG-B0) on 8173 training and 904 test images.
• Classified images into nine distinct categories: operation wound, laceration, abrasion, skin defect, infected wound, necrosis, diabetic foot ulcer, chronic ulcer, and wound dehiscence.
• Analyzed accuracy rates for each network.

Main Results:

• Overall accuracy ranged from 74.0% to 82.4%.
• The VGG19 network demonstrated the highest accuracy at 82.4%.
• Performance is comparable to existing studies in wound classification.

Conclusions:

• The VGG19 model shows significant potential for developing an AI-driven wound diagnostic system.
• Such AI systems could improve wound diagnosis and treatment for both professionals and the general public.
• Further development could lead to more comprehensive AI tools for wound care.