Automatic Classification of Wound Images Showing Healing Complications: Towards an Optimised Approach for Detecting...

Area of Science:

• Medical imaging
• Artificial intelligence in healthcare
• Digital wound care

Background:

• Maceration is a common wound healing complication impacting patient outcomes.
• Accurate and timely detection of maceration is crucial for effective treatment.
• Current diagnostic methods for maceration can be time-consuming and subjective.

Purpose of the Study:

• To develop and evaluate convolutional neural network (CNN) architectures for automated maceration classification in wound images.
• To identify the optimal CNN model for maceration detection, considering performance and complexity.
• To explore image processing techniques and model interpretability for enhanced diagnostic accuracy.

Main Methods:

• A dataset of 458 annotated wound images was utilized.
• Several CNN models, including MobileNetV2, were trained and evaluated for maceration classification.
• Image cropping strategies and Grad-CAM visualizations were employed to analyze model performance and decision-making.

Main Results:

• MobileNetV2 demonstrated the highest classification accuracy among the evaluated CNN models.
• Model performance was analyzed in relation to model complexity and dataset size.
• Grad-CAM visualizations provided insights into the CNN's feature focus for maceration identification.

Conclusions:

• CNNs, particularly MobileNetV2, show significant potential for accurate automatic classification of wound maceration.
• Automated maceration detection can enhance diagnostic accuracy and efficiency in digital wound care.
• Further research can optimize CNN models for improved wound assessment and patient outcomes.