Detection of Hypergranulation Tissue in Chronic Wound Images Using Artificial Intelligence Algorithms.

Area of Science:

• Medical technology
• Artificial intelligence in healthcare
• Wound healing research

Background:

• Hypergranulation in chronic wounds signifies impaired healing, leading to delayed recovery and increased infection risk.
• Early misidentification of hypergranulation hinders timely and effective clinical intervention.
• Current diagnostic methods for hypergranulation may lack precision, impacting treatment efficacy.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for distinguishing hypergranulated from non-hypergranulated wound tissue.
• To assess the performance of various deep learning architectures in identifying hypergranulation.
• To provide a tool that assists clinicians in the early detection of hypergranulation.

Main Methods:

• A dataset of 6235 wound images was compiled from two sources, ensuring balanced classes.
• Five deep learning architectures (ViT, VGG16, RepVGG, MobileViT, RepGhost) were evaluated using transfer learning.
• Performance was measured by accuracy and area under the receiver operating characteristic curve (AUC).

Main Results:

• The RepGhost model achieved the highest performance, with 81.4% accuracy and 89.4% AUC.
• Lightweight models like RepGhost and MobileViT showed high performance, suitable for mobile device implementation.
• The developed deep learning method outperformed both convolutional neural networks (CNNs) and transformer architectures.

Conclusions:

• Deep learning offers a promising approach for the accurate and early detection of hypergranulation in chronic wounds.
• The RepGhost model demonstrates significant potential for clinical application in wound management.
• This study utilized a large, balanced dataset, representing a novel contribution to deep learning research in hypergranulation.