Multi-Class Wound Classification via High and Low-Frequency Guidance Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Convolutional Neural Networks (CNNs) show promise for wound image classification.
• Classifying multiple wound types remains challenging due to image complexity and feature extraction limitations in existing CNNs.

Purpose of the Study:

• To develop an advanced deep learning model for accurate multi-class wound image classification.
• To address information loss in feature extraction by separating high- and low-frequency components.

Main Methods:

• Proposed the High and Low-frequency Guidance Network (HLG-Net) with two branches: High-Frequency Network (HF-Net) and Low-Frequency Network (LF-Net).
• Utilized pre-trained ResNet and Res2Net for HF-Net to capture high-frequency details.
• Employed a Multi-Stream Dilation Convolution Residual Block (MSDCRB) for LF-Net to extract low-frequency information.
• Integrated a fusion module to combine features from both branches for final classification.

Main Results:

• HLG-Net achieved maximum accuracies of 98.00% (2-class), 92.11% (3-class), and 82.61% (4-class) wound image classification.
• The proposed method significantly outperformed previous state-of-the-art techniques.

Conclusions:

• HLG-Net offers a superior approach for multi-class wound image classification.
• The network's ability to effectively process high- and low-frequency features enhances classification accuracy in intelligent medical systems.