Eff-ReLU-Net: a deep learning framework for multiclass wound classification

Area of Science:

• Medical technology
• Artificial intelligence in healthcare
• Computational biology

Background:

• Chronic wounds pose significant health risks, including infections and amputations.
• Increasing prevalence necessitates automated wound assessment to reduce reliance on manual methods.
• Accurate and rapid wound classification is crucial for effective treatment.

Purpose of the Study:

• To develop an efficient and reliable deep learning model for multi-class chronic wound classification.
• To improve upon existing EfficientNet-B0 architecture for enhanced feature extraction.
• To validate the model's performance on diverse wound datasets.

Main Methods:

• Proposed Eff-ReLU-Net, an EfficientNet-B0-based model incorporating ReLU activation and additional dense layers.
• Employed data augmentation techniques including rotations and translations to enhance model generalization.
• Evaluated model performance on the AZH and Medetec wound datasets with cross-corpora analysis.

Main Results:

• Eff-ReLU-Net achieved high performance metrics on both datasets.
• Achieved 92.33% accuracy, 97.66% precision, 95.33% recall, and 96.48% F1-score on the Medetec dataset.
• Attained 90% accuracy, 89.45% precision, 92.19% recall, and 90.84% F1-score on the AZH dataset.

Conclusions:

• The proposed Eff-ReLU-Net demonstrates significant effectiveness for classifying chronic wounds.
• The model's architecture and augmentation strategies contribute to robust performance and generalizability.
• Automated wound classification using Eff-ReLU-Net offers a reliable solution for clinical practice.