Efficient Wound Classification Using YOLO11n: A Lightweight Deep Learning Approach

Wound classification is a critical task in healthcare, requiring accurate and efficient diagnostic tools to support clinicians. In this paper, we investigated the effectiveness of the YOLO11n model in classifying different types of wound images. This study presents the training and evaluation of a lightweight YOLO11n model for automated wound classification using the AZH dataset, which includes six wound classes: Background (BG), Normal Skin (N), Diabetic (D), Pressure (P), Surgical (S), and Venous (V). The model’s architecture, optimized through experiments with varying batch sizes and epochs, ensures efficient deployment in resource-constrained environments. The model’s architecture is discussed in detail. The visual representation of different blocks of the model is also presented. The visual results of training and validation are shown. Our experiments emphasize the model’s ability to classify wounds with high precision and recall, leveraging its lightweight architecture for efficient computation. The findings demonstrate that fine-tuning hyperparameters has a significant impact on the model’s detection performance, making it suitable for real-world medical applications. This research contributes to advancing automated wound classification through deep learning, while addressing challenges such as dataset imbalance and classification intricacies. We conducted a comprehensive evaluation of YOLO11n for wound classification across multiple configurations, including 6, 5, 4, and 3-way classification, using the AZH dataset. YOLO11n acquires the highest F1 score and mean Average Precision of 0.836 and 0.893 for classifying wounds into six classes, respectively. It outperforms the existing methods in classifying wounds using the AZH dataset. Moreover, Gradient-weighted Class Activation Mapping (Grad-CAM) is applied to the YOLO11n model to visualize class-relevant regions in wound images.