HECA-YOLOv11: A High-Accuracy Hybrid Attention Network for Real-Time Monkeypox Skin Lesion Detection with Interpretability
Year: 2025
Type: article
Abstract: Introduction: The global outbreak risk of monkeypox necessitates efficient diagnostic tools. Current methods are slow and struggle to differentiate similar skin lesions, while deep learning models lack both accuracy and interpretability for multi-class classification. This study aims to develop a high-performance, real-time screening tool for early detection of monkeypox and similar conditions. Methods: This study proposes an enhanced architecture that incorporates the Highly Efficient Channel Attention (HECA) mechanism into YOLOv11 to improve feature selection. HECA performs Global Average Pooling (GAP) and Max Average Pooling (MAP) on input feature maps. Using spatial transformations and pixel-domain enhancements, the research establishes and augments a four-class dataset containing monkeypox, measles, chickenpox, and normal skin images. Gradient-Weighted Class Activation Mapping (Grad-CAM) generates heatmaps for lesion localization visualization, while a PyQt5-based interactive diagnostic interface streamlines clinical workflows. Results: Experimental results demonstrate the proposed model’s superior performance compared to existing approaches (EfficientNet-v2-l, GoogLeNet, ResNet50, VGG19, MobileNetv2, DenseNet201, and YOLOv11s) on the test set, with enhanced detection speed, accuracy (98.9%), precision (99%), recall (99.3%), and F1-score (99.1%). HECA-YOLOv11 outperforms SE-YOLOv11 and ECA-YOLOv11 by over 5.1% across all evaluation metrics. Heatmap analysis confirms the model’s focus on lesion regions, aligning with medical reasoning. Discussion: The integrated attention mechanism and interpretability techniques effectively address key limitations in automated skin lesion detection. The high performance and low latency signify a major advancement towards reliable clinical decision support, with important implications for early outbreak containment. Conclusion: The HECA-YOLOv11 framework provides a highly accurate and efficient tool for early monkeypox screening. Its standardized interface holds significant potential to accelerate public health responses and improve global control of emerging infectious diseases.
Authors: Xianghong Deng, Chonlatee Photong
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416895559