WoundNet-Ensemble: A Novel IoMT System Integrating Self-Supervised Deep Learning and Multi-Model Fusion for Automated, High-Accuracy Wound Classification and Healing Progression Monitoring

TL;DR

Chronic wounds impose heavy clinical and economic burdens, with subjective assessments contributing to misclassification and delayed care. The authors propose WoundNet-Ensemble, an IoMT system that fuses ResNet-50, DINOv2, and Swin Transformer in a weighted ensemble to classify six wound etiologies with 99.90% accuracy and to provide longitudinal healing tracking for remote monitoring. Leveraging self supervised pretraining on large image corpora and a robust multi architecture fusion, the approach delivers high accuracy and potential edge deployment for point of care. The work also includes a longitudinal healing tracker and clinical alert system, advancing objective wound assessment and telemedicine capabilities while offering reproducibility through public release of models and code.

Abstract

Chronic wounds, including diabetic foot ulcers which affect up to one-third of people with diabetes, impose a substantial clinical and economic burden, with U.S. healthcare costs exceeding 25 billion dollars annually. Current wound assessment remains predominantly subjective, leading to inconsistent classification and delayed interventions. We present WoundNet-Ensemble, an Internet of Medical Things system leveraging a novel ensemble of three complementary deep learning architectures: ResNet-50, the self-supervised Vision Transformer DINOv2, and Swin Transformer, for automated classification of six clinically distinct wound types. Our system achieves 99.90 percent ensemble accuracy on a comprehensive dataset of 5,175 wound images spanning diabetic foot ulcers, pressure ulcers, venous ulcers, thermal burns, pilonidal sinus wounds, and fungating malignant tumors. The weighted fusion strategy demonstrates a 3.7 percent improvement over previous state-of-the-art methods. Furthermore, we implement a longitudinal wound healing tracker that computes healing rates, severity scores, and generates clinical alerts. This work demonstrates a robust, accurate, and clinically deployable tool for modernizing wound care through artificial intelligence, addressing critical needs in telemedicine and remote patient monitoring. The implementation and trained models will be made publicly available to support reproducibility.

Figures (6)

• Figure 1: ResNet-50 training dynamics showing rapid convergence to optimal performance with minimal overfitting. The model achieved perfect classification by epoch 11 with early stopping.
• Figure 2: DINOv2 training progression demonstrating the effectiveness of self-supervised pre-training. The model shows stable convergence with validation performance reaching 99.81% by epoch 15, validating the transfer of robust visual features from natural to medical images.
• Figure 3: Swin Transformer training curves illustrating hierarchical feature learning. The model achieves smooth convergence with validation accuracy of 99.81%, benefiting from multi-scale attention mechanisms optimal for wound image analysis.
• Figure 4: Ensemble weight distribution based on validation accuracy. The optimized weighting scheme assigns proportional influence to each model (ResNet-50: 33.4%, DINOv2: 33.3%, Swin Transformer: 33.3%), ensuring balanced contribution while maximizing collective performance.
• Figure 5: Normalized confusion matrix for WoundNet-Ensemble demonstrating near-perfect classification across six wound types. Minimal confusion (<0.1%) occurs only between foot ulcers and venous ulcers—clinically similar presentations—with no misclassifications between malignant and non-malignant wounds or acute vs. chronic etiologies.