시각 지능 기반 ResNet 모델을 활용한 욕창 진단 및 분류 알고리즘 연구

Abstract

Stage-wise classification of pressure ulcer lesions was conducted using ResNet-18, ResNet-34, and ResNet-50 architectures. ResNet-18 demonstrated superior performance across accuracy, precision, recall, and F1-score. An optimized model was developed by fine-tuning its hyperparameters. To enhance interpretability, Grad-CAM was applied to visualize attention regions, confirming focus on clinically relevant features. These results support the model’s reliability and promote trust in AI-assisted decision-making. Future work will explore diverse backbone networks, validation with external datasets, and integration of advanced explainable AI techniques to improve generalizability and clinical applicability.