Exploration of an Intelligent ICU Pressure Ulcer Skin Assessment System Based on a Dynamic Reward Adaptive Framework
Summary: The system integrates a dual-path intelligent training framework with a dynamic reward mechanism, using an adaptive detector to classify limited samples into true and fake categories. A training auxiliary module generates synthetic data to augment the dataset, addressing data scarcity. Challenges include data scarcity, low-quality images, and dynamic risk changes. Results show 89.7% accuracy on CP-LFW dataset, outperforming SenseTime (76.2%) and MIT EmoPainNet (82.1%). Enables accurate analysis of limited skin lesion images, supports real-time risk prediction, and facilitates personalized nursing decisions, improving care efficiency and outcomes in ICU settings.
Key Highlights:
- Framework: Dual-path training, dynamic rewards, synthetic data gen.
- Challenges: Scarce high-quality images, low-res, dynamic risks.
- Results: 89.7% accuracy (CP-LFW); beats SenseTime (76.2%), EmoPainNet (82.1%).
- Implications: Real-time ICU assessment, personalized care.
- Authors: Han J, Lei Y, Qiu Q.
Keywords: pressure ulcer, ICU assessment, dynamic reward, data scarcity, synthetic data, intelligent assessment