AI-ASSISTED WOUND CARE TRIAGE: AN INTEGRATED FRAMEWORK FOR CLINICAL ASSESSMENT AND HEALTHCARE RESOURCE OPTIMIZATION

Abstract

Wound care is one of the most resource-intensive areas of hospital operations; however, most decisions that drive resource consumption, taken in this field, are made through visual inspection and individual clinical judgment. This variability in assessment creates unnecessary hospital congestion and misallocation of clinical resources, affecting the medical supply chain. This thesis proposes an integrated AI-assisted framework that combines clinical wound assessment and operational triage decision-making. The proposed framework is organized into four interconnected layers through which information flows from raw data to clinical pathway decisions. The vision layer applies deep learning models to extract visual characteristics from the wound image. The clinical layer translates the visual outputs and metadata into clinically meaningful indicators of wound severity and healing status. The temporal layer captures wound progression and identifies whether a wound is improving or deteriorating when sequential images are available. Finally, the operational layer transforms all outputs, using evidence-based triage protocols, into one of the three care pathway decisions: home care, mobile care, or hospital care, and detects cases requiring escalation. The vision layer was implemented using a U-Net architecture for wound detection and segmentation, and achieved a Dice coefficient of 0.822 and 0.65, respectively. The operational impact of the framework was evaluated through an optimization-based discrete-event simulation study; Results showed that deploying the AI triage tool with an optimized confidence threshold of 0.765 reduces average patient length of stay by approximately 16%, eliminates false negative triage decisions, and reduces the required clinical staffing configuration, demonstrating that the benefits of AI-assisted triage extend beyond individual clinical decisions to measurable system-level improvements. The primary contribution of this thesis is the integration of these components into one single framework. To our best knowledge, no existing system combines automated wound image analysis with a knowledge-based clinical inference model and an operational simulation-optimization framework within a single architecture. The results suggest that this integration is operationally meaningful, offering a foundation for more consistent, data-driven wound triage in hospital settings.

Date of Award	2026
Original language	American English
Awarding Institution	HBKU College of Science and Engineering