Towards Human-Centric and Resilient Hospital Ecosystems: An AI-Driven Facemask Compliance Detection Framework for Healthcare Workers and Patients

<title>Abstract</title> <p> The hospitals around the globe are undergoing a triple transformation driven by digitalization, resilience, and human-centricity. Hospitals are critical service-delivery sectors, where digital technologies need to be embedded in clinical operations to protect patients and healthcare workers (HCWs). This study reviews artificial intelligence (AI) and deep learning (DL)-based facemask detection techniques, examining how they can be embedded into hospital infrastructure to serve as personal protective equipment (PPE) compliance for both patients and HCWs. A systematic review was conducted of deep learning architectures for facemask detection including ResNet-50, MobileNetV2, CNN, the YOLO family (v1–v5), VGG-19, and Inception-V3 and evaluated against the transformation dimensions: (i) <italic>digitalization</italic> , assessed through real-time inference capability and integration with existing hospital CCTV; (ii) <italic>resilience</italic> , assessed through robustness to clinical-environment variability (occlusion, lighting, patient positioning); and (iii) <italic>human-centricity</italic> , assessed through the system's sensitivity to the differentiated needs of HCWs, patients, and visitors across hospital zones. The review shows that no single architecture optimally satisfies all three transformation dimensions simultaneously. YOLOv5-based models deliver the strongest digitalization and resilience performance, achieving precision and recall above 97% under occlusion and variable lighting, making them suitable for high-acuity zones such as ICUs and operating theatres. Embedding AI-based facemask detection into hospital infrastructure operationalizes the triple transformation at the level of frontline care delivery: it digitalizes a previously manual compliance process, strengthens institutional resilience against healthcare-associated infections (HAIs) and focuses on the wellbeing of HCWs and patients. This study extends the integration of triple transformation to hospital service ecosystems, an environment where the same transformation pressures apply but have received comparatively little systematic attention from an AI-engineering perspective. It provides one of the first frameworks to evaluate facemask detection architectures explicitly against these three transformation dimensions. </p>