Recognizing Nurse Care Activity

Nursing care activities are essential to healthcare delivery, directly impacting patient safety, recovery, and overall well-being. Accurate recognition and documentation of these activities are critical for assessing nurse performance, managing resources efficiently, and maintaining consistent care quality. However, nursing activities are inherently complex, influenced not only by the nurse's actions but also by patient behavior. Traditional documentation methods, which rely heavily on manual input, are labor-intensive and prone to errors, often leading to gaps in performance evaluation and care optimization. This paper addresses the necessity for advanced, automated systems to accurately recognize nursing care activities. We propose a novel artificial intelligence model featuring bidirectional interactive cross-attention based on Transformer, which leverages the complementary nature of multimodal data through mutual information exchange to enhance recognition accuracy and contextual understanding. We evaluate the performance of bidirectional interactive cross-attention, and experiments demonstrate that it performs excellently. Our method highlights the potential to significantly enhance nursing activity recognition, which is expected to improve nursing activity recognition and support better workload assessment, scheduling, and care quality.