Descriptor: Neonatal Intensive Care Unit Fine-Grained Care Video Dataset (NICU-Care)

The aim of this study is to construct a high-quality video dataset called neonatal intensive care unit (NICU)-Care to support visual tasks such as nursing action recognition, target detection, and semantic segmentation in NICU scenarios. By building a standardized simulated NICU environment and capturing multiview red-green-blue (RGB) videos of multiple professional nurses performing nursing operations such as diaper changes and bottle feeding on simulated infants, we constructed a dataset covering six types of basic nursing tasks with fine-grained action annotations and pixel-level target segmentation labels. The dataset is not only highly close to the actual clinical nursing process in terms of scene construction and operation content but also has a well-structured design in terms of annotation system. In the construction process, we first detail the data acquisition and annotation methods and then analyze and evaluate the data quality and task adaptability by training mainstream models to verify the performance and challenge of the dataset in different tasks. To ensure the structured management and reusability of the data, we also designed a unified record format and storage scheme. The public release of NICU-Care will provide important support for nursing process modeling, intelligent monitoring system development, and data-driven nursing training. In the future, we will continue to expand the data scale, enrich the annotation dimensions, and explore the introduction of multimodal sensory information to further enhance the scientific value and application potential of the dataset.

IEEE SOCIETY/COUNCIL Engineering in Medicine and Biology Society

DATA TYPE/LOCATION Movie/Video; Shanghai, China

DATA DOI/PID 10.21227/kwxh-v002