Abstract
Parents of infants in the Neonatal Intensive Care Unit (NICU) often face significant challenges due to a lack of clear information, compounding the stress of a highly emotional experience. To address this, an automated system capable of generating concise summaries of an infant's progress is proposed. Central to this system is a specialized NICU ontology, developed using Protégé 3.4.2 following a seven-step methodology. This ontology comprises 10 classes, 17 data properties, and 8 object properties, along with cardinality restrictions to ensure logical consistency. Validation through reasoning and SPARQL queries confirmed the ontology's effectiveness, paving the way for its use in natural language generation (NLG) applications within the neonatal domain.
Introduction: Bridging the Communication Gap
The NICU is a high-pressure environment where medical jargon and complex data can alienate parents. While medical staff provide updates, the technical nature of Electronic Medical Record (EMR) data often leaves parents feeling overwhelmed and powerless. A "data-to-text" system—software that transforms raw clinical data into natural language summaries—offers a solution. However, such systems require a formal knowledge structure, or ontology, to interpret the domain correctly. This paper details the development of a localized NICU ontology designed to facilitate these automated updates.
Methodology and Ontology Design
The researchers employed the established Noy and McGuinness methodology to construct the ontology. The process focused on defining the relationships between infants, parents, medical staff, and treatments to create a machine-readable map of the clinical environment.
Class Hierarchy
The ontology defines 10 core classes that represent the primary entities in the NICU environment. These include Preterm_infant, Nurse, Treatments, Nutrition, and Parents. This structure allows the system to categorize data points accurately.
Properties and Constraints
To define the attributes of these classes, 17 data properties (e.g., Weight, AdmittedDate, MRN) were established. Furthermore, 8 object properties (e.g., CaredBy, FedBy, HasParent) define the relationships between classes.
A critical aspect of the design is the implementation of cardinality restrictions. These logical constraints ensure that the data remains accurate and meaningful for the summary generator. For instance, the model specifies that a Preterm_infant must have exactly two parents and be cared for by at least one nurse.
Validation and Results
The ontology was validated using the FaCT++ Reasoner to ensure internal consistency. Initial data type errors were identified and corrected during this phase. Subsequently, the authors executed four SPARQL queries to test the model's utility in retrieving specific clinical information, such as identifying infants treated with specific medications (e.g., Folic Acid) or listing infants by bed number. All queries returned correct results, proving the ontology's readiness for integration into a summary generation system.
Conclusion and Future Directions
This ontology serves as a robust knowledge backbone for future NICU support systems. While currently focused on nutrition and basic treatments, future iterations will expand to include respiratory and neurological data. The ultimate goal is to integrate this ontology with a Natural Language Generation (NLG) module to provide parents with clear, automated, and empathetic updates on their child's health, reducing the emotional burden of the NICU experience.
Summary updated