The Intensive Care Unit (ICU) of Ghent University Hospital is responsible for the treatment of patients with acute, life-threatening disorders. The patient is constantly monitored by a team of doctors and nurses. During the hospitalization, the patient's relatives experience a lot of stress. The medical staff of the Ghent University Hospital tries to comfort and support the family members by offering specific information about the patients situation. Currently, this is being performed verbally, supported by paper brochures. However, during this stressful period the family members tend to forget the given information. The given brochures, which are also available on-line, are created for broad categories of pathologies, to ensure usability for a large group of patients. Consequently, the brochures contain a plethora of information that is not applicable to the Fig. 1. Architecture of the NAOMI platform particular patient's condition. As such, the family members tend to be drowned in non-specific and detail information. To enrich their knowledge about the patient's pathology, they then tend to search the Web for more information, e.g., using Symptom Checker3 . This often leads to misconceptions and more worry because the found material is not personalized and contains too much information that does not apply to the specific situation or treatment of the patient.
To counter these issues, we have designed a patient information system, called NAOMI, to support communication between the medical staff and the patient's relatives about the patient's condition. NAOMI allows the medical staff to upload brochures. These are semi-automatically annotated according to pre-defined, fine-grained categories. When a patient is admitted to the ICU or his/her condition is updated in the electronic patient record (EPD) of the hospital, his/her pathology and/or symptoms are automatically mapped on these annotations. As such, only relevant sections of the brochures are provided as information to the relatives. In this way, NAOMI provides a source of information that is personalized to the patients current condition. Moreover, to support the verbal communication, NAOMI also contains a question-and-answer service, allowing the medical staff to inquire about the (medical) history and lifestyle of the patient, and a diary, enabling the patient and relatives to share text and pictures with each other.
The high-level architecture of NAOMI is shown in Figure 1. The various components are highlighted below.
Annotation service: The annotation service provides the functionality to semi-automatically annotate the various paragraphs of the uploaded brochures according to fine-grained categories. First the text is pre-processed by: identifying the different paragraphs, performing tokenization that splits up the paragraph in single tokens consisting of 1 or more words that have joint meaning, e.g., non invasive ventilation, normalizing the text to remove capitalization and superfluous characters, e.g., punctuation and numbers, performing lemmatization to transform each word to its stem, removing stop words, and translating the remaining words to English. To identify the labels for the paragraphs, a medical ontology is used, namely the SNOMED CT (Systematized Nomenclature of Medicine -Clinical Terms) ontology4 . It represents a collection of medical concepts in a hierarchy, where each one of them is described by fixed medical terms. The tokens that remain after the pre-processing are looked up in the SNOMED terms by using Stardog5 and SPARQL. The most frequent occurring medical terms from SNOMED are used to label the paragraph. The brochure and its labels are stored in the database in a tree-based manner. The root is the title of the brochure. Each child of this root represents either a paragraph or a subtitle. Paragraphs below a subtitle are added as children of this subtitle node. As such the structure of the document is reflected in the shape of the tree, i.e., the more subtitles, the deeper the tree. A parent node also receives the labels of its children and the most frequent ones are used to label this parent (i.e. title). As such, the root contains all the most important labels that categorize this brochure and all the relevant paragraphs and titles can easily be identified that are linked to a particular term.
Rule-based service: The rule-based service, implemented using the Easy Rules system6 , is used to map the pathology, symptoms and treatment plans of the patient on the annotation categories. These rules are incepted and curated by the medical staff of Ghent University Hospital. This ensures that only the paragraphs of the brochures are selected and shown that are relevant for this particular patient's condition. An example of a Rule is: IF ICDS score > 4 AND ETT Tube present THEN linked categories are "invasive ventilation", "non-invasive ventilation", "sedation", "fixation", "tracheal cannula".
Information retrieval and storage: The platform connects to existing databases of Ghent University Hospital through the use of a data virtualisation system, namely jBoss Teiid7 . From these database, the necessary information about the patient is retrieved, e.g., his/her pathology, symptoms and current treatment. A dedicated database for NAOMI was also set-up, which contains, for example, the uploaded brochures and its annotations, account information of the users (e.g. the relatives) and the questionnaires of the medical staff towards the relatives.
Web applications: Two web application were developed in Angular, one for the medical staff and one for the patient and his/her relatives.
The demo will allow the visitors to operate both designed web applications. First, they will log in as a medical staff member. They will be able to choose from a number of provided brochures to upload into the platform. Next, they will use the first web application to semi-automatically annotate this brochure. Second, they will log in as a family member of a given patient, who has a pathology linked to the brochure they just annotated. They will then see how the second web application automatically fetches the parts of the brochure that are relevant for this patient. The two videos at the following link give an idea of the functionality of both web applications: https://www.youtube.com/playlist? list=PLFnSuWuVj0rFwVYp4o1d0bekLVYaq8n9J.