Serge Autexier
Deutsches Forschungszentrum Fur Kunstliche Intelligenz Gmbh

A key goal of the CrowdHEALTH project is to provide holistic health records (HHRs), which combine clinical health data and other health-relevant data such as lifestyle, physical and social activities, nutrition related activities or personal assessments, e.g. quality of life. While health data and especially clinical data is often available in electronic health records, there is lack for non-health but health-relevant data such as nutrition, physical or social activities.

For physical activities efforts are made to derive types of physical activities from sensor data. However, there are no standardized categories of physical activities. For nutritional information, most existing applications focus on nutritional data and less on food categories. Moreover, most standards in that domain are pure taxonomies and the links to nutritional information have no open access, which prevents free use of the data for analytics as envisaged in CrowdHEALTH. Similarly, information about dishes, their compositions and nutritional values are also not open access and dish descriptions from freely accessible recipe databases provide details on ingredient in non-standardized formats, mostly in natural language, mixing unsystematic volume or quantity information, ingredient descriptions, preparation status, provenance, production type or even brand names, which hampers linking of these with nutritional information in databases. Furthermore, to enable users to manually adjust dish compositions or provide new dishes is typically not supported by nutrition tracking apps, which is a necessary feature to support a detailed nutrition tracking beyond predefined dishes.

In order to integrate activities and nutrition information in HHR records DFKI has therefore developed a food and activity ontology, which can be aligned with existing taxonomies and also serves to link data for specific food types and activities. That ontology was included in the definition in the HHR categories. In order to include physical activities data into HHRs, DFKI has developed a simple mechanism to automatically include data obtained from fitbit trackers. This is used together with the project partner CareAcross to include physical activity data into of the CrowdHEALTH platform since April 2018.

For nutrition tracking, DFKI developed analytics tools to support (1) automatic mapping of nutritional data to the food category and (2) map ingredients from dish descriptions or recipes to the food ontology. To support a low-threshold collection of nutritional data, the development of a web application is currently finalized by which users can provide their data anonymously. That web application is planned to be released in German and English in April 2019 in order to populate the CrowdHEALTH platform also with nutrition data.