This chapter describes an ETL (extract, transform, load) approach for the generation of instructional resources from pre-existing collections of clinical data, in particular in the field of radiology. The approach advocates, on the one hand, the extraction of information from such sources and its representation in a unified and machine-processable format. On the other hand, the extracted information can be transformed to fit different instructional needs. This transformation process can involve both automatic transformations and transformations carried out by experts using specialized editors. Finally, the information resulting from the transformation process can be exported in standard formats in order to load it to learning management platforms (e.g., Moodle or Canvas). The chapter illustrates, with the help of a radiology clinical case collection, how this approach can be supported by a flexible digital collection management tool called Clavy.
TopIntroduction
Data sources in healthcare have been growing in an exponential way over the last years. The variety and volume of data sources coming from different health systems and clinical environments have boosted new technological platforms and tools based on AI (Artificial Intelligence) to processing this huge amount of information. This situation is particularly critical in the current pandemic context due to the increasing number of digital medical collections which are composed by myriads of radiology images, clinical reports, and text annotations over them. These multiple and heterogeneous data sources must be collected and processed by means of AI tools and they can also be used as instructional resources for training future healthcare students and practitioners. Therefore, there is a real educational demand for novel platforms and mechanisms to support the collection and processing of these multiple types of healthcare data and the current work proposes an ETL (Extraction-Transform-Load) approach (Ong et al, 2017) to cope with such a demand. The proposed ETL approach is based on a tool called Clavy (Gayoso-Cabada et al, 2016, Buendía et al. 2020, 2021), specialized in the management of data collections, and it is addressed to deal with the three stages of Extraction, Transformation and final Load of instructional resources which can be generated from existing digital medical collections.
The first challenge in the extraction of healthcare data sources is the high number of locations in which such data items can be found. That is particularly true in the radiology domain that uses public imaging repositories, publishing databases, specialized Web sites or dataset archives holding data associated to this domain. Moreover, these data sources can store different multimedia formats ranging from clinical text reports written in natural language, electronic health database records, spreadsheets measuring several biomarker values or images gathered by means of imaging techniques such as X-Ray, MR, CT or ultrasound. Their access and extraction require a wide range of mechanisms depending on the data source location, the structure of the collected data or the semantics associated to them.
Once this myriad of data sources has been extracted, the second challenge lies in obtaining an information model that provides a homogeneous schema of the required data structure. The aim of this global information model is to show the multiplicity of features and attributes characterizing the considered radiology data sources. A first transformation process will populate this model with the information gathered from all these information sources. Afterwards, other transformation processes should be able to convert this raw information in the integrative model to fit in more specific schemata according to the selected educational purposes which can range from merely expositive presentations to advanced interactive products (e.g., including questions or simulation exercises). These transformations can be supervised by experts who decide which data attributes will be integrated into the final instructional resources or the multimedia items attached to them. Moreover, cleaning or curation processes should be incorporated to guarantee the quality or safety of the data used during this transformation stage.
The final challenge of the ETL approach consists in generating the instructional resources from the schemata used during the previous transformation processes and loading them into platforms or environments which allow for the different healthcare agents their actual application. In particular, we have experimented with the generation of resources for the educational domain (Buendía et al. 2020, 2021). In this point, the use of standard educational specifications is crucial to enable the upload of these resources to the most popular e-learning platforms as well as improving their reusability and interoperability. In summary, the main objectives of the current work consist in providing a global framework that enables the extraction and transformation from medical digital collections to finally generate and load educational resources, which can be deployed in different kinds of e-learning platforms.