Experiments on the dielectric properties of biological tissues generate data that characterizes the interaction of human tissues with electromagnetic elds. This data is vital for designing electromagneticbased therapeutic and diagnostic technologies, and for assessing the safety of wireless devices. Despite the importance of the data, poor reporting and lack of metadata impede its reuse and forgo interoperability. Recently, the minimum information model for reporting Dielectric Measurements of Biological Tissues (MINDER) has been developed as a common framework. In this work, we have developed a metadata model and implemented a data sharing framework to improve ndability and reproducibility of experimental data inspired by FAIR principles. We de ne a process for sharing the reported data and present tools to support rich metadata generation based on existing community standards. The developed system is evaluated against competency questions collected from data consumers, and thereby proven to help to interpret and compare data across studies.
Data sharing is the release of research data for use by others [
One of the fastest growing areas for medical device development in Europe is
electromagnetic (EM) imaging and therapeutics. Within the context of an aging
population and exponential growth in healthcare costs, EM-based techniques
provide a very attractive solution for new therapeutics and diagnostic
technologies, since they are low cost, non-ionising and largely non-invasive. Many
European companies have attempted to commercialize their technology. However,
over 75% of medical device companies go out of business within the rst ve
years [
In this work, we aim to bridge the experimental data gap in device
development by proposing an approach and implementing tools to improve ndability
and reusability of dielectric measurement experiments. Achieving well
maintained, interoperable, and machine actionable data and metadata are the main
building blocks towards this goal. This can be tackled with Semantic Web
technologies. The Life Sciences domain, as an early adopter of the Linked Data
approach, provides many examples for integrating data from multiple sources and
making them queryable on the web through the SPARQL query language [
Recently, the FAIR guiding principles have been proposed for scienti c data
management and stewardship, and have had an impact on the scienti c
community at large. These principles, rather than prescribing a set of standards,
describe the qualities or behaviours required of data sources to achieve their
optimal discovery and reuse [
Accessible: By using a standardized communication protocol, data, as well as metadata, should be accessible, even if the actual data no longer exists. Interoperable: Applying controlled vocabularies and quali ed references for metadata to be used for knowledge representation.
Reusable: By having a clear and accessible data license and using a well-de ned set of accurate vocabulary, the data becomes easily re-usable.
Existing standards can be applied to ful ll the requirements of the FAIR guidelines in varying degrees. However, the need for developing further standards is apparent. Our speci c goal is to adapt some of these FAIR principles using Semantic Web standards to improve discoverability and reusability of experimental dielectric data sets to support EM device development.
We follow an incremental approach to achieve optimal FAIRness of the data sets. This paper reports our rst iteration by application of a set of Semantic technologies, and achieves some degree of FAIRness. The current implementation is neither complete in terms of coverage of all FAIR principles nor does it yet demonstrate the full bene ts of existing standards. However, this work provides a starting point and a good example of practical applications of Semantic Web technologies for FAIR data sharing, as well as providing a view on future directions.
The rest of the paper is structured as follows: Section 2 gives an overview of related works. In section 3, we discuss the modeling of Dielectric Measurements of Biological Tissues (DMBT) experimental data. Section 4 discusses our proposed pipeline for sharing DMBT data and metadata, domain speci c vocabulary development, and our RDFization technique. In section 5, we describe access mechanisms to data and metadata. Finally, we provide a future outlook and conclude the paper. 2
Although the knowledge discovery approach was invented more than 400 years
ago, the dissemination of knowledge is still mostly done in the same way as it was
when invented in the 16th century [
One of the rst attempts to tackle the problem was the 5 star OPEN DATA
by Tim Berners Lee5. This approach provides ve guidelines for data namely:
the data must be available on the web under an open license, must be structured
in a well-de ned format, accessible in a non-proprietary format like CSV, must
use URIs for denotation, and must be linked to other data to add contextual
information. Although, some organizations, for example, Thomson-Reuters [
The dielectric properties, namely, the relative permittivity ("r) and conductivity ( ), of biological tissues quantify the interaction of electromagnetic elds with the human body. Together, these properties characterize how EM waves are re ected at, absorbed by, and transmitted through the body. Knowledge of the dielectric properties of various tissues is vital to the eld of dosimetry (safety studies, such as for wireless communication devices), and for the implementation of EM-based medical technologies, such as microwave ablation and imaging.
Dielectric measurements are typically performed using an open-ended
coaxial probe connected to one port of a vector network analyzer (VNA) through a
specialized cable [
Although the process of conducting a dielectric measurement on a tissue sample appears rather straightforward, there are a multitude of confounders that can impact the measured data. These confounders are a likely source of inconsistencies in reported data. Both equipment-based measurement confounders and clinical confounders a ect the accuracy of dielectric data. Uncertainties in the dielectric data caused by these measurement confounders have been thoroughly investigated over the years and can now be reduced or eliminated by following good measurement practice. However, clinical confounders have been relatively little investigated to date and may introduce a signi cant level of additional uncertainty into the dielectric data.
The reusability of dielectric measurement data and reproducibility of experiments can be improved by capturing metadata that describes the confounders and by making this metadata a part of the data sharing practice. Moreover, having confounders' metadata together with the metadata about the study itself can help data consumers to de ne their data requirements and will improve the discoverability of datasets.
Recently as part of our earlier work [
model type selection, number of poles, and tting algorithms. The developed
reporting model is also compliant with the MIRIAM guidelines [
In this work, we report on the implementation of a framework to semantically express the metadata and data reported via the MINDER reporting schema and templates. We also developed a platform enabling the discovery of and access to data and metadata by both individuals and machines. To demonstrate the added value of our proposed solution, we collected a set of competency question that are commonly asked by data consumers, as follows:(i) is there any data for pancreas tissue? (ii) is there any data for porcine bladder tissue? (iii) are there kidney measurements available at 24:3 C? (iv) are there any measurements available on biological tissues at 18 GHz? (v) is there any liver data taken between 20 C and 25 C over the frequency range of 1{2 GHz? (vi) is there any tissue-mimicking phantom data? We present that our realized approach enables us to nd answers to these questions. 4
Currently there is no standard way to nd and access data on dielectric measurements of biological tissues (DMBT) generated in labs. In most cases, the metadata is only partially recorded, if at all, in lab books and the data is stored separately on hard disks. Some metadata is reported unsystematically in publications, without any controlled vocabulary, resulting in the lack of both human and machine discoverability. 4.1
In this work, we have created a pipeline to transform the semi-structured, nonstandardized DMBT data and metadata resulting from experiments in individual labs to a machine discoverable triple store, and developed a data portal to ful ll data access requirements of end users. Our semantic pipeline implementation follows a subset of the FAIR data principles, namely: { To make the scienti c data ndable, we assign unique and persistent identiers to metadata and data, and uploaded it in a publicly accessible repository. We assigned persistent identi ers for each data object. For the identiers we used dereferenceable URLs through the MIRIAM registry. { To make our data accessible, at rst we reviewed the data and metadata and decided on required access mechanisms. Metadata is served via machine interoperable, well de ned SPARQL protocol, whereas data can be accessed and downloadable as CSV les with prede ned headers. We did not implement any access control mechanism, since all data currently residing on the platform is freely available. { To make the scienti c data interoperable, we applied Semantic Web technologies. The metadata is transformed into the Resource Description Framework (RDF) format utilizing shared, domain speci c vocabularies and ontologies. { To make the scienti c data reusable, we provided rich and well de ned metadata. Moreover, reusing existing vocabularies and ontologies makes the shared data linkable to other data sources.
The basic work ow that results from this work is illustrated in g. 1. At rst, the metadata les are parsed and transformed into Resource Description Framework (RDF) according to the vocabulary. This process is further explained in section 4.3. The resulting triples are then transferred into a Virtuoso triple store9 that is accessible from the web. The web-application then enables the user to browse this data in various ways, for example, through guided drillingdown into the experiments or the computation of statistics. Further, the user can query the data either directly or using pre-speci ed query templates. This web application is served using an nginx web server and itself programmed in PHP. Also easy programmable direct access to the data is provided to enable machine access. 4.2
Vocabularies de ne the terms (concepts and relationships) used to describe and
represent a domain. In this work we developed a vocabulary with terms used
to de ne rich experiment metadata. The data model is based on the MINDER
minimum information model, which followed the ISA framework classi cation.
In order to optimize interoperability, we reuse a variety of existing terms from
established ontologies and vocabularies. Suitable candidates were discovered using
the ontology browse and search tools Ontobee [
We only reuse terms fully re ecting the semantic meaning of the term in the context of our application and chose de nitions from more commonly used ontologies when several suitable candidates were available to simplify integration with existing datasets. This content reuse enables us to develop a consistent representation of this domain, reusing content, deploy existing models and align them to other related datasets. Moreover, this helps us to increase the interoperability between other ontology-based applications. 4.3
RDF is a W3C standard for the description and modeling of data in a structured way. This standard also provides an abstract and conceptual framework for de ning and using metadata and metadata vocabularies by applying statements that consist of subjects, predicates and objects to an ontology. Consequently, the data model becomes easily searchable, ndable, and accessible.
The originally generated metadata for our DMBT experiments was stored in Excel format. To transform this data into RDF, we have developed a Java application that parses given metadata les into a data structure, which is then converted into RDF, adhering strictly to the developed vocabulary. The developed RDFization tool makes use of the Apache Jena framework. Metadata is stored in an online triple store hosted on Amazon web services. As described above, this endpoint can be accessed directly or through our web application10. This web application is designed for non-tech-savvy users to facilate uplaod of new data, however it can receive as input only a speci c template. 4.4
To allow reliable referencing of the entities and to enhance the interoperability between di erent controlled vocabularies and databases, we utilize persistent identi ers (PIDs). PIDs enable the combination of data concerning the same entity from multiple data sources via identi er matching. Thus they enhance the interoperability so that more than one entity can be combined across individual documents and data sources.
The identifier.org system directly employs dereferenceable URLs through
the MIRIAM registry11 which provides persistent identi cation for life science
data [
Overall this system provides a good basis for the persistent identi cation of data and metadata in our usage scenario and is thus employed as PID provider for all datasets in the Bio-MINDER tissue database. 5
To demonstrate the e ectiveness of our proposed approach, we evaluated our approach through our web application. The Semantic Web technologies used in our web application provide the encoding of the entities by assigning resolvable identi ers. The used vocabulary also de nes the purpose or interpretation of terms. Additionally, a high-level description of the ontology is available from the web interface so that user can reuse it. These functionalities help make the data accessible and enhances the re-usability. We provided access to data for both human and machine consumption through SPARQL query interface and web application respectively. Also, as part of the web application, there is a tab in which all of the above-mentioned competence questions are answered based on SPARQL queries. These can be executed and the result browsed. Further, when a user selects a speci c investigation ID, both the metadata and the description can be accessed. An example is shown in g. 2. The concrete experimental data can then be downloaded when the user agrees to the speci ed terms and conditions. 6
In this paper, we demonstrated a use case of Semantic Web and a subset of the FAIR principles to improve the repeatability of dielectric measurements of biological tissue and the reusability of the produced data. We showed how to adopt the MINDER speci cation and developed a domain speci c, controlled vocabulary which makes the data more ndable and reusable, setting rst steps towards the FAIR principles. We utilized persistent identi ers (PIDs) for both the data and metadata from experiments on the dielectric measurements of biological tissue. This allows the referencing of data in both a location-independent and resource-dependent manner. The provision of resolvable identi ers (URLs) ts well with the Semantic Web vision, and the Linked Data initiative. Moreover, we have reused existing ontologies for terms from our controlled vocabulary where appropriate. Then, we made the overall system available through a web interface which also speci cally answers the competence questions which were gathered from domain experts.
In this rst iteration, our approach still has several limitations. First, we currently have only met some aspects of the FAIR principles. In future work, we could look at whether it is reasonable and feasible to also address other aspects. For example, we did not deal with licensing properly. We have included a simple licence for the provided les, but these are not in a form which is machine interpretable. One reason we have refrained from de ning this is that there is currently no consensus on what a good way would be. This issue is, for example also discussed in working groups on the DCAT standard12, and several application pro les have chosen di erent ways to address this. Hence, it would be better to wait until a consensus is reached there before reinventing the wheel ourselves. We have currently also not speci ed any data access restrictions. For the datasets currently provided, there is no such need, but for data commercially provided, this has to be amended.
Currently, we also chose to only reuse existing vocabulary terms in case there was an exact match. We could consider also adding redundant vocabulary terms, which are broader as the ones we applied to increase potential reusability. A related issue is that we did not make use of, for example, the DCAT vocabulary for specifying our data catalog. The reason is that we were mainly focused on the domain vocabulary and not on the higher level from the start. This can be amended in future work.
This work already showed useful progress. In later work, we would like to further improve this approach for not only biological experiments but also other domain like agriculture, nance, marketing, etc. 12 https://www.w3.org/TR/vocab-dcat/