A Visual Medical Knowledge Explorer
|
If you are the presenter of this abstract (or if you cite this abstract in a talk or on a poster), please show the QR code in your slide or poster (QR code contains this URL). |
Abstract
Background: A second generation of UMLS based visual medical knowledge explorer is currently under development. Using new similarity measures and new visualization techniques, we have significantly advanced our previous prototype, Visual Concept Explorer (VCE) (http://research.ischool.drexel.edu/vce/). We will début this new system at this conference and introduce new visual features and interactive functionalities that will help users navigate and explore medical knowledge through concept maps and concept semantic relationships.
Objectives: There are three objectives of this project. The first is to investigate ways to generate beautiful, meaningful, and useful visual concept maps for medical and biological domains. The second is to implement a practical system that can be applied to multiple biomedical vocabularies and ontologies to generate instant concept maps based on user’s information needs as well as information content and context extracted from databases of medical records and medical literature (PUBMED, MEDLINE plus or search engines, etc.). The third is to study how the automatically generated concept maps will help user access medical records and literature through visual exploration and how the visual exploration process improves learning of biomedical concepts.
Methods: Multiple ways of calculating semantic relationships of terms are used and compared in this project. The base line is the co-occurrence of terms, which is what we used in the previous prototype. Using term co-occurrence counts to approximate semantic relationships of the terms has been studied and reported widely in the literature. The co-occurrence provides a good contextual and semantic relationship when the terms are used frequently and appeared in many documents. However, it also lacks of specificity and accuracy when the terms are ambiguous or with multiple meanings. This sometimes makes it difficult to understand the semantic maps created solely based on the co-occurrence counts.
With recent advancement in similarity measurement for medical concepts, many alternative measures might be used to replace or enrich the co-occurrence count for the purpose of identifying semantic relationships of vocabulary terms. We have tested several measures implemented in the open source software package UMLS::Similarity and used the resulting similarity measures to generate semantic maps in our system. Some interesting similarities and differences are found when comparing the maps generated by different similarity measures and by the co-occurrence counts.
On the visualization front, we have moved our flash-based visualization to JavaScript-based visualization using the open source package D3.js (http://mbostock.github.com/d3/) and some other JavaScript packages. This makes our system runnable on all the current browsers without the need to download any plug-ins. The visual mapping algorithms we implemented include the PathFinder algothrim, Kohonen mapping algorithm, Hierarchical clustering, direct-force algorithm, etc. For each map generated, the user has a choice of selecting different styles of displays, including the bubbling display, hierarchical display, network display, map display, etc. All the displays are interactive where the user can “click, move, zoom-in/out, or select†to either see more details/more information or to refocus on a new area of the map to explore. Furthermore, the user’s exploration process is directly linked to an underlying search engine (currently the system is linked to PubMed; it can be configured to link to other search engines or databases) to directly show the documents or records that match to the current view space of the semantic maps.
Results: The results we will present at the conference include a fully functional visual mapping system that are built on top of UMLS, PUBMED and other biomedical databases. The system will assist user to explore the medical concepts visually, conduct visual search and exploration, and map concept in one vocabulary (i.e. MeSH) and view terms in another (i.e. SNOWMED CT, etc.). The system can also be used to compare differences of semantic maps generated based on co-occurrence counts and based on selected similarity measures.
Conclusions: The innovative visualization system to be presented will demonstrate how semantic relationships can be visualized. The system will appear to researchers and medical workers as well as medical companies.
Objectives: There are three objectives of this project. The first is to investigate ways to generate beautiful, meaningful, and useful visual concept maps for medical and biological domains. The second is to implement a practical system that can be applied to multiple biomedical vocabularies and ontologies to generate instant concept maps based on user’s information needs as well as information content and context extracted from databases of medical records and medical literature (PUBMED, MEDLINE plus or search engines, etc.). The third is to study how the automatically generated concept maps will help user access medical records and literature through visual exploration and how the visual exploration process improves learning of biomedical concepts.
Methods: Multiple ways of calculating semantic relationships of terms are used and compared in this project. The base line is the co-occurrence of terms, which is what we used in the previous prototype. Using term co-occurrence counts to approximate semantic relationships of the terms has been studied and reported widely in the literature. The co-occurrence provides a good contextual and semantic relationship when the terms are used frequently and appeared in many documents. However, it also lacks of specificity and accuracy when the terms are ambiguous or with multiple meanings. This sometimes makes it difficult to understand the semantic maps created solely based on the co-occurrence counts.
With recent advancement in similarity measurement for medical concepts, many alternative measures might be used to replace or enrich the co-occurrence count for the purpose of identifying semantic relationships of vocabulary terms. We have tested several measures implemented in the open source software package UMLS::Similarity and used the resulting similarity measures to generate semantic maps in our system. Some interesting similarities and differences are found when comparing the maps generated by different similarity measures and by the co-occurrence counts.
On the visualization front, we have moved our flash-based visualization to JavaScript-based visualization using the open source package D3.js (http://mbostock.github.com/d3/) and some other JavaScript packages. This makes our system runnable on all the current browsers without the need to download any plug-ins. The visual mapping algorithms we implemented include the PathFinder algothrim, Kohonen mapping algorithm, Hierarchical clustering, direct-force algorithm, etc. For each map generated, the user has a choice of selecting different styles of displays, including the bubbling display, hierarchical display, network display, map display, etc. All the displays are interactive where the user can “click, move, zoom-in/out, or select†to either see more details/more information or to refocus on a new area of the map to explore. Furthermore, the user’s exploration process is directly linked to an underlying search engine (currently the system is linked to PubMed; it can be configured to link to other search engines or databases) to directly show the documents or records that match to the current view space of the semantic maps.
Results: The results we will present at the conference include a fully functional visual mapping system that are built on top of UMLS, PUBMED and other biomedical databases. The system will assist user to explore the medical concepts visually, conduct visual search and exploration, and map concept in one vocabulary (i.e. MeSH) and view terms in another (i.e. SNOWMED CT, etc.). The system can also be used to compare differences of semantic maps generated based on co-occurrence counts and based on selected similarity measures.
Conclusions: The innovative visualization system to be presented will demonstrate how semantic relationships can be visualized. The system will appear to researchers and medical workers as well as medical companies.
Medicine 2.0® is happy to support and promote other conferences and workshops in this area. Contact us to produce, disseminate and promote your conference or workshop under this label and in this event series. In addition, we are always looking for hosts of future World Congresses. Medicine 2.0® is a registered trademark of JMIR Publications Inc., the leading academic ehealth publisher.
This work is licensed under a Creative Commons Attribution 3.0 License.