Document Actions
Knowledge Media in Health Informatics: From Semantics to Social Software
Prof. Marc Eisenstadt, Chief Scientist
The Knowledge Media Institute
The Open University,
[PowerPoint slide notes]
1 Hello everyone, it gives me great pleasure to be here today, representing the Open University – in many ways the mother of distance learning, virtual education and elearning since 1969.
2 We have always used a mixture of real and virtual technologies to deliver education to over 3,000,000 students in 36 years, and there are five reasons behind our success:
3. Peer-reviewed and beautfully-produced material
· A strictly-controlled examination system that is highly-respected throughout
· over 8,000 part-time tutors all over
· Student life is varied and challenging, both online and offline.
· and research that is the life blood of any academic institution ...
· the 5th item, research, is what I am going to talk about today.
4. Just over twelve years ago we dreamed of bringing together the best minds in knowledge technologies and media technologies, and have now grown to over 75 members of staff working on 40 projects, with 60% of our funding obtained from external research grants, typically from the EU and UK science funding councils.
5. This is what we mean by knowledge media.
Knowledge Media is about the processes of generating, understanding and sharing knowledge using several different media, as well as understanding how the use of different media shape these processes.”
6. Among our many active areas of research, three in particular have something interesting to contribute to Health Informatics: semantic web services, social software, and multimedia information retrieval.
7. First, Semantic Web services... let me break this down into the semantic side and the services side, as follows...
8. The web today, as wonderful as it is, still has no way to understand what I really mean or need in a specific context – for example if I search for star wars the results will include the movie and also the Reagan controversial defense policy.
OK, you know this already, but let’s see what we can do about it!
9. The semantic web takes todays massive interconnected web that is mainly only readable by people...
[slide build]
... and adds a layer of hidden annotation to make the web readable by machines. This is very important, and in my opinion you should forget about other definitions of the Semantic Web... think of a web readable by computers and you’ll have a clearer picture of what is happening in today’s research.
10. To achieve pages readable by machine is not easy. Assuming you do not understand Chinese... THIS is what a typical web page looks like to a computer... very clear layout that the computer is of course good at, but meaningless content
11. With XML we can nicely provide the beginnings of meaning by structuring content so that it can be rendered in different ways on different devices... for example this is Curriculum Vitae or job resume with sections such as name, education, work history..
BUT WAIT A MINUTE, I am cheating... how does a machine know the meaning of ‘name’, ‘education’ and all these labels?
*IT DOES NOT*... these are equally mysterious to machine... meaningless junk
THAT is the real purpose of all the layers that you hear about, RDF, RDF Schemas, OWL ...
12. To provide clarity and meaning, what is called an ontology that is a formal, explicit specification of a shared conceptualization...
By *EXPLICT* we mean unambiguous definitions of concepts, attributes and relationships;
BY *FORMAL* we mean it is machine-readable
BY CONCEPTUALIZATION we mean model of some domain of study
and by SHARED we mean a commonly accepted understanding in some community
13. The formalization of knowledge that I just descirbed is only one part of the picture – and I’ll now link that up with the web services part.
14. A web service today is characterised by five things:
[as shown – these build up one at a time]
• A program accessible via other programs using standard internet protocols
• Loosely coupled, reusable components
• Encapsulate discrete functionality
• Distributed
• Add new level of functionality on top of the current web [e.g. Amazon, Google maps, etc.]
15. provides great functionality, but there are problems...
- The descriptions are still syntactic
- only people can do the jobs ofdiscovering new services, composing them together and running them
- which means that are limits to how far web services can spread
16. The Vision of Semantic Web Services is to take ...
[ANIMATED BUILD 1 AT A TIME]
the old web that is static and syntactic that we know very well,
to add the dynamic power of web services that is providing so many exciting features today;
to add the formal machine-readable descriptions of domains that I mentioned a moment ago, and
bring these together by providing formal descriptions of web services so that these can be discovered and composed together automatically by machine....
..in effect allowing us to automate all aspects of application developement through reuse of high-level componenets.
17. In the domain of breast cancer, we are looking at ways to provide automated decision support tools for the three key people involved in case histories: the radiologist who interprets the mammogram, the pathologist who interprets the biopsy, and the clinician who knows the patient history.
18. The goal of the software demonstration we built is as you see here: to show how “Heteregenous complex information can be integrated, annotated, and services deployed, in collaborative decision support contexts.”
19. We begin with a domain ontology as I described earlier, a formal specification of every aspect of the domain,
20. provided both textually and visually
21. ...allowing the decision makers even from their separate locations to bring together all the relevant data that is actually stored on separate servers... for example...
22. one specific patient visit from the past...
23. has several different results stored
24. including an already annotated so-called...
25. Region of Interest, that we can investigate in more detail... and let’s suppose that image in stored on server 1 in
26. and even better we can send these formal descriptions, to a natural language generation program sitting on a server 2 at the University of Sheffield 200 kilometres away... that generates the entire natural language description of the patient..
27. working only with the shared formal description as its input, and accessible to the diagnostician anywhere in the world
28. I must emphasize that NONE of the images I am showing you comes from a normal web site – everything you see is constructed dymically by magic from the semantic descriptions, depending on the context.
29. That includes other services such as finding patients with the same age... normally not a big deal, but in this case since the system has enough semantic descriptions to compose such a service automatically, if offers us the right menu at run-time, using a service composition program that comes from server 3 at another remote location, in fact our server at the Open University, another few hundred kilometres away.
30. The same is true for patient histories
31. and even the image data that is stored on server 4 in
32. and delivered using “tiling” to provide the highest resolution...
33. Annotations can of course be added, but the special trick here is that...
34. the region description is then sent off to feature modules on server 5 at
35. The same patient has a series of MRI scans in the history...
36. and the problem is to align the images properly, in case the patient moved for example...
37-39. No surprise... server 6 at another location does the image alignment, which is very CPU intensive
40-42. and sends back the aligned images, that can also be annotated
43. and sent off to the separate feature detection service as with the mammograms
44. Because of the semantic web service architecture, even rival analysis algorithms, as long as they fit the high level service description...
45. ... , can be automatically discovered and plugged in... delivering their own interpretations for the diagnostic team to discuss.
46. The distributed nature of the architecture means that it can be integrated with future hospital systems, but this is a research prototype, and I don’t pretend integration is easy. The key is to have rich high level service descriptions: painful to provide once, but after that it ensure scaleability and software reuse.
47. Our goal was as stated before, and it shows that high-level semantic descriptions can support automatic service construction. We can achieve both large scale and software reuse if the right high level descriptions are deployed.
48. A second more ambitious project is now underway using the same approach: LHP aims to create a model of the human musculo-skeletal apparatus which can predict how mechanical forces are exchanged internally and externally at any dimensional scale from the whole body down to the protein level.
49. This project has only just started – but the principle is exactly the same as the breast cancer example I just showed... integrated diverse data sources from the many partners using the Open University’s semantic web services tools
50. Now I want to describe what we’re doing with Social Software that you might also find interesting
51. Here is our global village ...
52. Wouldn’t it be nice if you could find someone out there who could help you solve an urgent problem right now?
53. To ask who is facing a similar problem, or where I can find complementary approaches, or has anyone found contrary evidence... and just light them up on a map and speak with them... we have the beginnings of such a tool already in place...
54. One part of our solution is called MSG, a free instant messenger that is much simpler than all the others, yet interoperates with all of them. it runs directly in any web browser, and allows a site administrator to set up group-wide contact lists.
55. Group maps can be added, including geographical layouts directly from Google Maps, and live presence information, in this case green dots, shows who is available.
56. We have an award-winning text mining tool that searches vast archives of text and returns names rather than documents... and we apply what we call ‘social triangulation’ so that even if your own pages do not have such good information, we can still find it using the pages of your buddies!
57. This enables us to highight the right person with the right knowledge in the right place at the right time: the holy grail of knowledge management!
58. On another aspect of social software, some of you may remember our videoconference tool called FlashMeeting...
it is extremely simple and extremely scalable: you just go to a web page, and it works!
59. Including various voting and whiteboard tools... but more importantly for us...
60. every meeting can be replayed in a timeline view to access any point in any meeting...
61. and we provide extensive analysis tools not only show who performed different actions such as voting and chatting, and where they were located, but also
62. a new kind of analysis tool that depicts the amount of time spent (the angle of each slice) and the number of turns taken (the radius of each slice) for both video and text chat, for every meeting.
63. We are now analysing thousands of meetings to see their characteristics in different communities of practice.
64. One final slide comes from a new area of research for us, multimedia information systems
65. We have a very high-performance image search algorithm that has been applied to medical imaging
66. And of great excitement to us right now is a new tool we call iBase that lets you search through millions of images that have no annotation at all. You can take any image as an example, drag it into the tray at the bottom, and say ‘get me more like this’... You can specify different filters such as colour or frequency distribution, or even abstract attributes such as ‘smiling face’, and it finds a family of similar-looking images.
67. Here are some relevant URLs, but please feel free to email me using the address at the bottom if you want to pursue anything in more detail, and I’ll put you in touch with the relevant people. Many thanks for your attention!
On Line Closing Notes for the Virtual Event