Project KHRESMOI: Challenges in ehealth search
Published Sep 13, 2010 by Katja Mayer, Henning Müller
Started on 1 September 2010, the project Khresmoi aims to develop a multi-lingual multi-modal search and access system for biomedical information and documents. IRF.net met with Henning Müller, the project coordinator (University of Applied Sciences, Western Switzerland) to talk about objectives and challenges of the project.
IRF: What is the main goal of the recently started project?
Henning Müller: The main goal of the project „Knowledge Helper for Medical Information Users“ (Khresmoi) is to analyze medical information that includes a large variety of different tools and techniques that need to be integrated. The integration of different information sources and different technologies to analyze information, including pure text retrieval, semantic and image analysis, image and multi-lingual retrieval, and even machine translation. So if you query in one language you might get documents from another language and can then translate them. This could also be coupled with question-answering techniques, with the final goal to help medical professionals, from general practitioners to radiologists, who need help in analyzing data and who might find this help in specific documents, or in fact from ontologies, which they have rarely access to at the moment.
IRF: What are the specific challenges of integrating so many different technologies?
HM: Most of the technologies to be integrated have already been used and evaluated, but in very closed and specific environments. In such specific areas they have proven to work well, except maybe the 3D and 4D data analysis, which has not been done before.
Semantic technologies, machine translation, information retrieval, those are established research fields that have been used for a long time, but mostly on closed corpora and very specific problems. What we target within Khresmoi is much larger. We have specialists in the consortium that can integrate all these technologies, and this is the challenge, because the experts now have to work together to combine these tools like text retrieval with semantic analysis, or semantic analysis with images, and they have to identify the complementary parts of these areas for a successful combination. So it is truly the integration and combination of tools that is the biggest challenge in my opinion.
There are so many different ways of treating information, of deploying standards, of evaluating a tool and most importantly not to forget about the integration of users in this loop. The Khresmoi evaluation will be user-centered, so we have a very strong focus on the use cases. We would like to hear from the users’ perspective what is important and needed. Interface design and human computer interaction are therefore also important cornerstones of the project.
IRF: Which user groups will be involved in the evaluation?
HM: The target user groups are defined in our use cases. To start with the most specialized one, we will work together with radiologists from the Geneva University Hospital and the Vienna Medical University. They will mainly evaluate the tools for image analysis and retrieval. These professionals know how to interpret their data, but they are so overwhelmed with data to analyze, that it can be rather stressful. So in bringing specific information at their fingertips we can help them evaluate their images. Another target group are general practitioners. They are represented by the Austrian Society of Physicians in Vienna. Their over 2.700 members will get access to Khresmoi technology via the portal of the society. This user group is not so specialized, but nevertheless has a demand for quick access to relevant information. The third and last user group in the project is represented by the Health On The Net Foundation. It is lay people searching for health related information in the internet. An increasingly large group of internet users is using search engines to find such information.
Currently, there are very few specialized search engines, and most of the common search engines do not rank their results according to the reliability of the information, they do not tell the searchers whether this is a good, trustworthy source, or not. Industries putting a lot of money into optimizing their search engine ranking are not necessarily the ones providing the most reliable information. The Health On The Net Foundation gives out a quality label, making it possible to search for information only on such labeled web pages, where certain quality standards are given.
Another goal of the Khresmoi project is to supply information based on the knowledge of the information seeker. Whereas the radiologist will require very specialized information, eg. not a very broad introduction to what a radiograph is, a general practitioner might require a more basic information, but on the other hand is used to medical terminology. In contrast my grandmother would need content very easy to understand. HON has already started to look for tools of estimating the level of knowledge that is required to understand information displayed on a web page to be able to distinguish information for medical professionals or for lay persons. The next step is then to display only this information that is fitting into the knowledge profile of the user.
IRF: Would you like to outline the technical challenges to the IRF community?
HM: Yes, certainly. We are talking about terabytes of data, billions of facts that have been made available by Ontotext, so in this respect we need to have very large storage, powerful computing capabilities. Atos, being not only software, but also hardware solution specialist, will also be of big help with such issues. The project will include cloud technology that we need to adapt our tools and algorithms to. But our tools also have to be optimized for the medical domain. Not all partners have a long experience in the medical field, which poses specific challenges.
We also hope to cooperate with the WHO on a regular basis. There is a lot of interest in multi-lingual retrieval, because all their documents need to be translated into 6 languages. Working with their translation memory / corpora to train our models could improve the translation aspect of the project.
There are many challenges that also might be interesting to members of the IRF community, such as information extraction, linking of extracted information to ontologies, machine translation, multi-lingual information retrieval, domain-specific retrieval and search interfaces. One of the aims of Khresmoi is to make available document corpora for research use, so the IRF community will be able to get access to the data required to work on these challenges in the biomedical domain.
IRF: What do we have to do to judge the quality of a web page?
HM: Health related information on the web is not a field where standards can be employed easily. Web content somehow escapes all attempts to be standardized.
Health on the Net (HON) bases their quality label on the metadata of a website: when was the information supplied, who wrote it, and also who sponsors the web page. This does not prevent all poor content but it gives indications to judge it. We do not want to censor the web but build minimum standards on this. There are several papers on trying to define what quality data is remaining open for different opinions at the same time.
IRF: Getting access to medical information is a highly politicized matter.
HM: Yes, indeed. We are constantly approached by people who are afraid of the misuse of their private records, even though we do only use fully anonymized data and data that is openly available on the Internet, in databases and scientific journals, for example. Khresmoi is not interested in the linking of personal data records. The idea is to improve access to external and reliable medical information for doctors, so that they have more time to talk and to treat patients. Furthermore, patients and information seekers should get better tools to identify the reliability or trustworthiness of a source. This is what we are working on. The Health on the Net search engine will be improved by Khresmoi technology.
Project Website: khresmoi.eu
For further reading: