For advancement in the medical field and making good decisions in terms of diagnosis, treatment and follow-up it is necessary that biomedical images be readily available and the content of those images be very easily detectable and searchable.
This will also help in data management and, as a secondary use, for biomedical research and assessment of care delivery.
During the last twenty years, an astounding repository of image data has been accumulated as a result of scientific experiments. This image data is used in advancing research and making critical decisions when providing medical care. This is why it is very necessary that efficient practices are designed to access biomedical images and publications. This will allow timely flow of knowledge from researches to peer investigators and other healthcare practitioners.
So where is the problem?
Traditional search engines that we use e.g. Google, Yahoo etc. don’t really see pictures/images, they only read the data that the image comes with. Also these search engines mainly encourage the syntactic keyword based search mechanism, which has a big limitation for the biomedical community.
As we all know, scientific data is maintained so that we can further investigate and find new and significant patterns from the datasets. This helps advance a scientific phenomenon in a way that had not been understood completely. The issue here is that at present, biomedical research data is mainly found in legacy formats (databases, XML etc.). This hinders data integration and also scientific discovery.
Enter the UNBSJ’s computer science department & Mr, Ahmad Bukhari
Now, University of New Brunswick in Saint John (UNBSJ) has developed their own imaging tagging systems by recognizing features within images. The interesting part here is that the project was successful because of contributions of a Pakistani PHD student, Mr. Ahmad Bukhari, at UNBSJ.
Mr. Ahmad is a Pakistani Computer Scientist who currently is a postdoc research associate at Yale University, USA. Mr. Ahmad already has a PhD in Computer Science from the University of New Brunswick, Canada and a Masters in Engineering from Gyeongsang National University, Korea. In addition to his distinguished education, Mr. Ahmad has vast experience in artificial intelligence, semantic web, machine learning, and information engineering.
To elaborate, the search issue came to light when Harvard and Yale had problems with searching images within their repository, which was escalated to UNBSJ.
Yale had an image finder portal that contained around two million images including medical illustrations such as x-rays, charts and graphs from scientific journals. They were finding it difficult to maintain and scale the portal.
In an interview, Mr. Bukhari explained how traditional archiving methods are not fully automated and require optimization. “This is a Big Data problem. Scientists want to be able to quickly find and reuse images so they can automatically integrate them with other scientific data. What we came up with was a new data management framework based on semantic technologies,” explains Bukhari.
In the same interview, Mr. Bukhari and project chair Mr. Baker explained that in their approach they took each image and added additional annotations based on features within the image. They identified features within the images, extracted these and then found out more about the images from other data sources online.
This information was then annotated to the images as tags. It is a large scale data integration exercise. They did face various problems including parsing through images that were bad in quality, however they found a way to enhance the quality of such images before tagging them.
Once the tags were added the images could now be searched through more precisely. In addition to this, they also developed an algorithm whereby related images can also be identified.
“We took this immense amount of image data and republished it with medical metadata. You can find the relevant image and documentation faster.” they mentioned. “The ultimate goal is what we call image first – Once you have found the right image you are looking for you can link to other images and the publications where the images were published. It’s here that the scientists will find what they are looking for. We help them get there faster. That’s the value we bring.”
So to sum it up, this new approach will permit users to identify a medical image they are interested in, find images that are related to the first image and finally locate the scientific documents from where the images originated. This approach will significantly cut down a scientist’s time finding and correlating relevant scientific information.
Ahmad Bukhari image credit: unb.ca