DTU search engine outmatches Google
Rare diseases/illnesses are difficult to diagnose. Now physicians can get help from a customized search engine, doubling the performance of the best on the market.
The symptoms of a patient points in all different directions – what is the doctor to do? Ten years ago he would grab the medical handbook – today he would more likely try searching on Google or PubMed (a database with scientific papers).
Sometimes it pays off, but it is often difficult to find information on very rare diseases. It exists/can be found on the internet, but the search engines we normally use spot the most popular sites. This is a challenge to physicians, when about it comes to a rare diagnose.
Ole Winther learned this, and as an expert in statistics and development of intelligent algorithms, he started thinking about how to develop a user-friendly tool. Designed for helping physicians find the needle in the haystack.
It took Ole Winther and his students two years to develop the search engine, FindZebra. Today, FindZebra has been freely available for more than two years, and has supplied millions of users all over the world with possible diagnoses.
The machine must think like a human
“Basically, it’s all about finding a humanly created search line and translating it to something the machine can understand,” Dan Sventrup, a PhD student on the project at DTU Compute, explains. He studies the construction of intelligent algorithms, enabling better searches on the internet.
This is exactly what Google tries to do, but our most popular search engine works on other conditions, not meeting the physicians’ needs. Google analyzes the search words and finds out, in which articles on-line they appear. The search words have to occur frequently on the page, or many sites must link to the page, in order for it to be ranked highly in a Google search.
“The search engine scans the words of the search line. But when you search for rare diagnoses, they seldom appear in the most popular pages. And the pages we are interested in are obviously not the most frequently seen or ones with many links. This is why even a highly intelligent search line on Google can completely overlook pages describing rare diagnoses,” Dan Svenstrup explains.
If you for instance search for pain in stomach, the machine has to know that it is stomachache as a symptom. And it must come up with the same results as when you search for abdominal pain.
“FindZebra is a search engine capable of translating information from relevant databases, supplied by experts. We have used a number of statistic methods to find a model suitable for the physicians’ needs,” Dan Svenstrup explains.
Because the search feature is not meant for a broader audience but exclusively for diagnoses, it has been possible to tailor it precisely. And this has given the developers high degrees of freedom to implement quite complicated search systems.
In a quarter of a second FindZebra searches 13 selected databases containing 35.000 articles in total. The choice of databases has of course been made in close cooperation with physicians.
Massive means to save
Some patients can use the search engine themselves, as Robert from England did. He later wrote to Ole Winther:
I've seen endless consultants who have all been at a complete loss as to what I have. I've been diagnosed with Bronchiectasis, Youngs syndrome, and cystic fibrosis, but were all proved negative. Then I saw your site in the Guardian and tried it and after tweaking the symptoms it came up with mastocytosis, which has since been confirmed. I am now on the right treatment and feeling a thousand times better.
Thank you, without your site I would still be lost.
“It’s cases like this one that motivate us to develop and spread the knowledge of FindZebra. Had the physicians from the beginning had access to proper search tools, it would have made a tremendous difference,” Dan Svenstrup tells.
It is easy to understand that a correct diagnosis enhances the quality of life for the patient in question, and it saves the heath care system massive expenses.
And we are talking a lot of cases. It is assumed that, at this moment, 10 million Europeans are affected by diseases, which their physicians are struggling to find a diagnosis for. 25 % of these wait for more than five years before their diagnose is determined. For some, it never happens.
To ensure optimal performance of the search engine, the scientists work closely together with Rigshospitalet and Harvard Medical School, among others.
A leap into business
At the beginning, when Ole Winther and his colleagues at the DTU started to really develop the search engine, the objective was entirely academic.
First, two PhD students developed FindZebra, version 1, which achieved huge media coverage. For Ole Winther, who is dedicated to bringing research out of the laboratories for public use, this was his encouragement to continue working with it.
“The project was captured by commercial interest. Ole found it appealing, and began seeing possibilities for developing the idea,” Dan Svenstrup explains.
He had taken his Master degree at DTU Compute, followed by employment at Danske Bank – but he missed the academic world. He then called his former supervisor, Ole Winther.
“I was allowed to return and continue working with the search engine, which had to be made better and faster. You don’t want to wait five seconds for a result – now it takes less than ¼,” Dan Svenstrup explains.
While he slaved with the algorithms, Ole Winther tried to settle the legal matters. This took him ages.
Rescued by Copenhagen Spin-outs
Ole Winther had neither the experience nor the energy to pursue potential business contacts. He had extensive dialogues with them, and on his desk there was even a contract, for which he couldn’t assess the consequences of signing. Therefore he contacted Copenhagen Spin-outs (CSO), who became a crucial sounding board in the time to come.
At the same time he was contacted by Mads Emil Matthiesen, a young entrepreneur. Mads Emil Matthiesen had been following the project through network and media, and he was looking for new challenges.
“I found FindZebra interesting, and felt like spending/wanted to spend some time on the project,” Mads Emil Matthiesen says. He had the exact qualifications, which Ole Winther lacked as a scientist: experience with business development.
Mads Emil Matthiesen started making a business plan in cooperation with CSO.
“The medical industry is very interested in FindZebra, but we try to keep them at an arm’s length, for the time being. Everything is going well, and right now we are taking the first steps towards establishing a physicians’ advisory board, to make sure that the project moves in the right direction,” says Mads Emil Matthiesen.
Our growth has to make a difference
“The search engine works, so far so good. But commercially there is evident unresolved potential – we would, of course, like much more traffic,” Mads Emil Matthiesen says.
Despite the fact that FindZebra is twice as good as Google at suggesting rare diagnoses, the physicians can be difficult to convince. They are not the easiest group of users to capture.
”We can make FindZebra more intuitive with relatively simple changes. Looking physicians over the shoulders can give us input on important changes in design, which we would never have thought of ourselves. We need to understand our users even better,” Mads Emil Matthiesen says, and points out fundraising as the natural next step of the process.
”With sponsorships, a bright development team can implement improvements to generate more traffic, and bring FindZebra a giant step forward within 12 months, opening the doors for additional funding,” Mads Emil Matthiesen promises.
Initially the object is to improve the life of people with rare diseases. Big business comes later …
Do you have a rare disease? Find out on www.findzebra.com
When you hear a gallop, you would think about a horse, not a zebra. Sometimes, however, physicians are confronted with rare diseases, the zebras. This is how the search engine received its name
- Harvard Medical School in USA
Founders of the company
Financial support: Copenhagen Spin-out
Blue book, Dan Svenstrup, 37 years old
2014-17: PhD student at DTU Compute
2013-14: Research assistant at DTU Compute
2013-14: Different projects in statistics and artificial intelligence
2009-13: Senior analyst at Danske Bank Markets
2009: Master in applied Mathematics
Dan Svenstrup spends his days at the DTU, where he works with mathematical language models. When he is not buried in algorithms, he puts on his hiking or jogging clothes and is very sporty. In the last couple of years he has been very interested in the Danish start-up environment, where he has been involved in several different projects. In five years he will still be busy with advanced mathematical models, either in his own company or in a newly established company – he hopes.
Blue book, Mads Emil Matthiesen, 28 years old
2010-11: R&D-student worker, Coloplast
2011-12: Research assistant, Harvard Medical School
2012: Master of Engineering, Medicine & Technology, DTU
2013-14: Management consultant, The Boston Consulting Group
2009-: Founder, CathVision ApS
Mads Emil Matthiesen has been hooked on entrepreneurship since he participated in the Stanford Entrepreneurship Week in 2009. He lives in a cohousing for entrepreneurs, Nest, and is the daily manager of CathVision, another start-up. He used to run at elite level, but his knee couldn’t handle the strain, so today he spends all of his energy on business development.