U.S. venture firm Khosla Ventures is leading an early round of investment of $8 million in Zebra Medical Vision Ltd., an Israel-based startup building a database of anonymous medical images.
The images — X-rays, CT scans and MRI scans — are of real patients but have been anonymized by leaving out names, addresses and other identifying details.
By using a large dataset of such images researchers hope to be able to create computer algorithms that can identify diseases more accurately, and in some cases earlier than is currently possible.
Computerized diagnosis that is based on visual analysis of images is currently used in some medical tests and procedures, but it is far from common practice.
Headquartered in Kibutz Shfayim north of Tel-Aviv, Zebra was founded in 2014 co-founders Eyal Toledano, Eyal Gura and Elad Benjamin. The three are tech entrepreneurs with backgrounds in image processing and medical equipment.
By partnering with large health organizations from Israel and internationally, Zebra said it was able to gather one of the largest existing databases of medical imaging files.
Zebra does not pay the health organizations for the images, as the latter hope to benefit from the data analysis provided by Zebra. The company does not require a certificate from health authorities because the data is anonymized.
“Zebra obtains the de-identified data via its HMO partnerships in accordance with their data privacy compliance policy,” the company said.
Initial digestion of the files coming from the health organization at Zebra is done using discs, with new information being transmitted over the web.
Alongside millions of files of medical scans with corresponding anonymized medical data about the patient, Zebra is also offering research teams image processing capacity on both the company’s servers and on secured private clouds, as well as research tools like software libraries researchers can access.
The database is already used by various research teams in Universities in Israel, Canada and the U.S., Zebra said.
One of the problems that have held back advancements enabling computers to analyze medical imaging files was the lack of available, large, and well-tagged data-sets of relevant images. “Today computer vision research teams are more likely to try and teach a computer to identify cats than to identify cancer,” said Gura.
“A patient can go through Mammography many years until something is found. By looking at hundreds and thousands of historical images of such patients much earlier anomalies can be found,” said Euan Thomson, an operating partner at Khosla Ventures, who will be joining Zebra’s board.
Zebra plans to use the funds to double its team, establish headquarters in the U.S. and integrate with the systems of more health organizations.