Israeli startup Zebra Medical Vision introduces a new algorithm for better diagnosing breast cancer


Israeli startup Zebra Medical Vision announced a new machine learning algorithm Wednesday that detects breast cancer earlier and more effectively. It’s the third major diagnostics algorithm release from the company in as many months.

“All early stage cancer is difficult to find, however breast tissue is inherently more difficult to diagnose because of its properties and diversity,” Zebra CEO Elad Benjamin told Geektime. He emphasized that they were trying to focus on algorithms for diseases that posed the highest danger to the public, meaning those with the highest mortality rates. Cancer, and with a special interest in breast cancer, falls under that rubric.

The announcement follows the introduction of two other algorithms in mid-August — one for detecting clogged arteries and another for fatty liver in CT scans. Their current roster of algorithms covers pulmonary (lungs and the right side of the heart) and bone health as well. Alongside the August announcement, Zebra told Geektime to expect new algorithms covering brain, abdomen and chest data. The company has said it will expand to MRIs and X-rays in the future.  

Zebra raised a $12 million funding round earlier this year led by Intermountain Healthcare, bringing its total funding to date to $20 million. That money seems to have been a boon to the company as it has rapidly made progress on new algorithms, which have to be fine-tuned for different diseases.

“Some [tissues] are dense, some with more fat content, some have calcium deposits which look like cancer but aren’t, etc.,” Benjamin continued. “In addition, the mammogram itself is a very large image — sometimes 5,000 by 5,000 pixels. And the cancerous region can be a few pixels squared — so there’s also a visual challenge of trying to find a ‘needle in a haystack.'”

Zebra CEO Elad Benjamin (Photo credit: Zebra via Twitter)
Zebra CEO Elad Benjamin (Photo credit: Zebra via Twitter)

Zebra Chief Medical Officer Dr. Eldad Elnekave announced the results of tests the company conducted with the new algorithm at the SIIM Conference on Machine Intelligence in Medical Imaging (CMIMI) last month, where he argued that the method was vastly better than other computer-aided detection (CAD) methods. Benjamin told Geektime he thought that given that conference’s very specific focus, there was no more ideal time to present an announcement that inevitably would be of massive public interest.

“CMIMI was a conference dedicated specifically to the topic of machine learning in medical imaging, and it was attended by the top academic institutions, solution providers and radiologists that are leading the discussion of how machine and deep learning will integrate into the business of radiology. We felt this was an ideal forum to announce our algorithm, which uses advanced and innovative machine learning techniques to reach superior results.”

The American Cancer Society estimates there will be 246,600 new cases of breast cancer in the US over the full calendar year 2016, resulting in 40,450 deaths. The high public profile for breast cancer has garnered it a lot of attention on better diagnosing the disease. CAD has become more prevalent in breast cancer screenings over the last several years, though some studies have concluded there are no clear advantages to using CAD to optimize (or hasten) diagnosis of the disease, or as the abstract of that.

“CAD is a very generic name for ‘computer-aided diagnosis,’ so our algorithms fall within that category as well. However, the clear distinction is the technology used and the results achieved. Traditional “Mammo CAD’ algorithms used older computer vision techniques and had relatively poor results (or results that didn’t improve upon radiologists reading the mammogram as is),” Benjamin asserted to Geektime

As a study in the European Journal of Cancer put it, “The evidence that double reading with arbitration enhances screening is stronger than that for single reading with CAD.” That means demand for a more effective CAD tool is still high, and Zebra thinks it has it. “Our ability to integrate thousands of exams into our learning process, combined with deep learning techniques, have (sic) yielded much higher quality results,” Benjamin explains.

BCC Research estimates the global breast cancer diagnostics and technology market to have reached a value of $20.8 billion already back in 2012. Since then, there have been increased efforts to bring startups and entrepreneurs into the breast cancer space, such as with the 2014 Breast Cancer Startup Challenge. Plenty of startups focus on particular cancers — Somatix focuses on lung cancer, SkinVision on skin cancer, and several startups have joined the battle against cervical cancer.

Without saying for sure how much earlier Zebra’s solution could beat the competition, Benjamin claimed to Geektime that, “Our clinical validation has shown that we detect malignant tissue 10%-15% better than human readers, or other algorithms, and we do so with over 50% less false positives. It is hard to estimate how much earlier we detect, but the results mentioned above lead to overall earlier detection and diagnosis.”

His point regarding false positives is a crucial one as it can cause considerable and needless distress to patients. Developing solutions that cut down on both false positives, and of course false negatives, remains a key goal for researchers.

Based in Kibbutz Shefayim, the company was founded in 2014 by CEO Elad Benjamin, Eyal Gura and Eyal Toledano. Besides Intermountain Healthcare, they count Khosla Ventures, OurCrowd, Marc Benioff, Deep Fork Capital, and Dolby Family Ventures among their many backers.


Please enter your comment!
Please enter your name here