The story behind Zebra’s recent Mammography Publication

phil-e1469945648791Just over three years ago, Zebra medical vision launched into the world with two fundamental convictions and one mission. We knew that the potential for every novel technology is defined by its most profound application- which can be discovered only by people with ample doses of creativity, persistence and good first aid skills. 

We felt the same about human potential:  defined by a person’s decisions rather than her ability- a view captured by Shimon Peres in a quote on our wall: “you are as great as the cause you serve.” 

Our single mission was to realize the full potential in medical imaging by harnessing advances in machine-learning – image analysis technology and enabling the best minds anywhere to tackle the challenge with us. 

Our publication in this week’s Journal of Digital Imaging, “Malignancy detection on mammography using dual deep convolutional neural  networks and genetically discovered false color input enhancement” (explained below) is a strong validation of our vision. 

Here’s the story behind the paper: two years ago, Phil Tear was working as a highly sought after software architect in Berkshire, UK. He heard about a crazy Israeli company called Zebra, which was bent on applying GPU’s (Graphic Processing Units) to improve medical imaging interpretation. GPUs  contained the technology which was powering the boundless growth of the gaming industry.  Zebra’s message was: “we have millions of medical imaging studies to train on, and we’ll give you all the GPUs you can use if you choose to take this journey with us.” 

The message resonated with Phil. Although still under 40, he had lost his wife to ovarian cancer three years earlier. Within six months, Phil became Zebra’s first full time scientist based outside of Israel. Within a year he made his first breakthrough in cancer detection and a year later here we are, publishing the vs 1.0 results of his mammography algorithm, which automatically and within seconds  detects breast cancer with an accuracy (sensitivity of 91% and specificity of 80%) similar to that of expert radiologists. 

For the small minority of you who will not be reading the entire paper (contact me for a copy), here’s an oversimplified version of how he did it. Phil understood the technology (Neural Networks) which has made it possible in the last few years to automatically identify almost any real life object (cars, street signs, individual faces, rare cat breeds…)  This technology is embedded everywhere from satellites to smart phones, but it’s nowhere to be found in medical imaging.  He set out to apply neural network training (also known as deep learning) to mammograms:  high resolution images of the breast obtained for breast cancer screening.

Phil had thousands of examples to learn from. He experimented to find the best processing filters which naturally brought out the difference between normal breast tissue and abnormal cancer. He then took an off-the-shelf neural network which had been pre-trained on millions of natural world images and was capable of identifying thousands of different objects. This network seemed to struggle to find meaning in the black-and-white shadows created by x-rays through the breast, in the same way a scholar of English literature might struggle with a Japanese restaurant menu. Phil sought a way to translate, ultimately converting black and white density to a spectrum of color on the RGB (Red Green Blue) scale.  

Along the way, Phil was supported by an all-Star team of radiologists, including Dr. Oshra Benzeqen of Rabin Medical Center (Tel Aviv University) and Dr. Michael Fishman of Beth-Israel Deaconess (Harvard University). 

The result is a remarkable coalescence of immense knowledge – clinical, radiologic and algorithmic – condensed into a code that can run on any digital mammogram anywhere. A radiologist can use it as a “second reader,” and the benefit may be even greater in the vast majority of the world where the radiology shortage limits how many mammograms can be interpreted at all. Congratulations to Phil and the entire Zebra team! 

-Eldad Elnekave, MD

Zebra receives CE Approval for its Analytics Engine

I happy and proud to share our latest news with all of you – the regulatory approval of our Imaging Analytics Engine in Europe, Australia and New Zealand. Receiving CE approval marks another milestone for Zebra, as we continue to provide physicians and hospitals with a growing number of analytics tools to help them improve care. Continue reading “Zebra receives CE Approval for its Analytics Engine”

Zebra’s CMO interviews Oxford Rheumatologist Dr. Kassim Javaid

eldad-photo Dr. Elnekave: What were the defining experiences which compelled you to dedicate your professional life as a consultant Rheumatologist to the issue of bone health & osteoporosis?


 Dr. Javaid: Rheumatology covers a wide range of diseases across the lifecourse. My interest in bone health was sparked by working with an inspirational professor that then led to me to complete a PhD in the field. Additionally osteoporosis allows interplay between epidemiology, biological processes, therapies and heatlh services research as well as interactions with multple other medical and surgical specialties. 

Continue reading “Zebra’s CMO interviews Oxford Rheumatologist Dr. Kassim Javaid”

Zebras & Streetlights part 2

In our last blog we discussed the need to search beyond the proverbial streetlight in medical imaging and presented two of Zebra’s first automatic imaging algorithms: coronary calcium quantification and liver density measurement. In that blog post, we pointed out that CT-chest lung cancer screening gained approval by demonstrating a 20% reduction of lung cancer deaths in the National Lung Cancer Screening Trial (NLCST). We argued that the merits of CT chest screening for long—term smokers might be markedly underestimated: more people in the NLCST died of cardiovascular disease than lung cancer, and the most powerful predictors of cardiovascular disease and mortality are coronary calcium and visceral fat, both of which can be screened for using the same Chest CT (hence our rationale for developing these fully automatic algorithms). Now we look even farther beyond the streetlight and mortality statistics to examine the factors most predictive of how well people live.

Smokers are at 13x increased risk of developing chronic obstructive lung disease (COPD); fully half of smokers develop some degree of COPD. For a person with COPD, life is defined mainly by how well they can breathe – “just” breathe. For healthcare providers, COPD is a mighty foe – prevalent, powerful and unpredictable. COPD exacerbations account for more than 12% of all ER admissions and 10% of inpatient hospital beds but those numbers can change precipitously: seasonal and even sporadic fluctuations  of COPD  exacerbations can overwhelm hospital wards [1]. The hospitalization course of any individual with COPD is just as precarious: according to the Royal College of Physicians [2], nearly 1 in 12 hospital admissions for COPD exacerbation resulted in death during that admission and 1 in 4 were associated with death within that year. The same study found that fully 30% of patients admitted with acute COPD required repeated hospitalization within 3 months.

After cardiovascular disease (25%) and lung cancer (24%), the next distinct cause of death among the NLCST population was lung disease (11%). Could the same screening CT data be used to change not only who dies of lung disease but, perhaps more important, how those with chronic lung disease live? Indeed, distinct CT- identifiable patterns of lung architecture have been correlated to different risks of clinical deterioration and can even be used to predict which of various therapy options would be most effective [3], [4] [5] [6] [7].

Interestingly, a single measurement of the size of the pulmonary artery on CT is the most powerful predictor of hospitalization and mortality in COPD patients[8], [9]. In fact, finding a pulmonary artery diameter which is larger than that of the adjacent aorta is associated with a greater than 3x risk of requiring hospitalization for acute respiratory failure within twelve months[9]. Although this finding is rare among the general population, it is seen in 10% – 30% of individuals with longstanding lung disease. Combining these two assessments of lung texture and pulmonary artery diameter could identify those in greatest risk of disease progression in order to more effectively deliver preventative and maintenance care [10].

COPD is a mighty foe – its prevalence will only increase in the next decades while resources to treat it remain relatively static. But COPD is not one disease – not a single diagnosis that follows a single predictable clinical trajectory. Its precariousness may be its greatest threat – and hence our target: to define the key features of COPD most predictive of disease course so that we can prevent and treat those who need it first. Two more automatic insights which every Chest CT should include: emphysema quantification and pulmonary artery diameter. To be continued…



[1] J. Kidney, T. McManus, and P. V Coyle, “Exacerbations of chronic obstructive pulmonary disease.,” Thorax, vol. 57, no. 9, pp. 753–4, Sep. 2002.

[2] B. T. S. and B. L. F. Royal College of Physicians, “Report of the National Chronic Obstructive Pulmonary Disease Audit 2008: Clinical audit of COPD exacerbations admitted to acute NHS trusts across the UK.,” 2008.

[3] O. M. Mets, M. Schmidt, C. F. Buckens, M. J. Gondrie, I. Isgum, M. Oudkerk, R. Vliegenthart, H. J. de Koning, C. M. van der Aalst, M. Prokop, J.-W. J. Lammers, P. Zanen, F. A. Mohamed Hoesein, W. P. Mali, B. van Ginneken, E. M. van Rikxoort, and P. A. de Jong, “Diagnosis of chronic obstructive pulmonary disease in lung cancer screening Computed Tomography scans: independent contribution of emphysema, air trapping and bronchial wall thickening.,” Respir. Res., vol. 14, p. 59, Jan. 2013.

[4] O. M. Mets, P. A. de Jong, and M. Prokop, “Computed tomographic screening for lung cancer: an opportunity to evaluate other diseases.,” JAMA, vol. 308, no. 14, pp. 1433–4, Oct. 2012.

[5] M. K. Han, B. Bartholmai, L. X. Liu, S. Murray, J. L. Curtis, F. C. Sciurba, E. A. Kazerooni, B. Thompson, M. Frederick, D. Li, M. Schwarz, A. Limper, C. Freeman, R. J. Landreneau, R. Wise, and F. J. Martinez, “Clinical significance of radiologic characterizations in COPD.,” COPD, vol. 6, no. 6, pp. 459–67, Dec. 2009.

[6] D. A. Lynch, J. H. M. Austin, J. C. Hogg, P. A. Grenier, H.-U. Kauczor, A. A. Bankier, R. G. Barr, T. V Colby, J. R. Galvin, P. A. Gevenois, H. O. Coxson, E. A. Hoffman, J. D. Newell, M. Pistolesi, E. K. Silverman, and J. D. Crapo, “CT-Definable Subtypes of Chronic Obstructive Pulmonary Disease: A Statement of the Fleischner Society.,” Radiology, p. 141579, May 2015.

[7] F. A. A. Mohamed Hoesein, M. Schmidt, O. M. Mets, H. A. Gietema, J.-W. J. Lammers, P. Zanen, H. J. de Koning, C. van der Aalst, M. Oudkerk, R. Vliegenthart, I. Isgum, M. Prokop, B. van Ginneken, E. M. van Rikxoort, and P. A. de Jong, “Discriminating dominant computed tomography phenotypes in smokers without or with mild COPD.,” Respir. Med., vol. 108, no. 1, pp. 136–43, Jan. 2014.

[8] A. Chaouat, R. Naeije, and E. Weitzenblum, “Pulmonary hypertension in COPD.,” Eur. Respir. J., vol. 32, no. 5, pp. 1371–85, Nov. 2008.

[9] J. M. Wells, G. R. Washko, M. K. Han, N. Abbas, H. Nath, A. J. Mamary, E. Regan, W. C. Bailey, F. J. Martinez, E. Westfall, T. H. Beaty, D. Curran-Everett, J. L. Curtis, J. E. Hokanson, D. A. Lynch, B. J. Make, J. D. Crapo, E. K. Silverman, R. P. Bowler, and M. T. Dransfield, “Pulmonary arterial enlargement and acute exacerbations of COPD.,” N. Engl. J. Med., vol. 367, no. 10, pp. 913–21, Sep. 2012.

[10] J. Garcia-Aymerich, E. Barreiro, E. Farrero, R. M. Marrades, J. Morera, and J. M. Antó, “Patients hospitalized for COPD have a high prevalence of modifiable risk factors for exacerbation (EFRAM study).,” Eur. Respir. J., vol. 16, no. 6, pp. 1037–42, Dec. 2000.

Intermountain Healthcare – Welcome to the Zebra-Med herd!

It gives me great pleasure to announce Zebras’ new funding round of $12m. The funding is being led by Healthbox, through the Intermountain Innovation Fund, with participation of Dolby Ventures, Ourcrowd, Marc Benioff and Khosla Ventures, who led our Series A. This investment strengthens Zebra for the foreseeable future and allows us to continue pushing hard to realize our vision.

Continue reading “Intermountain Healthcare – Welcome to the Zebra-Med herd!”