When You Hear Hoofbeats, Think of Zebras, Not Horses
I still remember the blaring siren, rolling beds, doctors in white coats, and a million tests that I had to pass. I was a young 5th grader watching my favorite NFL team the New York Giants play the Tampa Bay Buccaneers. The Giants were riding a three-game winning streak in Eli Manning’s second season under center. But this game had an added level of interest for me because it matched up twin brothers Tiki and Ronde Barber against each other. I have a very close younger brother in appearance, age, personality and people always confused us as twins. It didn’t help that our mom decided to just buy two of every outfit she liked and we would match on the daily. So, this game was really special for me as I was watching my dream play out in front of me: My brother and I were both going to be professional sports players and have an epic rivalry! My mother came into the room and made a very questionable decision that I still wonder about until this day. She demanded I go shower in the middle of the game! Now, I wasn’t dirty from playing sports or in desperate need of a shower (I had showered just that Friday), yet in order that I get to bed early, my mother told me she was turning off the game and forcing me to go shower. I realized that arguing would waste precious seconds of game time so I immediately jumped in the shower, shampooed my hair, soaped my body, and hopped right out. I didn’t even bother putting clothing on as I grabbed a towel and ran to go turn the game back on. In my haste, I never properly dried my feet (or really any part of my body) and as I was getting to the TV I slipped on the wood floor and hit my head on the corner of the TV stand. When I opened my eyes, I couldn’t remember anything. Why was my body wet? What time was it? Why was I lying on the floor? What happened to my clothing? I started crying just as my parents rushed into the room, after hearing the loud thump from downstairs. They quickly figured out I had fallen and were trying to calm me down. I was out-of-sorts and I couldn’t get over how my body was feeling. I laid down on their bed and had my entire family sit with me in the room as they dialed 911 to get an ambulance to bring me to the hospital. I thought I was dying. In the hospital, they ran a CAT Scan and later an MRI to determine the extent of the damage. I was diagnosed with a serious concussion and was told about the healing and recovery process. I also found out the Giants had won 17-3! (Thank God!) This was my first hospital experience and my first of four concussions. I have been the beneficiary of some amazing doctors and radiologists working behind the scenes to diagnose these head injuries and to make sure my brain and head would be ok. But just how hard and important is it to be a radiologist?
Radiologists today are looking at more scans, more slices per scan, and much higher resolution scans and it's only increasing as technology continues to improve. The demand for medical imaging services is continuously increasing, outpacing the supply of qualified radiologists and stretching them to produce more output, without compromising patient care. With two billion people expected to join the global middle class in the next decade a crisis is imminent. There are a lot of variabilities and it takes a tremendous amount of training to be able to consistently give the right answer, and these important behind the scenes doctors are playing a greater role in patient decision making and care. There is a clear need to maintain a high level of accuracy even as the volume and complexity continue to rise. Only by adopting new technology that significantly enhances the capabilities of radiologists can this crisis be mitigated. Zebra Medical Vision is trying to create machine-learning algorithms that will help existing radiologists detect diseases earlier and prevent illnesses and death. But on top of that, they want to make these expensive scans more affordable. By analyzing millions of patient scans, Zebra is teaching their software how to identify specific important medical findings in those scans to help the radiologist produce a better outcome for their patients. Additionally, as part of AI1, Zebra is revolutionizing health care by providing access to high-quality care at an affordable cost.
Zebra’s Radiology Assistant receives imaging scans from various modalities and automatically analyzes them for a number of different clinical findings. Findings are provided in real time to radiologists or other physicians and hospital systems as needed. Zebra uses a proprietary database of millions of imaging scans, along with machine and deep learning tools, to create software that analyzes the data with amazing accuracy. Providers use Zebra to alert them to patients at high risk of cardiovascular, lung, bone and other diseases. This facilitates preventative care programs, proper risk adjustments and allocation, as well as conformance with quality standards. Zebra’ goal is to provide focus to the right patients, at the right time, while saving overall costs and improving patient care.
Headquartered in Kibbutz Shefayim, Israel, the company was founded in 2014 by co-founders Eyal Toledano (currently CTO), Eyal Gura (currently Chairman), and Elad Benjamin (currently CEO). Benjamin went to Tel Aviv University, majoring in physics, and then to Stanford Graduate School of Business for his MBA. He had previously co-founded and was CEO of Vascular Precision, a Medical Device startup that developed a unique vascular closure device for large arteriotomies. He also previously worked as VP of Marketing and Product Management at Algotec Systems, General Manager of Healthcare Information Solutions at Carestream Health, and VP of Surgical Business at Lumenis before Zebra Medical Vision. Gura graduated his BA with the Zell Entrepreneurship Program of IDC Herzliya, where he also Co-Founded the IEC (first entrepreneurship club in Israel). He then graduated the Wharton Business School of the University of Pennsylvania. He served in the Israeli army as the CPO and Head of Navigation for the Submarine Flotilla and Submarine Academy. Gura is an Angel investor and venture capitalist with Pitango Venture Capital, the largest venture capital fund in Israel. He was also formerly the co-founder of TheScouter.com, PicScout (acquired by Getty Images), PicApp (acquired by Ybrant Digital), and The Gifts Project (acquired by eBay). In 2014, he was named a Young Global Leader by the World Economic Forum and sits on the board of many of the companies he invests in. Lastly, Toledano also graduated from the Zell Entrepreneurship Program of IDC Herzliya majoring in Computer Science before getting a Master of Science from MIT. He also served as a Developer and Team Leader for 6 years in the IDF. Zebra Medical vision has raised over 20 million dollars in funding and has been funded by Khosla Ventures, Marc Benioff, Intermountain Investment Fund, OurCrowd, and Dolby Ventures. Zebra’s 30 employees are based in Israel, the US, and the UK. They have won recent awards such as Gartner Cool Vendor 2017, Frost & Sullivan Technology Innovator 2017, Ayn Rand's Foundation Atlas Award for social impact, and FastCompany top-5 AI company. This company is equipped with the right people and funding to make a real difference in the AI Healthcare world.
Right now, Zebra’s algorithm focuses on 4 main areas, with more to come in the future. The first is Bone Health. Zebra’s bone density algorithm uses existing CT scans, performed for any reason, to output a result which is equivalent to the Bone Density T-Score generated by DEXA scans. Providers can use their existing CT data to conduct prescreening for people with increased risk of fracture, with no need for additional tests or radiation. These can then be targeted for Bone Health or Fracture Prevention programs, reducing overall fracture rates and associated costs. Just to put this in perspective, Osteoporosis affects 1/3 women and 1/5 men over 50 and is responsible for 9 million fractures globally every year. 80% of those at risk are not identified or treated, and patients who suffer from an Osteoporotic fracture experience a significant degradation in their quality of life. One example of this is that 25% of hip fracture patients end up in a nursing home within 12 months of their fracture. The costs of Osteoporosis treatment are estimated to be $17 Billion in the US alone. One of the parameters used to identify patients at risk of Osteoporosis is bone density. A DEXA scan provides a T-Score, which along with other risk factors gives an indication of the likelihood of Osteoporosis. Unfortunately, few people actively seek to monitor their bone density, and DEXA scans are only performed by a small percentage of the population. This perpetuates the low identification rate. The second aspect of Zebra’s bone health tests is compression fractures. Zebra’s compression fractures detection algorithm was developed utilizing a combination of traditional machine vision segmentation and convolutional neural net (CNN) technology and can be applied to any CT of the chest, abdomen and/or pelvis. The algorithm automatically segments the vertebral column, and then identifies and localizes compression fractures. This is just one of the ways in which Zebra can help review a patient's CT scan.
Zebra’s fatty liver algorithm analyses CT Chest and Abdomen data to automatically segment the liver, and calculates its average density. When detected in time, fatty liver can be reversible with lifestyle modifications involving diet, exercise and reduced alcohol intake. This algorithm can provide a ‘wake up’ call to pre-diabetics to spur lifestyle interventions. Fatty Liver is common as it’s found incidentally on CT in 11.4% of the adult population in the US and in 22% among diabetics. Fatty liver is a risk factor for several key preventable diseases. Presence of fatty liver is associated with subclinical cardiovascular changes, elevated inflammatory markers of atherosclerosis and heart dysfunction. In diabetics (type II), fatty liver is associated with coronary artery disease. It is also independently associated with increased coronary artery calcification and is a strong predictor of high-risk coronary artery plaque. The presence of fatty liver indicates 2.13x – 4.6x risk of having high-risk coronary artery plaque. People with fatty liver are nearly 2x as likely to experience a cardiovascular event (heart attack or sudden death) over a mean follow up interval of 7.3 years.
Zebra’s emphysema algorithm analyzes CT Chest studies, detects emphysematous regions in the lungs, and quantifies the volume of emphysema in comparison to the overall lung volume. A more accurate understanding of the prevalence of the disease within a given population can help patients manage the disease more effectively before it degrades to more severe, less treatable manifestations. Emphysema is one of the diseases that comprises COPD (chronic obstructive pulmonary disease). It is a long-term, progressive obstructive lung disease in which the alveoli (small sacs) that promote oxygen-carbon dioxide exchange between the air and the bloodstream become damaged or destroyed. There are approximately 12 million individuals in the US who carry a diagnosis of COPD and the American Lung Association estimates that there were twice as many patients with impaired lung function (indicative of early-stage COPD) than patients with diagnosed COPD. COPD is the third leading cause of death behind heart disease and cancer, and current estimates suggest that COPD costs the nation almost $50 billion annually in both direct and indirect health expenditures. Zebra’s Coronary Calcium Scoring algorithm automatically calculates Coronary Calcium Scores based on standard, non-contrast Chest CTs. This tool can provide early detection of people at high risk of severe cardiovascular events. Coronary artery calcium is a biomarker of coronary artery disease – and quantification of coronary calcification is a strong predictor of cardiovascular events such as heart attack or strokes.
Zebra plans to continue to develop a broad pipeline of clinical tools, as they drive towards their mission of AI-based radiology. Their engineers are developing algorithms to detect findings on plain radiography (X-Rays), Mammography, Head CTs, MRI and more. The system can detect 11 different ailments right now, and will be able to sniff out six more by the end of 2017. The company has 35 diagnostic products in total that it plans to release within a year. These will be added to their Imaging Analytics engine, that will grow to become a comprehensive analysis tool. Some future problems they hope to detect are malignant breast lesions, acute brain bleeds, the risk for pulmonary hypertension, the risk for aortic aneurysms, and lung nodules in chest CTs. There is a clear immediate impact that Zebra can make to a radiologist’s job but the future looks even brighter and more comprehensive.
So Why is an AI Healthcare Startup company working with Radiologists called “Zebra Medical Vision”? Well, any medical intern is taught a famous quote: "When you hear hoofbeats, think of horses, not zebras". For non-medical students, this means to look for more common diseases. Zebra is the medical slang for arriving at an exotic medical diagnosis when a more commonplace explanation is more likely. Like in many other fields of our lives, utilization was prioritized over precision. Now with the availability of computing power and data, this company hopes to finally be able to take care of the Zebras!
Using both platforms, and their partner network which reaches over 1,100 hospitals and providers, Zebra is able to leverage the tremendous powers of the global research community and channel subsequent insights to healthcare institutions. They have teamed up with providers such as Carestream Health, Intermountain Healthcare, and Telerad Tech. Carestream Health is a worldwide provider of medical imaging systems and IT solutions; X-ray imaging systems for non-destructive testing; and advanced materials for the precision films and electronics markets. Intermountain Healthcare is a Utah-based not-for-profit system of 22 hospitals, 185 clinics, a Medical Group with about 1,500 employed physicians and advanced practitioners, a health plans group called SelectHealth, and other medical services. Lastly, Telerad Tech is the technology arm of India’s first and largest teleradiology company - Teleradiology Solutions (TRS) corporation. They signed a partnership in July to bring Zebra-Med’s cloud-based deep learning analytics engine to over 20 countries and 150 hospitals and healthcare organizations. In July 2017, Zebra announced that the company had been granted the CE approval and subsequent release of its Deep Learning Analytics Engine in Europe, as well as regulatory clearance and product release in Australia and New Zealand. These partnerships are just the beginning of impact Zebra wants to have in the radiology world as it hopes to eventually be in every country of the world.
Zebra’s newly announced program called AI1 is a new suite that offers all its current and future algorithms to healthcare providers globally for $1 USD per scan. “With this new model, we hope to facilitate adoption globally, especially in countries where access to radiology is difficult,” says Elad Benjamin, “We are making a commitment to provide our current tools, and all future ones, for a flat $1 USD per scan. By doing so we believe that a true difference can be made in the provision of radiology services worldwide.” The business model is designed so that Zebra can be accessible to all customers, even those outside of the Western world.
Just how good can their product be? Zebra recently went through testing at an Oxford University hospital, and the system was sensitive in 95 percent of 100 CT scans, with a specificity of 100 percent. (Specificity measures how well a test can distinguish those with a disease from those without, while sensitivity is the number of true positive tests compared to all patients who actually have the disease.) Also, Gura said that Zebra Medical Vision algorithms were more accurate than expert radiologists. He said that the company’s breast cancer algorithm, which was developed and trained using 350,000 mammograms of diagnosed breast cancer, detects breast cancer with a 92 percent accuracy, in comparison to the 87 percent accuracy rate of radiologists. Gura said, “machine and deep learning will help the finite amount of radiologists we have currently to be 10 times better and analyze 10 times more reports per hour.”
The cherry on top is that Zebra Medical Vision just announced a partnership with Google to offer its algorithms on the search giant's cloud. Currently, Zebra's software is installed on-site, which can be costly for hospitals having to pay for servers to store the imaging. That's why Zebra has partnered with Google to offer its algorithms as part of the Google Cloud. Major hospitals and healthcare systems are slowly moving their data storage to the cloud, which offers a cheaper way to store large amounts of information. Companies like Microsoft, Google and Amazon offer such services. But these technology giants also provide added services on top of their clouds, such as data analytics and AI capabilities. Both Microsoft and Amazon have special tools in the cloud-focused on areas of medicine such as genomics or drug development. Google has been trying to push the healthcare side of its cloud platform in the past few months. The partnership with Google can help the search giant bolster its cloud efforts as it competes with Amazon and Microsoft in the fast-growing space. And for Zebra, the deal could mean the ability to scale its business faster. This partnership might be just the deal Zebra is looking for to vault it to the top of the AI healthcare world as we close out 2017. So next time you see a Zebra don’t be afraid, it's probably just your friendly Israeli startup making the world a better place!