Zebra Medical Vision: Machine Learning Meets Healthcare Imaging

Zebra Medical Vision: Machine Learning Meets Healthcare Imaging

Advancements in machine learning and artificial intelligence (AI) have drawn numerous companies to the field of healthcare radiology, a medical specialty historically defined by technical challenges and process inefficiencies. One such company is Israel-based Zebra Medical Vision (Zebra-Med), which was founded in 2014 to apply machine learning to the identification of pathogens in medical images.

For background, the average hospital generates 50 petabytes of data each year in the form of clinical notes, lab tests, and medical images.[1] Yet less than three percent of this data is used is analyzed and/or used to inform medical decisions.[2] This market opportunity drives product development at companies like Zebra-Med that leverage machine learning and AI to make sense of hospitals’ data and help improve provider efficiency, increase diagnostic accuracy, personalize treatment, and improve the patient experience.

For example, while timely detection of brain bleeds is critical, research has shown that such bleeds are missed anywhere between 12% and 51% of the time[3], and approximately six million people worldwide die every year of brain bleed related conditions[4]. Similarly, chest x-rays are among the most common medical images ordered, but also some of the most difficult for radiologists to interpret. Deploying machine learning and AI-based technologies has the potential to improve accuracy by replacing mundane, highly time-intensive activities currently performed by radiologists unassisted by image detection technologies. This would in turn allow radiologists to become much more efficient and accurate, and thus decreases variability in diagnostic decisions across healthcare organizations.

In the short term (i.e., the next two years), Zebra-Med is building a pipeline of targeted AI-based products including a chest x-ray, brain bleed computed technology (CT), and mammography lesion readers. For reference, the company’s chest x-ray AI product was trained using nearly two million images to identify 40 different common clinical findings. Simultaneous with product development, Zebra-Med is focused on obtaining requisite regulatory approvals, which it has done for seven of its products in countries across Europe, the United States, Latin America, and Asia.

Another step that Zebra-Med could take in the near term is working directly with radiologists and hospital groups to integrate the company’s offerings into physician workflow. For healthcare technology companies like Zebra-Med, developing clinically relevant products is just the first step. Integrating these products into physician workflow is an entirely separate challenge, made more difficult by physician time constraints and technology fatigue on behalf of hospital groups inundated with other point solutions that rarely offer full integration with a hospital’s existing technology stack. It’s unclear from press coverage exactly how Zebra-Med plans to approach this challenge but one tactic might be working directly with professional groups (e.g., American Association of Medical Dosimetrists, Association for Medical Imaging Management) to familiarize radiologists with the technology. Zebra-Med might also focus its business development efforts on marquee teaching hospitals (e.g., Mayo Clinic, Massachusetts General Hospital) as getting buy-in from one of these brands would build significant industry credibility for the company.

In the medium term (i.e., two to ten years out), Zebra-Med intends to offer an “All-In-One” imaging analytics package that bundles its targeted product offerings into a $1 per any scan offering to hospitals. Additional features of this offering that Zebra-Med has yet to develop include algorithms to detect low bone mineral density, vertebral fractures, fatty liver, coronary artery calcium, and emphysema. Once operational, the company’s software offerings will provide a second set of eyes for physicians in acute care settings and further its founders’ vision for scalable, high quality healthcare imaging solutions available globally.

An additional step that Zebra-Med could take in the medium term to bring its offerings to market include building out its own teleradiology practice. Rather than sell its solutions to hospital groups (as currently contemplated) this would entail Zebra-Med hiring its own radiologists and providing AI-assisted medical image reading services to hospital groups on an outsourced/contracted basis. While Zebra-Med would still face the challenge of convincing hospitals to trust it technologies this would ensure that radiologists using its software were fully trained/properly using the product.

Given that Zebra-Med operates in a nascent industry, one important open question relates to the willingness of hospital groups and health insurers to trust diagnoses driven by machine learning and AI algorithms. This remains a critical question as healthcare has historically been a slow adaptor of new technologies.

Word count: 757

Sources: Company press releases, TechCrunch articles, and Israeli Innovation News.

[1] Press statement by the European Society of Radiology (ESR), October 2018.

[2] Ibid.

[3] Missed Diagnosis of Subarachnoid Hemorrhage in the Emergency Department, American Heart Association.

[4] Facts and Figures, World Stroke Organization.


Envision Energy: Use machine learning technology to achieve an energy efficient and low-carbon future in China


Stitch Fix: The Fashion Maven of Machine Learning

Student comments on Zebra Medical Vision: Machine Learning Meets Healthcare Imaging

  1. Thanks for sharing this story, this is amazing how progressive Israel medicare is. Did you consider the implications on the cost for this project? How it differs from the other less technically advanced health providers?

Leave a comment