Zebra Medical Vision

The finding no one was looking for

A medical scan is done to answer one question. Someone fell, so doctors check for a fracture. Someone is coughing, so they take a chest X-ray. The radiologist reads the image to answer that question, and that is what the job requires.

But the same image often contains much more. An abdominal CT scan also shows the liver and the vertebrae. A chest X-ray also shows the outline of the heart and calcification in the arteries. These findings are physically there, but no one was looking for them, so in many cases no one sees them.

This is the opportunity the company was built around, and it is the kind that is hard to get any other way. A scan that was already taken, already paid for and already stored in the system can be used to detect diseases early, even diseases that had nothing to do with the reason for the scan. There is no need to send anyone for another test, expose them to more radiation or spend another shekel. The data already exists, waiting in the system. Zebra Medical Vision, founded in 2014, built a series of algorithms around this idea that scan existing images and flag such findings for the medical team's attention.

Nine FDA clearances in four years: what it took

Between 2018 and 2021 the company received nine 510(k) marketing clearances from the US Food and Drug Administration. To judge that number, remember the context: until the middle of the last decade, there were almost no approved medical algorithms in the world at all.

The clearances covered findings that are completely different from one another: quantifying calcium in the coronary arteries, suspected pneumothorax (collapsed lung) on a chest X-ray, suspected bleeding inside the skull, vertebral compression fractures, and mammography. The eighth clearance, for the coronary artery calcium solution, was announced in September 2021.

The point that is easy to miss is that there is no single clearance that covers everything. Each intended use needs its own path: its own evidence, its own comparison to an existing device, and its own submission. Nine clearances are not nine algorithms but nine completed regulatory processes. With them, the company showed that this path can be done at a steady pace, not just in theory or once by luck.

What the clearances actually allow: triage, not diagnosis

Most of the company's clearances were for triage and alert tools, and this is a real distinction, not legal wording. The software scans the queue, marks a case as urgent or draws attention to it, and so changes the order in which the radiologist gets to it. It does not make a diagnosis, does not replace a reading and does not take responsibility away from anyone.

Why the distinction matters so much: a tool that changes priorities shortens the time until a doctor sees a critical case, and in a brain bleed or a collapsed lung that can save lives. A tool that makes a diagnosis carries a completely different responsibility and needs a different level of evidence.

It is also worth remembering what a 510(k) clearance is not. It is not proof of broad clinical benefit. It is a finding that the device is close enough to an existing approved device. Whether routine use of such tools actually improves patient outcomes is tested in separate studies, and in some fields the question was still open years after the clearance itself.

A dollar per scan: the model that tried to break the price barrier

In October 2017 the company introduced an unusual pricing model called AI1: all of its algorithms, current and future, at a flat rate of one dollar per scan. Elad Benjamin, one of its founders and its CEO at the time, framed it as a commitment, not a launch price but a policy.

The logic addresses a real barrier. When each finding is priced separately, a hospital has to decide in advance which diseases it can afford to look for, and that is a strange choice from a medical point of view. A flat rate removes that decision: once the scan comes in, all the algorithms run on it.

The offer was built into Google's cloud, and at the time the company reported that its engine was deployed in more than 50 hospitals and had analyzed more than a million scans. The model was bold, and also risky as a business. It bases revenue on volume, so it needs mass adoption to pay off. Not all of the industry followed that bet, and whether it succeeded remained an open question.

Billing codes: the obstacle few people talk about

There is a step between "the regulator cleared it" and "a hospital uses it," and it is almost never mentioned in news stories: who pays. In the United States, a medical service is billed to an insurer under a standard code. Without such a code, a hospital that runs the software pays for it out of its own pocket and it becomes an expense. With a code, a path to reimbursement opens.

According to the acquisition announcement, Zebra obtained a first-of-its-kind billing code for an AI-based product, together with the American College of Radiology. This partnership is no accident: a new code is set through a long professional process, not by a request from a single vendor.

This achievement is hard to appreciate without the context, but it is often the difference between a cleared product and a product in use. You can develop an excellent algorithm, get past the regulator, and find that no institution adopts it because it has no way to bill for it. In the US health care industry, the way the money flows is no less decisive than whether the technology itself works as it should.

2021: the merger with Nanox, and what was actually agreed

On 10 August 2021 Nano-X Imaging announced an agreement to merge with Zebra, which was headquartered at Kibbutz Shefayim. The headline at the time, a deal of "up to $200 million," was made of two parts: $100 million in shares up front, and up to $100 million more, subject to meeting milestones and to deductions.

When the deal actually closed, on 8 November 2021, the numbers were different: the merger was valued at about $110 million in shares at closing, plus up to $84 million in additional shares subject to milestones. This gap is worth keeping in mind every time the word "up to" appears next to a deal amount.

Since then the technology has operated under the name Nanox AI. More clearances were granted under the new name, including HealthFLD for detecting fatty liver in February 2024. Still, it is worth keeping proportion: in the fourth quarter of 2025, Nanox's AI and software solutions brought in only about half a million dollars.