Stephanie Cole

Your Lung Cancer Screening Software Is Now an FDA-Cleared Medical Device - and That Changes Everything

March 3, 2026

Last month, I was having lunch with a friend who works in hospital administration - the kind of lunch where you go in thinking it'll be an hour and walk out three hours later, slightly confused about where the afternoon went. She kept using a phrase I'd never heard before: Software as a Medical Device. She said it like it was obvious. I nodded like I understood. I went home and looked it up.

What I found was a story that I think the business press has almost entirely missed - and that people running any kind of health-adjacent operation need to understand right now.

In early February 2026, a French company called Median Technologies received FDA 510(k) clearance for something called eyonis LCS. Median Technologies describes itself as a developer of AI-powered Software as a Medical Device (SaMD) for early cancer diagnosis, and eyonis LCS is its AI/ML-powered computer-aided detection and diagnosis tool intended specifically for lung cancer screening. On the surface, that sounds like a niche medical announcement. It is not. It is a signal about where AI software regulation is going - and what happens when a piece of software stops being just software.

What Actually Happened

eyonis LCS is the first end-to-end detection and diagnosis device FDA cleared specifically targeting lung cancer screenings. That distinction - end-to-end - is important. There are other AI tools that help flag nodules, help with measurements, help radiologists prioritize. eyonis LCS is a computer-aided detection and diagnosis (CADe/CADx) Software as a Medical Device designed to analyze imaging data generated from CT scans to assist radiologists in detecting and characterizing parenchymal pulmonary nodules on the lungs. It does both jobs. Detection and diagnosis. In one piece of software. Cleared by the FDA as a medical device.

In manufacturer performance testing, eyonis LCS demonstrated 93.3% sensitivity, 92.4% specificity, and 99.9% negative predictive value. That last number is the one I keep coming back to. A 99.9% negative predictive value means that when this software says there's nothing to worry about, it is almost never wrong about that. The high negative predictive value aims to reduce false positives, with the company reporting one false positive per 1,000 cases.

The market Median is targeting isn't small. The USPSTF eligible population for lung cancer screening using LDCT in the United States includes adults aged 50 to 80 years with a 20-pack-year smoking history, and based on current eligibility criteria, 14.5 million individuals qualify for screening. And the stakes are not abstract. Lung cancer remains the leading cause of cancer death in the US, with stage 1 disease yielding approximately 80% long-term survival compared to 15% five-year survival when detected after symptoms appear.

The stock market noticed immediately. Following the announcement before stock markets opened on February 9th, Median's Paris-listed stock climbed by 19% at market open versus a previous close of €4.28, and by 15:00 CET the company's stock had risen by around 50%. A 50% single-day jump for a company that makes software. That's not a niche story.

The Reimbursement Part Is Actually the Whole Story

Here's where I think most coverage of this event falls completely flat. Everyone is focused on the AI angle - which, fine, yes, it's impressive. But the genuinely transformative part of this clearance is what it unlocks financially for healthcare systems that adopt it.

The device benefits from existing reimbursement under Category III CPT codes 0721T and 0722T, both assigned to New Technology APC 1508, with Medicare payments ranging from $601 to $700, and this established framework is expected to accelerate adoption among imaging providers.

Think about what that means. An imaging center that deploys eyonis LCS isn't just getting an AI tool - it's getting a tool that Medicare will pay $601 to $700 for, per use, under an already-existing billing code. The reimbursement infrastructure was already sitting there waiting. That is not a small thing. That's the difference between a promising technology that takes fifteen years to reach patients and one that could be in clinical workflows within the next eighteen months.

My travel agent who handles the European itineraries - I have two, one for Europe and one for everywhere else, which apparently surprises people - she once told me that the difference between a trip that works and a trip that doesn't is almost never the destination. It's whether the logistics are already in place when you arrive. That's what a pre-existing reimbursement code does for a new medical device. The path is already paved.

The Radiologist Shortage Is the Context Nobody Is Providing

You cannot talk about this FDA clearance without talking about the workforce crisis that makes it genuinely urgent.

The United States is facing shortages in a myriad of medical fields, including diagnostic radiology, and the increasing number of imaging studies, owing to advancing technology and an aging population, is outgrowing the capacity of radiologists. This isn't a future problem. It's present tense and getting worse. Imaging use is increasing 3-4% annually due to aging populations, chronic disease, and expanded access to preventive care, while the number of practicing radiologists is only growing approximately 1% annually.

Only 16% of the eligible people are able to get lung cancer screening scans, which means there is a big void here and a need for improvement to have lung cancer screening reaching the general masses to a greater extent. Sixteen percent. Of fourteen and a half million eligible Americans, barely two million are getting the screening they're supposed to get. The capacity simply doesn't exist to do more - not without something changing.

Given the comparable projected levels of growth in supply and demand, the present radiologist shortage is projected to persist unless steps are taken to grow the workforce and/or decrease per person imaging utilization. The experts are essentially saying: train more radiologists or find a way to make each radiologist more productive. The second option has a solution available today. The first option takes a decade.

Derek forwarded me something this week about a hospital in the midwest turning away imaging patients because there wasn't anyone to read the scans. He'd framed it as a general interest article. I think he was trying to make a point. He was right.

Software as a Medical Device Is a Category Shift, Not a Feature

This is where I want to be direct about why I think this matters beyond radiology departments and hospital procurement committees.

The FDA's concept of Software as a Medical Device - SaMD - represents a fundamental reclassification of what software can be. For years, software that assisted doctors was called a "clinical decision support tool" and largely existed in a regulatory grey zone. It helped clinicians, but it wasn't accountable in the way a drug or a device was accountable. FDA clearance changes that. It means the software has been evaluated for safety and effectiveness. It means there's a performance standard. It means there's liability attached in ways there wasn't before.

By mid-2025, the FDA had added 115 radiology AI algorithms to its approved list, with approximately 873 total - making medical imaging the single largest AI target among all medical specialties. Eight hundred and seventy three. The volume of AI medical software going through formal FDA review has become extraordinary, and it's accelerating.

Tempus AI, which has a market cap of several billion dollars and operates one of the more ambitious AI imaging platforms in the country, plans to embed eyonis LCS capability within its FDA-cleared, CE-marked Pixel imaging platform, an AI-driven solution that supports advanced radiology image analysis and automated reporting and lesion quantification. When a company that size moves fast to integrate a newly cleared tool, it's worth paying attention to why.

The reason is simple. FDA clearance isn't just regulatory permission - it's commercial legitimacy. It's the thing that lets a hospital's legal team stop blocking a procurement. It's what gets something onto an approved vendor list. Median plans to commercialize eyonis LCS in the United States through a combination of direct enterprise sales, strategic distribution partnerships, and integration into existing clinical environments. That's a go-to-market strategy that only works after FDA clearance because before clearance, most enterprise hospital systems won't touch it.

What People Running Businesses Should Actually Take From This

I had someone from the office pull together a few comparable clearances for context - she spent most of a morning on it. What she found was that this isn't a one-off event. It's a pattern. The FDA is clearing AI-based diagnostic software at an accelerating pace, and the companies receiving those clearances are not acting like startups hoping to eventually find a business model. They're structured for enterprise deployment, with reimbursement pathways, distribution partnerships, and integration strategies ready to go at the moment clearance is announced.

For anyone in the broader healthcare business ecosystem - whether you're running a medical imaging center, a hospital administration technology company, a healthcare SaaS business, or even an employer managing self-insured health plans - the category of Software as a Medical Device is going to become one of the most important procurement categories in the next five years. And the companies that understand how FDA clearance intersects with reimbursement codes will have a structural advantage over the ones that are still treating AI tools like software subscriptions.

Tory mentioned something in a team meeting last week about the importance of staying ahead of regulatory shifts rather than reacting to them. He was talking about something else entirely - I think he was trying to apply it to his own situation, honestly - but he wasn't wrong. The companies that are going to win in health AI are not the ones with the best algorithm. eyonis LCS was evaluated in two pivotal clinical studies required for marketing approvals in the US and Europe, both successfully completed. The companies that win are the ones that cleared the regulatory hurdle, built the billing infrastructure, and showed up with a product hospitals could actually deploy without fear of legal exposure.

That's not an AI story. That's a business strategy story. And it's one that's playing out right now, in a space most business publications are still treating as a niche medical news item.

I'm going to keep watching this space closely. The intersection of AI software regulation and enterprise healthcare procurement is where some genuinely important business dynamics are developing - and most of the people who should be paying attention are currently focused somewhere else entirely. Those two things won't remain true at the same time for long.

For more on the broader dynamics of software companies navigating regulation and market legitimacy, see our take on HSBC's SaaSpocalypse framing and what we wrote about Wall Street's complicated relationship with software stocks. And if you're thinking about what AI-driven changes mean for how your own business operates day to day, the software-defined everything question is one worth sitting with.