I Sold My Startup A Year After Founding It. Here's Why That Was The Fastest Way To Build Real-World Healthcare AI

Overview

Louis Blankemeier, co-founder and CEO of healthcare AI startup Cognita, explains why he sold his company less than a year after founding it. Cognita built AI models that interpret medical images such as X-rays and CT scans across tens of thousands of potential diagnoses and generate full radiology reports. Faced with a choice between raising venture capital to stay independent or accepting an acquisition by Radiology Partners, the world's largest radiology practice, he argues that joining a large, established company was the fastest path to deploying reliable clinical AI at scale.

Key Takeaways

Cognita was founded in October 2024 to turn the founders' Ph.D. research into real-world medical imaging AI.
Its models interpret X-rays and CT scans across tens of thousands of potential diagnoses and write full radiology reports.
Less than a year in, the team chose acquisition by Radiology Partners over raising venture capital to stay independent.
Blankemeier argues research success is not the same as clinical readiness because real-world radiology is full of edge cases.
He compares the challenge to self-driving cars, where only a few companies reached reliability after a decade of investment.
He says reliable clinical AI needs scale, integrated systems, large diverse datasets and continuous edge-case collection.

Stats & Key Facts

#A single CT study can contain 10 high-resolution volumetric series, effectively 3D videos.
#Adding prior studies for the same patient can produce roughly a billion pixels of data.
#Cognita's research models were trained on tens to hundreds of thousands of studies.

I Sold My Startup A Year After Founding It. Here's Why That Was The Fastest Way To Build Real-World Healthcare AI

Why Cognita was founded

The company started to move academic research into real clinical use.

›Blankemeier and his co-founders set out in October 2024 to make their Ph.D. research useful in the real world.
›They had built AI models interpreting medical images across tens of thousands of potential diagnoses.
›The models generate comprehensive radiology reports that mirror how radiologists reason in clinical practice.

At a time when AI in radiology was limited to flagging a handful of specific conditions, Blankemeier describes Cognita's broad diagnostic approach as a fundamental shift in what radiology AI could attempt.

The fork in the road

Within a year the team faced a major decision about its future.

›The choice was to raise venture capital and stay independent or accept an acquisition by Radiology Partners.
›Radiology Partners is described as the world's largest radiology practice.
›Conventional tech wisdom holds that real ambition means staying independent.

Blankemeier writes that clinical AI is highly regulated with long sales cycles and complex stakeholder dynamics, where structural advantages tend to harden market positions over time. The team decided that joining forces, structured to protect their velocity, would improve the odds of realizing their mission of increasing access to healthcare.

Research success is not clinical readiness

Blankemeier draws a sharp line between academic demonstrations and production care.

›His Ph.D. models were trained on tens to hundreds of thousands of studies, large for research scale.
›Those models make strong academic demonstrations but would not meet production safety and consistency standards.
›Real-world radiology is defined by edge cases where rare but critical pathologies appear regularly.

He notes that a single CT study can contain 10 high-resolution volumetric series, effectively 3D videos, and adding prior studies for the same patient can reach about a billion pixels encoding entire medical textbooks worth of information. Models that work in controlled research often fall apart when exposed to real-world complexity.

The self-driving car analogy

Blankemeier uses autonomous vehicles to illustrate the reliability gap.

›A decade ago, self-driving progress looked impressive.
›The real world kept introducing new failure modes.
›After more than a decade of significant capital investment, only a handful of companies approached true reliability.

What reliable clinical AI requires

He identifies the patterns that let the leading self-driving companies succeed and applies them to radiology.

›The companies that progressed most controlled the entire system and achieved scale early.
›They owned the vehicles, sensor stack, data pipeline, simulation environments and deployment infrastructure.
›That integration at scale let them continuously collect rare edge cases, retrain, validate and redeploy safely.

Blankemeier argues radiology is no different. Success in the real world requires massive, diverse historical datasets and live data feeds that surface rare edge cases and distributional shifts, plus the clinical resources to redesign workflows around AI, secure regulatory clearance, refine models safely and monitor performance after deployment.

Frequently Asked Questions

What does Cognita's AI do?

It interprets medical images such as X-rays and CT scans across tens of thousands of potential diagnoses and generates comprehensive radiology reports that mirror how radiologists reason in clinical practice.

When was Cognita founded and when was it sold?

The founders set out in October 2024 to commercialize their Ph.D. research, and they accepted an acquisition less than a year later.

Who acquired Cognita?

Radiology Partners, described in the article as the world's largest radiology practice, acquired the company.

Why did Blankemeier choose acquisition over staying independent?

He argues clinical AI is highly regulated with long sales cycles, and that joining a large established company, structured to protect velocity, would better realize the mission of increasing access to healthcare.

Why is research-grade radiology AI not ready for clinical use?

Models trained on research datasets often fall apart against real-world complexity and edge cases, and they would not meet the safety and consistency standards required for production patient care.

Blankemeier's argument is that for highly regulated, data-intensive clinical AI, an acquisition that provides scale and integrated infrastructure can be the fastest route to a real-world impact.