VC's funding for AI tools for healthcare was projected to reach $11 billion last year. This is the headline figure that speaks to the broad belief that artificial intelligence is transformative in key sectors.
Many startups applying AI in healthcare are trying to promote efficiency by automating some of the management that will get them on track and allow for patient care. Hamburg-based Elea fits widely into this type, but it starts with a pathology lab, a relatively unnoticed, underserved niche. We believe that it can scale workflow systems equipped with voice-based AI agents, accompanied by analyzing disease samples. Includes porting a workflow-centric approach to accelerate the output of other healthcare departments.
ELEA's first AI tool is designed to overhaul how clinicians and other lab staff work. This is a complete alternative to legacy information systems and other ways of working (such as using Microsoft Office to enter reports), reducing the time it takes to move workflows to an “AI operating system” and output diagnostics to “substantially” unfold the automation of audio-to-text transcription and other forms of automation.
Elea says that after operating with its first user for about six months, the system has reduced the time it took the lab to produce about half the report in about two days.
Gradual automation
Dr. Christoph Schröder, CEO and co-founder of Elea, says that the step-by-step manual workflow in the Pathology Lab means there is a good range of productivity gains by applying AI. “We basically turn this all around – and every step is much more automated… [Doctors] Please talk to MTA Elea [medical technical assistants] Talk to Elea and tell him what they are seeing and what they want to do with it,” he explains.
“ELEA is an agent, performs all the tasks in the system and prints things. For example, prepares stains and all of those things – [tasks] Much faster, much smoother. ”
“It really doesn't increase anything. It replaces the entire infrastructure,” he adds the lab's legacy systems and cloud-based software that wants to replace more silent ways of working. The idea of AI OS is that you can adjust everything.
Startups are built on a variety of large-scale language models (LLMs) by tweaking specialized information and data to enable core functionality in the context of pathology labs. The platform burns text from speech to transcribe the staff's audio notes and also “from text to structure.” The system can turn these transcribed audio notes into active orientation, allowing AI agent actions to work.
Elea also plans to develop a unique basic model for slide image analysis for each Schröder. But for now, we're focusing on scaling our first product.
The startup's pitch to lab suggests that what can take two to three weeks using traditional processes can replace the boring things that can be stacked up and stacked up productivity gains and surrounded reports in manuals, thus replacing the boring things that human error and other workflow keys can inject friction.
This system can be accessed by lab staff via the iPad, Mac, or web app. It offers a variety of touch points for different types of users.
The business was established in early 2024 and launched in October in its first laboratory, spent time in stealth in 2023, with a background in applying AI to autonomous driving projects at Bosch, Luminar and Mercedes.
Another co-founder, Dr. Sebastian Cass (startup CMO), brings a clinical background, has worked in the intensive care, anesthesiology and emergency departments for over a decade, and was previously a medical director for a large hospital chain.
So far, ELEA has inked partnerships with major German hospital groups (which has not yet disclosed how many hospital groups) and says it handles around 70,000 cases per year. That's why the system has had hundreds of users so far.
With more customers set to launch “Soon,” Schröder also says he is looking at international expansion, with a special look at entry into the US market.
Seedbacking
The startup is the first to disclose the 4 million euros of seeds grown last year (led by fly ventures and giant ventures). This is used to build engineering teams and get products into the first lab.
This figure is a fairly small amount of billions of dollars mentioned above, and the funding that currently roams the space every year. However, Schröder argues that AI startups don't need an army of engineers or hundreds of millions of troops to succeed. Also, this healthcare context means taking a department-centric approach and maturing the target use case before moving to the next application area.
Still, at the same time, he makes sure his team is considering raising a (larger) Series A round – perhaps this summer, Elea is actively marketing to buy more labs, rather than resorting to the word-of-mouth approach they started.
He says the competitive landscape of AI solutions in healthcare:
“Many of the tools you see are add-ons on top of existing systems [such as EHR systems] …It's something [users] You need to do something else that people who don't really want to work with another tool, another UI, or something else they have to do.
“What we instead built is actually deep integration into our own lab information system — or we call it a pathology operating system. This means that ultimately the user doesn't even need to use a different UI, and you don't need to use another tool. And it's just talking to Elear, saying what it's watching, saying what it wants to do, and what Elear should do with the system.”
“We also don't need some engineers. We need a dozen, two dozen, really, really good,” he insists. “We have roughly 20 engineers in our team. They can do amazing things.”
“The fastest growing companies we've seen these days don't have hundreds of engineers. They have one or two dozen experts, and those guys can make amazing things. And that's also the philosophy we have, which is why we don't really need to raise hundreds of millions of millions,” he adds.
“It's definitely a change in paradigm… in the way you build a company.”
Scaling workflow thinking
Choosing to start with a pathology lab was a strategic choice for ELEA, as it represents the pathology space as “very global” as it represents the addressable market, worth billions of dollars, according to Schröder. In particular, global lab companies and suppliers are increasing scalability as service play compared to situations that are fragmented enough to supply hospitals.
“For us, it's very interesting because you can build one application and actually expand it in real life, so it's very interesting – from Germany to the UK to the US,” he suggests. “Everyone thinks the same thing, acts the same way, has the same workflow. And if you solve it in German, if you do great things with current LLMS, you solve it in English [and other languages like Spanish] …So it opens up many different opportunities. ”
He also praises the Pathology Lab as “one of the fastest growing regions in medicine.” It points out that medical development, such as molecular pathology and increased DNA sequencing, is creating demand for more types of analysis and more analysis. All of this means more work in the lab and more pressure on the lab is more productive.
As ELEA matures its lab use cases, he says it could be that they might try to move to areas where AI is more commonly applied to healthcare, such as supporting hospital doctors to capture patient interactions, but other applications they develop also focus on workflows.
“What we want to bring is this workflow mindset where everything is treated like a workflow task, and you need to have a report at the end and send that report,” he says.
Image processing is another area where ELEA is interested in other future healthcare applications, such as speeding up data analysis in radiology.
assignment
How accurate is it? Because healthcare is a very sensitive use case, these AI transcription errors can have serious consequences if there is a mismatch between what a human physician says and what the ELEA hears and reports to other decision makers in the patient care chain.
Currently, Schröder says he evaluates accuracy by examining the number of characters that users change in reports provided by AI. Now, he says it's between 5-10% of cases where some manual interactions are occurring for these automated reports that could indicate errors. (He also suggests that doctors may need to make changes for other reasons, but they say they are working to “drive” the rate at which manual interventions occur.)
Ultimately, he insists he will stop along with doctors and other staff who are asked to review and approve the AI output. ELEA's workflow suggests that it is not at all different to the legacy processes that it is designed to replace (e.g., physician voice memos are typed by humans, and such transcription is incorrect, typist”).
However, automation can result in higher throughput volumes. This could put pressure on checks that require human staff to address more data and reports than before.
Schröder agrees that there may be risks on this. However, he says it has incorporated a “safety net” feature that will allow AI to try to find potential problems. “We call it the second eye,” he says. [the doctor] Tell him now and give him a comment and suggestions. ”
Patient confidentiality can be another concern that comes with agent AI that relies on cloud-based processing (similar to ELEA) rather than data that remains on-premises and is under lab control. In addition to this, Schröder claims that the startup has resolved its “data privacy” concerns by separating patient identity from the diagnostic output, and thus essentially relies on pseudonymization of data protection compliance.
“We're always anonymous on the way, every step just does one thing. And we're going to combine the data on the device that the doctor is looking at,” he says. “So, basically, I have a pseudo ID to use in all the processing steps – it's temporary and then removed – but when the doctor sees the patient they're paired with the device for him.”
“We will work with our servers in Europe to make sure everything is data privacy compliant,” he says. “Our main customer is a public hospital chain called Germany's critical infrastructure. From a data privacy perspective, we had to make sure everything was safe. And they gave us a thumbs up.”
“In the end, we probably over-achieved what we need to do, but it's always better to be on the safe side, especially when processing medical data.”