Paris-based startup Nabla fair Announced which has raised a $24 million Series B financing round led by Cathay Innovation, with the participation of ZEBOX Ventures, CMA CGM's corporate venture capital fund. This funding round comes just months after Nabla signed a large-scale partnership with Permanente Medical Group, a division of US healthcare giant Kaiser Permanente.
According to a source, Nabla has reached a valuation of $180 million following today's funding round. The company could also end up raising more money from US investors as part of this round.
Nabla has been working on an ai co-pilot for doctors and other medical staff. The best way to describe him is that he is a silent coworker who sits in the corner of the room, taking notes and writing medical reports for you.
The startup was originally founded by Alexandre Lebrun, Delphine Groll and Martin Raison. Lebrun, CEO of Nabla, was CEO of Wit.ai, an ai assistant startup acquired by Facebook. He later became the head of engineering at Facebook's FAIR artificial intelligence research lab.
A few weeks ago, I saw a live demonstration of Nabla with a real doctor and a fake patient pretending to have back pain. When a doctor starts a consultation, he presses the start button on the Nabla interface and forgets about his computer.
In addition to the physical exam portion, a consultation also includes a long discussion with lots of questions about what brings you here and your medical history. At the end of the consultation, there may also be recommendations and prescriptions.
Nabla uses speech-to-text technology to convert the conversation into a written transcript. It works with both in-person consultations and telehealth appointments.
Once the patient has left, the doctor presses the stop button. Nabla then uses a large language model refined with medical data and health-related conversations to identify important data points in the query: medical vital signs, drug names, pathologies, etc.
Nabla generates a complete medical report in one to two minutes with a consultation summary, prescriptions and follow-up appointment letters.
These reports can be customized to the doctor's needs with custom formatting for their notes. For example, you can add instructions to make the note more concise or more detailed. Or you can ask to generate notes that follow the Subjective, Objective, Assessment, and Plan (SOAP) note pattern that is widely used in the US.
During the demo I saw, I was very surprised by the overall effectiveness of Nabla. Even though we were in a crowded room and Nabla was running on a laptop a couple of meters from the demo presenters, the tool was able to produce an accurate transcription and a useful report.
With Nabla Copilot, as the name suggests, the startup does not try to take humans out of the medical circuit. Doctors still have the final say, as they can edit reports before filing them in their electronic health record (EHR) system.
Instead, the company believes it can help doctors save time on administrative work so they can spend more time focusing on patients.
“What we know in the near future is that we don't want to try to replace doctors. You've seen companies, like Babylon in the UK, spend a billion dollars trying to create chatbots and automate things immediately and take doctors out of the loop. And we decided a long time ago with Nabla Copilot that (doctors) are the pilots and we work alongside them,” said Nabla co-founder and CEO Alexandre Lebrun.
“It's a bit like the automation of autonomous vehicles. We are still at level two today. Very soon we will begin level three with clinical assurance support. Then level four is clinical decision support, but with FDA approval, because decisions are made that cannot really be explained,” he added.
At some point, one could even imagine a level five of autonomous healthcare, which would mean taking doctors out of the room. But Lebrun remains very cautious on this front.
“For some situations in some markets, like in some countries where they don't have access to healthcare, it would be something relevant,” he said. In the long term, he sees the diagnosis process as a “pattern matching problem” that could be solved with ai. Doctors would focus on empathy, surgical procedures, and critical decisions.
While Nabla is based in France, the majority of the company's clients are in the US following its implementation at Permanente Medical Group. Nabla is not only a work in progress, but is actively used by thousands of doctors every day.
The Nabla Privacy Model
Nabla is currently available as a web app or as a Google Chrome extension. The company is very aware that it handles sensitive data. That is why it does not store audio or medical notes on its servers, unless both the doctor and the patient give their consent.
Nabla focuses on data processing rather than data storage. After a consultation, the audio file is discarded and the transcript is stored in the EHR that doctors already use for their patient files.
In more technical terms, when a doctor starts a recording, the audio is transcribed in real time using an optimized speech-to-text API. The company uses a combination of a commercially available speech-to-text API from Microsoft Azure and its own speech-to-text model (a refined model based on the open source Whisper model).
“When you have just a normal speech-to-text algorithm, they may or may not be good for medical data. But we have one tuned. And, as you've probably seen, the text is very light at first and then turns dark. And when it gets dark, it means that we check it with our own model and correct it with names of drugs or medical conditions,” Nabla ML engineer Grégoire Retourné said during the demonstration I saw.
The transcript is first pseudonymized, meaning that personally identifiable information is replaced with variables. Pseudonymous transcripts are processed using a large language model. Historically, Nabla has been using GPT-3 and then GPT-4 as its main large language model. As an enterprise customer, Nabla can tell OpenAI that it cannot store its data and train its large language model on those queries.
But Nabla has also been playing with a refined version of Llama 2. “In the future, we plan to use increasingly narrower models instead of general models,” Lebrun said.
Once the LLM has processed the transcript, Nabla removes the pseudonym from the result. Doctors can view the note, which is stored on the computer in the local web browser's archive file. Notes can be exported to EHR.
However, doctors can give approval and ask for patient consent to share medical notes with Nabla so they can be used to correct transcription errors. And since Nabla is on track to process more than 3 million queries per year in three languages, Nabla is likely to improve very quickly thanks to real-world data.
Image credits: Romain Dillet/TechCrunch
