Subscribe to learn more about this topic
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Ultrasound AI for Sleep Apnea Detection

AI can predict the risk of Apnea from a single ultrasound, so patients no longer need 24 hours in a sleep lab

Markus Schmitt
Markus Schmitt

Sleep apnea affects nearly a billion people, causing them to stop breathing in their sleep. Sometimes for minutes at a time. It’s connected to Alzheimer’s disease and affects up to 30% of elderly people.

Common as it is, sleep apnea is very difficult to diagnose. Patients generally need to spend at least one night at a “sleep lab”, where doctors analyse them while they sleep. As the day-to-day symptoms are relatively mild (snoring and disturbed sleep), many people simply live with it in ignorance.

We chatted to James Lee from AmCad about how they are pioneering AI on ultrasound images for the early detection of obstructive sleep apnea (OSA). AmCad’s solution can detect OSA in as little as 10 minutes, while the patient is awake.

James spoke about:

  • What the challenges are when selling to doctors;
  • How AmCad stays relevant in an ever-evolving market;
  • Why their competitors often fail.

Selling AI solutions to doctors

Telling doctors they need help from partially automated solutions isn’t easy. Doctors are usually confident in their own skills, having undergone extensive training to make accurate diagnoses. On one hand, they don’t trust AI to make decisions they make themselves; and on the other hand, they’re reluctant to support something that could one day replace them.

AmCad takes special effort in it’s positioning, emphasizing that it’s tools are to support doctors, not replace them. Helping doctors appreciate a distinction between support and replacement is hard, especially for those unfamiliar with AI tooling in practice, and who know it only as a higher level concept.

Automating boring tasks, like reporting, is one selling point that doctors find attractive, but James also demonstrates how AI-powered tools can help in other ways. Doctors don’t need to strain their eyes looking for tiny signals in an image when an AI can quickly highlight potentially interesting regions, for example.

A clinical team that works with doctors, experiencing their frustrations and needs first-hand, is vital to ensuring the solution aligns with what doctors want. And having doctors experience a solution and see how it can help them goes a long way to winning them over to the value of AI-powered solutions.

Staying relevant to the market 

While AmCad’s team includes doctors who are closely involved in envisioning new products, it’s still hard to be confident that other doctors will appreciate the product.

And because approval processes can take several years, keeping up with a changing market is challenging.

For AmCad, a lot of the process involves being out in the field and sitting down with clinicians: not just for events, but in their workspaces, observing their workflows, and making sure that the R&D people are on the same page.

A lot of similar companies fail at the key stages

For an AI company to succeed in a medical field, it needs to clear three distinct phases.

  • Phase 1 – R&D and vision: requiring one or several brilliant minds who each have a unique outlook on a specific problem.
  • Phase 2 – FDA approval: depending on someone to deal with bureaucracies and handle paperwork.
  • Phase 3 – Sales and marketing: ensuring people know about the solution and are willing to buy it.

James emphasizes the rarity of the same people being good at all three stages. The brilliant scientists key to a successful phase 1 often have little interest in phase 2. And the people who excel with the long and slow approval process are not always the best at generating excitement about the product.

This means that team effort is very important; that no single person or team puts their ego first or thinks they’re best placed to dictate the entire process. Instead, each person needs to know their role and understand the right time to speak up.

In James and AmCad’s own experience, this also meant that different stages of the company’s development made sense for people with different skill sets to be in charge. At the beginning, the company was completely managed by people with deep experience in R&D. Only once they went to market did the owner (who has a background in corporate finance) take over as general manager.

Most companies fail at the FDA stage – phase 2. AmCad was the first company to get an FDA-approved ultrasonic solution to market, and this head start has contributed to its success.

Once you succeed with one product, it’s easier to expand into others

It took AmCad seven years from concept to production for their first medical AI product. This was a solution to detect Thyroid nodules, the most common thyroid disease, and James acknowledges that it wasn’t easy.

But from here, AmCad was able to take its learnings to deliver five other ultrasound products. Initial experiences in dealing with the FDA process, working in clinics to ensure product-market fit, and iterating on a solution allowed the team to accelerate the process for the following products.

Are you working on medical diagnostics?

We’re a machine learning agency that specialises in machine learning for medical diagnostics. If you’re working on similar problems, we’d love to hear from you. Contact us.

Get Notified of New Articles

Leave your email to get our weekly newsletter.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.