In this episode, I talk with Dean Freestone, co-founder and CEO of Seer, about epilepsy. Seer uses home monitoring to diagnose and manage neurological conditions, relieving bottlenecks in the healthcare system.

Highlights:

• Using machine learning to summarize data to reduce the labor intensive search for episodic events.
• Handling imbalanced datasets.
• Controlling the workflows to enable gathering and annotating huge datasets.
• Working with technicians to speed up review of EEG data.
• Using machine learning to capture features that doctors can’t describe.
• Dealing with low inter-reviewer agreement from clinicians.
• How bias can manifest is neurological data.
• Do not underestimate the cost and amount of work to build a healthcare AI startup.

Quotes:

“We specialize in home clinical monitoring with a special focus on helping people with seizures and blackouts and attacks related to epilepsy.”

“In order to do this well in the home environment, we need to work on the technology stack. And so that involves computing at the edge. So we’re talking about wearable devices, mobile networks, cloud computing, and AI, and bringing these things all together to create really large infrastructure to be able to support getting data from the home and environment to the people who need to see it in the clinical environments and look in home.”

“Epilepsy and these episodic disorders like epilepsy really benefit from machine learning and maybe the most out of any particular condition. And that’s because seizures can be quite episodic, as I said before, and a little bit intermittent. And that means that the symptoms that we need to record and examine can be like looking for a needle and a haystack. And so it can be very labor intensive for somebody to read the data. If we can use machine learning to try and just reduce the data or, or edit it down, just something like a highlight reel, that means that we can be a lot more efficient, a lot more accurate and effective, and we can scale up our systems so that we can make it more accessible to folks that wouldn’t normally get.”

“In reality, in real life, the training sets are anything but balanced. And so it’s very hard to assess the performance to these algorithms once they’re deployed out into a real clinical environment. And it means that we need vast amounts of data as well to be able to get to a level of performance that’s starting to get to and surpass human skill.”

“At Seer in Australia, we monitor more patients for epilepsy than every hospital in Australia combined together and then tripled.”

“The second I left that room, nobody would use the algorithms and it become apparent to me that without me being there, they just didn’t trust it. They didn’t know how to use it and added an extra degree of risk and added a small degree of time to get over that learning curve to understand what was happening. And so when we started Seer, we had banked on this insight and instead of trying to push our technology out for external people to use in the clinical world, we took the approach how about if you guys send your patients our way and we’ll take care of the heavy lifting in terms of using the technologies.”

“We want these systems to scale up globally and build planetary scale infrastructure for healthcare. How is that going to translate to people living in Africa or India or other areas around the world where we really do need to help these people?”

“There’s a lot of parallels with autonomous driving and it’s really interesting to see that the rate of incidents with autonomous vehicles may be lower than what there would be with a person at the helm of a car. But still any incident is just absolutely unacceptable. And I think about this a lot and the parallels to healthcare. I think that already we can outperform the standard medical system in a lot of cases. But still any incident when we’re using AI in healthcare is still unacceptable. And perhaps one of the reasons why this is the case is that when a person is at a helm it’s considered an accident. And when an AI or a algorithm is at the helm, it’s considered a systematic bias within the procedure. And that’s unacceptable. And so we need to be really conscious of that.”

“The biggest piece of advice I would give to anyone starting up right now is absolutely do not underestimate the cost and the amount of work it’s going to take. It’s not your typical sort of SaS world or social networks and things like that. Healthcare is a lot more rigorous. It takes a lot more capital, a lot more time, and a lot more people to do it well and do it in a safe and secure way and get all the regulatory compliances that we need. And so do not, absolutely do not, under finance.”

Links:

• Seer’s website
• Seer on LinkedIn
• Dean Freestone on LinkedIn