AI is yet to revolutionize sleep, but it is happening soon…
Unless you are living under a rock, you must have heard of AI or Artificial Intelligence. More specifically machine learning — a suite of technologies that allow machines to learn skills on their own using data. From your Tik Tok feed to autonomous cars, AI plays a major role in every aspect of our lives. Arguably, one of the most profound impacts of AI will be on health.
All the hype surrounding AI and machine learning is not new. We have gone through many cycles of hype which ultimately culminated in disappointment and disillusionment. Unlike the last cycles, where AI could solve a small scale toyish problem really well, garnering sudden interest from researchers but ultimately fail to scale up to real-world problems. This time AI not only scaled well it also beat the experts in many areas. In less than a decade, these AI systems outperform the experts in diagnosing breast cancer, reading a chest X-ray, diagnosing diabetic retinopathy and reading a brain MRI or CT-Scan.
Despite all the progress in various aspects of health, the impact of AI on sleep both from a consumer point of view as well as a clinical point of view has been limited. On the consumer front, there was a huge hype around wearables and sleep trackers which turned out to be a major disappointment. Not only are the trackers inaccurate and unreliable, but they also fail to provide any actionable insights to actually have an impact. After all, saying that you are not getting enough deep-sleep is interesting but useless information unless there is a way you can actionably increase it.
On the clinical front, things are not any better. Take for example sleep apnea, a type of sleep-disordered breathing that affects nearly a billion people worldwide. Although treatments are effective, only a tiny minority are even diagnosed. This is shocking, to say the least, as this is not a benign ailment. Severe apnea can seriously affect the quality of life, reduce lifespan by over 20 years and increase the chances of accidents by 11 times! The only thing benign about the disease is that it is sometimes accompanied by loud snores. Ironically, that is the only symptom that your partner notices and nudges you to go see a doctor.
You might be quick to suggest that this is an issue of awareness. There is no denying that for a long time sleep has been a neglected part of our life. While scientific progress in the understanding of Sleep and books like Why we sleep? by Matthew Walker and The Sleep Revolution by Arianna Huffington are changing public perceptions. I would argue that this is a failure of technology. After all, you do not need to know about a disease to diagnose it early. How many terms and measurements did you know in your regular health check-up? The real barrier to scalable, affordable and accessible sleep diagnosis is the current state of sleep test itself. All of you unlucky (or perhaps lucky) fellow who underwent a sleep study would know what I am talking about.
If I tell you that you have to be hooked to 20 electrodes all across your body, with two belts across your chest, a nose cannula, and a finger clip. Furthermore, you will have to sleep in a new environment with people watching you and video cameras recording all your moves. Will you agree to the test? Ohh…I forgot; the test costs 1500$ and you may have to wait a few months before you get a slot. If you fail to sleep on D day or you catch a cold you will have to repeat the study or reschedule it. This is ridiculous.
Nonetheless, being an engineer working at the junction of AI and Sleep, I also realized why this has been the case. To reap the advantages of AI, enormous amounts of high-quality data is a necessity. Until a few years back, this was a major challenge. Add to this the fact that sleep itself is not very well understood, with scientists coming up with new guidelines to measure them every few years (the last update by the American Academy of Sleep Medicine was on 2017). Consequently, the engineers who know AI lack a firm understanding of sleep and the doctors who understand sleep lack a comprehensive understanding of AI. Finally, there is a lot of skepticism within the Doctors as well. They have burned their hands trying out technologies delivered by the previous AI hype cycles which failed to meet expectations.
This brings me to the future, which I strongly believe is going to be profound. Initiatives like the national sleep research resource (NSRR) are making access to data easier (we still have a long way to go), sparking new collaborations between engineers and doctors. And every problem is an opportunity in disguise. The huge problems surrounding sleep is creating huge opportunities for disruption. Many people and organizations are realizing this opportunity and building exciting technologies to solve the problem. Two years back I too left my job as a post-doc at Duke-NUS Medical School. I joined forces with my co-founder Kishan and started Neurobit with a mission to make high-quality sleep health accessible and affordable for everyone. While the journey has been arduous with most people and investors being skeptical, all you need is a few people who believe in your vision and a few early adopters. In the next section, I will paint a picture of the future based on my expertise in both Sleep and AI as well as the work that we are doing within Neurobit to be a part of this future.
Doctors and Sleep Technologists will spend less time with data and more time with patients.
Even today the vast majority of clinical sleep data collected is analyzed manually or outsourced to developing countries. Manual sleep scoring is a complex process that even well-trained experts spend 1 to 2 hours on for one night of data. As the technologists are involved in conducting the sleep test and interacting with the patient throughout the night, manual scoring presents an undue burden on them. Consequently, technologists are consistently overworked and are forced to spend less time with patients. Some sleep clinics choose to cut corners which negatively affects the quality of care. Fortunately, this area is mature for disruption. In fact, the American Academy of Sleep Medicine recently published a position statement and an associated publication welcoming and setting guidelines and suggestions to incorporate this change. At Neurobit, we have been working in this area for some time now. The biggest challenge is to ensure that the AI truly works at the highest levels of accuracy and reliability across datasets and demographics. Something that previous generations of AI failed to do. Having now worked with over 65 Universities and research labs across the world, our technology has demonstrated its utility processing over 100,000+ hours of clinical-grade sleep data. The seed of the technology came out of Prof Michael Chee’s lab at Duke-NUS Medical School. Over the past two years, at Neurobit, we have now reached a level where it is fully featured and can reduce the workload of the sleep technician by 90%. In-fact we are happy to announce that today on the 13th of March 2020, World Sleep Day, we are opening up access to our cloud-AI sleep scoring product — Z3Score to everyone. We are welcoming pilots as we await FDA clearance for the software.
Clinical grade sleep diagnostics will move out of sleep labs into the hands of the patients.
There is already a rapid shift towards home sleep tests — a reduced version of the sleep study conducted in the lab. Unfortunately, in its current form, they are far from comfortable and far from accurate. They are bulky, costly and more of a confirmation device rather than a screening tool. AI brings new hope in this direction. During our product development and R&D, we noticed that the accuracy and fidelity of our AI systems were barely affected as we reduced the number of channels. To our surprise, we could reduce the form factor of current home sleep tests to a simple chest patch. This is not an isolated observation. Researchers around the world are slowly realizing the power of AI in extracting clinical-grade sleep information from few physiological channels. We are most excited about this technology which we call Pulse, as it has the power to impact the lives of over a billion people.
Treatments will become personalized
Lack of large scale data and the ability to analyze them at scale is limiting the treatments to one size fits all solutions. AI is going to have a major impact on personalized sleep medicine. Soon it will be possible to isolate subject-specific factors that are negatively affecting sleep and attack them in a targetted fashion.
Consumer sleep devices will move beyond measurement to intervention
As clinical-grade devices become more consumer-friendly, it is natural that consumer devices will inch towards clinical-grade. But the real value would not come from measurement, but how these devices take these measurements and act on it, sometimes in real-time. Future consumer sleep devices will be inconspicuous. Embedded inside the bed, hidden on the wall or stitched into your clothes. They will not only measure but control every aspect of your environment — temperature, sound, light, and smell to continuously improve your sleep. If that sounds like science fiction, let me tell you that our technologies are already demonstrating this albeit in an academic setting. Many researchers across the world are using our tech to control sound to improve the depth of deep-sleep and to selectively activate different memories during sleep.
The scientific literature will move away from small cohorts to population-based statistics. Expert definitions to outcome-based statistics.
Due to the limited sharing of data and lack of access to scalable high-quality sleep measurements, most of the scientific literature is limited to small homogenous cohorts. Most of the definitions and recommendations within sleep medicine are driven by experts. This will change as larger population-level data becomes available. AI will then drive data and outcome-based decisions to better inform the academics which will better inform clinicians to achieve value-based sleep healthcare.
TLDR: I am very optimistic about the future of Sleep and Sleep Medicine. There is no doubt in my mind that AI is going to play a major role in it. Even though the field has been lagging compared to other health aspects, when it catches up it is going to revolutionize society. I am hopeful that Neurobit will play a major role in it. I am curious to hear your thoughts and feedback, you can reach me at amiya@neurobit.io
Sleep Well…
