by Leland Brewster11-15-2017
This article is a continuation of one first published in Healthbox’s September Newsletter. If you haven’t already, you can read part one of the article here.
Having looked at the history of AI and machine learning, as well as some of the strategies employed by today’s data scientists, in the first part of this article, we’ll now explore how three companies are using these techniques to transform the healthcare industry. I recently spoke with three companies - TAO Connect, Zebra Medical Vision, and Hindsait - each of whom are using AI and machine learning to power their products, push boundaries, and improve care for patients. While brought to life through unique experiences, each founder largely shared a common perspective on how the field has evolved, the challenges of applying it in healthcare, and what the future might hold.
The Power of Recent Technological Advances
At Healthbox, we’ve seen the number of companies leveraging AI and machine learning as part of their product offerings explode in recent years. To Dr. Sherry Benton, Founder & Chief Science Officer at TAO Connect, this is not a coincidence. At TAO, Dr. Benton has built a platform of digital behavioral health solutions that streamline therapy and improve patient outcomes. She highlighted that the most significant advances in the space are recent, noting: “The tools we use, especially deep learning, to power our Mind Elevator simply weren’t powerful enough five years ago.” Mind Elevator is a companion chatbot that provides feedback to patients on the sentiment of their feelings and helps patients modify them such that they are not distortive or ruminative.
The recent advances in AI and machine learning that Dr. Benton referred to have driven powerful improvements to TAO’s products. Techniques like natural language processing, a foundational component of the Mind Elevator, “were first based on relatively basic linear models to rate individual thoughts or feelings but our ability to help patients continues to explode as we are able to layer in deep learning in a production environment,” Dr. Benton added. As a result of these technological advances, TAO has seen consistent 20% quarter-over-quarter growth and is dedicated to using technology to help millions of patients access the behavioral health care they need in a timely and affordable manner.
The Unique Challenges of Healthcare
In what seems a universal truth, building for healthcare is harder than building for other industries; AI- and machine learning-powered tools are no exception. Necessary, but onerous privacy, regulatory, and security requirements lengthen sales cycles and implementation times. That is, of course, if the company can access and interpret the data in the first place. For Eyal Toledano, Co-Founder & Chief Technology Officer at Zebra Medical Vision, “Messy data is a way of life. Our data comes from hospitals and, especially in healthcare, data contains mistakes; it’s not always labeled, it’s full of free text, and the same data might have two different interpretations.” Zebra, which has developed a diverse portfolio of algorithms that help radiologists interpret imaging studies and diagnose patients, faces the added challenges created when ingesting images. “Neural networks are easy to train on well-centered, low-resolution, same-scale, and normalized images but that’s the opposite of what you find in healthcare. We have to do a very significant amount of work to normalize our imaging data before we can even start using it to help patients,” Mr. Toledano added.
All of this assumes you can access the data in the first place which, with few publicly available datasets and intense security due to HIPAA regulations, is far from a guarantee. Even once the data is pre-processed and Zebra can apply its powerful algorithms, the nature of healthcare means that knowing whether or not the algorithm is ‘right’ becomes a problem in and of itself. Mr. Toledano noted radiology is particularly challenging: “It’s hard enough to get a set of radiologists to agree on specific findings in a single study but it’s even harder to get them to arrive at the same diagnosis repeatedly when shown a series of similar images.” As a result, Zebra holds itself to exacting standards when developing new algorithms such that FDA, hospitals, and clinicians can be confident in their output. With a new flat price of $1 per interpreted scan, Zebra is helping providers manage their ever-increasing workload without compromising quality, despite the challenges operating in healthcare presents.
The Future of Machine Learning & AI in Healthcare
Healthbox has witnessed the steady expansion in the number of problems for which AI and machine learning are being implemented to solve through our consulting engagements with and fund management work on behalf of hospitals and health systems. While still nascent, this toolbox is being turned to repeatedly and ever more often, just as it has in recent years in other industries. Pinaki Dasgupta, Founder & CEO at Hindsait, comes from a background in aviation and defense where predictive analytics have a considerable history and track record: “Modern turbine engines have hundreds of sensors collecting data in real time that can be analyzed to predict, quite accurately, the lifespan of a particular part and the uptime of the engine more broadly; the use of AI is a cost of doing business there.” Hindsait is bringing some of those same principles to healthcare by, among other things, streamlining utilization management for payors.
Hindsait augments expensive and highly-trained human resources with powerful algorithms that can handle the routine work, keeping the experts focused where they are most valuable. Mr. Dasgupta believes the need for such tools will only grow over time, noting “Medical knowledge is predicted to double every 73 days by 2020 and it’s simply not possible to retrain a doctor every 73 days. AI and machine learning have to be tools to help nurses and doctors make more informed decisions that incorporate that latest information.”
While not without setbacks caused by some recent headlines, “the C-suite in healthcare has really started to embrace the reality that AI is no longer a far-fetched dream but something that is generating very real results that they can point to today,” he added. Hindsait has gone all-in on AI and machine learning and has results to show for it, saving millions of dollars for clients and cutting administrative review times from three days to a single day, all with fewer appeals. Mr. Dasgupta postulates: “The use of AI in healthcare is inevitable. There will be roadblocks and stumbles along the way, of course, but I have no doubt these tools will lead to healthier patients, happier doctors, and savings all around.”
AI & Machine Learning Will Transform Healthcare
While each of the three companies is tackling a unique problem in healthcare, each relies heavily on modern machine learning and AI techniques and, in doing so, illustrates the breadth of their applicability. Much as the computer did in the 1980s, the internet in the 1990s, and the smartphone in the aughts, AI and machine learning burst onto the scene in the 2010s. And as all three founders agree, medicine will never be the same.
Have questions about the work any of these companies is doing? Get in touch with them by emailing firstname.lastname@example.org.