Enes Hoşgör, Ph.D., CEO at Gesund writes:
Widely held visions of the future see artificial intelligence (AI) driving cars, working alongside humans in factories, and even flipping burgers. Arguably, the most impactful vision of AI is in healthcare where it holds tremendous potential to greatly improve human health.”
Indeed, AI is already at work in healthcare use cases like lifestyle/health monitoring and even drug research and is expected to grow at 48% annualized between 2017 and 2023. But the true promise of AI is in expanding clinical-grade applications like radiology – already showing potential with some early steps – expanding access to value-based care and precision medicine.
In spite of AI’s vast potential, a critical limiting factor to the still-nascent technology, according to a recent report, is the risk of injuries and errors resulting from poor access to diverse and equitable data sources needed to train and validate algorithms. In the face of this bottleneck, the industry must establish a superhighway to speed access to the necessary data coupled with an ML tool chest and provide scalable development and validation pathways like those in other areas of clinical treatment.
AI at Work
Physicians and scientists have long advocated for healthy lifestyle choices as a way to improve health outcomes and even lengthen patients’ lives. The constant drumbeat around prevention as the best route to good health helps explain the mainstream popularity of health trackers, like FitBit and Apple Watch. Powered by sophisticated algorithms, this class of gadgets can track baseline physiological data like pulse rates and ECG measurements, as well as a broader set of data that includes brain activity, glucose and sleep.
There’s debate around the accuracy of fitness trackers but they’re widely considered good enough for the average fitness-focused consumer tiring to stay active and healthy. It’s also important to note that since more data typically improves the accuracy of an algorithm, fitness trackers are likely to improve and expand their usefulness in helping to keep people healthy.
A much more advanced example of AI currently at work helping to improve human health is in the research and development of new chemicals and drugs. Companies like Schrödinger and Exscientia are using AI to model new chemicals at the molecular level to help pharmaceutical companies drastically reduce the time and cost of developing new drugs, with some already in clinical trials.
AI’s Long Road
The most exciting and impactful applications of AI are clinical-grade use cases that promise benefits including the improvement of the timeliness and accuracy of diagnostics, expansion of access to care and delivery of care precisely tailored to individuals at scale.
The progress of AI in diagnostic areas – radiology in particular – is impressive and represents the widest application of the technology in a clinical setting so far. About 30% of radiologists are currently using AI in their practice with 20% of those not currently using it planning to purchase AI tools in the next few years, according to a recent survey.
Imaging data’s relative uniformity is a good fit for AI and companies like Gleamer and Sirona Medical are seeing early success in helping radiologists cope with workloads that doubled over the last two decades.
Of course, increasing workloads are not unique to radiologists. The Black Swan effects of the global pandemic aside, experts project a strong increase in healthcare demand and workloads – particularly in the U.S. – as the population ages.
AI-enabled process automation offers a way to cope with surging demand and workloads by helping to streamline both administrative and clinical workflows freeing up resources to handle more patients.
Major healthcare providers are already moving ahead with automation pilots and programs with 89% of healthcare executives indicating that the approach is creating efficiencies in their systems and 91% saying that it’s increasing access to care.
As AI expands access to healthcare it also holds “unprecedented opportunities” to customize treatments and provide precision medicine at scale by combining personalized medical details like clinical history and symptoms with other social, psychological and socio-economic data that exists outside of the medical system. Experts say that over the average person’s life they generate the equivalent of 300 million books packed with personal data that could provide health-improving insights.
It’s exciting to consider the promise of such broad applications but AI’s current road to widespread clinical use is frustratingly blocked by a lack of access to data needed to both train and validate their algorithms.
Building a Superhighway to Better AI
The brains behind AI applications are algorithms – complex mathematical equations – that pick through vast amounts of data with superhuman speed and accuracy to deliver analytical insights. However, before an algorithm starts churning out insights it must be trained just like human intelligence, which requires diverse and copious data. The more and more diverse the data, the better the insights.
In a high-stakes, highly regulated clinical setting, the accuracy and safety of an algorithm must be validated, which also requires diverse and copious data. But there’s a mismatch between the data suppliers – hospitals and clinical providers – and AI developers.
Hospital IT systems are high-compliance environments that don’t fit with the often cloud-based, third-party services – referred to as MLOps stacks – required to build and train algorithms.
The tools used by AI developers in other industries simply don’t work in the healthcare industry, which makes building clinical-grade AI immensely less efficient and much more costly. Without the appropriate technology stack available to facilitate compliant and ML-friendly data access, AI companies have thus far resorted to anonymized data dumps that are not aligned with the broader lifecycle of an AI algorithm.
The dichotomy between hospitals as data custodians and technology companies as algorithm developers has underscored the need for a new infrastructure – a superhighway – that can orchestrate this multi-stakeholder collaboration in a secure and meaningful fashion.
On the validation side, traditional healthcare-related sectors like pharmaceuticals and medical devices make use of contract research organizations (CROs) that gather and manage the data needed to conduct clinical trials and validate safety and efficacy on the way to regulatory clearance from the FDA.
CROs don’t yet exist for medical algorithms, which leaves AI developers and regulatory bodies lacking the appropriate tools and data access for evaluation purposes.
Solving this two-tiered problem requires a platform – like the one we’re building at Gesund.ai – that untangles the bottleneck of limited data by connecting curated and diverse data sets governed by an ML toolbox to innovative companies in a HIPAA-compliant fashion so they can more quickly and efficiently build and validate algorithms.
It’s still early days for clinical-grade AI but the promise of the technology is profound for both healthcare as an industry and human health in general. For AI to mature toward fulfilling its promise, however, we must build a superhighway that removes the gridlock restricting access to the data that’s critical for creating effective and safe applications.
Editor’s Note: Enes Hoşgör, Ph.D. is an engineer by training and an AI entrepreneur by trade-driven to unlock scientific and technological breakthroughs having built AI products and companies in the last 10+ years in high compliance environments. After selling his first ML company based on his Ph.D. work at Carnegie Mellon University, he joined a digital surgery company named caresyntax to found and leads its ML division. His penchant for healthcare comes from his family of physicians including his late father, sister and wife. Formerly a Fulbright Scholar at the University of Texas at Austin, some of his published scientific work can be found in Medical Image Analysis; International Journal of Computer Assisted Radiology and Surgery; Nature Scientific Reports and British Journal of Surgery, among other peer-reviewed outlets.