The global pandemic has raised awareness of the importance of our health and the fragility of healthcare systems around the world. It has become apparent how archaic many of our health processes are. Healthcare technology is at a point of convergence, where artificial intelligence is being incorporated into all areas of the industry. With the influx of big data, AI has the opportunity to reshape the practice of healthcare. Its applications have the potential to improve existing technologies, ignite personalized medicine, and benefit historically-underserved populations.
According to The Harvard Gazette, medical errors kill an estimated 200,000 people and cost $1.9 billion annually. Without major structural change, healthcare systems will struggle to remain sustainable. So how does AI present a solution?
Healthcare and machine learning are in many respects uniquely complementary. Medicine is largely based on pattern recognition, an activity that machine learning excels at given increasing data. An example of a machine learning application is to detect deviations from normal bodily functions, represented as laboratory values or reported symptoms, across a population of users. Such telemetry can enable providers to predict health trends and aid in accurate diagnosis and prescribed treatments, which leads to better care outcomes and improves the efficiency of care delivery. But, in order to accomplish this, AI algorithms must first organize and make sense of the data available so that they can be trained properly.
Analyzing and interpreting medical images is a highly specialized skill that requires a qualified practitioner to go through extensive and rigorous training and practice. Similarly, AI applications require extensive training to recognize patterns and to then productively correlate new patterns with previously seen medical image data and corresponding outcomes. To achieve this, AI applications and algorithms need access to correctly labeled medical data so they can optimally utilize existing information when analyzing new scenarios. Simply put, accurate data labels are a key pre-requisite for AI applications and algorithms to function; and the better the quality of the data labels associated with the AI training datasets, the better the performance of the resulting AI program.
Centaur Labs, one of Omega Venture Partners’ portfolio companies, is innovating a transformative solution to solve for data labeling across healthcare AI applications. Centaur Labs was founded by Erik Duhaime, CEO, while he was a PhD student of OMEGA’s Thomas Malone at MIT. Over 30% of the world’s data is generated by the healthcare sector but most is unstructured or poorly labeled. Centaur seeks to improve the quality of health data used to train AI algorithms and transform the tools and workflows needed to place high quality data at the center of developers’ efforts.
“Omega Venture Partners’ track record, acumen, and high-impact network make them a very attractive partner for us. Omega has a nuanced understanding of how AI stands to impact digital healthcare, and they have added immediate value through prospective customer introductions.” – Erik Duhaime, CEO and Founder, Centaur Diagnostics, Inc
In his experiments, Erik asked people of various expertise, even novices, to annotate medical images. He discovered that by intelligently combining their opinions, he could get extraordinarily accurate results. Duhaime states, “To create real, big impacts, we need scalable human intelligence – the human intelligence to accurately annotate millions of medical images, and this is what we are providing at Centaur Labs.”
To train AI algorithms to properly and accurately label complex medical data, Centaur pioneered DiagnosUs, a digital community and platform that collects the opinions of medical professionals from 140 countries and blends them with artificial intelligence and machine learning recommendations—this results in data labels that are far more accurate and higher quality than those of either experts or computers working alone. The platform takes advantage of the power of collective intelligence—leveraging AI to augment and amplify human potential as well leveraging digital workflows and clever incentives to do so with speed and volume.
Advances in medical data labeling have the potential to improve healthcare delivery worldwide, turbo-charging the development of more advanced and more accurate AI solutions. Having access to this technology not only improves the state of the art in digital healthcare but also has the potential to reduce persistent health inequities. Even hospitals with limited resources can have access to expert opinions from all specialties. Even the most obscure medical cases have a better chance of being solved.
Indeed, what inspires us at Omega Venture Partners about the intersection of AI and digital health is its immense potential to democratize healthcare access and improve healthcare outcomes to benefit society at large.