UK Medtech start-up Babylon Health recently developed an artificial intelligence service which passed England’s general practitioner exam with a score of 81 percent, beating the average score achieved by all graduating medical students by 9 points.
The success, while preliminary, raises questions about the role AI can play in the healthcare industry, as well the direction of its inevitable evolution on a global scale.
How has medicine already been impacted by AI?
Contrary to what many might think, artificial intelligence has already been integrating itself into the medical industry for the better part of the decade.
As early as 2015, Enlitic, a Bay Area startup, sent out 80 technicians to imaging centers across Asia and Australia to test a deep learning algorithm compatible with currently existing software. They hoped to develop this technology to be capable of detecting disease across all major imaging platforms (MRI, CT scan, ultrasound, X-Ray) with near-perfect accuracy. The contributions Enlitic and similar companies have made towards enhancing patient outcomes have proven a real benefit to the radiology industry, which is expected to increase in efficacy as AI technology continues to improve.
In Finland, the cloud computing software, Aiforia, can perform laborious image analysis tasks in a fraction of the time traditional methods would take, accurately identifying which images correspond to diseased tissue at a rate equivalent to a qualified observer. Use of this AI-driven software could free up precious time for doctors to engage in more specialized work such as analyzing rare tissue samples or developing multifaceted care regimens for individual patients.
Aiforia is already in use by some 6,000 pathologists, researchers, and pharmaceutical R&D teams to manage and share digital slide collections. Its creators––senior healthcare scientist Anna Knuuttila and CEO of Fimmic Kaisa Helminen––plan to expand its capabilities even further still.
How can we expect the role of AI to grow in MedTech?
The first significant trend we can (likely) expect to see is a more extensive adoption of patient-interfacing software.
In the past, this notion seemed like a moonshot given computational limitations, however with rapidly developing IoT technology, it’s not inconceivable that one day the initial screening process for medical services will be handled entirely by AI programs. The increasing popularity of self-diagnostic sites such as WebMD suggests that users are becoming accustomed to entrusting a rudimentary algorithm (i.e., a machine) to receive expedient medical advice, regardless of accuracy.
Supplemental research also establishes that global support for diagnostic chatbots among millennial parents is currently at 60%––a promising outlook for the future of med-based AI software.
Most experts predict AI will enjoy “widespread use” throughout all fields and specialties by 2023. They base their observation not only on the tremendous potential to hone and develop existing software within that time frame but also on the already widespread use of AI MedTech by the clinical and research spaces, which often function as bellwethers for the adoption of new technology within the medical space as a whole.
Still, there are challenges to such widespread proliferation. And, in fact, the most pressing one can be found by returning to a brief consideration of AI’s role in aiding pathologists and diagnosticians.
AI needs to be trained in order to do its job. While deep learning machines can largely determine how to do that job through trial, analysis, and error, they still require initial configurations to distinguish between relevant information and noise. For many diseases – especially ones with varying, overlapping symptoms and complex causes – no existing machines would even know where to begin.
This issue belies an even larger problem with AI diagnosis. Namely, even machines on the cutting edge of MedTech cannot yet replicate the human intuition of an experienced medical practitioner. To date, this is why human doctors continue to outperform automated detection programs and the reason that even IBM’s Watson – the most well-established, well-supported AI diagnostic aid – has consistently failed to deliver on its promises.
What do You Think?
Do you know of any advantages or disadvantages to healthcare AI which we missed? Maybe you have an idea for a new use of existing technology in the MedTech space?
Let us know! We’d love to hear from you in the comments below!