How can we help?
Avoid blind spots in your market. Get the insights you need to accelerate innovation.
2019 was another strong year for machine learning and artificial intelligence within the diagnostic space, and the acceleration of progress in ML / AI is clearly going to continue within healthcare. We expect that at least 3 major trends from 2019 will continue to shape the space in 2020 and beyond. 1) The conversation will continue to shift away from the total replacement of providers within fields such as radiology, and towards programs that augment the abilities of healthcare providers within the clinical setting. 2) Big tech will continue to influence and disrupt the space through multiple routes including in-house product development, partnerships and acquisitions. 3) Both governmental and medical bodies will increase their involvement in providing both guidance and regulation to the use of ML and AI in healthcare.
While publications on algorithms outperforming radiologists such as Google’s use of an AI system for breast cancer screening1 are becoming relatively frequent, experts are acknowledging that complete automation is unlikely for multiple reasons. First, these algorithms are incapable of performing all analyses that are regularly performed by physicians. While algorithms excel at specific tasks that they are trained to perform (e.g., differentiating between healthy cells and cancerous cells, or identifying pneumothorax on an x-ray), their design may result in them missing key findings that an experienced healthcare provider would be able to detect. There are thousands of potential findings that can be identified by radiology, and some may be so rare or novel that algorithms either improperly classify the findings or fail to recognize them at all. Second, algorithms are often seen as “black boxes,” where it is difficult to identify “why” the algorithm classifies results into certain categories. One of the critical responsibilities of physicians involved in the interpretation of medical images in both radiology and pathology is providing additional information to physicians should there be questions regarding the interpretation of the patient’s results. Organizations as well as providers may be reluctant to adopt programs to replace these experts if the program is unable to explain “why” it diagnosed a patient in a specific manner. Third, multiple factors influence the ability of an algorithm to interpret results. Factors such as image capture by different machines and / or operators within institutions can result in image variability that may be misinterpreted as a diagnostic factor by ML / AI algorithms. Publications such as NYU’s study on identifying breast cancer from mammograms2 showcases a more likely near-term future in which physicians reap the benefits of combining their expertise with advanced analytics to improve their overall accuracy and the quality of care they provide to their patients.
Amazon continues to build out and improve their AWS offerings for institutions such as Amazon Textract3, and their medical transcription service4. In parallel, Haven5, the joint healthcare venture between Amazon, Berkshire Hathaway and JPMorgan Chase launched their website earlier this year, and more developments are expected to come in the future regarding their plans. Additionally, AWS continues to work with organizations such as the Pittsburgh Health Data Alliance6 through machine learning research sponsorships to advance innovation in multiple areas including cancer diagnostics, precision medicine, voice-enabled technologies and medical imaging. Google continues to heavily integrate into organizations including Mayo Clinic7 and Ascension8 via strategic partnerships, moving their documents to the cloud in order to improve care and provide advanced analytics to improve organizational efficiency. Google will also be offering Suki’s digital clinical assistant technology9 to their AI / ML and cloud computing healthcare products. While Suki’s technology currently focuses on administrative tasks (e.g., documentation, information retrieval from EHRs), this product may impact entire healthcare systems by saving providers time and reducing their daily administrative burdens. Google also continue to be involved in the development of ML / AI tools for medical diagnostics. They published articles on SMILY10 (similar image search for histopathology), a tool that can be utilized to search for similar images within archives to help providers identify and classify complex cases as they encounter them as well as deep learning models to detect pneumothorax, nodules and masses, airspace opacities, and fractures within chest x-rays11 that achieved an overall expert-level accuracy. Other organizations including NVIDIA, Apple, Facebook and more are all getting involved in various facets of the space as well, and it is likely that the competition will continue to heat up over the upcoming decade.
While some regulatory bodies may have been skeptical of ML / AI in the past, organizations are accepting that it is here to stay and exploring the best ways to incorporate it into standard clinical practice. Organizations such as the Canadian Medical Protective Association12 have begun to support the appropriate use of AI in healthcare. Additionally, organizations such as the American Society for Gastrointestinal Endoscopy13 have included certain ML / AI devices within their practice guidelines, and the American College of Radiology14 has launched a list of FDA-cleared AI algorithms to enable easier access for radiologists and developers. Finally, the FDA15 has released discussion papers and requests for feedback to improve their understanding and ability to provide guidance on these tools, and continues to hold public workshops16 to discuss the evolving role of artificial intelligence in healthcare.
In the near-term, ML / AI is expected to become an invaluable tool that will be incorporated into the toolbox of providers to improve and enhance their abilities to provide care to patients. ML / AI is going to continue to increase in its importance within the healthcare space, and we look forward to seeing the impact of these innovations in the upcoming decade.
References / Notes
Seth is an Associate at DeciBio with experience in identifying novel opportunities within ML & AI, molecular diagnostics, immuno-oncology and research tools. He is passionate about supporting disruptive technologies that can be used in the clinic, and is the curator of the Big Data & AI weekly newsletter. Connect with him on LinkedIn or email him at [email protected]
Disclaimer: Companies listed above may be DeciBio clients and/or customers