By David Talby, CTO, John Snow Labs
Healthcare organizations’ interest in artificial intelligence (AI) has given rise to technologies and tools to help deploy machine learning products and services that achieve everything from streamlining processes to saving lives. But despite significant strides in AI adoption in recent years, healthcare differs quite a bit from other industries in its approach. Strict regulatory and security guidelines require tight controls and processes in place for working with data and models regarding sensitive personal information.
While this has caused some barriers, it’s also helped create a culture of careful innovation that other industries should learn from. Healthcare has been at the forefront in many areas that are now collectively referred to as Responsible AI, a governance framework that addresses and documents the ethical and legal challenges of AI. The healthcare, biotechnology, and pharmaceutical industries have long had regulations governing data privacy as well as guidelines for deploying AI and machine learning.
This has not hindered innovation in the field. Rather, a lot can be learned from how healthcare organizations have evolved their approach to AI and a new industry survey from John Snow Labs and Gradient Flow titled, “The 2021 AI in Healthcare Survey Report,” aims to shed light on this. The global survey seeks to understand more about the current trends and approaches healthcare organizations are taking with AI. According to the data, here are some of the top trends to watch.
NLP is Gaining Steam, Even in its Infancy
Natural Language Processing (NLP), along with business intelligence (BI) and data integration, have topped the list of foundational healthcare AI technologies. While text mining approaches to build AI applications are nothing new, NLP provides the contractual and regulatory support specific to healthcare that make it extremely valuable. For example, in the past year, there has been a cascade of research papers, news articles, social media, and other literature about the COVID-19 pandemic. This volume of information extends far beyond human’s abilities to make sense of it all. There is also a need to identify and understand harmful disinformation surrounding the topic and how it propagates. NLP can shed light on these information gaps that no other technology can.
Beyond the pandemic and research, NLP is also being used in more practical ways, such as painting a more holistic view of the patient. From provider notes, to electronic health records, to diagnostic imaging, NLP can link them, cutting through the noise to create a comprehensive, longitudinal, and personal view of each patient. Between NLP, BI, and data integration, it’s clear that healthcare organizations are motivated to put their troves of data to work.
AI Users Shift from Data Scientists to Clinicians and Patients
Outside of consumer applications that use AI-enabled technology on the backend, it would seem that enterprise AI is reserved for data scientists and IT professionals. However, a majority of survey respondents from mature organizations—referring to those that have had AI models in production for at least two years—stated that clinicians and healthcare providers were users of their AI technologies. In fact, over half of respondents (59%) from mature organizations indicated that patients were also users of their AI technologies.
This is interesting and exciting to see, as it points to how increasingly automated processes, like a patient interfacing with a chatbot before being directed to a provider, can free up resources for healthcare organizations and streamline the information gathering process for users. Back to responsible healthcare AI solutions, it also indicates a level of trust in the technology—both in terms of accuracy and user experience —that will only democratize the technology further in the coming years.
Early Adopters and Mature Organizations Differ in their Approach to AI
It’s clear AI is demonstrating both its value and usability in a healthcare setting, but not all organizations are approaching it the same, and that has a lot to do with where they are in their journey. One example of this is how respondents validate their AI and machine learning models. A majority of those with experience deploying models to production choose to rely on their own data rather than on third-party or software vendors’ data. In comparison, 31% of companies that are still in the exploratory phase are more open relying on evaluation from software vendors.
Another difference is how organizations evaluated technologies, tools, and services. While factors such as state-of-the-art accuracy, no data sharing with vendors, and scalability were top priorities for all respondents, technical leaders valued no data sharing with vendors and ability to train their own models more than generalists. This implies that those further along are keeping secure practices and customization to fit their needs, while novices are more open to using other resources to get their AI initiatives running—and although different, both will further progress in the field.
With many machine learning technologies still in early stages of development and production, it’s amazing how healthcare has been able to apply AI to improve processes, patient care, and save headaches and costs in a number of areas. Even in the wake of the global pandemic, the growth of AI hasn’t slowed, and either has innovation. In fact, some might argue 2020 has even accelerated AI adoption, or at the very least, proven what it’s capable of. Healthcare is on the forefront of this new and exciting technology, and I look forward to seeing where it’s headed this year and beyond.