Artificial Intelligence In Health Care Is Exploding – What Does It Mean?
The devices, tests, technologies and pharmaceuticals used by the medical profession to diagnose and treat human illness and disease are constantly evolving set of tools. The scientists and researchers behind this evolution are always looking to upgrade current technology and knowledge while thinking about new ideas to better maintain and improve our health and the care needed to live longer, better lives.
While it did not begin with health care, artificial intelligence (AI) is one of those ideas that has taken off in the last several years.
How AI is used to improve health and health care is a varied menu of choices. Driven by the explosion of health care data now available to scientists, medical professionals and technology companies, we have gone well beyond the traditional analytics of crunching the numbers to see trends in disease and illness. Now, in addition to using data to look backwards to see what worked and how it worked, and to develop best practices for the future, AI also allows the world of medicine to use data to provide real time decisions about care.
One of the best and simplest definitions of AI comes from a recent Pew white paper, which defined it as “the ability of a machine to perform a task that mimics human behavior, including problem-solving and learning.”
As it is currently being used, AI is traditionally viewed as being in one of several categories (though there are nuances to each):
- Machine Learning. This is a common form of AI and is much like it sounds. The FDA defines machine learning “as an artificial intelligence technique that can be used to design and train software algorithms to learn from and act on data. Software developers can use machine learning to create an algorithm that is ‘locked’ so that its function does not change, or ‘adaptive’ so its behavior can change over time based on new data. In health care the most common form of machine learning AI is precision medicine or predicting what treatments are likely to be successful based on patient attributes and the treatment context. Other examples of ML include an imaging system that uses algorithms to give diagnostic information for skin cancer in patients, and a smart sensor device that estimates the probability of a heart attack.
- Rule Based Expert System. This is the earliest AI, based on a collection of “if-then” rules in a specific area. This type of AI has its limits, as the more rules that are created the greater the likelihood conflicts will occur, which in turn limits its usefulness. Early use of this form of AI was in clinical decision support and it is still used widely today. Examples include an imaging system using algorithms to provide recommendations on what type of imaging technology a set of symptoms suggest (i.e., X-Ray, CT, PET).
- Natural Language Processing. NLP is much like it sounds. Based on human language, NPL aims to organize, make sense of, and provide useful information to physicians and researchers of data and information that is unorganized, spread across databases and in some cases (physician notes for example) hard to decipher. Examples and use of NLP in health care include the creation, understanding and classification of clinical documentation and published research, the analysis of unstructured clinical notes on patients, the preparation of reports (e.g., on radiology examinations) and the transcription of patient interactions from the clinical setting.
In an article in Health IT Analytics from 2018, that stands up well to scrutiny, AI is further broken down in 12 treatment and diagnostic areas that it is being used in. These are arranged in a best guesstimate of what how AI is being most used today:
- Clinical decision support at the bedside;
- Creating more precise analytics for pathology images;
- Developing the next generation of radiology tools;
- EHR as a reliable risk predictor;
- Bringing intelligence to medical devices and machines;
- Smartphones as diagnostic tools;
- Monitoring health through wearables and personal devices;
- Machine learning to advance the use of immunotherapy for cancer treatment;
- Containing the risks of antibiotic resistance;
- Reduce the burden of EHR use;
- Expanding access to care in underserved or developing regions; and
- Unifying mind and machine through brain-computer interfaces.
As in what should not be a surprise, the development and use of AI in health care is outpacing government and the private sector in oversight and regulation (whether government or self-regulation). This may prove to be the most challenging issue for AI.
At the center of this debate is how best to regulate/oversee AI and its use while allowing for discovery and development that is not held back or stopped before its value can be evaluated. The FDA has written that “medical device manufacturers are using these technologies to innovate their products to better assist health care providers and improve patient care. One of the greatest benefits of AI/ML in software resides in its ability to learn from real-world use and experience, and its capability to improve its performance.” But the FDA has not yet put regulations in place at this point; its focus has primarily been on medical devices, as AI is used in many other applications. And while it may be clear the government understands the promise, there is ongoing scrutiny of AI Therefore, it seems clear that what we need now in the United States is a partnership between the FDA, Congress and the private sector to develop a pathway that will ensure the safety and effectiveness of AI enabled health care technology. The challenges ahead include but are not limited to creating standards and regulations to:
- Limit data bias;
- Create a common understanding of the data used;
- Ensure AI tools are built and trained on large and diverse datasets; and
- Ensure the appropriate deployment of AI (not used for what it is not designed).
We are on a pathway toward great discovery and advancements in the diagnosis and treatment of human diseases. AI is at the beginning of this pathway and is expected to play a crucial role, but like any emerging technology there is much we do not yet know about its value and benefit and to truly use it to our benefit we must get ahead of the technology or we will be forever doomed to trail behind.