Using Artificial Intelligence in Healthcare

The adoption of artificial intelligence (Al) in healthcare is rising and delivering positive signs in assisting and solving a variety of problems for patients, providers and hospitals. Forbes Magazine reports a 14x increase in Al startups since 2000. With investment in the industry up six-fold, topping out at over $3 billion, spending on Al is likely to surpass all the outlays on EHRs combined. This chapter explores the current uses of Al, e.g., listening devices and Web applications, facial recognition and its use in the exam room, and ethical questions that arise from the use of Al in healthcare. Lastly, we consider how to establish a strategy for implementing Al in your own practice or hospital.

What Is Artificial Intelligence?

First, let’s define artificial intelligence.

■ The essential requirement of Al is intelligence, defined as the ability to acquire and apply knowledge and skills. It is the capacity to interact (speech, vision, motion, manipulation), reason, learn, adapt and think abstractly, as measured by objective criteria, such as test taking. Also, the Al must be capable of adapting to the outcomes or variables on its own.

■ The term artificial intelligence is an umbrella term for machines capable of perception, logic and learning. Today, there are two types of Al:

  • - Machine learning is a form of Al that employs algorithms capable of learning from data to make predictions or decisions; as the machine’s exposure to data increases, the performance capabilities are improved. In some cases, this may be a simple “if, and, do what” programming logic. For example, the “if-then statement” is set as the most basic of all the programing controls. It tells your program to execute a certain code only if a statement is true. For example, the heart monitor will send an alert if blood pressure drops below a certain threshold, but only if the program detects that the monitor is connected to a patient. An example of the program language could look like the following
  • • Alert Monitor Alarm () {

■ // the "if" clause: blood pressure dropping

■ if (connected to patient){

■ // the "then" clause: sound the alarm

A single Al program could have millions of IF-AND-DO-WHAT programing lines of code/scripts for making split-second decisions like how the human brain might work. For example, if you see a person running toward you, your reaction will vary considerably depending on data already stored in your brain. Do you know the person? Do they look excited to see you? Are they running up to you to give you a hug? If all are true, the do what outcome would likely be to stay calm. If the person running toward you looks unfamiliar, angry and is carrying a knife, the do what outcome is to RUN!!!! Should such an event occur, the brain (or Al) quickly responds and adapts based on the incoming data. (See Figure 5.1 for an example of machine learning logic.)

- Deep Learning Al is also based on the IF-AND-DO-WHAT programing concept but will act more like the human brain and adjust behavior based on prior outcomes. Using the same example of a person running toward you, the brain will record a massive amount of data from that event (e.g., location, time, images, facial recognition, etc.), then use this stored information to make better decisions in the future, such as avoiding dark alleys at 2:00 AM. Deep learning uses many-layered neural networks (computer systems based on the human brain and nervous system) to build algorithms that find the most efficient way to perform a task based on vast sets of data. Deep learning will typically improve over time by adding all

A hypothetical example of multilayer perceptron network

Figure 5.1 A hypothetical example of multilayer perceptron network.

past outcomes to the logic for future decisions. (See Figure 5.2.) One major limitation of the deep learning method is the inability to apply a conscience to decision-making. A major concern with the deep learning method is the possibility for bias and prejudice to seep into the algorithms. Bias in Al happens when the data used are unrepresentative of reality, or reflect the existing prejudices of the developer/ programmer. An example of this was recently seen in Al software used to help judges in sentencing criminals based on the probability of the person being a repeat offender. Although Al is becoming better every day, the algorithms we see every day still have a long way to go before being safely applied to the criminal justice system.

While we do not foresee Al replacing all humans, the study of the ethics and risks of machine involvement in patient care, compared to traditional methods, has yet to catch up to the technology. More concerning is the overall impact on our society and the shifts in inequalities that Al is expected to cause. Specifically, Al is expected to eliminate 40% of all repetitive jobs over the next 20 years. Examples include call centers, patient check-in, registration, triage, collections, accounts receivable (AR) follow-up and campus delivery services. In March of 2019, UPS launched a new service using drones to transport blood and other medical supplies between the various buildings at the WakeMed Health and Hospitals medical campus in Raleigh,

An illustration of a deep learning neural network.Source

Figure 5.2 An illustration of a deep learning neural network.Source: 03875065.

North Carolina.1 The speed at which the drones delivered the samples was significantly faster compared to humans, which in this case could mean the difference between life and death. While this use of Al is a positive factor, the initiative instantly eliminated several courier jobs on campus. A few drone developer jobs were created, but the number of jobs eliminated compared to jobs created was far greater. However, expanded technology in any industry can be expected to create similar repercussions. This matter should not be seen as a deterrent from continuing the use of Al in the healthcare industry, as the benefits and capabilities of the technology are just beginning to surface. The possibility of eliminating jobs is always a by-product of innovation and should never be a justification for the status quo. For example, take a look at what Uber has done to the taxicab industry.

< Prev   CONTENTS   Source   Next >