Artificial Intelligence in Healthcare: Promise and Reality

Thomas H. Davenport

Artificial intelligence (AI) became a topic of academic research in the 1950s, and, by the 1970s, there were several AI programs for healthcare. These systems, including MYCIN (for diagnosis of blood diseases), CADUCEUS, and INTERNIST-I (the latter two for more general diagnosis in internal medicine), were developed at universities and became useful research tools. They were also heralded as the advance guard of a new approach to medicine.

However, like many Al-based healthcare systems since that time, they remained research tools and were never employed in clinical practice. As Edward Shortliffe, the principal creator of the MYCIN system in the 1970s, wrote thirty or so years later with regard to technology in healthcare, “One might guess that the potential for a ‘revolutionary’ change in how medicine is practiced is at hand, but similar observations have been made for years.”1

This situation persists today. AI in healthcare is primarily found in the research lab, but not at the clinical bedside. There are many research studies suggesting that AI can perform as well or better than humans at key healthcare tasks, such as diagnosing disease. Deep learning neural networks are already outperforming radiologists at spotting malignant tumors, and guiding researchers in how to construct cohorts for costly clinical trials. However, for a variety of reasons, it is likely to be many years before AI replaces or substantially augments humans for broad medical process domains. In this chapter, I describe both the potential that AI offers to automate or augment several different aspects of care, some implications of AI for the healthcare workforce, and some of the barriers to rapid implementation of AI in clinical practice. I conclude with a discussion of ethical concerns and an overview of the future of AI in healthcare.

AI Key Inputs and Healthcare

AI is a collection of technologies that makes use of three key inputs, each of which is relevant to and used within healthcare. Each type of input—leading to different types of processing and outputs—is described below.

Statistical Inputs

Machine learning fits statistical models to structured numerical data. It “learns” by training models with data having known outcomes, and then applies the resulting models to predict or classify unknown outcomes. Machine learning is one of the most common forms of AI; in a 2018 Deloitte survey of 1100 U.S. managers whose organizations were already pursuing AI, 63% of companies surveyed were employing machine learning in their businesses.2 It is a broad technique at the core of many approaches to AI, and there are many versions of it.

In healthcare, the most common application of traditional machine learning is precision medicine—predicting what treatment protocols are likely to succeed on a patient based on various patient attributes and the treatment context.3 These models require a training dataset for which the outcome variable (e.g., onset of disease) is known; this is, by far, the most common form of machine learning and is called supervised learning.

Machine learning can also be used to analyze epidemiological information about disease outbreaks. Blue Dot, the Canada-based firm that was among the firsts to predict the spread of the novel coronavirus, used machine learning to analyze news sources, airline passenger itineraries, and other data to produce an early pandemic warning.4 It also made an early prediction of the spread of the Zika virus to the United States.

A more complex form of machine learning is the neural network—a technology that has been available since the 1960s, has been well established in healthcare research for several decades, and is frequently employed for categorization applications such as determining whether a patient will acquire a particular disease.5 It views problems in terms of inputs, outputs, and weights of variables or “features” that associate inputs with outputs. It has been likened to the way that neurons process signals, but the analogy to the brain’s function is largely conceptual.

The most complex forms of machine learning involve “deep learning," or neural network models with many levels of features or variables that predict outcomes. There may be thousands of hidden features in such models, which are uncovered by the faster processing of today’s GPU (graphics processing units) and cloud architectures. Deep learning is often used for image recognition; a common application of deep learning in healthcare is recognition of potentially cancerous lesions in radiology images/’ It is increasingly being applied to radiomics, or the detection of clinically relevant features in imaging data beyond what can be perceived by the human eye.7 Both radiomics and deep learning are most commonly found in oncology-oriented image analysis, and, typically, require many (thousands or even millions) of the labeled images. Their combination appears to promise greater accuracy in diagnosis than the previous generation of automated tools for image analysis, known as computer-aided detection or CAD.

Deep learning is also increasingly used for speech recognition, and as such is a statistically based approach to natural language processing or NLP. It requires large amounts of validated speech from which to learn. Deep learning can also be used to analyze bodies and characters of text. Statistical NLP has been largely responsible for recent substantial improvements in certain areas of AI language processing, including speech recognition, machine translation, and sentiment analysis. These applications, however, are not frequently used in healthcare outside of generating transcripts from dictated notes or patient conversations.

Unlike earlier forms of statistical analysis, each feature in a deep learning model typically has little meaning to a human observer. As a result, the explanation of the model’s outcomes may be very difficult or impossible to interpret. This is a challenging issue in healthcare, because patients may often require explanations of why a particular diagnosis is indicated.

 
Source
< Prev   CONTENTS   Source   Next >