Ten simple rules for engaging with artificial intelligence in biomedicine

Abstract

The first industrial revolution led to mechanical production and steam power; the second, mass production and electrical power; and the third, electronics and computers. Today, as most sectors of the world move forward into the fourth industrial revolution, one centered around data and artificial intelligence (AI), biomedicine finds itself still in the third, lagging behind the rest [1]. Only recently, the exponential growth of technology has facilitated the widespread integration of computers into the biomedical domain from electronization of medical data analysis to automated detection of biomedical images [2–3]. Rather than merely automating time-consuming processes within healthcare, AI stands to reduce medical errors, expand upon the relationships between basic science and clinical medicine, and improve our analysis of existing datasets too large and complex for traditional statistics [3]. Despite these potential benefits, many biomedical facilities are hesitant to incorporate such systems into their practices due to the liability associated with AI making decisions that impact the health of patients [4], such as misdiagnosis (see Rule 8). Additionally, there exists a computational “black box,” a phenomenon describing the difficulty of understanding how AI algorithms arrive at a particular result (see Rule 3). Without a clear means of understanding how these machines generate their output, biomedical facilities are often skeptical of incorporating these “black boxes” into their work practices. As such, the “explainability” issue is an important barrier to overcome before applying these powerful technologies in biomedicine [5]. The lack of understanding around AI and the tantalizing benefits of this new wave of technology suggest the need for professionals in biomedical fields to acquire a basic understanding of AI and its medical applications in order to understand its clinical utility and engage with cutting-edge research. As such, there is a clear need for literature that explains AI in a way that is digestible to professionals in other fields [5]. Without a fundamental understanding of data science models and AI methods, modern biomedical experts who are not well versed in these fundamentals may be intimidated. Introduction to the basics of AI, such as big data analysis, data mining, machine and deep learning, and computer vision, would allow for the expansion of innovative designs in biomedicine. The importance of biomedical involvement in emerging technology is highlighted in the flaws of contemporary electronic medical records (EMR) that are widely used across the healthcare system. The ideal AI adaptation of EMR would be able to facilitate patient care through a variety of features like tracking changes in medical history of a patient and alerting caretakers of concerning health patterns; however, with the current state of EMR, tasks as simple as sharing medical records between healthcare facilities are burdensome. Though it has many virtues, most biomedical professionals agree that the current implementation of EMR is less than ideal, in part because it was developed and implemented with minimal consideration for the flow of information in the biomedical field [6]. The current weaknesses of EMR should serve as a warning, illustrating the importance of biomedical involvement in the deployment of new technologies. When these advancements inevitably make it to the forefront of clinical medicine, biomedical professionals should feel like they are in the driver’s seat rather than helpless passengers along for the ride. We propose the following rules to allow biomedical professionals to attain some measure of control and strap down their panic at the sight of words such as “algorithms,” “AI,” “machine learning” and the like. Rule 1: Don’t panic Computation-based technologies are ubiquitous in our lives, touching almost every facet of our day-to-day interactions. Nevertheless, the vast majority of us do not understand how these systems operate, let alone how to troubleshoot them when problems become apparent. Quickly overwhelmed and frustrated by error messages, constant update reminders, and pop-up advertisements, many of us have an adverse reaction to the increased incorporation of technology into our daily lives. As healthcare begins to adopt a new language unfamiliar to most people, there will be pushback. When there are words that we do not understand, such as “machine learning,” our immediate response is to experience an internal error message and shut down. While it is only natural to have an uneasy and uncomfortable feeling when approaching anything unfamiliar, this sensation can be debilitating and prevent the exploration of the unknown. Now is the time to resist the urge to fall back into something comfortable and learn how to embrace that feeling to allow yourself to grow from new experiences. It should be reassuring that, although many of the terms used in AI seem exotic, they are often deviations on reasonably simple statistical concepts that many biomedical professionals already understand. For example, “a multivariate predictive model using three knots of nonlinearity for continuous values” is fundamentally a linear regression model with some extra bells and whistles. “Deep learning” is a specific method to train neural networks, which are based upon different layers of computational “neurons” that recognize patterns (see Rule 3), much like neurons in the brain firing in response to specific visual inputs [7]. In learning about these new techniques, biomedical professionals will find that they are already familiar with many of the underlying algorithms. Your preexisting knowledge of statistics will serve as the foundation for your understanding of AI, because AI builds on statistics. Both statistics and AI manipulate data with similar algorithms and differ only in the purpose—inference versus prediction, respectively.

Ähnliche Arbeiten