The application of AI in the field of medicine has been widely forecasted since the term was first coined by John McCarthy more than 60 years ago.¹ In 1955, McCarthy originally envisioned AI as “the science and engineering of making intelligent
machines.” He predicted that these machines would be capable of performing feats previously thought to be exclusively in the domain of human intelligence, such as abstract thought, advanced problem-solving, and iterative self-improvement. At that time, this extremely innovative concept sparked much discussion over potential applications of such a technology and the implications such an innovation would have on the worldwide economy. In 1976, Jerrold S. Maxmen, a professor of psychiatry at Columbia University, predicted that AI would bring about the “postphysician era” by the 21st century,² describing the change as “possible, inevitable, and desirable”.³ Although AI has not replaced the role of physicians, this technology has already profoundly affected other industries. Some examples of the utilization of AI can be seen in the development of autonomous self-driving cars, online purchase recommendations, targeted advertisements, and high-frequency stock trading. Although the end user is generally insulated from seeing its direct employment, AI has already been established and fundamentally ingrained within many facets of today’s society. However, the employment of AI in medicine has lagged behind that of other industries.⁴,⁵ Despite initial excitement over the possibilities of AI in the medical field, practical applications of AI have only recently begun to materialize.

Topol⁶ outlined several large-scale factors that are likely playing a role in the recent acceleration of AI implementation in health care. The first factor is economic: it is becoming progressively apparent that health care in the United States is a failing business model. Rapidly increasing expenditures are paradoxically paired with deteriorating key outcomes and decreasing reimbursements for health care providers. Despite having the highest health care expenditure per capita among developed countries, the United States consistently ranks poorly in key quality metrics such as average life expectancy, maternal and infant mortality, and health equity.⁷,⁸ Innovation in the form of AI offers exciting potential in both improving health care outcomes and reducing inefficiencies that currently plague modern medicine globally. The second factor is the generation of patient data at an unprecedent large scale. From high-resolution medical imaging, continuously evolving electronic medical records, genome sequencing, and numerous diagnostic testing capabilities, each patient encounter produces a considerable number of discrete points of information, generating big data that cannot be effectively analyzed with human processing or standard statistical methods. One study of electronic medical records found that a single patient’s health record was associated with an average of approximately 32,000 data elements.⁹ In an age of information overload, a physician is tasked with integrating this overwhelming amount of data and synthesizing a clinical decision, a seemingly impossible task for a physician already given “insufficient time, insufficient context, and insufficient presence,” as described by Topol.⁶ Judicious employment of AI and its predictive abilities may provide a solution to these problems of economic sustainability and overwhelming data.

ML is a subset of AI that involves the use of computational algorithms that can analyze large data sets in order to classify, predict, or gain useful inference.⁶,¹⁰ In its most simplistic form, ML models are given inputs and outputs of training sets of
real-world data to analyze and make connections from using various methods of pattern recognition.¹¹ The models are then tasked with creating predictions given inputs from a testing set, and its predictions are compared with actual known outcomes in order to quantify the accuracy of the algorithm. These algorithms exhibit the same experiential learning associated with human intelligence, having the capacity to continually assess and improve the quality of its analyses given an adequate amount of data inputs, with the potential to continue learning after implementation as new data are available.¹¹,¹²,¹³

Deep learning can be thought of as an additional subset of ML (Figure 1). Made possible with increasingly powerful computational processing capabilities, deep learning models are more sophisticated algorithms that require less human supervision for development. Also known as deep neural networks, these models can mimic the structure and function of the biologic neuronal brain. Unlike traditional ML algorithms, which generally require human expertise and the predetermined transformation of raw data into a suitable format, deep learning models are a form of representation learning.¹⁴ They work autonomously, allowing the system to discover alternative representations with differing levels of abstractions (Figure 2). The neural network begins with an input tier that receives the raw data. The network then progresses to a number of hidden tiers that each respond to different features of the input.¹⁵ Through this process of developing multiple hidden layers, the model continues to develop more and more abstract representations of the
data. Similar to the way the human brain functions, the machine is able to make “neuronal” connections from “dendrites” on multiple hierarchical data levels.¹² Eventually, the model learns to appreciate a concept on multiple layers and dimensions, building on itself to create a web of interconnected relationships.¹¹

In the field of orthopaedics, ML technology is still somewhat new, with limited studies detailing potential uses. ML can be used in orthopaedics for the identification of fractures from plain radiographic images, the identification of orthopaedic
implants for hardware removal and/or modular revision,¹⁶,¹⁷,¹⁸,¹⁹,²⁰,²¹ the prediction of postoperative opioid use following total hip arthroplasty (THA),²⁰ and the evaluation of remote patient monitoring systems.²² Some of these topics are reviewed in greater detail later in this chapter. Although the maturity of AI in the field of orthopaedics has lagged behind fields such as radiology, dermatology, and ophthalmology, research interest in ML in orthopaedics has increased rapidly in the past 2 decades.

In 2018, Cabitza et al²³ conducted a systematic review and reported massive growth in the number of studies detailing ML strategies applied to the field of orthopaedics in the years immediately preceding. Most papers focused on the areas of osteoarthritis (OA) detection and prediction, bone and cartilage imaging, and spine pathology detection. Evaluating the development of ML in the field of orthopaedics in the manner of traditional health technologies, the application of ML is still limited to phase 2 studies. It is vitally important for orthopaedic surgeons to gain a fundamental appreciation for ML and the paradigm shift it represents, not just in orthopaedic surgery, but in the practice of medicine as a whole.

Remote patient monitoring systems are an avenue to increasing the value of care, applied either in the preoperative and postoperative period. Although many companies have developed software to monitor step counts and activity level, the application of ML allows patients and health care providers to track their participation in home exercise programs and general activity levels. The surgical team can therefore track rehabilitation and intervene with calls or additional office visits if postoperative milestones are not being met. Although patients and practitioners have embraced digital technology (eg, glucose monitors), wearable fitness devices are a relatively untapped resource for patients and physicians to access personal analytics that can contribute to preventive care, postoperative care, or aid in the management of a chronic disease. This new technology understandably raises questions concerning the effect on users’ health and well-being: the margin of error may be high when patients without medical training attempt to interpret the data from these devices. With appropriate frameworks in place, wearable devices that are integrated into health care systems could improve postoperative care.

In the context of research, these wearable technologies present an opportunity to collect data that can further support the goals outlined previously: collecting more data on patients’ activity level before and after surgery to stratify risk levels of future patients based on their specific demographics, comorbidities, and injury or disease.²⁴ Of course, these wearable fitness devices may, as with other technologic gadget trends, drift out of popular favor. Furthermore, concerns about
privacy and security of personal data are valid as users do not “own” the data that they create when wearing these devices. However, given their current popularity, the adoption of these technologies has potential.

Remote patient monitoring systems have been proven to be effective for patients undergoing primary total knee arthroplasty (TKA) for OA. In one cohort study of 25 patients,²² patients who underwent this procedure downloaded a mobile application onto their personal iPhones and recorded preoperative mobility and patient-reported outcome measures, beginning 2 to 4 weeks before surgery. A knee sleeve was paired with the patient’s iPhone via Bluetooth and the application notified the patient to complete weekly exercises. The knee sleeve and phone collected data on mobility (daily step count), a weekly range of motion checks, weekly patient-reported outcomes, daily home exercise program compliance, and self-reported daily opioid consumption. Some of these data points required active engagement by the patient/user and others were collected passively through the smartphone’s native sensors. For this group, this system was found to be reliable, low maintenance and well received during the process of recovery from TKA.²² Although this small cohort may not represent a broadly generalizable experience, the experience guides an ongoing effort to reduce cost and physician resources required to efficiently distinguish a patient who is thriving after surgery from one who needs additional intervention. It may be a feasible option to engage patients, communicate procedural value, and survey patients postoperatively. More studies are required to evaluate the clinical significance of the intervention and its effect on population health.