Artificial Intelligence (AI) and Machine Learning (ML)are interwoven in our everyday lives: from the way our email inbox is organized to the algorithms that dictate our Netflix preferences. In comparison, the adoption of ML in the medical field has been relatively slower, but there has been a surge in growth during recent years. AI and ML have been applied in various fields of medicine such as mental health, cardiology, dermatology, and radiology, where it has specifically seen the greatest use [1, 2]. AI and ML are gaining increasing interest given their success in these fields of medicine, where in some cases they are able to outperform human specialists [3]. The application of AI and ML in the orthopedic field is still in an earlier stage compared to other areas of medicine [1]. Structured research frameworks such as cohort studies and randomized controlled trials, as well as more experimental research are still needed for AI and ML to be widely accepted in the field of orthopedics [1].
ML utilizes computer algorithms and statistics to identify complex patterns and trends within the data that otherwise would not be distinguishable by humans [4].ML can “learn” patterns from data and produce models linking covariates to a target variable of interest and build models to describe the behavior of a system [4, 5]. In the field of medicine, ML can compile data from imaging and laboratory tests, and electronic medical records to guide physicians in formulating more efficacious and productive decisions [1].
Two broad categories of ML are generally employed in medicine, depending on the task: supervised learning and unsupervised learning [5]. Supervised learning focuses on choosing among subgroups to describe a new instance of data and estimating an unknown parameter [5].For example, an automated interpretation of an ECG where a specific pattern is linked to a set of diagnoses or how a lung nodule from a chest radiograph is detected automatically [5]. In contrast, unsupervised learning centers on the patterns or groupings within the data rather than predicting an output [5]. The goal of unsupervised learning is to uncover the hidden structure in the data and learn its pattern [2]. For example, endomyocardial biopsies can be taken and histologically examined to identify cellular composition that can aid in developing targeted therapy for myocarditis [5]. What category of machine learning is applied depends on the needs of the patient and the physician.
ML holds tremendous potential to improve the quality of life for patients in a plethora of medical specialties including mental health. The Kyoto Prefectural University of Medicine utilized Simple Linear Regression and L1-Sparse Canonical Correlation Analysis ML algorithms to identify a clinical biomarker for Obsessive Compulsive Disorder (OCD) [7]. Using MRI, the biomarker was able to distinguish between patients with OCD and non-affected human controls with 73% accuracy [7]. Repeating the exam with a different MRI machine and subset of patients led to 70% accuracy [7].While the sample size for this study was the limiting factor (N = 108), the reproducibility of the data holds promise for future clinical applications. Furthermore, another study analyzed tens of thousands of Instagram photos to identify markers of depression [8]. Researchers utilized color analysis, Instagram metadata components, and algorithm face detection to create predictive models for depression screening. These predictive models were able to outperform general practitioners in diagnosing depression [8]. Results held true even for patients who did not have an initial diagnosis of depression [8]. The simple utilization of such ML models in the primary care office could greatly increase the early and successful diagnosis of depression.
Radiology is another field that has been quick to adopt ML. In some applications, ML performed as well or even better than orthopedic surgeons in fracture detection of the upper extremity, ankle, and spine [9]. ML can also be integrated into current imaging systems making them ‘intelligent,’ leading to faster imaging speeds and the ability to offer modifications to ongoing magnetic resonance imaging sequences to visualize a lesion more accurately [2]. This can also be done by integrating the use of information from patient’s medical records, allowing the program to determine the most appropriate patient-specific imaging examination and protocol [9]. It even has the potential to automatically detect incidental findings on imaging and learn how to identify critical findings, such as a pneumothorax [2]. The use of ML does not aim at replacing the radiologist but augment their workflow and enhance their diagnostic accuracy [2] These algorithms are able to identify findings that might not be so easily seen by the human eye, such as using the variations in intensity on MRI to predict O6-methylguanine methyltransferase gene promoter methylation in glioblastoma multiforme tumors [10].
Orthopedic surgery is one of the most technologically innovative fields in medicine.AI and ML adoption is still in a preliminary phase in orthopedics [1]. ML can be used to provide a patient-specific predicted rate of post-operative complications, predict injury risk patterns, and guide clinical decision making [9, 11].Considering the recent growth of ML in the specialty and the quantity of new research that has been updated, a systematic review is required. There are over 3,300 published articles relating to AI and ML in orthopedics, with over 1,100 of those having been published in the last two years alone. Given the marked increase in publications, the primary objective of this review is to provide an update on the advances of AI and ML in the field of orthopedic surgery. The secondary objectives of this review are to evaluate the applications of AI and ML in providing a clinical diagnosis and predicting post-operative outcomes and complications in orthopedic surgery.