The Evolution of Machine Learning and Its Applications in Orthopaedics: A Bibliometric Analysis

Artificial intelligence (AI) and machine learning (ML) are computational systems designed to perform tasks that typically require human intelligence, with the capability to learn and improve by processing real-time data. Ongoing advancements in these models have led to their growing application in the field of medicine, leveraging their capabilities to enhance outcomes. To explore their impact in orthopaedics, the 100 most-cited articles on ML applications were identified through a comprehensive search of all databases within Web of Science, limited to English-language publications but with no restriction on publication year. Data were extracted to analyse distinct key aspects of ML methodology and applications within different orthopaedic subspecialties. The level of evidence (LoE) of included studies was also assessed. The included articles collectively accounted for a total of 10,886 citations. Citation count per article ranged significantly from 57 to 428 (mean: 108.9 ± 56.1). The majority of the studies were classified as LoE V (n = 46; mean citations = 108 ± 41.9), with 43 of them being experimental in terms of study design. Only one study achieved level I status, highlighting a significant gap in methodological quality research within the field. Musculoskeletal imaging was the most prominently represented subspecialty (n = 44), followed by trauma (n = 23) and arthroplasty (n = 21). Convolutional neural networks (CNNs) were predominant in terms of ML technique (n = 37), while deep learning (DL) was the most common ML field discussed. A total of 17% of studies included a human comparison group, with AI in orthopaedics generally demonstrating performance close to, but seldom surpassing, that of human experts. ChatGPT (versions 3.5 and 4.0) did not demonstrate superior performance compared to orthopaedic surgeons in four separate studies where direct comparisons were made. Overall, most of the highly influential articles on machine learning applications in orthopaedics are based on lower levels of evidence. These models require more critical evaluation and strong human oversight to ensure their effective integration into routine orthopaedic practice and to support a productive collaboration between humans and AI systems.