Response to “Regarding ‘Editorial Commentary: Artificial Intelligence in Sports Medicine Diagnosis Needs to Improve’”

Abstract

The recent letter to the editor written by Ramkumar et al.1Ramkumar P.N. Karnuta J.M. Nwachukwu B.U. Williams R.J. Regarding “Editorial commentary: Artificial intelligence in sports medicine diagnosis needs to improve”.Arthroscopy. 2021; 37: 1365-1367Abstract Full Text Full Text PDF Scopus (4) Google Scholar concerned several points raised by Dr. Nikolaos Paschos in a recent editorial commentary2Paschos N.K. Editorial commentary: Artificial intelligence in sports medicine diagnosis needs to improve.Arthroscopy. 2021; 37: 782-783Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar on our systematic review entitled “Diagnostic Performance of Artificial Intelligence for Detection of Anterior Cruciate Ligament and Meniscus Tears: A Systematic Review.”3Kunze K.N. Rossi D.M. White G.M. et al.Diagnostic performance of artificial intelligence for detection of anterior cruciate ligament and meniscus tears: A systematic review.Arthroscopy. 2021; 37: 771-781Abstract Full Text Full Text PDF PubMed Scopus (5) Google Scholar In particular, Ramkumar et al.1Ramkumar P.N. Karnuta J.M. Nwachukwu B.U. Williams R.J. Regarding “Editorial commentary: Artificial intelligence in sports medicine diagnosis needs to improve”.Arthroscopy. 2021; 37: 1365-1367Abstract Full Text Full Text PDF Scopus (4) Google Scholar sought to clarify the distinction between the appropriate use and current interpretation of artificial intelligence (AI) on the basis of several findings in our systematic review; or, in other words, between what AI is and what it is not. Although we are grateful for their well-intended defense of AI given their expertise in this growing area of research and concur with the majority of their primary disputes, we believe that the purpose and conclusion of our systematic review was misconstrued to align with their intended message. We also believe that the editorial commentary written by Dr. Paschos contained several statements that were imprecise as it pertains to the applications of AI. The primary concern raised by Ramkumar et al.1Ramkumar P.N. Karnuta J.M. Nwachukwu B.U. Williams R.J. Regarding “Editorial commentary: Artificial intelligence in sports medicine diagnosis needs to improve”.Arthroscopy. 2021; 37: 1365-1367Abstract Full Text Full Text PDF Scopus (4) Google Scholar was that our concluding message stated that AI did not outperform clinical experts and that Dr. Paschos was overzealous in his subsequent commentary. Specifically, the authors directed attention to Dr. Paschos’ discussion of whether AI is “ready to take over a part of the diagnostic process” and subsequent cautioning that enthusiasm for AI models should be resisted until more data are acquired.2Paschos N.K. Editorial commentary: Artificial intelligence in sports medicine diagnosis needs to improve.Arthroscopy. 2021; 37: 782-783Abstract Full Text Full Text PDF PubMed Scopus (4) Google Scholar Their third point states “most importantly, no one should be advocating, suggesting, or conceiving AI replace or compete with the role of the physician.” Our study did not insinuate that this is a possibility nor suggest that this represents the function that AI is intended to assume. Furthermore, Dr. Paschos did not claim this either, as his statement was that AI is currently not ready to take over a part of the diagnostic process. Suggesting that AI compete with or completely replace the role of a physician and expert is unfounded; this is a notion with which we entirely agree. The authors discuss why our study was inadequate to (1) evaluate the ability of AI to diagnose anterior cruciate ligament (ACL) and meniscus pathology and (2) compare experts to AI performance. This is an inherent limitation based on the principles of any systematic review. We did not in fact evaluate the ability of AI to diagnose ACL and meniscus pathology, nor did we compare experts to AI performance; rather, we qualitatively described the existing literature pertaining to the topic. In opposition to the statement that we evaluated the quality and quantity of magnetic resonance imaging scans inputted into various AI-based models (which we did not), we simply evaluated and presented the quality and quantity of studies that examined AI-based task performance for meniscus and ACL pathology diagnosis. We did not infer that AI can or cannot be used as a diagnostic adjunct but rather presented the existing data in an unbiased manner. As with all systematic reviews, the level of evidence and strength of recommendations are only as strong as the available literature permits them to be.4Cote M.P. Lubowitz J.H. Rossi M.J. Brand J.C. Reviews pooling heterogeneous, low-evidence, high-bias data result in incorrect conclusions: But heterogeneity is an opportunity to explore.Arthroscopy. 2018; 34: 3126-3128Abstract Full Text Full Text PDF PubMed Scopus (25) Google Scholar Dr. Paschos states, “These problems highlight the need for complete transparency and independent AI research with a high level of evidence prior to use in clinical practice.”3Kunze K.N. Rossi D.M. White G.M. et al.Diagnostic performance of artificial intelligence for detection of anterior cruciate ligament and meniscus tears: A systematic review.Arthroscopy. 2021; 37: 771-781Abstract Full Text Full Text PDF PubMed Scopus (5) Google Scholar Although AI clearly holds much potential through the ability to integrate patient-specific data into prediction models, perform task automation, and mitigate administrative and physician burden,5Ramkumar P.N. Kunze K.N. Haeberle H.S. et al.Clinical and research medical applications of artificial intelligence.Arthroscopy. 2021; 37: 1694-1697Abstract Full Text Full Text PDF PubMed Scopus (11) Google Scholar the literature on this topic in sports medicine remains limited at present. Presenting the heterogeneity of the current literature pertaining to AI should be viewed as an opportunity to identify areas to improve rather than a need to defend AI.4Cote M.P. Lubowitz J.H. Rossi M.J. Brand J.C. Reviews pooling heterogeneous, low-evidence, high-bias data result in incorrect conclusions: But heterogeneity is an opportunity to explore.Arthroscopy. 2018; 34: 3126-3128Abstract Full Text Full Text PDF PubMed Scopus (25) Google Scholar High quality inputs and data sources are essential, as the application of AI not only necessitates large amounts of data, but unbiased and high-quality data, and this should not be a point of contention. A systematic review is not capable of accounting for this limitation if its goal is merely to provide a subjective synthesis of the literature.4Cote M.P. Lubowitz J.H. Rossi M.J. Brand J.C. Reviews pooling heterogeneous, low-evidence, high-bias data result in incorrect conclusions: But heterogeneity is an opportunity to explore.Arthroscopy. 2018; 34: 3126-3128Abstract Full Text Full Text PDF PubMed Scopus (25) Google Scholar However, there is much merit to the letter written by Ramkumar et al.,1Ramkumar P.N. Karnuta J.M. Nwachukwu B.U. Williams R.J. Regarding “Editorial commentary: Artificial intelligence in sports medicine diagnosis needs to improve”.Arthroscopy. 2021; 37: 1365-1367Abstract Full Text Full Text PDF Scopus (4) Google Scholar and we support the authors’ argument that some statements presented in the editorial commentary by Dr. Paschos risk being misinterpreted by those unfamiliar with AI. The commentary by Dr. Nikoloas Paschos was primarily a narrative on the current ineptitude of AI in sports medicine literature, especially as it pertains to diagnostic processes. This commentary, too, was not without its limitations. It is clear that AI systems have several limitations and obstacles as Dr. Paschos stated, although we should not let previous mishaps, such as those pertaining to the referenced IBM Watson, preclude us from investigating the applications of AI further. Ultimately, messages should be taken from both sides: AI has not yet demonstrated reproducible and efficacious results in the sports medicine literature, but we should be advocating for the continued use of AI. AI has demonstrated proficiency and its potential for value in many other realms. The application of AI in sports medicine represents one tree in its infancy; accordingly, we should start by looking at the forest and recognizing AI as an established set of statistical processes with an irrefutable history, considerable potential, and an entity that is here to stay. Download .pdf (.1 MB) Help with pdf files ICMJE author disclosure forms

Ähnliche Arbeiten