Regarding “ <i>Editorial Commentary</i> : Artificial Intelligence in Sports Medicine Diagnosis Needs to Improve”

Abstract

We read the Editorial Commentary, “Artificial Intelligence in Sports Medicine Diagnosis Needs to Improve,” by Dr. Nikolaos Paschos with great interest.1Paschos 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 This commentary was written with respect to the systematic review performed by Kunze et al.2Kunze 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 the review, 11 studies that used artificial intelligence (AI)–based techniques for the detection of anterior cruciate ligament and meniscus pathology from magnetic resonance imaging (MRI) scans were evaluated.2Kunze 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 The systematic review reported that across 11 studies, AI-based techniques were able to identify anterior cruciate ligament (ACL) and meniscus pathology with impressive efficacy. Among the 5 studies comparing the computer to a clinical expert head-to-head, the human outperformed the model in 2 of the studies. As such, the concluding message was “AI did not outperform clinical experts.” The editorial amplified this message by distilling the value of AI to whether AI was “ready to take over a part of the diagnostic process” and cautioned that “we need to resist our enthusiasm for these novel tools until we have robust, long-term, and replicated data.” Although few will argue the importance of scientific rigor and remaining critical of new process adoption, it is important to understand what AI is and what it is not. AI is not the rendering science fiction has portrayed wherein humans battle machines for dominance. AI is simply an analytic technique theorized over 50 years ago postulating that computers could recognize patterns to automate human tasks. AI is only now popular because the 2 missing ingredients have become commercially available: computing power and large datasets (i.e., “Big Data”). Of those 2 ingredients, computer processing power is a well-indoctrinated staple of society as a reliable tool. Just like clinical experience to a surgeon or specimens to a translational experiment, the quality and quantity of inputted data are the critical inflection point and therefore the key ingredients in determining the conclusion of any experiment, Netflix movie suggestion, or other AI-based prediction. Although nuanced differences among AI architectures exist, the quality and quantity of inputted data reflects the performance and accuracy of any AI-based model. Thus the studies in the systematic review did not actually evaluate AI itself; instead, they evaluated the quality and quantity of MRI images inputted into various AI-based models across 11 heterogeneous studies with variable ground truths and inclusion criteria. In the same spirit that transitioning from horse-drawn carriage to automobile did not require a randomized controlled trial to demonstrate superiority, the potential of AI need not be defended. Instead, our process acquisition of inputted data requires continued scrutiny to appreciate its strengths and limitations. Second, if studies truly sought to compare experts to the computer on a level playing field, it would be difficult to make the case that an AI-based model trained on a mere 1,370 ACL MRI screenshot images was outperformed when it performed similarly to 3 musculoskeletal radiologists with a combined expertise exceeding 36 years.3Bien N. Rajpurkar P. Ball R.L. et al.Deep-learning-assisted diagnosis for knee magnetic resonance imaging: Development and retrospective validation of MRNet.PLoS Med. 2018; 15e1002699Crossref PubMed Scopus (154) Google Scholar Third and most importantly, no one should be advocating, suggesting, or conceiving AI replace or compete with the role of the physician. Instead, AI exists as an adjunct, whether automating the documentation burden or expediting clinic triage in settings with limited expert access to immediately interpret ACL tears on MRI. The great volume of repetitive administrative burden thrust on physicians could be lessened with such automation AI offers, allowing more face-to-face patient care and paradoxically “making healthcare human again.”4Topol E.J. Deep medicine: How artificial intelligence can make healthcare human again. Basic Books, Hachette, UK2019Google Scholar If we are able to acquire training data with high fidelity and identify areas of meaningful use for task automation, we may be able to harness the potential of AI by mitigating the current demands and constraints of our dynamic healthcare environment. In so doing, we stand on the precipice of rendering leaner, higher-value care with newfound economies of scale that liberate both the orthopaedic surgeon on an individual level and the healthcare sector on a macro level.5Makhni E.C. Makhni S. Ramkumar P.N. Artificial intelligence for the orthopaedic surgeon: An overview of potential benefits, limitations, and clinical applications.J Am Acad Orthop Surg. 2021; 29: 235-243Crossref Scopus (1) Google Scholar, 6Ramkumar 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, 7Rajan P.V. Karnuta J.M. Haeberle H.S. Spitzer A.I. Schaffer J.L. Ramkumar P.N. Response to letter to the editor on "Significance of external validation in clinical machine learning: Let loose too early?.Spine J. 2020; 20: 1161-1162Abstract Full Text Full Text PDF Scopus (1) Google Scholar, 8Myers T.G. Ramkumar P.N. Ricciardi B.F. Urish K.L. Kipper J. Ketonis C. Artificial intelligence and orthopaedics: An introduction for clinicians.J Bone Joint Surg Am. 2020; 102: 830-840Crossref PubMed Scopus (17) Google Scholar, 9Helm J.M. Swiergosz A.M. Haeberle H.S. et al.Machine learning and artificial intelligence: Definitions, applications, and future directions.Curr Rev Musculoskelet Med. 2020; 13: 69-76Crossref PubMed Scopus (33) Google Scholar, 10Ramkumar P.N. Haeberle H.S. Bloomfield M.R. et al.Artificial intelligence and arthroplasty at a single institution: real-world applications of machine learning to big data, value-based care, mobile health, and remote patient monitoring.J Arthroplasty. 2019; 34: 2204-2209Abstract Full Text Full Text PDF PubMed Scopus (25) Google Scholar, 11Haeberle H.S. Helm J.M. Navarro S.M. et al.Artificial intelligence and machine learning in lower extremity arthroplasty: A review.J Arthroplasty. 2019; 34: 2201-2203Abstract Full Text Full Text PDF PubMed Scopus (24) Google Scholar Download .pdf (.11 MB) Help with pdf files ICMJE author disclosure forms

Ähnliche Arbeiten