A practical guide to the implementation of artificial intelligence in orthopaedic research—Part 3: How orthopaedic research benefits from the implementation of artificial intelligence

James A. Pruneski, Ayoosh Pareek, Bálint Zsidai, Jacob F. Oeding, Jonathan D. Hughes, Felix C. Oettl, Philipp W. Winkler, Thomas Tischer, Elmar Herbst, Alberto Grassi, Michael T. Hirschmann, Christophe Ley, Yinan Yu, Kristian Samuelsson, Esska Artificial Intelligence Working GroupJournal of Experimental Orthopaedics 2025 · DOI

TL;DR

This paper, the third in a series, explores the transformative potential of Artificial Intelligence (AI) in orthopaedic research. It highlights how AI, particularly machine learning and deep learning, can significantly enhance diagnostic accuracy, surgical planning, outcome prediction, and administrative efficiency, ultimately advancing patient care and our understanding of musculoskeletal pathology.

What Problem Does This Paper Address?

The paper addresses the growing integration of AI into orthopaedic research and practice, aiming to provide a comprehensive overview of its benefits and practical implementation. While AI has shown immense promise, its adoption in orthopaedics has lagged behind other specialties. This paper seeks to bridge that gap by detailing how AI can be leveraged across various aspects of orthopaedic research, from initial patient assessment to long-term outcome tracking. It also implicitly addresses the need for a structured approach to AI implementation to ensure high-quality, reliable research and clinical application.

Methods

The paper doesn't introduce novel methods but rather surveys and categorizes existing AI applications within orthopaedics. It discusses various AI techniques, including:

  • Machine Learning (ML): Algorithms that learn from data without explicit programming, used for tasks like prediction and classification.
  • Deep Learning (DL): A subset of ML utilizing neural networks, particularly effective for complex tasks like image and speech recognition.
  • Natural Language Processing (NLP): AI that enables computers to understand and interpret human language, useful for synthesizing large datasets from electronic medical records (EMRs).

The paper categorizes AI applications into several key areas: injury and outcome prediction, imaging interpretation, surgical planning, and administrative tasks. It also touches upon generative AI for synthetic data creation and privacy protection.

Datasets

The paper does not explicitly detail specific datasets used for training or evaluation in a dedicated section. However, it frequently refers to the use of:

  • Electronic Medical Records (EMRs): A primary source of data for AI models, containing patient demographics, clinical notes, and treatment histories.
  • Radiographic Data: Including MRI, X-rays, and CT scans, crucial for image interpretation and diagnostic AI models.
  • Surgical Data: Information related to surgical procedures, outcomes, and patient-specific variables.
  • Registry Data: Large-scale collections of patient data used for research and model development.

The paper emphasizes the importance of high-quality, diverse, and large datasets for training effective AI models.

Key Insights and Ideas

  • AI as an Augmentation, Not Replacement: A recurring theme is that AI should augment, not replace, the physician's role. It's a tool to enhance diagnostic accuracy, efficiency, and decision-making.
  • The Power of Pattern Recognition: AI excels at identifying subtle patterns in complex, high-dimensional data that might be missed by human observation, leading to improved predictive capabilities.
  • Transformative Potential in Surgical Planning: AI can automate tedious manual measurements and calculations in surgical planning, leading to increased efficiency and potentially more predictable outcomes.
  • Streamlining Administrative Burdens: AI can significantly reduce the time and resources spent on administrative tasks like coding, billing, and documentation, freeing up clinicians' time for patient care.
  • The Need for High-Quality Data and Validation: The success of AI in orthopaedics hinges on the availability of clean, representative data and rigorous external validation of models to ensure generalizability and trustworthiness.
  • Interdisciplinary Collaboration is Crucial: Effective AI implementation requires collaboration between clinicians, data scientists, and engineers to bridge the gap between technical capabilities and clinical needs.
  • Generative AI for Privacy and Data Augmentation: Generative AI offers a promising avenue for creating synthetic data to protect patient privacy and augment limited datasets, thereby improving model training.

Key Results

The paper focuses on the potential and demonstrated successes of AI across various orthopaedic research areas rather than presenting specific quantitative results from a single study. It highlights that:

  • Injury and Outcome Prediction: AI models have shown success in predicting various osseous and soft tissue pathologies, as well as outcomes for procedures like ACL reconstruction and joint arthroplasty.
  • Imaging Interpretation: DL models have been successfully implemented to aid in diagnosing pathologies across different subspecialties, often achieving accuracy comparable to or exceeding human experts.
  • Surgical Planning: AI platforms have demonstrated success in automating measurements for surgical planning, with early positive results for arthroplasty templating.
  • Administrative Tasks: AI pipelines have been published for predicting operative time, length of hospital stay, costs, and billing codes, with models achieving notable accuracies (e.g., 0.765 and 0.79 for length of stay and cost in one study).

Limitations and Future Work

Limitations:

  • "Black Box" Nature: The lack of interpretability in some advanced ML models (especially deep neural networks) can undermine trust and complicate error correction.
  • Data Quality and Bias: High amounts of missing data and nonrepresentative samples can lead to performance bias and limited generalizability.
  • External Validation: Models often require external validation on diverse population data and prospective evaluation before widespread clinical implementation.
  • Ethical Considerations: Patient privacy, data security, and ensuring generalizable training populations are critical ethical considerations.
  • Inappropriate Vernacular and Overstated Claims: The paper cautions against the misuse of AI terminology and the overstatement of AI capabilities in research publications.

Future Work:

  • Continued Development of AI Models: The paper anticipates further advancements in AI, particularly in areas like autonomous clinical support, precision surgical planning, and personalized patient care.
  • Creation of Large Registries: The development of comprehensive, large-scale data registries is essential for training and validating novel AI algorithms.
  • Federated Learning: This paradigm is highlighted as a way to address data governance and privacy concerns by training algorithms collaboratively without direct data exchange.
  • Interdisciplinary Collaboration and Standardization: Future efforts will focus on fostering better communication between disciplines, establishing quality assurance protocols, and ensuring external validation.
  • Generative AI Applications: Further exploration of generative AI for data synthesis, privacy protection, and overcoming language barriers in academic writing is expected.

Summary updated

Abstract

Level IV.
BibTeX
@article{pruneski2025practical,
  author = {James A. Pruneski and Ayoosh Pareek and Bálint Zsidai and Jacob F. Oeding and Jonathan D. Hughes and Felix C. Oettl and Philipp W. Winkler and Thomas Tischer and Elmar Herbst and Alberto Grassi and Michael T. Hirschmann and Christophe Ley and Yinan Yu and Kristian Samuelsson and ESSKA Artificial Intelligence Working Group},
  title = {{A practical guide to the implementation of artificial intelligence in orthopaedic research—Part 3: How orthopaedic research benefits from the implementation of artificial intelligence}},
  journal = {Journal of Experimental Orthopaedics},
  year = {2025},
  doi = {10.1002/jeo2.70481},
}
Figure