Raab Dominik, Heitzer Falko, Kocks Christine, Kowalczyk Wojciech, Kecskeméthy Andrés, Jäger Marcus
Lehrstuhl für Mechanik und Robotik, Universität Duisburg-Essen, Lotharstraße 1, 47057, Duisburg, Deutschland.
Lehrstuhl für Orthopädie und Unfallchirurgie, Universitätsklinikum Essen, Essen, Deutschland.
Orthopadie (Heidelb). 2025 Sep 2. doi: 10.1007/s00132-025-04712-w.
Artificial intelligence (AI) is considered a key technology for alleviating the burden on the healthcare system. For instrumental gait analysis, AI-based evaluations promise a direct and intuitive access to clinically relevant information in orthopaedics and trauma surgery, avoiding the challenging and time-consuming manual evaluations of large amounts of patient data.
The objective of this work is to investigate the specific challenges and limitations of using AI for clinical evaluation of gait analysis data and to propose effective solutions to address these limitations.
This work combines a systematic literature review on AI in gait analysis with practical experiences from applications of AI in the authors' own published research projects.
Six key challenges have been identified. While AI methods work best when extensive training data, a limited number of influencing factors, and a clearly defined target variable are available, instrumental gait analysis is characterised by opposite conditions (little training data, multiple influencing factors, and fuzzy target variables). To address these contradicting characteristics, a catalogue of possible solution approaches focusing on integrating clinical expert knowledge into AI development and operation is outlined.
It is shown that AI offers significant potential for improving the efficiency and quality of gait data exploitation. However, current AI approaches from other fields are only partially transferable to gait analysis due to insufficient fitting. By addressing the specific challenges for AI in gait analysis, it can be expected that specialized procedures and best practices can be developed, which will boost AI assistance in IGA clinical evaluation.
人工智能（AI）被视为减轻医疗系统负担的关键技术。对于仪器化步态分析，基于人工智能的评估有望直接、直观地获取骨科和创伤外科临床相关信息，避免对大量患者数据进行具有挑战性且耗时的人工评估。
本研究旨在探讨将人工智能用于步态分析数据临床评估的具体挑战和局限性，并提出有效解决方案以克服这些局限性。
本研究结合了关于人工智能在步态分析中的系统文献综述以及作者自身已发表研究项目中人工智能应用的实践经验。
已确定六个关键挑战。虽然当有大量训练数据、有限数量的影响因素以及明确界定的目标变量时，人工智能方法效果最佳，但仪器化步态分析的特点却相反（训练数据少、影响因素多、目标变量模糊）。为应对这些矛盾特征，概述了一系列可能的解决方法，重点是将临床专家知识整合到人工智能开发和操作中。
研究表明，人工智能在提高步态数据利用效率和质量方面具有巨大潜力。然而，由于拟合不足，目前其他领域的人工智能方法仅部分可转移到步态分析中。通过应对人工智能在步态分析中的具体挑战，有望开发出专门的程序和最佳实践，从而提升人工智能在仪器化步态分析临床评估中的辅助作用。
