适应性体力活动中的机器学习：慢性病个性化锻炼的临床应用、监测和实施途径：叙述性综述

摘要(中文译文)

机器学习（ML）正日益影响运动与锻炼的评估和实施，但其在针对慢性疾病患者的适应性体育活动（APA）中的作用尚未得到全面综合。基于机器学习的方法有望增强功能评估、支持个体化运动处方，并促进在预防性、社区性和长期适应性运动环境中实现可扩展的监测，尤其是在功能异质性显著且对运动反应多变的人群中。本叙述性综述旨在综合并批判性地讨论当前与APA实践核心专业流程相关的机器学习应用。采用结构化叙述性综述方法，检索了PubMed/MEDLINE、Scopus和Web of Science，并辅以针对工程学来源的定向检索，以捕获生物医学数据库中未被一致索引的机器学习方法。检索覆盖了当代机器学习方法日益应用于人体运动和锻炼研究的时期，最后更新于2026年1月。证据按主题综合为与APA实践相关的应用导向领域。机器学习在APA中的应用包括无标记运动和步态分析、可穿戴传感器数据处理、平衡与跌倒风险评估以及功能分类。预测性和自适应模型支持运动强度、进阶和负荷的个体化调控，包括远程和混合实施模式。应用涵盖肿瘤学、心脏代谢、呼吸、神经肌肉疾病以及适应性运动情境。伦理、法律和治理问题，如算法偏见、数据隐私和职业问责，成为安全和公平实施的核心考量。机器学习代表了APA中一个有前景的决策支持层，通过增强评估、个性化和监测来补充专业知识。其有效整合需要稳健的验证、可解释性和负责任的治理，以确保机器学习在APA实践中增强而非取代专业判断。

Abstract

Machine learning (ML) is increasingly influencing the assessment and delivery of movement and exercise, yet its role within adapted physical activity (APA) for individuals with chronic conditions has not been comprehensively synthesized. ML-based approaches have the potential to enhance functional assessment, support individualized exercise prescription, and facilitate scalable monitoring across preventive, community-based, and long-term adapted exercise settings, particularly in populations characterized by functional heterogeneity and variable responses to exercise. The aim of this narrative review is to synthesize and critically discuss current ML applications relevant to the core professional processes of APA practice. A structured narrative review was conducted using searches in PubMed/MEDLINE, Scopus, and Web of Science, complemented by targeted searches in engineering-oriented sources to capture ML methods not consistently indexed in biomedical databases. The search covered the period in which contemporary ML approaches have been increasingly applied to human movement and exercise research and was last updated in January 2026. Evidence was synthesized thematically into application-oriented domains relevant to APA practice. ML applications in APA include markerless motion and gait analysis, wearable-sensor data processing, balance and fall-risk assessment, and functional classification. Predictive and adaptive models support individualized regulation of exercise intensity, progression, and workload, including remote and hybrid delivery models. Applications span oncology, cardiometabolic, respiratory, neuromuscular conditions, and adapted sport contexts. Ethical, legal, and governance issues, such as algorithmic bias, data privacy, and professional accountability, emerge as central considerations for safe and equitable implementation. ML represents a promising decision-support layer for APA, complementing professional expertise through enhanced assessment, personalization, and monitoring. Its effective integration requires robust validation, interpretability, and responsible governance to ensure that ML augments, rather than replaces, professional judgment in APA practice.

如何引用