Park Seul Gi, Mun Sae Byeol, Kim Young Jae, Kim Kwang Gi
Department of Nursing, Gachon University, Incheon, Republic of Korea.
Department of Health Sciences and Technology, Gil Medical Center, GAIHST, Gachon University, Incheon, 21999, Republic of Korea.
Sci Rep. 2025 Jun 6;15(1):20012. doi: 10.1038/s41598-025-04065-6.
Incomplete tetraplegia, incomplete paraplegia, and cauda equina syndrome are major neurological disorders that significantly reduce patients' quality of life, primarily due to impaired motor function and gait instability. Although conventional neurological assessments and imaging techniques are widely used for diagnosis, they are limited by temporal constraints and physical accessibility. This study explores the integration of machine learning and 3D motion capture gait data for effective classification of these conditions. Gait data from 214 patients were analyzed, and key features were identified using recursive feature elimination. Machine learning models, including support vector machine, random forest, and XGBoost, were trained and validated. The XGBoost model achieved the highest accuracy (74.42%) and F1-score (74.27%), with age, cadence, and double support emerging as the most influential features. Sex-based differences revealed that males exhibited greater dynamic gait variables, while females showed higher stability-oriented metrics. Age-based analysis indicated significant gait changes after 60 years, highlighting the role of stability-related features. These findings demonstrate the potential of integrating 3D motion capture and machine learning as a scalable, noninvasive diagnostic tool. By detecting subtle gait variations, this approach can aid in early diagnosis and personalized treatment planning for individuals with neurological impairments.
不完全性四肢瘫、不完全性截瘫和马尾综合征是主要的神经系统疾病，主要由于运动功能受损和步态不稳，严重降低了患者的生活质量。尽管传统的神经学评估和成像技术被广泛用于诊断，但它们受到时间限制和身体可达性的限制。本研究探索将机器学习与3D运动捕捉步态数据相结合，以有效分类这些病症。分析了214例患者的步态数据，并使用递归特征消除法确定了关键特征。对包括支持向量机、随机森林和XGBoost在内的机器学习模型进行了训练和验证。XGBoost模型实现了最高准确率（74.42%）和F1分数（74.27%），年龄、步频和双支撑期成为最具影响力的特征。基于性别的差异表明，男性表现出更大的动态步态变量，而女性则表现出更高的稳定性指标。基于年龄的分析表明，60岁以后步态有显著变化，突出了与稳定性相关特征的作用。这些发现证明了将3D运动捕捉和机器学习相结合作为一种可扩展的非侵入性诊断工具的潜力。通过检测细微的步态变化，这种方法可以帮助对神经损伤患者进行早期诊断和个性化治疗规划。
