Venerito Vincenzo, Manigold Tobias, Capodiferro Marco, Markham Deborah, Blanchard Marc, Iannone Florenzo, Hügle Thomas
Rheumatology Unit, Department of Precision and Regenerative Medicine and Ionian Area (DiMePre-J), University of Bari Aldo Moro, Bari, Italy.
Department of Rheumatology, Inselspital, University Hospital Bern, Bern, Switzerland.
Rheumatol Adv Pract. 2025 Apr 18;9(2):rkae143. doi: 10.1093/rap/rkae143. eCollection 2025.
To investigate the association between hand motion tracking features obtained through computer vision from smartphone cameras and disease activity in patients with RA.
The PyPI package of MediaPipe (version 0.9.0.1) was used for key landmark detection. Finger joint angles were calculated in each frame using the normalized dot product of the vectors (equations). RA patients were instructed to perform a rapid repetition of five fist closures. Hand movements were captured using standard smartphone cameras. Kinetic features time to maximum flexion for MCP, PIP and DIP joints were correlated with RA disease activity and disability outcomes. Logistic regression was used to investigate associations of range of motion and kinetic features with 28-joint DAS (DAS28) low disease activity/remission.
Our model showed promising performance in predicting low disease activity/remission in RA patients. Internal validation using 5-fold cross-validation on the training dataset ( = 81) yielded a mean accuracy of 0.72 (s.d. 0.09), specificity of 0.65 (s.d. 0.17), recall of 0.86 (s.d. 0.05) and area under the receiver operating characteristics curve (AUROC) of 0.80 (s.d. 0.09). External validation on the test dataset ( = 19) demonstrated improved performance with an accuracy of 0.84, specificity of 0.75, recall of 0.91 and AUROC of 0.89. Greater PIP and DIP joint angle changes, along with faster time to maximal flexion, were associated with lower disease activity. Significant correlations were observed between kinetic metrics and standard clinical measures, including DAS28, swollen joint count, tender joint count and HAQ Disability Index.
Single-camera motion capture of repeated fist closure may serve as an accessible digital biomarker for disease activity in RA.
研究通过智能手机摄像头的计算机视觉获得的手部运动跟踪特征与类风湿关节炎（RA）患者疾病活动度之间的关联。
使用MediaPipe的PyPI包（版本0.9.0.1）进行关键地标检测。在每一帧中使用向量的归一化点积（公式）计算手指关节角度。指导RA患者快速重复进行五次握拳动作。使用标准智能手机摄像头捕捉手部运动。掌指关节（MCP）、近端指间关节（PIP）和远端指间关节（DIP）达到最大屈曲的时间等动力学特征与RA疾病活动度和残疾结局相关。采用逻辑回归研究运动范围和动力学特征与28关节疾病活动评分（DAS28）低疾病活动度/缓解之间的关联。
我们的模型在预测RA患者低疾病活动度/缓解方面表现出良好的性能。在训练数据集（n = 81）上使用五折交叉验证进行内部验证，平均准确率为0.72（标准差0.09），特异性为0.65（标准差0.17），召回率为0.86（标准差0.05），受试者操作特征曲线下面积（AUROC）为0.80（标准差0.09）。在测试数据集（n = 19）上进行外部验证显示性能有所提高，准确率为0.84，特异性为0.75，召回率为0.91，AUROC为0.89。PIP和DIP关节角度变化越大，以及达到最大屈曲的时间越快，与疾病活动度越低相关。在动力学指标与标准临床测量之间观察到显著相关性，包括DAS28、肿胀关节计数、压痛关节计数和健康评估问卷残疾指数（HAQ）。
重复握拳动作的单摄像头运动捕捉可作为RA疾病活动度的一种便捷数字生物标志物。
