Context/Objective: This is a preliminary study of movement finalities prediction in manual wheelchairs (MWCs) from electromyography (EMG) data. MWC users suffer from musculoskeletal disorders and need assistance while moving. The purpose of this work is to predict the direction and speed of movement in MWCs from EMG data prior to movement initiation. This prediction could be used by MWC to assist users in their displacement by doing a smart electrical assistance based on displacement prediction.

Design: Experimental study.

Setting: Trained Subject LAMIH Laboratory.

Participants: Eight healthy subjects trained to move in manual wheelchairs.

Interventions: Subjects initiated the movement in three directions (front, right and left) and with two speeds (maximum speed and spontaneous speed) from two hand positions (on the thighs or on the handrim). A total of 96 movements was studied. Activation of 14 muscles was recorded bilaterally at the deltoid anterior, deltoid posterior, biceps brachii, pectoralis major, rectus abdominis, obliquus externus and erector spinae.

Outcome Measures: Prior amplitude, prior time and anticipatory postural adjustments were measured. A hierarchical multi-class classification using logistic regression was used to create a cascade of prediction models. We performed a stepwise (forward–backward) selection of variables using the Bayesian information criterion. Percentages of well-classified movements have been measured through the means of a cross-validation.

Results: Prediction is possible using the EMG parameters and allows to discriminate the direction / speed combination with 95% correct classification on the 6 possible classes (3 directions * 2 speeds).

Conclusion: Action planning in the static position showed significant adaptability to the forthcoming parameters displacement. The percentages of prediction presented in this work make it possible to envision an intuitive assistance to the initiation of the MWC displacement adapted to the user's intentions.

背景/目的：这是一项基于肌电图 (EMG) 数据的手动轮椅 (MWC) 运动最终预测的初步研究。MWC 用户患有肌肉骨骼疾病，在移动时需要帮助。这项工作的目的是在运动开始之前根据 EMG 数据预测 MWC 中的运动方向和速度。MWC 可以使用此预测，通过基于位移预测的智能电气辅助来帮助用户进行位移。

设计：实验研究。

环境：受过训练的受试者 LAMIH 实验室。

参与者：八名健康受试者接受过手动轮椅训练。

干预：受试者在三个方向（前、右和左）以两种速度（最大速度和自发速度）从两个手的位置（大腿或手边）开始运动。总共研究了96个动作。在三角肌前部、三角肌后部、肱二头肌、胸大肌、腹直肌、外斜肌和竖脊肌的双侧记录了 14 块肌肉的激活。

结果测量：测量了先前的幅度、先前的时间和预期的姿势调整。使用逻辑回归的分层多类分类用于创建级联预测模型。我们使用贝叶斯信息标准对变量进行了逐步（向前-向后）选择。通过交叉验证的方式测量了分类良好的运动的百分比。

结果：使用 EMG 参数可以进行预测，并允许在 6 个可能的类别（3 个方向 * 2 个速度）上以 95% 的正确分类来区分方向/速度组合。

结论：静态位置的动作计划对即将到来的参数位移显示出显着的适应性。这项工作中提出的预测百分比使得可以设想一种直观的帮助，以适应用户的意图来启动 MWC 置换。