Stroke Rehabilitation Exercise Data Utilizing 3D Depth Sensors and IMU Sensors

There is a need for datasets for physical rehabilitation exercise for stroke. The current most prominent datasets available are the University of Idaho - Physical Rehabilitation Movements Data Set (UI-PRMD), which consists of only skeleton data and KInematic Assessment of MOvement and Clinical Scores for Remote Monitoring of Physical Rehabilitation (KIMORE) containing RGB depth video, along with skeleton joint position and orientations. But, as for our knowledge, so far only the KIMORE dataset has defined features and physician’s scoring. In order to add more datasets besides the KIMORE dataset we have produced a setup that uses two methods: multi-scale graph convolution disentanglement and a unified spatial-temporal graph aggregation with a convolution operator named G3D that provides a scoring that is ethically approved by Independent University Bangladesh (IUB)’s institutional review board (IRB).

This dataset’s main directory consists of two directories and two .csv files -
IMU_Data Directory: Contains 128 files of 128 participants in .csv format. Example: P1.csv, P2.csv…P128.csv.

KINECT_Skeleton_Data Direcory: Contains 631 files of 128 participants in .skeleton format. Example: P1_Female_21_Exercise Type 1.skeleton…P128_Male_24_Exercise Type 5.skeleton.

Participants_Information.csv: Contains participants’ ID, age, and gender.

Participants_with_Performaces_Scores.csv: Contains participants’ control factor (CF) and primary outcome (PO) scores with exercise types.

The IMU_Data directory contains data from 2 IMU sensors. There are 128 .csv files consisting of the X, Y, Z axis data of Accelerometer, Gyroscope and Magnetometer present in the IMU sensors. Each individual .csv file consists of 3-5 sheets for the 5 defined exercises for physical rehabilitation. The KINECT_Skeleton_Data direcory contains 631 skeleton files generated using the RGB-D sensors present in the Microsoft Kinect v2. The skeleton files are generated for 3-5 exercises for the 128 participants. Participants_Information.csv file also contains the age, gender, participant ID of 128 individuals. The Participants_with_Performaces_Scores.csv file contains the scores for each exercise of each individual participant. The scores are classified into two categories, Primary Outcome (PO), and Control Factor (CF). POs and CFs signify the movement of upper limbs and physical constraints during the exercises.

The 5 exercises:
Exercise Type 1: Lifting of the arms features
Exercise Type 2: The lateral tilt of the trunk with the arms in extension
Exercise Type 3: Trunk rotation
Exercise Type 4: Pelvis rotations on the transverse plane
Exercise Type 5: Squatting

脑卒中物理康复训练领域亟需专用数据集。当前主流可用数据集包括仅包含骨骼数据的爱达荷大学物理康复运动数据集（University of Idaho - Physical Rehabilitation Movements Data Set, UI-PRMD），以及集成RGB深度视频、骨骼关节位置与姿态数据的远程监测物理康复运动学评估与临床评分数据集（KInematic Assessment of MOvement and Clinical Scores for Remote Monitoring of Physical Rehabilitation, KIMORE）。但据现有文献来看，目前仅有KIMORE数据集定义了特征体系与医师评分标准。为补充KIMORE之外的康复数据集，我们搭建了一套方案，采用多尺度图卷积解耦与名为G3D的卷积算子实现统一时空图聚合两种方法，生成的评分已通过孟加拉国独立大学（Independent University Bangladesh, IUB）机构审查委员会（Institutional Review Board, IRB）的伦理审批。

本数据集的主目录包含两个子目录与两个.csv格式文件：
1. IMU_Data目录：收纳128名受试者的128个.csv格式数据文件，命名规则为P1.csv、P2.csv……P128.csv。
2. KINECT_Skeleton_Data目录：收纳128名受试者的631个.skeleton格式数据文件，命名规则为P1_Female_21_Exercise Type 1.skeleton……P128_Male_24_Exercise Type 5.skeleton。
3. Participants_Information.csv：记录受试者编号、年龄与性别信息。
4. Participants_with_Performaces_Scores.csv：记录受试者的控制因子（Control Factor, CF）与主要结局（Primary Outcome, PO）评分，以及对应训练动作类型。

IMU_Data目录下的数据来自2个惯性测量单元（Inertial Measurement Unit, IMU）传感器，128个.csv文件分别包含各传感器中加速度计、陀螺仪与磁力计的X、Y、Z轴采集数据。每名受试者的.csv文件包含3至5个工作表，对应5种预设康复训练动作。KINECT_Skeleton_Data目录下的631个骨骼数据文件由Microsoft Kinect v2的RGB-D传感器生成，对应128名受试者的3至5种训练动作。Participants_Information.csv记录了全部128名受试者的编号、年龄与性别信息。Participants_with_Performaces_Scores.csv则记录了每名受试者各训练动作的评分，评分分为主要结局（PO）与控制因子（CF）两类，分别代表上肢运动状态与训练过程中的身体约束情况。

本次研究涵盖的5种训练动作如下：
训练类型1：上肢抬举动作
训练类型2：躯干侧倾伴上肢伸展动作
训练类型3：躯干旋转动作
训练类型4：骨盆水平面旋转动作
训练类型5：深蹲动作

搜集汇总

背景与挑战

背景概述

该数据集是一个用于中风康复训练的多模态数据集，包含128名参与者的IMU传感器和Kinect v2传感器数据，以及参与者的个人信息和表现评分。数据集旨在补充现有的康复训练数据集，提供更多样化的数据来源和评分标准。