Dataset for Gait Analysis of Cerebellar Ataxic Patients and Healthy Adults Using MediaPipe Pose
Description
This dataset is composed of human gait analysis data extracted using the human pose estimation framework MediaPipe Pose applied on two types of videos: 1. Cerebellar Ataxic Gait Videos. 2. Normal Gait Videos. Each one of the "Cerebellar Ataxic Gait " and "Normal Gait" folders contains 20 feature datasets extracted from 20 videos. The datasets are numbered from 1 to 20 with each numbering representing a healthy subject for the normal gait dataset and the same subject mimicking the cerebellar ataxic gait. Each dataset consists of 47 gait analysis features: 1. 3D coordinates (x, y, z) of 12 joints in the human body: right and left shoulders, elbows, wrists, hips, knees, and ankles (36 features). 2. Kinematic parameters representing joint angles: right and left shoulders abduction and adduction, hips flexion and extension, and knees flexion and extension (6 features). 3. Spatiotemporal parameters: step length, step width, feet clearance, and right and left stride speed (5 features). This dataset can be used for training medical diagnosis machine learning models, which can be beneficial for physicians in assisting patients diagnosed with cerebellar ataxia or other similar neurodegenerative diseases with their gait rehabilitation process. For those who want to use this dataset for their research, make sure to cite the following paper associated with this dataset: @inproceedings{khalil2022diagnosis, author={Khalil, Hisham and Saad, Ahmed Mohamed Saad Emam and Khairuddin, Uswah}, booktitle={2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)}, title={Diagnosis of Cerebellar Ataxia Based on Gait Analysis Using Human Pose Estimation: A Deep Learning Approach}, year={2022}, pages={201--206}, doi={10.1109/IECBES54088.2022.10079396}}
Files
Steps to reproduce
In order to reproduce this dataset, MediaPipe Pose Python API can be used on videos of different types of human gait. The videos are captured in the view normal to the frontal plane of human motion with a resolution of 1920 × 1080 (1080p). The features extracted and constructed can be referred to from the paper mentioned in the description and provided in the link below.
Institutions
Categories
Funding
University of Technology Malaysia
The Japanese Chamber of Trade & Industry, Malaysia