E-Posters-A Machine Learning Based Guidance Method for Postpartum Rehabilitation Exercises

A Machine Learning Based Guidance Method for Postpartum Rehabilitation Exercises - Da He, MD

Pdf Summary

The objective of this study is to develop a machine learning-based guidance method for postpartum rehabilitation exercises. The proposed method aims to provide automatic guidance for these exercises, as they have been shown to contribute to postpartum recovery but are often limited by the lack of professional guidance and the need for pose correction at home. The method consists of three main components: 2D key joints estimation, a classifier using the XGBoost algorithm, and quasi-real-time feedback. The 2D key joints estimation extracts key joints of the human body from images. The classifier is based on the extracted key joints and is tolerant of missing partial points. It performs a multi-class classification task with n standard poses and one other class. The quasi-real-time feedback compares the currently classified pose with the standard one and provides guidance for pose correction. <br /><br />The evaluation of the method on a test set of 124 samples showed high accuracy (>99%) and F1-score (>99%), as well as robustness in various conditions. The method also demonstrated effective feedback for pose correction. The study concludes that the developed system achieves high accuracy and precision in exercise posture classification, provides effective feedback and pose correction, and enables professional rehabilitation guidance. The method utilizes inexpensive and convenient marker-free devices.<br /><br />The experimental platform for the method includes two RGB cameras on mobile phones and a workstation with 16 GB RAM and a Nvidia 1080Ti GPU. The software used includes Python 3.5, Keras (a deep learning framework), and Ubuntu (an operating system).<br /><br />Overall, this study presents a machine learning-based method that can automatically guide postpartum rehabilitation exercises with high accuracy and provide effective feedback for pose correction, contributing to postpartum recovery and reducing the probability of pelvic organ prolapse and urinary incontinence.

Keywords

machine learning

postpartum rehabilitation

guidance method

automatic guidance

pose correction

2D key joints estimation

XGBoost algorithm

quasi-real-time feedback

multi-class classification

exercise posture classification