Individualised Responsible Artificial Intelligence for Home-Based Rehabilitation.

Area of Science:

• Rehabilitation engineering
• Artificial Intelligence in healthcare
• Machine Learning for health informatics

Background:

• Post-COVID-19 socioeconomic factors necessitate unsupervised, home-based rehabilitation solutions.
• Patient engagement and motivation in rehabilitation require personalized support.
• Artificial Intelligence (AI) in healthcare must adhere to Accountability, Responsibility, and Transparency (ART) principles for acceptance.

Purpose of the Study:

• To present a patient-centric, individualized home-based rehabilitation support system.
• To develop an AI model that meets individualization, interpretability, and ART requirements.
• To evaluate the system's performance in supporting daily living activity tests.

Main Methods:

• Utilized Timed Up and Go (TUG) and Five Time Sit To Stand (FTSTS) tests for activity assessment.
• Developed a hybrid machine learning algorithm combining ensemble learning and stacking (gradient boosted trees, k-nearest neighbors).
• Generated synthetic datasets to complement experimental data and mitigate bias.

Main Results:

• Achieved up to 100% accuracy in predicting patient medical conditions for FTSTS and TUG.
• Reached 100% and 83.13% accuracy in predicting areas of difficulty for FTSTS and TUG, respectively.
• Demonstrated a 5% (FTSTS) and 15% (TUG) improvement over previous intrusive monitoring methods.

Conclusions:

• The developed AI system effectively supports individualized, unsupervised home-based rehabilitation.
• The system accurately predicts patient conditions and identifies functional limitations.
• This approach offers a non-intrusive, computationally efficient alternative to camera-based monitoring.