ESTIMATION OF BALANCE STATUS IN PATIENTS WITH HEMIPARESIS: AN ARTIFICIAL NEURAL NETWORK IMPLEMENTATION

Area of Science:

• Neurology
• Rehabilitation Medicine
• Biomedical Engineering

Background:

• The Balance Evaluation Systems Test (BESTest) is crucial for assessing balance deficits in hemiparetic individuals.
• However, the BESTest is time-consuming and can be fatiguing for patients.
• Artificial neural networks (ANNs) offer a potential solution for efficient balance status estimation.

Purpose of the Study:

• To compare the predictive accuracy of ANNs with manual BESTest results.
• To identify the most influential BESTest sections using ANN analysis.
• To explore the potential for a streamlined BESTest using ANNs.

Main Methods:

• 66 hemiparetic individuals participated in the study.
• Balance was assessed using the BESTest, with data split for ANN model training, evaluation, and testing.
• Artificial neural networks (ANNs) were developed and compared against multiple linear regression models (MLRs).

Main Results:

• ANNs demonstrated superior performance in estimating balance status compared to MLRs (RMSE: 4.993 vs. 7.031).
• "Stability in Gait" was the highest contributing section, while "Stability Limits/Verticality" had the lowest contribution.
• Low error (RMSE) values confirmed the success of the ANN modeling for BESTest sections.

Conclusions:

• ANNs provide a more accurate and efficient method for assessing balance in hemiparetic patients than traditional models.
• This study highlights the potential for developing a shorter, more practical "mini-BESTest" subset.
• The findings support using ANNs to refine balance assessments for clinical practice.