Optimized Multiscale Entropy Model Based on Resting-State fMRI for Appraising Cognitive Performance in Healthy...

Area of Science:

• Neuroimaging
• Biostatistics
• Machine Learning

Background:

• Entropy models capture dynamic characteristics of resting-state functional MRI (rfMRI) signals.
• Current entropy models for rfMRI analysis suffer from subjective parameter selection and lack of standardized methods.
• Objective parameter selection is crucial for reliable analysis of rfMRI data.

Purpose of the Study:

• To propose an optimized multiscale entropy (MSE) model for objective parameter selection in rfMRI analysis.
• To identify brain regions with significant entropy differences as potential biomarkers for cognitive performance.
• To utilize these biomarkers for classifying cognitive scores in healthy elderly individuals using machine learning.

Main Methods:

• Healthy elderly volunteers were divided into 'excellent' and 'poor' cognitive groups based on scale tests.
• The MSE model parameters were optimized using sensitivity analyses, including receiver operating characteristic (ROC) and area under the ROC curve (AUC).
• Brain regions with significant entropy differences were identified as biomarkers, and their entropy values served as feature vectors for a probabilistic neural network classifier.

Main Results:

• The optimized MSE model successfully identified brain regions sensitive to cognitive performance.
• Machine learning classification of cognitive scores achieved an accuracy of 80.05%.
• Objective and fixed parameters for MSE were established, enhancing model reliability.

Conclusions:

• The optimized MSE model offers an objective approach to analyzing rfMRI data and identifying cognitive biomarkers.
• This method provides a reliable basis for using entropy measures to assess cognitive function in the elderly.
• The findings support the potential of optimized MSE for future cognitive health assessments.