A Novel Approach Analysing the Dynamic Brain Functional Connectivity for Improved MCI Detection

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Dynamic functional connectivity (dFC) is crucial for detecting mild cognitive impairment (MCI) and preventing Alzheimer's disease progression.
• Current deep learning methods for dFC analysis are computationally intensive and lack explainability.
• Existing methods like Root Mean Square (RMS) show insufficient accuracy for reliable MCI detection.

Purpose of the Study:

• To explore novel features for dFC analysis to enhance the accuracy and reliability of MCI detection.
• To develop a more explainable and computationally efficient alternative to deep learning for dFC analysis.
• To validate the proposed features in distinguishing healthy controls (HC) from early MCI (eMCI) and late MCI (lMCI) patients.

Main Methods:

• Utilized a public resting-state functional MRI dataset comprising HC, eMCI, and lMCI individuals.
• Extracted nine novel features (amplitude, spectral, entropy, autocorrelation, time reversibility) from dFC pairwise correlations, alongside RMS.
• Applied Student's t-test and LASSO regression for feature selection, followed by Support Vector Machine (SVM) classification.

Main Results:

• Identified a significant number of differentiating features between HC and MCI groups (6109 for lMCI, 5905 for eMCI).
• Achieved excellent classification performance for both HC vs. lMCI and HC vs. eMCI tasks.
• Demonstrated superior performance compared to most existing dFC analysis methods.

Conclusions:

• The proposed novel features offer a promising and generalizable framework for dFC analysis.
• This approach provides a reliable and potentially more explainable tool for detecting neurological brain diseases like MCI.
• The framework holds potential for application across various brain signals and neurological conditions.