Neural features underlying high-accuracy classification of amnestic mild cognitive impairment using...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Cognitive Science

Background:

• Early detection of amnestic mild cognitive impairment (aMCI) is critical for effective intervention.
• Distinguishing aMCI from healthy cognition is challenging using current methods.

Purpose of the Study:

• To develop and validate a machine learning approach for differentiating aMCI from healthy controls.
• To investigate the utility of multimodal sensory-evoked potentials for aMCI detection.

Main Methods:

• Utilized scalp recordings of auditory, visual, and somatosensory event-related potentials (ERPs) and functional connectivity (FC).
• Employed a support vector machine learning pipeline for classification.
• Analyzed ERP amplitude and phase-locking value-based FC across delta to alpha frequencies.

Main Results:

• Multimodal sensory data achieved high classification accuracy (96.1%), sensitivity (97.7%), and specificity (94.3%) for aMCI.
• Reduced ERP amplitude and altered frontal/posterior FC patterns in aMCI were significant predictors.
• Combining data from all sensory modalities outperformed single-modality analysis.

Conclusions:

• Sensory-evoked potentials offer significant clinical potential for early aMCI detection.
• Integrating amplitude and oscillatory FC measures from multiple sensory modalities enhances diagnostic performance.
• This multimodal approach provides a robust, interpretable method for identifying aMCI.