Applying Multiple Machine Learning Models to Classify Mild Cognitive Impairment from Speech in Community-Dwelling...

Area of Science:

• Neurology
• Computer Science
• Artificial Intelligence

Background:

• Cognitive impairment, particularly Mild Cognitive Impairment (MCI), poses a significant challenge for early detection.
• Developing accessible and accurate screening tools is crucial for timely intervention and management of cognitive decline.

Purpose of the Study:

• To develop and evaluate machine learning models for early detection of Mild Cognitive Impairment (MCI) using speech features.
• To integrate optimized speech classification features with machine learning algorithms for enhanced diagnostic accuracy.

Main Methods:

• Collected audio data from 65 MCI patients and 55 healthy controls (HCs) during a picture description task.
• Utilized the Librosa library for extracting 41 speech classification features.
• Applied Sequential Forward Selection (SFS) for feature optimization in Random Forest (RF) and Support Vector Machine (SVM) models.
• Trained an Extreme Gradient Boosting (XGBoost) model on preprocessed speech data.

Main Results:

• The optimized SVM model achieved an accuracy of 0.825 (AUC: 0.91).
• The optimized RF model reached an accuracy of 0.88 (AUC: 0.86).
• The XGBoost model attained an accuracy of 0.92 (AUC: 0.91), demonstrating superior performance.

Conclusions:

• Speech-based machine learning models significantly enhance the accuracy of distinguishing MCI patients from healthy older adults.
• These findings support the potential of speech analysis integrated with machine learning as a reliable tool for early cognitive deficit identification.