A Proof-of-Concept Development on Speech Analysis for Concussion Detection.

Area of Science:

• Neurology
• Speech Science
• Machine Learning

Background:

• Objective clinical decision-making for neurological disorders is increasingly reliant on advanced analysis techniques.
• Concussion detection currently lacks objective biomarkers, necessitating innovative diagnostic approaches.

Purpose of the Study:

• To evaluate the feasibility of using speech signal analysis for concussion detection.
• To develop and assess machine learning models for discriminating between concussed and healthy individuals based on speech features.

Main Methods:

• A dataset of 82 concussed and 82 healthy participants' speech was collected.
• Mel Frequency Cepstral Coefficients (MFCCs) were extracted to characterize speech articulation.
• Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Decision Tree (DT) classifiers were employed.

Main Results:

• All three machine learning classifiers achieved a Matthew's correlation coefficient score above 0.5 using MFCC-based features.
• The Decision Tree (DT) model demonstrated 78% sensitivity and 75% specificity in identifying concussions.
• These results indicate a significant correlation between speech characteristics and concussion status.

Conclusions:

• Speech analysis, particularly using MFCCs and machine learning, is a feasible approach for concussion detection.
• This study provides proof-of-concept for developing objective, speech-based tools for concussion diagnosis.
• Further research is warranted to refine these methods for clinical application.