A deep learning-based approach to diagnose mild traumatic brain injury using audio classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Mild traumatic brain injury (mTBI), or concussion, is prevalent in contact sports.
• Current diagnostic methods for mTBI are often subjective and lack reliability.
• Undiagnosed mTBI can lead to severe short-term and long-term health issues, emphasizing the need for objective diagnostic tools.

Purpose of the Study:

• To develop and validate a novel, computationally robust, and objective diagnostic method for mTBI.
• To explore the efficacy of Mel Frequency Cepstral Coefficient (MFCC) features from speech recordings for mTBI detection.
• To assess the performance of a particle swarm optimized bidirectional long short-term memory attention (PSO-Bi-LSTM-A) deep learning model in classifying mTBI.

Main Methods:

• Collected audio recordings of speech from rugby union athletes diagnosed with or without mTBI.
• Extracted Mel Frequency Cepstral Coefficient (MFCC) features from the speech data.
• Trained a novel particle swarm optimized bidirectional long short-term memory attention (PSO-Bi-LSTM-A) deep learning model using the extracted MFCC features.

Main Results:

• The PSO-Bi-LSTM-A model demonstrated high diagnostic performance with sensitivity of 94.7% and specificity of 86.2%.
• An Area Under the Receiver Operating Characteristic Curve (AUROC) score of 0.904 was achieved, indicating strong classification capability.
• The model exhibited minimal overfitting, suggesting robust reliability for current and future datasets.

Conclusions:

• The proposed deep learning approach using speech MFCCs shows significant potential as a pragmatic and objective diagnostic tool for mTBI.
• Further research with larger participant cohorts and model enhancements could improve classification accuracy.
• This method offers a promising avenue for improving mTBI diagnosis, particularly in athletic populations.