Detecting Post-Stroke Aphasia Via Brain Responses to Speech in a Deep Learning Framework

Area of Science:

• Neuroscience
• Computational Linguistics
• Medical Technology

Background:

• Aphasia diagnosis traditionally relies on time-consuming behavioral tests with limitations in accuracy and ecological validity.
• Existing methods can be influenced by co-occurring motor and cognitive deficits in individuals with aphasia (IWA).

Purpose of the Study:

• To develop and validate an automated screening tool for speech processing impairments in aphasia using neural tracking.
• To assess the feasibility of deep learning models for objective and efficient aphasia detection.

Main Methods:

• Modeled electroencephalography (EEG) responses to acoustic, segmentation, and linguistic speech features using convolutional neural networks.
• Trained models on healthy participants and evaluated on a cohort of 26 IWA and 22 healthy controls.
• Employed a support vector machine classifier with neural tracking measures for aphasia detection.

Main Results:

• Individuals with aphasia (IWA) exhibited significantly decreased neural tracking across all speech representations.
• The automated tool achieved 85.42% accuracy in individual-level aphasia detection.
• The screening process required only 9 minutes of EEG data, demonstrating time efficiency.

Conclusions:

• The developed deep learning framework accurately detects aphasia by analyzing neural tracking of speech.
• This automated, time-efficient, and robust method shows significant potential for clinical application in aphasia screening.
• The approach offers a more objective and generalizable alternative to traditional diagnostic methods.