Non-invasive stroke diagnosis using speech data from dysarthria patients

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Acute Ischemic Stroke (AIS) is a leading cause of long-term disability and mortality.
• Dysarthria, a speech impairment, is a common symptom of AIS, significantly affecting patient quality of life.
• Current diagnostic methods for stroke can be invasive and costly.

Purpose of the Study:

• To develop a cost-effective and non-invasive deep learning model for diagnosing Acute Ischemic Stroke (AIS).
• To utilize speech features, specifically dysarthria, for stroke detection.
• To evaluate the performance of various deep learning models in classifying stroke indicators from speech data.

Main Methods:

• Employed deep learning architectures including ResNet50, InceptionV4, ResNeXt50, SEResNeXt18, and AttResNet50.
• Trained models on speech data to extract and classify features indicative of stroke symptoms.
• Evaluated model performance using key metrics: Sensitivity, Specificity, Precision, Accuracy, and F1-score.

Main Results:

• The developed deep learning models achieved high performance metrics.
• Sensitivity reached 96.77%, Specificity 96.08%, Precision 92.82%, Accuracy 95.52%, and F1-score 93.82%.
• The models effectively classified speech features associated with AIS.

Conclusions:

• The proposed deep learning approach offers a promising non-invasive and cost-effective method for early stroke detection.
• This technology has the potential to improve patient outcomes through rapid diagnosis.
• Further research may lead to enhanced accuracy and broader clinical application for stroke diagnosis.