Non-invasive stroke diagnosis using speech data from dysarthria patients.

Abstract

Acute Ischemic Stroke (AIS) is a major cause of disability and can lead to death in severe cases. A common symptom of AIS, dysarthria, significantly impacts the quality of life of patients. In this study, we developed a deep learning model using dysarthria data for cost-effective and non-invasive brain stroke diagnosis. We utilized models such as ResNet50, InceptionV4, ResNeXt50, SEResNeXt18, and AttResNet50 to effectively extract and classify speech features indicative of stroke symptoms. These models demonstrated high performance, with Sensitivity, Specificity, Precision, Accuracy, and F1-score values reaching 96.77%, 96.08%, 92.82%, 95.52%, and 93.82%, respectively. Our approach offers a non-invasive, cost-effective alternative for early stroke detection, with potential for further accuracy improvements through additional research. This method promises rapid, economical early diagnosis, which could positively impact long-term treatment and healthcare options.