Early Stroke Prediction Methods for Prevention of Strokes

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Stroke is a leading cause of death, often preceded by transient ischemic attacks (TIAs).
• Current early stroke detection methods like MRI and CT scans are expensive and not always accessible.
• There is a critical need for cost-effective, noninvasive diagnostic tools for early stroke detection, especially in regions with rising stroke cases like India.

Purpose of the Study:

• To develop and evaluate noninvasive, time-series-based algorithms for early stroke detection using processed electroencephalography (EEG) data.
• To compare the performance of Long Short-Term Memory (LSTM), bidirectional LSTM (biLSTM), Gated Recurrent Unit (GRU), and Feedforward Neural Network (FFNN) for stroke prediction.
• To establish a computationally efficient and accurate method for forecasting stroke indicators from brainwave patterns.

Main Methods:

• Utilized processed electroencephalography (EEG) data as input for time-series prediction models.
• Implemented and compared four deep learning algorithms: LSTM, biLSTM, GRU, and FFNN.
• Evaluated the predictive accuracy of each algorithm for early stroke detection based on experimental outcomes.

Main Results:

• All proposed time-series algorithms demonstrated effectiveness in predicting early stroke signs.
• The Gated Recurrent Unit (GRU) algorithm achieved the highest accuracy at 95.6%.
• Bidirectional LSTM (biLSTM) achieved 91% accuracy, LSTM achieved 87%, and FFNN achieved 83% accuracy.

Conclusions:

• The study successfully demonstrates the efficacy of noninvasive EEG analysis for early stroke detection.
• GRU algorithm shows superior performance, offering a promising, accurate, and cost-effective tool for clinical application.
• These findings can significantly aid physicians in timely stroke diagnosis, potentially saving patient lives.