Stroke Prediction Using Bi-directional LSTM

Abstract

Abstract—The greatest cause of disability in adults and the elderly—including many social and financial challenges—is stroke. A stroke can result in death if it is not addressed. Patients who have had a stroke typically have aberrant biosignals, such as an altered ECG. Individuals can therefore immediately obtain the right therapy if they are observed and have their bio-signals precisely analyzed and analysed in real-time. However, the majority of stroke diagnostic and prediction systems rely on pricey, challenging to use image processing technologies like CT or MRI. In this study, we created an artificial intelligence (AI)-based stroke prediction system that uses real-time biosignals to identify stroke. Our system made use of both deep learning (Long Short-Term Memory) and machine learning (Random Forest) methods. Real-time EMG (Electromyography) bio-signals from the thighs and calves were gathered, the key characteristics were identified, and prediction models based on regular activities were created. For our suggested system, prediction accuracy values of 90.38% for Random Forest and 98.958% for LSTM were found. This approach may be viewed as an alternative, affordable, real-time diagnosis tool that can accurately anticipate strokes and may one day be used to other illnesses like heart disease. Keywords— Bone Mass Density (BMD), feature extraction, evolutionary algorithm - genetic algorithm, Deep learning modules, Jaccard index