Combination of Genetic Algorithm and Spiking Neural Network Leaky Integrate-And-Fire Model in Analyzing Brain Ct Scan Image for Stroke Disease Detection

Abstract

Abstract. Stroke is a condition where there is impaired brain function due to lack of oxygen caused by blockage, breakdown, or blood clots inside brain. Diagnosis of stroke is usually based on symptoms, but symptoms are not always the correct measure. In examining a stroke, the most common way to examine a patient is to perform a CT scan of the brain.

Purpose: This study was conducted with the aim of predicting brain scan images consisting of normal brain, ischemic stroke brain, and hemorrhagic stroke brain. It is also to understand how an algorithm works to recognize and predict an image.

Methods/Study design/approach: The image data is trained using machine learning algorithm of neural network, specifically spiking neural network (SNN) using leaky Integrate-and-Fire (LIF) method, which practices the biological performance of human nerves. SNN offers an alternative way of a computational algorithm that mimics the workings of the human brain, especially the nerves in the brain at a low computational cost. In addition, this research optimizes SNN parameters using genetic algorithm (GA). GA is proven to be a successful optimization algorithm from many sources. GA is performed after going through the process in the SNN LIF algorithm, then the parameters in SNN are entered into the algorithm operations in GA until it reaches the most optimal parameter value. Although it requires a large amount of computational time and cost, combining it with SNN will be precise and less labor-intensive.

Result/Findings: Calculation of accuracy results in this study using confusion matrix, conducted on SNN test with LIF method resulted in 90.27%. While with parameter optimization with GA, the final result of the SNN LIF algorithm produces 96.3% accuracy.

Novelty/Originality/Value: This study was conducted to predict stroke disease with human brain images as training data, using the SNN LIF model to train the model and identify patterns that help in predicting stroke risk. For comparison, this research also uses optimization of the model using GA which is useful for determining the core parameters in the training process of the SNN LIF model.