Alcoholism detection via GLCM and particle swarm optimization

Abstract

Alcoholism refers to the addiction to alcohol abuse from which lots of patients around the world suffer. Most of the patients with alcoholism cannot control themselves from consuming too much alcohol. Therefore, alcoholism could damage human bodies, including important organs like livers, eyes, especially brains. Scientists have observed through magnetic resonance imaging (MRI) on brains that the gray matter and white matter of alcoholism patients tend to decrease compared to normal healthy people. Based on this foundation, methods of alcoholism detection using computer-aided diagnosis techniques have been proposed in recent years. Unlike those methods like support vector machine (SVM) or convolutional neural networks (CNN), in this paper, we proposed a novel structure for alcoholism detection. Our structure applied gray level co-occurrence matrix (GLCM) as the feature extractor and adopted particle swarm optimization (PSO) training single-hidden-layer neural network as the classifier. It attained a sensitivity of 92.82±1.93%, a specificity of 91.31±1.71%, a precision of 91.35±1.47%, an accuracy of 92.06±0.87%, a F1 score of 92.06±0.89%, a MCC of 84.17±1.71%, and a FMI of 92.07±0.88%. Our proposed structure not only showed convincing performance via experiment datasets but also presented superiority of speed and simpleness to other strategies. It beat selected six state-of-the-art algorithms in almost every measure except for specificity and precision. From our perspective, our proposed structure for brain image classification is potential for similar fields and tasks.