A Scheme Based on Fractal Dimension with SVM Classifier and Genetic Algorithm for Rat Neuron Smear Images Classification

• Main Authors: Jia-Han Lin, 林佳翰
• Other Authors: Shyr-Shen Yu
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/dy76u8

碩士 === 國立中興大學 === 資訊科學與工程學系 === 106 === Stroke is cerebrovascular incident(CVI). It referred to the brain cell death caused by cerebral ischemia. It can divide into ischemic stroke and hemorrhage stroke. They make the brain dysfunction, and lead to a variety of dangerous symptoms, so stroke is the top ten causes of death in several years. The experimental images provided by Taichung Veterans General Hospital. The experimental body was a rat with cerebral artery occlusion. We used NeuN antibody staining rat brain neurons smear image, and all images divided into five levels by days. Each levels contains 20 images, and the total of images contains 100 images. On the other hand, we changed images size into same size because of images size are different. This preprocessing can obtain better recognition rates. In this paper, proposed method not only combined gamma correction with DBC (Differential Box Counting) method of fractal dimension but also conversion of color space. As the test results show, CIELab of color space was be use, and L component have better results. Finally, we used SVM(Support Vector Machine) to do machine learning and classification. On the other hand, we used GA（Genetic Algorithm）to find parameters of SVM. It make SVM have better capability in pattern recognition. In addition, we used k-fold cross validation to get results of classification. The accuracy of the experimental results achieved 96.8 ± 0.4. In addition, this paper also combines the decision tree method to obtain a high recognition rate of 97.8±0.0075. As the experimental results show, the proposed method can effectively recognize different stages of stroke images. This method can help doctor to do diagnosis and treatment of stroke, and improve the health of people.