A neural-network-based white blood cell classification system

• Main Authors: Jun-yan Zheng, 鄭俊彥
• Other Authors: Mu-chun Su
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/12825731447939930863

碩士 === 國立中央大學 === 生物醫學工程研究所 === 99 === This thesis presents a new white blood cell classification system. The system involves in two steps. While the first step is the segmentation of a white blood cell from an image, the second step focuses on the recognition of the types of white blood cells. We propose a new segmentation algorithm for the segmentation of white blood cells. First of all, we use the principal component analysis (PCA) to define an elliptical region in HSI color space. Pixels with color in the elliptical region will be regarded as the nucleus and granule of cytoplasm of white blood cell. Through a morphological process, we can segment the white blood cell from the image. Then, several features (e.g., color, shape, local directional pattern)are extracted from the cell. These features are fed into a neural network to recognize the types of the white blood cells. In this thesis, three different neural networks (i.e., multi-layer perceptron (MLP), support vector machines (SVM) and hyper-rectangular composite neutral networks (HRCNN)) are used to recognize white blood cell type. To test the effectiveness of the proposed white blood cell classification system, a total of 450 white blood cells images were used. The image size is 360 × 360 with the JPEG format. Overall, the processing time of each image from the start of segmentation to the end of recognition is about one second. The MLP could achieve the best results with the recognition rate 99% of accuracy. As for the SVM, it could achieve 97% of accuracy in overall data. Although the HRCNN could achieve 100% correct ratio for the training data set, the accuracy for the testing data set was not as high as the other two types of neural networks.