HEp-2 Sample Picture Yin-positive Classification Algorithm Research

Human Epithelial type 2(HEp-2)cells play an important role in the diagnosis of autoimmune disorder.Currently,the diagnoses of specimens image is generally made by the professionals through visual observation,which is time-consuming and easily influenced by subjective factors.In recent years,researchers have tried to use the computer vision algorithm to make the HEp-2 specimens diagnoses process automatic.The process can be divided into two parts,i.e.positive/negative classification and staining pattern decision.However,few of work has contributed for the first part.In order to fill the gaps in the field,this thesis proposed two methods for positive and negative HEp-2 specimen image classification.The first method classifies the HEp-2 specimen images by combining global and local features.First,the obvious samples are identified according to the global feature.The remaining specimens are processed by preprocessing,image enhancement and object segmentation to separate the target cells from the background.Local features are extracted thereafter for positive/negative classification.The second method solves the problem using deep learning.Firstly,the training dataset are augmented by rotation and scale transformation.Then,the two networks,i.e.VGG-16 and GoogLe Net,are trained with the augmented datasets.Finally,this paper attempts to combine the deep learning method with SVM(support vector machine)by inputing the features from convolution neural network to SVM for classification.The proposed framework was evaluated with using SZU dataset containing 877 positive samples and 413 negative samples.The results indicate that the deep learning method is better than the method combining global and local features,and it can achieve as high as 99.87% accuracy.Finally,this paper also introduces the positive and negative HEp-2 specimen image classification software developed based on the proposed algorithm.