Microscopic spectral imaging system for Automated Segmentation of WBC

Abstract

By using Image fusion technique, combining of both spatial newlineand spectral images, the resulted segmented image provides absolute accurate result and it is an newlineuseful tool for medical diagnosis. newlineThe methodology adopted for automatic segmentation of WBC are Gram-Schmidt newlineorthogonalization segmentation process for extracting spatial features of white blood cells whereas newlineImproved spectral angle mapper (ISAM) for extracting spectral features. Experimental results newlineprove that segmented white blood cells that are classified into five groups with absolute accuracy. newlineMoreover the result extracted from spectral features gives higher accuracy compared to spatial newlineonly cases that we observed in our earlier works i.e. segmentation with Fuzzy-C with snake newlinealgorithm and segmentation by K-means clustering with Gram-Schmidt process. By combining newlinespectral and spatial features, our proposed method produce higher accuracy than conventional newlinemethods and this multispectral imaging procedure is an optimistic approach in Biomedicine. newlineWhite blood cells are classified in to five groups (Neutrophil, Basophil, Eosinophil, Monocytes newlineand Lymphocyte), segmentation is a major step and is crucial for final classification. Quality newlinesegmentation gives better result. A bad quality segmentation leads a poor quality of classification. newlineFor obtaining a best result, it is important to perform segmentation process using quantitative newline newlineanalysis. To obtain better segmentation results, choosing right input image data is so important. newlineIdeally in an object of interest the object boundary is exactly match with segmented boundary, so newlineit is easy to identify that particular cell is a good one or cancerous (Leukemia). But in practice newlineobject of interest referred by multiple segment or one segment referred multiple objects in an newlineimage. So it can lead a poor accuracy of the results. So it is most appropriate to collect the newlineinformation based on both spatial and spectral analysis. In this regards our proposed work newlinesuccessfully able to segment and provide accurate WBC image and is an asset for medical newlinediagnosis. newlineAt last, we discuss the segmentation strategy for the classification of white blood cells and their newlinehealth state and separate the blast cell in to L1, L2 and L3 type. For spatial analysis we select newlineGram-Schmidt process, for spectral analysis we select improved spectral angle mapper (ISAM) newlinemethod. By combining spectral image with spatial image by wavelet transform based image newlinefusion technique. After segmenting WBC from this fused image we separate nucleus and newlinecytoplasm of WBC, by considering pixel intensity of the cell and their mean, kurtosis, skewness newlineand standard deviation we select its group and its health condition. Comparing our result with newlinesource provider result we group them as True positive(TP), True Negetive(TN), False newlinePositive(FP) and False Negative.. from this our accuracy range is around 97% with error rate of newline3%. Specificity and sensitivity are nearly 98% with less false negative and false positive Rate. newline