Development Of An Automated Diagnosis System For Classifying Tumor Cells In Multi Stained Cytological Images

Abstract

newlineAny abnormal growth that forms a swelling of the thyroid gland is newlineknown as a thyroid nodule Although the majority of thyroid nodules are newlinebenign about 5 10 of nodules are identified as malignant The fine needle newlineaspiration biopsy FNAB is the most common procedure to determine benign newlineand malignant types of tumor cells present in the thyroid nodules During newlineFNAB procedure a small needle is inserted into the thyroid nodule to collect newlinethe sample cellular material Then the smears are prepared on glass slides newlineusing sample material and screened by a pathologist under a microscope newlineWhile examining such samples the pathologist typically assesses the changes newlinein the distribution of the cells across the sample under examination The result newlineof fine needle aspiration biopsy is dependent on the experience of the newlinephysician performing the procedure However this judgment often leads to newlineconsiderable variation Although the standard manual screening techniques newlineare successful in identifying benign and malignant states of thyroid nodules newlinethey still have serious drawbacks among which misdiagnosis is the most newlinesignificant To overcome these problems and improve the reliability of newlinediagnosis it is necessary to develop an efficient automated diagnosis system newlinefor screening cytological images using image processing techniques newlineFurthermore the automated diagnosis system provides an advantage of newlineprocessing vast amount of medical images in a short period of time newlineAn automated diagnosis system for thyroid cancer is developed to newlinesegment and classify benign and malignant thyroid nodules using multistained newlineFNAB microscopic cytological images Initially the image newlinesegmentation is performed to remove the background staining information newlineand retain the appropriate foreground thyroid cell regions in multistained newlinethyroid FNAB cytological images using mathematical morphology and newlinewatershed transform segmentation methods statistical features newlineare extracted using twolevel discrete wavelet transform DWT newlinedecomposition graylevel cooccurrence matrix GLCM newline newline