تشخيص بيماري كم خوني فقرآهن بر مبناي ويژگي هاي ريخت شناسي با استفاده از روش هاي پردازش تصوير

چكيده انگليسي :

111 Iron Deficiency Anemia Detection Based On Morphological Features By Using Image Processing Techniques Mahsa Lotfi Mahsa lotfi@ec iut ac ir Date of Submission December 2014 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisors Behzad Nazari nazari@cc iut ac ir Saeid Sadri sadri@cc iut ac ir Abstract Iron deficiency Anemia disease is caused by the lack of iron in the human body Iron helps theconstruction of Hemoglobin in the body Consequently the shortage of this material causes the shortage ofHemoglobin protein in the blood By the reduction of Hemoglobin in the blood the number of red blood cellswill decrease and the size of these cells will become smaller than the normal blood cells Also the generalshape of the red blood cells will change Poikilocyte cells Routine medical procedure that deals with thediagnosis of iron deficiency anemia is based on the complete blood cell count CBC a method which isquite time consuming In this thesis a fully automated algorithm is suggested for the diagnosis of irondeficiency anemia The recommended approach has appropriate accuracy and precision and it also does notrequire sophisticated equipments The algorithm is based on the difference between the number ofPoikilocyte cells in normal and Iron deficient blood samples The automatic detection procedure contains 6steps 1 collecting database 2 Preprocessing methods 3 Binarization and Segmentation 4 FeatureExtraction 5 Classification of Poikilocyte cells and 6 The detection of Iron deficient samples In the firststep about 100 Normal and Iron deficient blood samples were collected and the images of these sampleswere taken by a digital camera connected to the microscope The quality of cell images are enhanced byhistogram equalization in the preprocessing step In addition incomplete cells in the margins of cell imagesare eliminated from the images by the use of region labeling Threshold based Binarization method is usedfor the segmentation of cells in the images Methods including Sauvola Niblack Histogram basedBinarization and Otsu are applied to the cell images and the best approach local Otsu is chosen as the finalBinarization method for the segmentation of cell images After the segmentation step 33 proper features areextracted from cell images for the detection of Poikilocyte cells By using extracted features in SupportVector Machine SVM Neural Network Decision Tree DT and K Nearest Neighbor KNN classifiers Poikilocyte cells are classified and counted The classification is based on One Against All theory and thefinal classification decision is determined by using the information of all of the classifiers through maximumvoting algorithm Accroding to maximum voting theory the final class of a test sample is the one thatreceives more votes from the classifiers Finally for discriminating Iron deficient samples from normal ones the number of each type of Poikilocye cells in cell images is considered as a feature for classification Thefeature vector contains 9 different features which determine the number of distinct cell types in a cell image The mentioned feature vector is given to the four already mentioned classifiers Final results show that theautomatic detection of Iron deficiency Anemia disease by the use of the suggested algorithm is accompaniedby 100 precision and 96 67 accuracy Keywords Iron deficiency Poikilocyte cell Support Vector Machine Neural Network Decision Tree K nearest neighbor

لطفي، مهسا

تشخيص بيماري كم خوني فقرآهن بر مبناي ويژگي هاي ريخت شناسي با استفاده از روش هاي پردازش تصوير

سلول پويكيلوسايت , ماشين بردار پشتيبان , شبكه عصبي , درخت تصميم گيري , K-6- امين نزديك ترين همسايه