استخراج ويژگي ها و دسته بندي سلول ها در تصاوير تست پاپ اسمير سرطان دهانه رحم

چكيده انگليسي :

Cervical Pap Smear Feature Extraction and Classification of Cells in Test Images Nahid Ghofrani nghofrani@ec iut ac ir Date of Submission 16 03 2009 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisors Saeed Sadri sadri@cc iut ac ir Rassul Amirfattahi fattahi@cc iut ac ir Abstract After breast cancer cervical cancer is the second cause of death among women Also this cancer has the tenth rank in the world among women and men The statistics show that early diagnosis and treatment in the early stages can decrease mortality rate of this disease So the number of women tests is very large in each city and throughout the country There are various methods for diagnosing this disease At present in many countries including our country Iran one of the most common and low cost methods for detection and classification of cervical cancer is Pap Smear test In Pap Smear test a layer of cells is picked up from the cervix and is put on a slide Diagnosis of cancer or pre cancerous stages depends on finding abnormal cells on the slide by a cytologist Each of the slides may include hundreds or thousands of cells therefore usually it is very hard to evaluate cells on the slide completely because of the large number of cells or inattention and fatigue of specialist So it is appropriate to use a computer aided diagnosis system to evaluate existence of cancer or pre cancerous lesions for use of cytologist In this research we try to classify cells which are obtained from Pap Smear test into two normal and abnormal groups In cancer and pre cancerous cases normal cells change to abnormal cells First various feature groups such as morphometric densitometric brightness statistical and textural features including Co occurrence Matrix Run Length Matrix and Statistical Geometric features are extracted from cell images Our study shows that among the above mentioned morphometric and co occurrence matrix features are the best feature groups The number of these features is very large In the next step we propose a method to select effective features by means of Genetic Algorithm and Recursive Feature Elimination Method for reducing feature numbers and improving algorithm performance For classifying cells among the existing methods such as different forms of Neural Networks and Support Vector Machines SVM we use SVM as one of the most reliable classification methods SVM is very suitable for classifying data which can not be classified by hyperplanes and need non linear classification methods To evaluate the effect of each feature we use Receiver Operating Curve ROC ROC curve present True Positive Rate versus False Positive Rate We calculate ROC curves for different feature groups The results show that area under ROC 0 9632 is obtained for a special group of features which are selected by Genetic algorithm Also True Positive rate and False Positive rate obtained by SVM are 96 and 4 respectively Key Words Cervical Cancer Pap Smear Test Computer Aided Diagnosis System Feature Extraction Feature Selection Classification pdfMachine Is a pdf writer that produces quality PDF files with ease Produce quality PDF files in seconds and preserve the integrity of your original documents Compatible across nearly all Windows platforms if you can print from a windows application you can use pdfMachine Get yours now

غفراني، ناهيد

استخراج ويژگي ها و دسته بندي سلول ها در تصاوير تست پاپ اسمير سرطان دهانه رحم

تشخيص كامپيوتري , انتخاب ويژگي , كرنل , SVM