A fluid dynamics based image segmentation approach and pap-smear image data classification

Today, cancer is one of the most important health issues of humanity. Diagnosis of cancer is based on examination of tissue samples, taken from patients by pathologists. One of the most common types of cancer is cervical cancer and it?s diagnosis is made by Pap-smear test which is based on visual examinations of biopsy samples under microscope for detection of anomalies. In order to reduce human error and accelerate the Pap-smear test, computer based decision and detection systems are required. In this study, development of a computer aided analyses system is aimed which is capable of detect anomalies on Pap-smear samples. Morphological features of cells on the samples, such as growth of cell nucleus or deformation on shape of cytoplasm area may provide crucial information about existence of cancer to such a computer aided system. However, in most cases of pap-smear image data, cytoplasm areas are overlapped and there are several types of artifacts on the samples which makes it difficult to extract features from cells for detection of abnormalities. Therefore a new segmentation method is developed and used in this study in addition to conventional morphological methods. Proposed method is based on fluid dynamics. Modeled fluid during the segmentation process is capable of penetrate entire cytoplasm areas even where high degree of cell overlapping occurs. In this way, extracted features form cell are used for classification after segmentation process. Besides, some of the machine learning algorithms are examined and compared in this study for classification of Pap-smear samples for development of a analyses system which is capable of recognize abnormalities on sample cells.