Mohd. Salleh, Nuryanti
(2015)
Pemprosesan Imej Dan Pengekstrakan Ciri-Ciri Morfologi Secara Automatik Untuk Sel Lesi Payudara Teraspirasi Jarum Halus.
Masters thesis, Universiti Sains Malaysia.
Abstract
Fine needle aspiration cytology (FNAC) has been known as one of the compatible method for detecting breast cancer at its early stage. This method is not only simple and fast; it is also low in cost. However, the process needs to be handled by highly skilled and experienced pathologists. Application in medical imaging has been found to be very useful to assist this process. For solid tumours, the morphological features have been shown to be more sensitive and very useful to diagnose breast cancer. Identification of these morphological features has often been performed either by visualization or using manual computational system. The main objective of this research is to identify automated processing techniques to extract morphological features from fine needle aspirated cell images of breast lesion. The processing techniques studied were firstly, image preprocessing techniques to enhance focal areas. Secondly, image segmentation techniques which concentrate on focal areas. Finally, feature extraction techniques to extract morphological features from the images. In this research, 13 morphological features were extracted from FNAC images which contain single cell while another 40 morphological features were extracted from images which contain cluster cells. The information based on area, boundary, grey level concentration of pixel, compactness, diameter, roundness, nuclear pleomorphism index (NPI), nuclear pleomorphism measurement (NPM), optical density (OD) and integrated nuclear density of breast cells in each image were used to obtain the morphological data. Those image morphological data were then analyzed and compared with manual data through correlation test. Based on the test correlation results obtained show that 9 out of 13 types of correlation data extracted from images of single cells has a strong linear correlation of more than 0.8. As for the image of the cells cluster, only 11 out of 40 types of data extracted has a strong linear correlation relationship.
Actions (login required)
![View Item View Item](/style/images/action_view.png) |
View Item |