1. UKM Learning and Research Repository
2. Master Thesis / Tesis Master
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Mammogram breast cancer nodule localization based on massive training support vector machine
Authors:	Ashwaq Mukred Saeed Qasem (P56124)
Supervisor:	Siti Norul Huda Sheikh Abdullah, Prof. Madya Dr.
Keywords:	Mammogram Breast cancer Universiti Kebangsaan Malaysia -- Dissertations Dissertations, Academic -- Malaysia
Issue Date:	29-May-2014
Description:	Computer Aided Detection system has been gradually used to speed up the decision- making process made by radiologists in detecting cancerous nodules in the human body. At present, the task of detecting nodules in breast mammogram images has become very complicated due to the complex and non-linear dense texture of the nodules. Active Contour is a popular contour-based segmentation method which uses energy function such as Chan-Vese to define edges of an object in the image segmentation phase. However, Chan-Vese which employs level-set algorithm depends on the divergence and convergence of the intensity values of the image pixels. It tends to segment the outlier component as part of the contour components, which increases the false-positive rate of the selected contour pixels. In overcoming these weaknesses, this research aims to improve the active contour algorithm based on machine learning for the localization of nodules in breast cancer mammograms. Thus, a region-based rejection model is developed and introduced based on the massive training of nodules and non-nodules and Support Vector Machine (SVM). Firstly, Marker-Controlled Watershed Segmentation algorithm is used to extract the initial contour. Then, the Chan-Vese segmentation algorithm is applied to segment the mass in the mammogram. The segmentation output is then passed through the SVM rejection model to reject the non-correctly segmented region. The SVM rejection model has been trained using a set of nodules and non-nodule teaching images with window size of 9x9. To evaluate the proposed work, about 46 mammogram images with their ground truth markers have been collected and verified from the UKM Medical Center. A comparison between the proposed and the standard methods has been conducted. Based on the experimental result, the proposed method with the introduction of the region rejection model based on the massive training of SVM, has outperformed the standard approach. The false positive rate has decreased from 0.19 before using the rejection model to 0.04 after using ejection model. It has also shown significant results in localizing breast cancer tumor.,Master of Information Technology
Pages:	74
Call Number:	Q325.5.Q287 2014 3 tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.