A Novel Approach for Removal of Pectoral Muscles in Digital Mammogram

Abstract

Removal of noise and pectoral muscles are the two important pre-processing steps in CAD system for the diagnosis of breast cancer. This work combines Robust Outlyingness Ratio (ROR) mechanism with extended NL-Means (ROR-NLM) filter based on Discrete Cosine Transform (DCT) for the detection and removal of noise. This method removes Gaussian and impulse noise very effectively without any loss of desired data. For segmenting and removing pectoral muscles, this paper uses global thresholding to identify pectoral muscles, edge detection processes to identify the edge of the full breast and connected component labelling to identify and remove the connected pixels outside the breast region. The result shows that our approach removes Gaussian and impulse noise effectively without any loss of desired data and overall gives 90.06% accuracy.

Efficient Denoising Framework for Mammogram Images with a New Impulse Detector and Non-Local Means

“…But these methods has a primary limitation that each pixel is determined based on the similar decision rule without considering how much impulse-like each pixel is (Marghny and Taloba, 2014). Moreover the performance of these methods are poor for higher noise density (Sreedevi and Sherly, 2015).…”

Mamogramların Sınıflandırılmasında Dokusal Özelliklerin Etkileri

“…Veri tabanı 208'i normal, 63'ü iyi huylu ve 51'i kötü huylu olmak üzere 322 görüntü içermektedir. (Suckling vd., 1994 (Saltanat vd., 2010;Ganesan vd., 2013;Sreedevi ve Sherly, 2015;Pak vd., 2015). Etiketler, ışık patlamaları, mamografi cihazından kaynaklı gürültüler ve pektoral kas alanı yüksek yoğunluk değerine sahip oldukları için ve BDT sistemlerinde yanlış sonuçlara neden olabileceklerinden dolayı mamografi görüntülerinden temizlenmeleri gerekmektedir (Saltanat vd., 2010;Ganesan vd., 2013;Pak vd., 2015).…”

Content-Based Image Retrieval Techniques for Mammographic Images Using Soft Computing Techniques

“…Then the Kalman filter is used to refine the ragged edge [17]. Pectoral muscles can be also segmented by means of global thresholding, followed by edge detection processes to identify the edges of full breast and connected component labeling to identify and remove the connected pixels outside the breast region [18,19].…”