Breast Cancer Diagnosis in Mammograms Using Multilevel Wavelet Analysis Mohamed Meselhy Eltoukhy tokhy2478@yahoo.com Brahim Belhaouari Samir Ibrahima FAYE (brahim_balhaouari, ibrahima_faye)@petronas.com.my Electrical and Electronic Engineering University Teknology PETRONAS Abstract- A computer aided diagnosis system based on multi- resolution analysis using the wavelet functions for interpreting digital mammograms is introduced and tested. The goal is to increase the diagnostic accuracy as well as the reproducibility of mammographic interpretation. This system is based on extracting an amount of the biggest coefficients in each level of a multilevel decomposition. A set of images from MIAS (Mammographic Image Analysis Society) database is used in evaluating the system. I. INTRODUCTION Breast cancer afflicts more than one million women in the world each year, and is the 2 nd leading cause of cancer deaths among women worldwide, after lung cancer. Almost 45% of those women live in developing countries. In Malaysia Breast cancer affects one in 19 women. In 2003, 64% of women diagnosed with a breast cancer in Malaysia were between the ages of 40 to 60 years old, the detection of breast cancer in an earlier stage will be better to successful treatment of the disease. It is unfortunate in Malaysia that nearly 40% of the new cases identified each year were already in the very advanced stages of the disease [1]. Developing a computer-aided diagnosis system for cancer diseases, such as breast cancer, to assist physicians in hospitals is becoming of high importance and priority for many researchers and clinical centers. However, achieving this early detection of cancer is not an easy task. The most accurate detection method in the medical environment is biopsy [2]. It is an aggressive invasive procedure which involves some risks, patient discomfort and high cost. Mammography is currently the best technique for reliable detection of early, non-palpable, potentially curable breast cancer. However, Radiologists vary in their interpretation of mammograms. In addition, the interpretation is a repetitive task that requires much attention to minute detail. Therefore, in the past decade, there has been tremendous interest in the use of image processing and analysis techniques [3,4] for Computer Aided Diagnosis (CAD) in digital mammograms. The goal has been to increase diagnostic accuracy as well as the reproducibility of mammographic interpretation. Computer aided detection in digital mammograms usually consists of preprocessing of mammogram image and feature extraction followed by classification. Designing an effective diagnosis system for digital mammograms is still a challenging problem that needs more investigation. Two main functions should be included in such a system. The first is to distinguish between the normal tissues and the different types of tumors such as Microcalcification clusters, Spiculated lesions, Circumscribed masses, Ill-defined mass, Architectural distortion and Asymmetry. The second function is to differentiate between benign and malignant tumors. Wavelet theory comprises a powerful tool for multi- resolution and texture analysis and it can be used very effectively for image processing. By using the multi-resolution capability, the wavelet transform can separate small objects, such as Microcalcifications, from large objects, such as large background structures. In their mammogram analysis study, Liu et al. [5] proved that the use of multi-resolution analysis of mammograms improve the effectiveness of any diagnosis system based on wavelets coefficients. They used a set of statistical features with binary tree classifier in their diagnosis system. Ferreira and Borges [6] indicated that the biggest wavelets coefficients in the low frequency coefficients of wavelets transform could be used as a signature vector for the corresponding mammogram. In their study, they used Haar and Daubechies-4 wavelets with selection of the biggest wavelet coefficients. This approach is applied on some images selected from MIAS dataset. A ROI (Region of Interest) of (64x64) pixels are selected from the original mammograms. Essam et al. [7] used a multi-resolution mammogram analysis in multilevel decomposition to extract a ratio of the biggest coefficients. They used Daubechies-4, -8 and Daubechies-16 wavelets with four level decompositions. Sakka E et al, [8] did a comparative study on some methods which are widely used in Microcalcification detection and feature extraction. The detection of Microcalcifications was achieved by decomposing the mammograms into different frequency sub-bands, and reconstructing the mammogram from the sub-bands containing only high frequencies, due to the fact that Microcalcifications correspond to high frequencies in the frequency domain of the image. The objective of this paper is to construct and evaluate a classifier for mammograms using multilevel wavelet decomposition. We test three different wavelets that are Daubechies-8 (db8), (sym8), and (bior3.7) wavelets. Firstly, a