Detection of Breast Mass in Digital Mammogram from Variable Hidden Neuron Ensemble Based Technique of Mass Classification Using Region Growing Segmentation

Abstract
Digital mammograms are the best method to detect the breast cancer in earlier stage using image processing methods. In this paper a new technology that enhances the variable hidden neural network for detecting the location of breast mass is proposed. First, the pre-processing methods are performed over digital mammogram image. Then the ROI is extracted from the pre-processed image. The Region Growing Segmentation is implemented to separate the part of the image that having the same pixel values from the mammogram. After that the features such as density, mass shape, mass margin, Abnormality Assessment rank, patient age, Subtlety value are extracted. The next process starts off with the creation of the neural networks by varying the number of neurons in the hidden layer. These are then trained, tested and ranked according to the classification of accuracy. To create an ensemble network, the Ten-Fold Cross validation which produces the classifiers, is used. The classifiers are then fused together to create the final ensemble network which reveals whether the image is malignant or benign.
Keyword:Breast cancer, Mammogram, Ensemble neural network, region growing segmentation, adaptive histogram equalization.
I.Introduction
Breast cancer affects large number of women population. The breast cancer is mainly occurred in the inner lining of the lobules which supplies the milk and the milk ducts which carries the milk from the lobules to the nipples and they called as lobular carcinoma and ductal carcinoma respectively. There are many factors that cause breast cancer that are calcium deposition in the breast, radiation exposure, obesity, getting aged, genetic problems and consumption of alcohol. According to the survey taken by the National Cancer Institute, 232240 females and 2240 males affected by breast cancer yearly in USA. Among them 39620 were died. In Australia, one in nine women is diagnosed with the breast cancer in their lifetime [1]. The various methods such as examining the breast, breast ultrasound, breast MRI, biopsy, mammograms, 2D combined with 3D mammograms are used to detect the breast cancer. The mammogram is taking an X-ray by compressing the breast between the two plastic plates [2]. The mammogram gives the better visibility at the skin, greater image flexibility, shorter exam times and more confidence in the results. The image processing is a physical process that takes an image as the input and produces an image or the parameters related to the image as the output. Many computer vision and computational intelligence based techniques are developed in past 20 years. It has a main disadvantage that a consistent and acceptable accuracy has not been achieved. Then Artificial Neural Networks (ANN) has been successful and demonstrated better than the other traditional methods [3]. The ANN is proposed as a simulation of the central nervous system of the human. The artificial neural networks are non-linear information processing device that are built from the interconnected neurons. The modern computers use an algorithmic approach to solve a specific problem but ANN process the information in a similar way that the human brain does. But it has low classification accuracy [4]. Recently, the ensemble techniques have been applied and shown that they have achieved higher accuracy than a single neural network. The ensemble technique is mainly based on the diverse base classifiers which produces better result. The performance of the ensemble technique is improved by the diversity. The diversity is introduced by varying the number of neurons in the hidden layer of the neural network [5]. The ensemble technique distinguishes the similar characteristics of benign and malignant breast masses.