Fast detection of masses in computer-aided mammography

We present a complete method for fast detection of circumscribed mass in mammograms employing a radial basis function neural network (RBFNN). This method can distinguish between tumorous and healthy tissue among various parenchyma tissue patterns, making a decision whether a mammogram is normal or not, and then detecting the masses' position by performing sub-image windowing analysis. In the latter case, with the implementation of a set of criteria, square regions containing the masses are marked as regions of suspicion (ROS). Fast feature extraction significantly reduces the overall processing time, allowing implementation of the method on low-cost PCs. A detailed description of the proposed method is given. The computational efficiency of the feature extraction module and the neural classifier is derived, followed by an analysis of the effects of various factors such as the size of the window shifting step, the digitization quality, and the mammogram size on the computational effort. In addition, the data set and experimental results are presented.