Modification of mammograms for early diagnosis of breast cancer using wavelet and neural networks

In this paper, some modifications of mammogram for the detection of microcalcifications are explained. This can be achieved in two steps. At first stage, wavelet transformation and two statistical properties are used to detect the abnormal pixels and at second stage, a set of ten characteristics are applied to images to cancel some erroneous detected pixels and improving the final result. Therefore, in both steps, a neural network with a hidden layer trained by some vectors is used The results are shown in FROC curves (Free Response Operating Characteristics). For more details, a typical process is explained in the paper. It should be mentioned that the main difference of this research with previous once, in addition to improving results, is use of a real database which is the mammograms of the local patients at a Shiraz hospital.