Improved Correction for Signal Dependent Noise Applied to Automatic Detection of Microcalcifications

Abstract

We are developing a computer-aided diagnostic method to assist radiologists in detection and classification of microcalcification clusters. In this study we focus on automated detection of clustered microcalcifications. Previously, a method for detection of microcalcifications in digital mammograms was developed at our institute [1]. It became evident that the detection performance depends strongly on a preprocessing step, in which images are rescaled to equalize image noise which appears to be highly dependent on the grey level. An adaptive approach was used, in which for each image separately high frequency noise was determined as a function of the grey level, and from this information each image was rescaled. For a small database of 40 images this appeared to be the basis for far better detection results than could be obtained by using a scale conversion derived from a phantom recording. In this study these results were validated using a much larger database.