Fig. 3
From: Convolutional neural network -based phantom image scoring for mammography quality control
A selection of noise power spectrums of unmodified test and modified training images for GE (A), Planmed (B), and Siemens (C) systems. NPS curves of the four artificially degraded and three noise-reduced images are shown for each system (gray). NPS of four lower than regular QC exposures and three higher exposures are shown in black. The uppermost NPS curves with the highest noise content represent modified images with the highest noise mixing values and unmodified images with the lowest exposure values. Likewise, the lowermost NPS curves correspond to the highest number of image averages for modified images and the highest exposure for the unmodified images. Pv denotes pixel value and mm denotes millimetres. The curves were smoothed with 5-point moving average