Application of wavelets to the evaluation of phantom images for mammography quality control

The main goal of this work was to develop a methodology for the computed analysis of American College of Radiology (ACR) mammographic phantom images, to be used in a quality control (QC) program of mammographic services. Discrete wavelet transform processing was applied to enhance the quality of images from the ACR mammographic phantom and to allow a lower dose for automatic evaluations of equipment performance in a QC program. Regions of interest (ROIs) containing phantom test objects (e.g., masses, fibers and specks) were focalized for appropriate wavelet processing, which highlighted the characteristics of structures present in each ROI. To minimize false-positive detection, each ROI in the image was submitted to pattern recognition tests, which identified structural details of the focalized test objects. Geometric and morphologic parameters of the processed test object images were used to quantify the final level of image quality. The final purpose of this work was to establish the main computational procedures for algorithms of quality evaluation of ACR phantom images. These procedures were implemented, and satisfactory agreement was obtained when the algorithm scores for image quality were compared with the results of assessments by three experienced radiologists. An exploratory study of a potential dose reduction was performed based on the radiologist scores and on the algorithm evaluation of images treated by wavelet processing. The results were comparable with both methods, although the algorithm had a tendency to provide a lower dose reduction than the evaluation by observers. Nevertheless, the objective and more precise criteria used by the algorithm to score image quality gave the computational result a higher degree of confidence. The developed algorithm demonstrates the potential use of the wavelet image processing approach for objectively evaluating the mammographic image quality level in routine QC tests. The implemented computational procedures could also enable the performance of advanced analyses to study potential dose reduction in a routine service.