Automatic Scoring of American College of Radiology (acr) Phantom Image Using Image Processing Technique By

A phantom image is used to perform quality control in mammographic facilities. The interpretation of the ACR phantom images is depend upon the experience, visual threshold, scoring criteria and state of the observer. This paper summarizes the first step of automating this score using digital image processing, namely template matching via cross correlation, and suggests a metric for the visibility of the test objects in the images. A set of ACR phantom images was obtained. These images were presented to a medical physicist, who indicated the number of circular test objects visible in each image. The radiographs of the phantom are scanned with 8 bits depth resolution at 600 dpi. These images are then filtered using a Gaussian filter having a width that is optimized to the scale size of the feature. A new circle edge detector was developed since the conventional Sobel and Roberts edge detectors are omnidirectional local operators, which do not take into account the geometry of the feature. The results were compared with our template matching technique. Our proposed visibility metric was consistent with the physicist’s scoring results and has the added advantage that it quantitatively scores the ACR phantom. In addition a critical value of the visibility could be defined which can then be applied automatically for quality control purposes to digital mammography systems.