A synthetic image phantom for evaluation of the performance of numerical algorithms for comparison of noisy medical images

The comparison of images which contain significant amount of noise is a common task in the nuclear medicine. The quantitative comparison, performed either by human observer or by numerical algorithms, is not a straightforward task and requires performance evaluation. The performance evaluation necessitates test images which contain noise and a known ground truth. Ideally these images must be applicable to evaluation of the performance of both human observer and numerical algorithms and to their comparison. The objective of this work is to present a concept for a synthetic image (hereafter called "clock phantom"), which is intended to serve as a phantom for evaluation of the performance of tasks dedicated to the problem of comparison of noisy (medical) images. The clock phantom contains both a known ground truth and a significant amount of noise and provides a measure for the performance and accuracy of the algorithms for comparison of images. The clock phantom image resembles a watch dial superimposed on a noisy background. The ticks on the dial have intensities, proportional to the hour they point to and are the objects that need to be identified by the algorithm. The intensities of the background pixels are sampled by Monte Carlo sampling and can follow virtually any known distribution that represents the noise properties of the compared images. The proposed concept for a synthetic image phantom has several advantageous features: it allows fast and accurate quantitative evaluation of the performance of image comparison algorithms which operate on the difference image; It provides a ground-truth against which the results of an observer or algorithm can be verified; The results of numerical algorithms can be easily compared or validated against the performance of a human observer; It can be used in the calculation of receiver operating characteristic (ROC) curves in problems related to the identification of objects on a noisy background.