Statistical approach for image quality evaluation in daily medical practice.

The ROC method usually used for image quality evaluation in medical diagnostics has a lot of advantages, however it is too complicated and inconvenient for daily medical practice. In this paper, a simple, rapid, and unbiased statistical approach is suggested as a method for evaluation of detectability of pathology simulators with small size and low contrast. The method takes into consideration both erroneous and true interpretations; within each are two types: false-positive P(Y/n) and false-negative P(N/s), and true-positive P(Y/s) and true-negative P(N/n). The methodology consists of the following steps: obtaining an image of a phantom, evaluation of the image by estimating the likelihood of test element presence in each of the numbered phantom areas (in percentage), comparing the real disposition of test elements with the results of the estimation, and calculating the various interpretation probabilities with very simple formulas. Assuming the distribution of photons corresponding to test elements and background are both Gaussian, one can obtain receiver operating characteristics (ROC) curves and curves of observer bias for over- and under-reading of test elements presence. Each of these curves is determined by the measurement of a single point. The suggested statistical method can be used for checking and adjusting imaging systems, selecting the physical and technical conditions of a diagnostic procedure, as well as for training and testing operators and physicians. Various designs of statistical phantoms are considered, and the merits of the suggested method and possibilities for its usage in quality assurance medical practice, particularly in x-ray diagnostics, are described.