Methods for evaluation of diagnostic imaging instrumentation

Abstract

The Rose model of signal-to-noise ratios in images has been applied to diagnostic imaging instrumentation, extending it by taking into account effects due to finite spatial resolution, background effects and texture. The model leads to an understanding of the impact of the physics of the total system (object and instrument) on output contrast and signal-to-noise ratios. Principally, it leads to an understanding of the fact that nuclear images are not (by and large) statistics limited. The model can also be used to compare different imaging techniques. To the extent that the signal-to-noise ratio is a necessary but not sufficient estimator of the detectability of a lesion, the model does not provide a complete description of the imaging characteristics of an instrument.

Realistic Expectations for the near Term Development of Clinical NMR Imaging

“…These are the S/N level (over a certain area), the contrast between features to be differentiated, and the spatial resolution. Another factor that needs to be included is the texture present in the image [8]. This represents the nonnormal noise.…”

Technology Needs in Medical Imaging

“…This view may not be accurate. We have previously presented an analysis [4,6] based on an extension of Rose's model of signal-to-noise (S/N) in images [5]. This analysis starts by noting that spatial resolution is a system parameter not dependent on the statistics of the image.…”

“…The exact dependence varies for different shapes of point spread function (PSF) and object. For a Gaussian PSF of a full width half maximum (FWHM), and a cylindrical object of diameter D, from [4],…”

“…If the actual granularity is measured in an image, in general we find that it is not flat and that it exceeds statistical noise. The difference between the actual and statistical noise is the texture of the instrument [4][5][6]. For instance, smoothing algorithms produce decreased values of granularity for small values of area.…”

“…[4][5][6] Basically, an imaging system can be considered in terms of its spatial resolution, noise transfer and generation characteristics and the statistics content of the images. These factors act on the subject, which can be considered in terms of its object contrast (Co) between an abnormal feature and the normal anatomy from which it needs to be distinguished.…”