Quantitative effects of stationary linear image processing on noise and resolution of structure in radionuclide images.

radionuclide image data in an at tempt to improve its diagnostic usefulness in evitably changes both the noise and the resolu tion of structures in the image. Although at present little conclusive evidence exists to mdi cafe whether or not image processing is bene ficial, an understanding of the quantitative effects of various processing techniques on pa rameters of resolution, noise magnitude, and noise texture seems to provide a useful begin ning towards solution of the more difficult problem of understanding the effects of image processing on diagnostic image quality. Quan titative effects of stationary linear image process ing procedures, which include all convolutions and â€oefiltering†• operations, can be predicted rather simply. In the present work, expressions for predicting these quantitative effects are de veloped and are applied to computer-synthesized image data to provide illustrative examples of the effects of typical noise-smoothing and reso lution-enhancing operations. Image processing changes both the noise and the resolution of structures in images. Smoothing the image of a radionuclide distribution to reduce noise, for example, also tends to degrade resolution and to change the â€oetexture†• of the remaining noise. Simi larly, processing to improve spatial resolution tends to increase image noise and to change its texture. In order to understand the effects of various image processing techniques on image quality, it seems reasonable to expect that we must understand their quantitative effects on both resolution and noise. Although the relationship between image quality in the clinical sense and quantitative measures of reso lution and noise is not at all clear at present, an un derstanding of the quantitative effects of various processing techniques seems to be a useful beginning towards solution of the more difficult problem of understanding the effects of image processing on diagnostic image quality. It is the purpose of this paper to point out funda mental mathematical relationships which enable one to predict the effects of stationary linear processing procedures on measures of spatial resolution, noise magnitude, and noise texture†• or â€oecharacter.†• Stationary linear image processing techniques (1†" 11) include all operations on the data describ ing the image that can be achieved by replacing each image element value by a weighted linear combination of the element of interest and sur rounding elements (linearity), in which the com bination rule employed (involving addition, sub traction, and/or differentiation) is independent of position on the image plane (stationarity …