Improvement of scintigrams by computer processing

Computer processing can improve the quality of scintigrams in several ways. It can increase the accuracy with which the image approximates the activity distribution by reversing degradation. It can selectively enhance normal or abnormal structures of interest. It can optimize the use of the display system presenting the image. The usefulness of computer processing must be determined by observer testing and clinical experience. The need to correct distortion in both intensity (nonuniformity) and space can be avoided by attention to calibration and to the setup of the imaging device employed and by use of the sliding energy window technique. Nonuniformity correction, especially for quantitative studies, should not be done using a flood field as this may actually decrease accuracy. Instead, any necessary correction should employ the sensitivity matrix, which measures the variation of sensitivity to a point source with the position of the source. Statistical fluctuations (noise) and degradation of resolution are commonly corrected using linear, stationary techniques [concepts which are defined and developed in the text], but nonstationary techniques appear to be frequently more successful at the expense of increased processing time. Techniques of choice for pure smoothing are nine-point binomial smoothing and variable shape averaging, and those for both sharpening and smoothing (preferred for most modern, high-count scintigrams) are unsharp masking, Metz or Wiener filtering, and bi-regional sharpening. Structures of interest can be enhanced by methods which detect and emphasize changes in local distributions of slope and curvature of intensity. High quality display devices are essential to reap any benefits from degradation correction. Those devices, which must have appropriately high sensitivity and must avoid display artifacts, have become available only recently. Use of the display should be matched to the processing done. Contrast enhancement, e.g. by histogram qualization, for optimal use for each image of the display intensity range, is often helpful. Most scintigram processing is done using computers with about 32K 16-bit words. Floating point hardware is often useful. Most processing methods require 1-30 seconds on such computers and usually under 15 seconds. Processing time tends to be negligible compared to time for user specification of the processing to be done, so the quality of command languages should be of concern. Careful observer studies using phantoms have shown processing to improve detectability of lesions when a single display is used for both processed and unprocessed images, but not when unprocessed images on standard analog displays are compared to processed images on common computer displays...