SPATIAL AND TEMPORAL CONVOLUTIONS FOR FREQUENCY FILTERING OF RADIONUCLIDE VENTRICULOGRAPHY (RNV)

Fourier filtering has been shown to be an effective means of improving nuclear medicine image quality. A convolution operator can be used in the space and time domains to emulate the characteristics of a Fourier frequency filter. In this study, multiple convolutions were tested to determine the optimal combination for visual analysis of exercise RNV images. The width of the convolutions varied from 3 to 9 point (pt.) matrix size, with cut-off frequencies of 0.1 to 0.25 cycles/sample. 4 spatial filters (SF), 5 temporal filters (TF) and 6 combinations of both SF and TF were compared. 5 observers compared the filtered studies to reach a consensus, based onlack of graininess, chamber separation and detail, edge sharpness, image contrast, and freedom from artifacts. After in vitro red cell labeling with 20 mCi of Tc-99m, 2 mins. of test data were obtained (64 x 64 matrix, 28 frames/cycle, .04 sec./frame, 55 max. cts./pixel) using a 10” camera with no zoom and a 15° slant hole collimator interfaced to a minicomputer (MDS A), TF alone were far more effective than SF when used alone; the TF with the sharpest roll-off (9 pt., CO = .1 (2.5 Hz)) was only marginally better than the 7 pt., CO = .1; SF stronger than 5 x 5, CO = .25 caused edge blurring. A similar test was done after the study was reformatted with double the number of frames54 frames/cycle; .02 sec./frame; and 34 max. cts./pixel. Results were comparable to those noted before but the improvement obtained using the sharper TF (9 pt., C) = .1 (5 Hz)) was more marked. Finally, when acquisition time was reduced from 2 mins. to 30 sec., the same filters performed similarly if the max. cts./pixel were increased from 14 to 34 by multiplication before filtering. Conclusions1) The spatial resolution, max. cts./pixel, and framing rate should be considered when choosing a filter; 2) TF perform better than SF when used alone; 3) While strong TF yields excellent results, light SF should be used to avoid edge blurring; 4) For the usual 2 minute exercise RNV study (28–32 frame) the combination of 7 pt., CO = .1 TF and 5 x 5, CO = .25 is recommended.