COMPARISON OF 4 SCATTER CORRECTION METHODS USING MONTE-CARLO SIMULATED SOURCE DISTRIBUTIONS

Scatter correction in SPECT is important for improving image quality, boundary detection and the quantification of activity in different regions. This paper presents a comparison of four scatter correction methods, three using more than one energy window and one convolution-subtraction correction method using spatial variant scatter line-spread functions. Methods: The comparison is based on Monte Carlo simulated data for point sources on- and off-axis, hot and cold spheres of different diameters, and a clinically realistic source distribution simulating brain imaging. All studies were made for a uniform cylindrical water phantom. Since the nature of the detected photon is known with Monte Carlo simulation, separate images of primary and scattered photons can be recorded. These can then be compared with estimated scatter and primary images obtained from the different scatter correction methods. The criteria for comparison were the normalized mean square error, scatter fraction, % recovery and image contrast. Results: All correction methods significantly improved image quality and quantification compared to those obtained with no correction. Quantitatively, no single method was observed to be the best by all criteria for all the source distributions. Three of the methods were observed to perform the best by at least one of the criteria for one of the source distributions. For brain imaging, the differences between all the methods were much less than the difference between them and no correction at all. Conclusion: It is concluded that performing scatter correction is essential for accurate quantification, and that all four methods yield a good, but not perfect, scatter correction. Since it is hard to distinguish the methods consistently in terms of their performance, it may be that the choice should be made on the basis of ease of implementation.