Preprocessing of SPECT projection data: benefits and pitfalls

The Pixon method, a statistically rigorous procedure for adaptive noise suppression that avoids the generation of spurious artifacts yet preserves all the statistically justifiable image features resident in the raw counts, is applied to nuclear studies. The present work focuses on the denoising of projection data at various count levels for subsequent SPECT iterative reconstructions, where each projection is denoised independently. The pitfall of applying such preprocessing to projection images is that tomographic information could be lost, resulting in the loss of weak or small sources. The goal is to investigate the benefits and pitfalls of noise suppression of projection data on the resulting reconstruction, with the ultimate goals to (i) increase sensitivity for detection of lesions of small size and/or of small activity-to-background ratio, (ii) reduce data acquisition time, and (iii) reduce patient dose. We use simulated and measured data and human observer studies, which are analyzed using quantitative measures. An accurate reconstruction at reduced counts using view-independent, noise-reduced projection images can result in significant gain in detectability based on simple SNR measures, but only minor improvements as tested with human observers. At the same time, conservative denoising of the projections results in the loss of small and weak sources, particularly cold lesions. Further analysis and clinical feedback may be warranted, yet it seems that such an approach contains serious pitfalls, likely outweighing the benefits.