An automatic method to determine cutoff frequency based on image power spectrum [SPECT]

We present an algorithm for automatically choosing filter cutoff frequency (F_c) using the power spectrum of the projections. The method is based on the assumption that the expectation of the image power spectrum is the sum of the expectation of the blurred object power spectrum (dominant at low frequencies) plus a constant value due to Poisson noise. By considering the discrete components of the noise-dominated high-frequency spectrum as a Gaussian distribution N(μ,σ), the Student t-test determines F_c as the highest frequency for which the image frequency components are unlikely to be drawn from N(μ,σ). The method is general and can be applied to any filter. In this work, we tested the approach using the Metz restoration filter on simulated, phantom, and patient data with good results. Quantitative performance of the technique was evaluated by plotting recovery coefficient (RC) versus NMSE of reconstructed images