Tuning Butterworth filter's parameters in SPECT reconstructions via kernel-based Bayesian optimization with a no-reference image evaluation metric

TL;DR

This work tackles objective parameter tuning in SPECT reconstruction by introducing a no-reference, perception-based PIQUE score and applying kernel-based Bayesian optimization to select Butterworth filter parameters in filtered backprojection. By coupling PIQUE with a gradient-free BO surrogate built on positive definite kernels, the authors demonstrate robust optimization without ground-truth references, validated on a StarGuide CZT SPECT/CT system using phantom and patient data. The results reveal a plateau of near-optimal configurations and show that the method yields consistent improvements in perceived image quality across different acquisition scenarios, without requiring reference images. The work also discusses practical initialization and exploration–exploitation strategies and points to future directions, including convergence analysis for PIQUE-guided BO and extensions to iterative reconstruction frameworks.

Abstract

In Single Photon Emission Computed Tomography (SPECT), the image reconstruction process involves many tunable parameters that have a significant impact on the quality of the resulting clinical images. Traditional image quality evaluation often relies on expert judgment and full-reference metrics such as MSE and SSIM. However, these approaches are limited by their subjectivity or the need for a ground-truth image. In this paper, we investigate the usage of a no-reference image quality assessment method tailored for SPECT imaging, employing the Perception-based Image QUality Evaluator (PIQUE) score. Precisely, we propose a novel application of PIQUE in evaluating SPECT images reconstructed via filtered backprojection using a parameter-dependent Butterworth filter. For the optimization of filter's parameters, we adopt a kernel-based Bayesian optimization framework grounded in reproducing kernel Hilbert space theory, highlighting the connections to recent greedy approximation techniques. Experimental results in a concrete clinical setting for SPECT imaging show the potential of this optimization approach for an objective and quantitative assessment of image quality, without requiring a reference image.

Figures (13)

• Figure 1: The StarGuide CZT SPECT/CT system. Left: an overview of the scanner. Right: the multi-detector system (from Zorz24).
• Figure 2: The Shepp-Logan phantom (left) and its corresponding sinogram (right) obtained through the Radon transform.
• Figure 3: Values varying $\rho$ and $\omega_0$. The obtained minimum value of PIQUE is $49.966$ at $\rho=9$ and $\omega_0=0.17$.
• Figure 4: FBP reconstructions for a fixed z-slice of the phantom. Left: $\rho=6$, $\omega_0=0.9$. Right: $\rho=9$ and $\omega_0=0.17$ (the optimal one in Figure \ref{['fig:sfere_chart']}).
• Figure 5: Jaszczak phantom, constant $\beta_m$, three initializations. The elements of $X_0$ are depicted in red, while in blue we show the points selected by the algorithm. The blue lines connect subsequent iterations of the scheme.

Theorems & Definitions (1)

• Remark 1