Tuning Butterworth filter's parameters in SPECT reconstructions via kernel-based Bayesian optimization with a...

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Processing

Background:

• Single Photon Emission Computed Tomography (SPECT) image reconstruction involves critical parameters affecting clinical image quality.
• Current image quality assessment methods, like Mean Squared Error (MSE) and Structural Similarity Index (SSIM), are subjective or require a ground-truth image.
• Objective and quantitative assessment of SPECT image quality is needed.

Purpose of the Study:

• To investigate the application of a No-Reference Image Quality Assessment (NR IQA) method, specifically the Perception-based Image QUality Evaluator (PIQUE) score, for SPECT imaging.
• To propose a novel approach for optimizing SPECT image reconstruction parameters using PIQUE.
• To enable objective and quantitative image quality assessment without a reference image.

Main Methods:

• Utilized filtered backprojection with a parameter-dependent Butterworth filter for SPECT image reconstruction.
• Employed a kernel-based Bayesian optimization framework, rooted in reproducing kernel Hilbert space theory, for optimizing filter parameters.
• Investigated connections to greedy approximation techniques like P- and f-greedy.

Main Results:

• Demonstrated the potential of the proposed NR IQA approach for evaluating SPECT images.
• Showcased the effectiveness of the Bayesian optimization framework for tuning reconstruction parameters.
• Achieved objective and quantitative image quality assessment in a clinical SPECT setting.

Conclusions:

• The novel application of PIQUE offers a promising solution for objective image quality assessment in SPECT.
• Bayesian optimization provides an effective framework for tuning reconstruction parameters in SPECT.
• This approach overcomes the limitations of traditional subjective and full-reference methods in SPECT image quality evaluation.