Characterization of statistical prior image constrained compressed sensing (PICCS): II. Application to dose reduction

Area of Science:

• Medical Imaging
• Image Reconstruction
• Radiology

Background:

• Computed tomography (CT) scans involve ionizing radiation, raising patient safety concerns.
• Current CT reconstruction methods like filtered backprojection (FBP) may not optimally balance dose reduction and image quality.
• Prior image constrained compressed sensing (PICCS) is an advanced reconstruction technique offering potential for dose reduction.

Purpose of the Study:

• To evaluate the performance of dose reduction using prior image constrained compressed sensing (DR-PICCS).
• To characterize the impact of statistical modeling of x-ray detection within DR-PICCS.
• To assess the effects of statistical modeling on the spatial resolution of reconstructed CT images.

Main Methods:

• Utilized numerical simulations with a Poisson noise phantom and in vivo animal datasets.
• Compared conventional filtered backprojection (FBP) with DR-PICCS reconstructions.
• Investigated local spatial resolution using the pseudopoint spread function (pseudo-PSF) metric.

Main Results:

• DR-PICCS without statistical modeling reduced pseudo-PSF width to <1.1 voxel width.
• Statistical modeling in DR-PICCS introduced anisotropic resolution (0.75-1.5 voxel width) but improved contrast-to-noise ratio (CNR) by fourfold in vivo.
• Diffusion filtering effectively minimized anisotropy, reducing pseudo-PSF width to below one voxel.

Conclusions:

• DR-PICCS effectively reduces CT noise compared to FBP, with manageable spatial resolution loss.
• Statistical modeling enhances noise characteristics but can introduce resolution anisotropy.
• Directional diffusion filtering combined with DR-PICCS and statistical modeling mitigates spatial resolution anisotropy.