Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Direct Low-Dose CT Image Reconstruction on GPU using Out-Of-Core: Precision and Quality Study

M. Chillarón, G. Quintana-Ortí, V. Vidal, G. Verdú

TL;DR

This work tackles the challenge of high-quality, low-dose CT reconstruction with tight clinical time constraints by deploying a dense QR-factorization–based algebraic method on GPU hardware using Out-Of-Core (OOC) techniques. It compares single- and double-precision implementations, showing that single precision substantially reduces reconstruction time while maintaining acceptable image fidelity, with SSIM near 1 and major structural preservation. The method is validated on COVID-CT-MD and DICOM-CT-PD datasets, across 512×512 and 768×768 resolutions, demonstrating the practicality of halving runtime to roughly 1.3 minutes for 2048 slices on modern GPUs, albeit with higher noise in air regions. The findings suggest single-precision OOC QR reconstruction is a viable option for sparse-view CT when combined with adequate projection counts, offering a competitive alternative to traditional iterative methods and enabling near real-time clinical workflows.

Abstract

Algebraic methods applied to the reconstruction of Sparse-view Computed Tomography (CT) can provide both a high image quality and a decrease in the dose received by patients, although with an increased reconstruction time since their computational costs are higher. In our work, we present a new algebraic implementation that obtains an exact solution to the system of linear equations that models the problem and based on single-precision floating-point arithmetic. By applying Out-Of-Core (OOC) techniques, the dimensions of the system can be increased regardless of the main memory size and as long as there is enough secondary storage (disk). These techniques have allowed to process images of 768 x 768 pixels. A comparative study of our method on a GPU using both single-precision and double-precision arithmetic has been carried out. The goal is to assess the single-precision arithmetic implementation both in terms of time improvement and quality of the reconstructed images to determine if it is sufficient to consider it a viable option. Results using single-precision arithmetic approximately halves the reconstruction time of the double-precision implementation, whereas the obtained images retain all internal structures despite having higher noise levels.

Direct Low-Dose CT Image Reconstruction on GPU using Out-Of-Core: Precision and Quality Study

TL;DR

This work tackles the challenge of high-quality, low-dose CT reconstruction with tight clinical time constraints by deploying a dense QR-factorization–based algebraic method on GPU hardware using Out-Of-Core (OOC) techniques. It compares single- and double-precision implementations, showing that single precision substantially reduces reconstruction time while maintaining acceptable image fidelity, with SSIM near 1 and major structural preservation. The method is validated on COVID-CT-MD and DICOM-CT-PD datasets, across 512×512 and 768×768 resolutions, demonstrating the practicality of halving runtime to roughly 1.3 minutes for 2048 slices on modern GPUs, albeit with higher noise in air regions. The findings suggest single-precision OOC QR reconstruction is a viable option for sparse-view CT when combined with adequate projection counts, offering a competitive alternative to traditional iterative methods and enabling near real-time clinical workflows.

Abstract

Algebraic methods applied to the reconstruction of Sparse-view Computed Tomography (CT) can provide both a high image quality and a decrease in the dose received by patients, although with an increased reconstruction time since their computational costs are higher. In our work, we present a new algebraic implementation that obtains an exact solution to the system of linear equations that models the problem and based on single-precision floating-point arithmetic. By applying Out-Of-Core (OOC) techniques, the dimensions of the system can be increased regardless of the main memory size and as long as there is enough secondary storage (disk). These techniques have allowed to process images of 768 x 768 pixels. A comparative study of our method on a GPU using both single-precision and double-precision arithmetic has been carried out. The goal is to assess the single-precision arithmetic implementation both in terms of time improvement and quality of the reconstructed images to determine if it is sufficient to consider it a viable option. Results using single-precision arithmetic approximately halves the reconstruction time of the double-precision implementation, whereas the obtained images retain all internal structures despite having higher noise levels.

Paper Structure

This paper contains 14 sections, 3 equations, 12 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Resolution on data stored in main memory
  4. Resolution on data stored in secondary storage
  5. Experimental Results
  6. Dataset COVID-CT-MD
  7. Simulated scanner parameters
  8. Precision and time efficiency
  9. Image quality
  10. Dataset DICOM-CT-PD
  11. Simulated Data
  12. Quality of images with $512 \times 512$ pixels
  13. Quality of images with $768 \times 768$ pixels
  14. Conclusions

Figures (12)

  • Figure 1: Slices per second versus number of slices being simultaneously computed for single-precision implementations (float) and double-precision implementations (double) on Alinna GPU.
  • Figure 2: Reference and the two reconstructed images of the slice studied (slice with worst metrics in the entire collection). Left image, reference; center image, reconstruction with double-precision; right image, reconstruction with single-precision.
  • Figure 3: Profile of the pixel values along the main diagonal of the slice studied (slice with worst metrics in the entire collection) for the three images: reference, double precision, and single precision.
  • Figure 4: Errors in the reconstructed images of the slice studied (slice with worst metrics in the entire collection). Left, double-precision; right, single-precision.
  • Figure 5: Cropped image
  • ...and 7 more figures