pith. sign in

arxiv: 1907.01891 · v1 · pith:55ZRKRBOnew · submitted 2019-06-27 · 📡 eess.IV · cs.DC

Computed tomography medical image reconstruction on affordable equipment by using out-of-core techniques

M\'onica Chillar\'on , Gregorio Quintana-Ort\'i , Vicente Vidal , Gumersindo Verd\'u This is my paper

Pith reviewed 2026-05-25 14:18 UTC · model grok-4.3

classification 📡 eess.IV cs.DC
keywords computed tomographyimage reconstructionQR factorizationout-of-core techniquesalgebraic reconstructionmedical imagingmulticore processorssolid-state drives
0
0 comments X

The pith

A QR factorization method with out-of-core techniques reconstructs high-quality CT images quickly on standard hardware.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper proposes a reconstruction method for computed tomography images that uses QR factorization together with out-of-core techniques. The approach runs on ordinary multicore processors and solid-state drives without requiring large RAM or specialized accelerators. It targets the main drawback of algebraic reconstruction methods, which produce high-quality results from fewer views but have historically been too slow for routine clinical use. If the method works as described, it would make these accurate reconstructions practical on affordable equipment in daily medical practice.

Core claim

The paper claims that a new reconstruction method based on the QR factorization, implemented with out-of-core techniques, is very efficient on affordable equipment consisting of standard multicore processors and standard Solid-State Drives, allowing it to boost the performance of reconstructions and implement a reliable and competitive method that reconstructs high-quality CT images quickly.

What carries the argument

QR factorization implemented with out-of-core techniques to process large matrices without loading them entirely into RAM.

If this is right

  • Algebraic CT reconstruction becomes fast enough for routine clinical workflows on affordable equipment.
  • High-quality images remain obtainable while using fewer projection views than traditional analytical methods.
  • The approach eliminates the need for specialized accelerators or very large memory banks during reconstruction.
  • Daily clinical practice can adopt algebraic methods without incurring high hardware costs or long wait times.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same out-of-core strategy could be applied to other matrix factorizations used in algebraic tomography.
  • Medical imaging centers without access to GPUs or large servers might adopt algebraic techniques more readily.
  • Further tuning of the out-of-core block size could yield additional speedups on consumer-grade SSDs.

Load-bearing premise

The out-of-core version of QR factorization keeps the image quality of algebraic reconstruction while delivering the claimed speed gains on ordinary computers.

What would settle it

Execute the out-of-core QR method on standard multicore hardware with limited RAM and compare both the peak signal-to-noise ratio of the resulting CT images and the wall-clock runtime against a conventional in-core algebraic solver on the same data.

Figures

Figures reproduced from arXiv: 1907.01891 by Gregorio Quintana-Ort\'i, Gumersindo Verd\'u, M\'onica Chillar\'on, Vicente Vidal.

Figure 1
Figure 1. Figure 1: An illustration of the first tasks performed by an algorithm-by-blocks for computing the QR factorization. The ‘•’ symbol represents a non-modified element by the current task, ‘∗’ represents a modified element by the current task, and ‘◦’ represents a nullified element (by the current task or by a previous task). The continuous lines surround the blocks involved in the current task. output is matrix D (co… view at source ↗
Figure 2
Figure 2. Figure 2: CT images B. Performance study The computer used in the performance experiments featured one Intel i7-7800Xr CPU (6 physical cores) and 128 GiB of RAM in total. The clock frequency of the processor was 3.50 GHz, and the so-called Max Turbo Frequency was 4.00 GHz. In addition to one small SSD for storing the operating system and programming tools, the computer had two disks that were employed in the experim… view at source ↗
Figure 3
Figure 3. Figure 3: Overall times and decomposed times of the initial configuration (B-OOC + HDD) for solving a linear system with A of dimension 266, 500 × 262, 144, and B of dimension 266, 500 × k, where k is the number of slices. they are stored in RAM was 128, since this size usually renders good performances when processing matrices of size 10240. In the rest of the codes not developed by us (matrix-matrix products, etc.… view at source ↗
Figure 4
Figure 4. Figure 4: compares the performances of the four configurations described above: Basic OOC AB on HDD, Overlap￾ping OOC AB on HDD, Basic OOC AB on SSD, and Overlapping OOC AB on SSD. The top subplot shows the times in seconds (lower is better), whereas the bottom subplot shows the speedup (higher is better) with respect to the initial configuration (basic OOC AB on HDD). The speedup is computed as the quotient of the … view at source ↗
Figure 5
Figure 5. Figure 5: Overall times and decomposed times of three con￾figurations for solving a linear system with A of dimension 266, 500 × 262, 144, and B of dimension 266, 500 × k, where k is the number of slices.     "      [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
read the original abstract

As Computed Tomography (CT) scans are an essential medical test, many techniques have been proposed to reconstruct high-quality images using a smaller amount of radiation. One approach is to employ algebraic factorization methods to reconstruct the images, using fewer views than the traditional analytical methods. However, their main drawback is the high computational cost and hence the time needed to obtain the images, which is critical in the daily clinical practice. For this reason, faster methods for solving this problem are required. In this paper, we propose a new reconstruction method based on the QR factorization that is very efficient on affordable equipment (standard multicore processors and standard Solid-State Drives) by using out-of-core techniques. Combining both affordable hardware and the new software, we can boost the performance of the reconstructions and implement a reliable and competitive method that reconstructs high-quality CT images quickly.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a QR-factorization-based algebraic method for CT image reconstruction that employs out-of-core techniques to achieve practical performance on commodity multicore CPUs and standard SSDs, thereby enabling high-quality low-dose reconstructions without specialized accelerators or large RAM.

Significance. If the performance and quality claims are substantiated, the work would lower the barrier to algebraic reconstruction in clinical settings by demonstrating that out-of-core linear-algebra techniques can deliver usable run times on affordable hardware.

major comments (2)
  1. [Abstract] Abstract: the claims that the method is 'very efficient' and reconstructs 'high-quality CT images quickly' are unsupported by any quantitative data (wall-clock times, RMSE/PSNR values, or comparisons against FBP or other algebraic solvers).
  2. [Results / Experiments] The central performance claim requires that the out-of-core QR implementation on the system matrix A preserves numerical stability while overcoming I/O latency on SSDs; no section supplies timings on the exact hardware class advertised or quantifies the I/O overhead for the sparse projection matrices involved.
minor comments (1)
  1. Notation for the system matrix A and the block partitioning used in the out-of-core QR is not introduced until late; an early diagram or pseudocode would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the major comments point by point below and will revise the manuscript to better substantiate the performance claims with quantitative data.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claims that the method is 'very efficient' and reconstructs 'high-quality CT images quickly' are unsupported by any quantitative data (wall-clock times, RMSE/PSNR values, or comparisons against FBP or other algebraic solvers).

    Authors: We agree that the abstract would benefit from explicit quantitative support. The results section of the manuscript reports wall-clock times, quality metrics, and comparisons, but these were not summarized in the abstract. We will revise the abstract to include key numbers (e.g., reconstruction times on the target hardware and RMSE/PSNR values versus FBP) while preserving the overall length. revision: yes

  2. Referee: [Results / Experiments] The central performance claim requires that the out-of-core QR implementation on the system matrix A preserves numerical stability while overcoming I/O latency on SSDs; no section supplies timings on the exact hardware class advertised or quantifies the I/O overhead for the sparse projection matrices involved.

    Authors: The experiments were run on standard multicore CPUs with SSDs as described. We will expand the results section to explicitly report numerical stability indicators (e.g., residual norms or condition-number checks) and provide a breakdown of I/O versus compute time for the sparse matrices, including additional tables if needed to quantify overhead. revision: yes

Circularity Check

0 steps flagged

No circularity: engineering application of known QR and out-of-core techniques

full rationale

The paper proposes an out-of-core QR factorization method for algebraic CT reconstruction on commodity hardware. No derivation chain, equation, or claim reduces to its own inputs by construction. The abstract and description present the approach as a direct combination of standard linear-algebra routines with I/O optimizations; no self-definitional relations, fitted parameters renamed as predictions, or load-bearing self-citations appear. Performance claims are externally falsifiable on the advertised hardware class and do not rely on internal redefinitions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Assessment limited to abstract; no free parameters, invented entities, or non-standard axioms are explicitly introduced in the provided text.

axioms (1)
  • domain assumption Algebraic methods such as QR factorization can reconstruct high-quality CT images from fewer views than analytical methods.
    This premise underpins the decision to use fewer views and is stated in the abstract as the motivation for algebraic approaches.

pith-pipeline@v0.9.0 · 5686 in / 1076 out tokens · 29104 ms · 2026-05-25T14:18:57.853307+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages

  1. [1]

    A. C. Kak and M. Slaney, Principles of Computerized Tomographic Imaging . SIAM, jan 2001, vol. 33

    work page 2001

  2. [2]

    Projected cancer risks from computed tomographic scans performed in the united states in 2007,

    A. B. De Gonz´ alez, M. Mahesh, K.-P . Kim, M. Bhargavan, R. Lewis, F. Mettler, and C. Land, “Projected cancer risks from computed tomographic scans performed in the united states in 2007,” Archives of internal medicine , vol. 169, no. 22, pp. 2071–2077, 2009

    work page 2007

  3. [3]

    Cancer risks from diagnostic rad iology,

    E. Hall and D. Brenner, “Cancer risks from diagnostic rad iology,” The British journal of radiology , vol. 81, no. 965, pp. 362–378, 2008

    work page 2008

  4. [4]

    A three-dimensional-weighted cone beam filte red backprojection (CB-FBP) algorithm for image reconstruction in volumetric CT helical scanning,

    X. Tang, J. Hsieh, R. A. Nilsen, S. Dutta, D. Samsonov, and A. Hagiwara, “A three-dimensional-weighted cone beam filte red backprojection (CB-FBP) algorithm for image reconstruction in volumetric CT helical scanning,” Physics in Medicine & Biology , vol. 51, no. 4, p. 855, 2006

    work page 2006

  5. [5]

    Fan-beam a nd cone-beam image reconstruction via filtering the backpro jection image of differentiated projection data,

    T. Zhuang, S. Leng, B. E. Nett, and G.-H. Chen, “Fan-beam a nd cone-beam image reconstruction via filtering the backpro jection image of differentiated projection data,” Physics in Medicine & Biology , vol. 49, no. 24, p. 5489, 2004

    work page 2004

  6. [6]

    A combination-weighted Feldkamp-based reconstr uction algorithm for cone-beam CT,

    S. Mori, M. Endo, S. Komatsu, S. Kandatsu, T. Y ashiro, and M. Baba, “A combination-weighted Feldkamp-based reconstr uction algorithm for cone-beam CT,” Physics in Medicine & Biology , vol. 51, no. 16, p. 3953, 2006

    work page 2006

  7. [7]

    Iterative re construction techniques for computed tomography part 1: technical princ iples,

    M. J. Willemink, P . A. de Jong, T. Leiner, L. M. de Heer, R. A . Nievelstein, R. P . Budde, and A. M. Schilham, “Iterative re construction techniques for computed tomography part 1: technical princ iples,” European radiology, vol. 23, no. 6, pp. 1623–1631, 2013

    work page 2013

  8. [8]

    Iterative re construction techniques for computed tomography part 2: initial results in dose reduction and image quality,

    M. J. Willemink, T. Leiner, P . A. de Jong, L. M. de Heer, R. A . Nievelstein, A. M. Schilham, and R. P . Budde, “Iterative re construction techniques for computed tomography part 2: initial results in dose reduction and image quality,” European radiology, vol. 23, no. 6, pp. 1632–1642, 2013

    work page 2013

  9. [9]

    Algebraic reconstruction in CT from lim ited views,

    A. H. Andersen, “Algebraic reconstruction in CT from lim ited views,” IEEE Transactions on Medical Imaging , vol. 8, no. 1, pp. 50–55, 1989

    work page 1989

  10. [10]

    Simultaneous algebraic re construction technique (SART): a superior implementation of the ART algorithm,

    A. H. Andersen and A. C. Kak, “Simultaneous algebraic re construction technique (SART): a superior implementation of the ART algorithm,” Ultrasonic Imaging, vol. 6, no. 1, pp. 81–94, 1984

    work page 1984

  11. [11]

    A Novel Weighted Total Difference Bas ed Image Reconstruction Algorithm for Few-View Computed To mography,

    W. Y u and L. Zeng, “A Novel Weighted Total Difference Bas ed Image Reconstruction Algorithm for Few-View Computed To mography,” PLoS ONE , vol. 9, no. 10, 2014

    work page 2014

  12. [12]

    Iterative reconstruction from few-view projections,

    L. Flores, V . Vidal, and G. V erd ˜Ao, “Iterative reconstruction from few-view projections,” Procedia Computer Science, vol. 51, pp. 703–712, 2015

    work page 2015

  13. [13]

    Parallel CT image reconstruction based on GPUs,

    L. A. Flores, V . Vidal, P . Mayo, F. Rodenas, and G. V erd´ u , “Parallel CT image reconstruction based on GPUs,” Radiation Physics and Chemistry, vol. 95, pp. 247–250, 2014. 12

    work page 2014

  14. [14]

    Combining grid computing and Docker containers for the study and parametrization of CT image reconstruction methods,

    M. Chillar´ on, V . Vidal, D. Segrelles, I. Blanquer, and G. V erd´ u, “Combining grid computing and Docker containers for the study and parametrization of CT image reconstruction methods,” Procedia Computer Science , vol. 108, pp. 1195–1204, 2017

    work page 2017

  15. [15]

    Total V ariation-Stokes Strate gy for Sparse- View X-ray CT Image Reconstruction,

    Y an Liu, Zhengrong Liang, Jianhua Ma, Hongbing Lu, Ke Wa ng, Hao Zhang, and W. Moore, “Total V ariation-Stokes Strate gy for Sparse- View X-ray CT Image Reconstruction,” IEEE Transactions on Medical Imaging , vol. 33, no. 3, pp. 749–763, mar 2014

    work page 2014

  16. [16]

    Performance compar ison between total variation (TV)-based compressed sensin g and statistical iterative reconstruction algorithms,

    J. Tang, B. E. Nett, and G.-H. Chen, “Performance compar ison between total variation (TV)-based compressed sensin g and statistical iterative reconstruction algorithms,” Physics in Medicine and Biology , vol. 54, no. 19, pp. 5781–5804, oct 2009

    work page 2009

  17. [17]

    Low-Dose Micro-CT Imaging for V ascul ar Segmentation and Analysis Using Sparse-View Acquisitions,

    B. V andeghinste, S. V andenberghe, C. V anhove, S. Stael ens, and R. V an Holen, “Low-Dose Micro-CT Imaging for V ascul ar Segmentation and Analysis Using Sparse-View Acquisitions,” PLoS ONE , vol. 8, no. 7, p. e68449, jul 2013

    work page 2013

  18. [18]

    Improved compressed sensing-based algorithm for spa rse-view CT image reconstruction

    Z. Zhu, K. Wahid, P . Babyn, D. Cooper, I. Pratt, and Y . Car ter, “Improved compressed sensing-based algorithm for spa rse-view CT image reconstruction.” Computational and mathematical methods in medicine , vol. 2013, p. 185750, mar 2013

    work page 2013

  19. [19]

    A review of breast tomosynthesis. Par t II. Image reconstruction, processing and analysis, and ad vanced applications,

    I. Sechopoulos, “A review of breast tomosynthesis. Par t II. Image reconstruction, processing and analysis, and ad vanced applications,” Medical physics , vol. 40, no. 1, 2013

    work page 2013

  20. [20]

    CT Image Reconstruction fro m Sparse Projections Using Adaptive TpV Regularization

    H. Qi, Z. Chen, and L. Zhou, “CT Image Reconstruction fro m Sparse Projections Using Adaptive TpV Regularization.” Computational and mathematical methods in medicine , vol. 2015, p. 354869, may 2015

    work page 2015

  21. [21]

    QR-factorization Al gorithm for Computed Tomography (CT): Comparison with FDK and Conjugate Gradien t (CG) Algorithms,

    M. J. Rodr´ ıguez-Alvarez, F. Sanchez, A. Soriano, L. Mo liner, S. Sanchez, and J. M. Benlloch, “QR-factorization Al gorithm for Computed Tomography (CT): Comparison with FDK and Conjugate Gradien t (CG) Algorithms,” IEEE Transactions on Radiation and Plasma Medical Sciences, vol. 2, no. 5, pp. 459–469, 2018

    work page 2018

  22. [22]

    CT image recons truction with SuiteSparseQR factorization package,

    M. Chillar´ on, V . Vidal, and G. V erd´ u, “CT image recons truction with SuiteSparseQR factorization package,” Radiation Physics and Chemistry, 2019

    work page 2019

  23. [23]

    CT Medical Imaging Reconstruction Using Direct Algebraic Method s with Few Projections,

    M. Chillar´ on, V . Vidal, G. V erd´ u, and J. Arnal, “CT Medical Imaging Reconstruction Using Direct Algebraic Method s with Few Projections,” in Computational Science – ICCS 2018 . Cham: Springer International Publishing, 2018, pp. 334–3 46

    work page 2018

  24. [24]

    An improved algorithm for reprojecting ray s through pixel images,

    P . Joseph, “An improved algorithm for reprojecting ray s through pixel images,” IEEE Transactions on Medical Imaging , vol. 1, no. 3, pp. 192–196, 1982

    work page 1982

  25. [25]

    Image Quality Metrics: PSNR vs. SSI M,

    A. Hore and D. Ziou, “Image Quality Metrics: PSNR vs. SSI M,” in 2010 20th International Conference on Pattern Recognition . IEEE, aug 2010, pp. 2366–2369

    work page 2010

  26. [26]

    The design and implementat ion of solar, a portable library for scalable out-of-core li near algebra computations,

    S. Toledo and F. Gustavson, “The design and implementat ion of solar, a portable library for scalable out-of-core li near algebra computations,” Proceedings of the Annual W orkshop on I/O in Parallel and Dis tributed Systems, IOPADS , 08 1998

    work page 1998

  27. [27]

    Design and implementati on of the parallel out-of-core ScaLAPACK LU, QR, and Cholesk y factorization routines,

    E. D’Azevedo and J. Dongarra, “Design and implementati on of the parallel out-of-core ScaLAPACK LU, QR, and Cholesk y factorization routines,” Concurrency - Practice and Experience , vol. 12, pp. 1481–1493, 12 2000

    work page 2000

  28. [28]

    POOCLAPACK: Paralle l Out-of-Core Linear Algebra Package,

    W. C. Reiley and R. A. van de Geijn, “POOCLAPACK: Paralle l Out-of-Core Linear Algebra Package,” Department of Compu ter Sciences, The University of Texas at Austin, Tech. Rep. CS-TR-99-33, 1 999

    work page

  29. [29]

    Parallel out-of-cor e computation and updating the QR factorization,

    B. C. Gunter and R. A. van de Geijn, “Parallel out-of-cor e computation and updating the QR factorization,” ACM Transactions on Mathematical Software, vol. 31, no. 1, pp. 60–78, March 2005

    work page 2005

  30. [30]

    Rapid development of high-performance out-of-core solv ers,

    T. Joffrain, E. S. Quintana-Ort´ ı, and R. A. van de Geijn , “Rapid development of high-performance out-of-core solv ers,” in Proceedings of PARA 2004, ser. LNCS, no. 3732. Springer-V erlag Berlin Heidelberg, 2 005, pp. 413–422

    work page 2004

  31. [31]

    Parall el out-of-core Cholesky and QR factorizations with POOCLAP ACK,

    B. C. Gunter, W. C. Reiley, and R. A. van de Geijn, “Parall el out-of-core Cholesky and QR factorizations with POOCLAP ACK,” in Proceedings of the 15th International Parallel and Distrib uted Processing Symposium (IPDPS) . IEEE Computer Society, 2001

    work page 2001

  32. [32]

    A runtime system for progr amming out-of-core matrix algorithms-by-tiles on multithreaded architectur es,

    G. Quintana-Ort´ ı, F. D. Igual, M. Marqu´ es, E. S. Quint ana-Ort´ ı, and R. A. V an de Geijn, “A runtime system for progr amming out-of-core matrix algorithms-by-tiles on multithreaded architectur es,” ACM Transactions on Mathematical Software (TOMS) , vol. 38, no. 4, p. 25, 2012

    work page 2012

  33. [33]

    Using desktop computers to solve large- scale dense linear algebra problems,

    M. Marqu´ es, G. Quintana-Ort´ ı, E. S. Quintana-Ort´ ı, and R. van de Geijn, “Using desktop computers to solve large- scale dense linear algebra problems,” The Journal of Supercomputing , vol. 58, no. 2, pp. 145–150, Nov 2011

    work page 2011

  34. [34]

    Out-of-Core Computation of the QR Fac torization on Multi-core Processors,

    M. Marqu´ es, G. Quintana-Ort´ ı, E. S. Quintana-Ort´ ı, and R. A. van de Geijn, “Out-of-Core Computation of the QR Fac torization on Multi-core Processors,” in Lecture Notes in Computer Science 5704, Euro-Par’2009, (Ed s. H. Sips, D. Epema, H. -X. Lin) , 2009, pp. 809–820

    work page 2009

  35. [35]

    FORBILD Head Phantom

    G. Lauritsch and H. Bruder, “FORBILD Head Phantom.” [On line]. Available: http://www.imp.uni-erlangen.de/phan toms/head/head.html

    work page

  36. [36]

    Deeplesion: au tomated mining of large-scale lesion annotations and unive rsal lesion detection with deep learning,

    K. Y an, X. Wang, L. Lu, and R. M. Summers, “Deeplesion: au tomated mining of large-scale lesion annotations and unive rsal lesion detection with deep learning,” Journal of Medical Imaging , vol. 5, no. 3, p. 036501, 2018

    work page 2018