Pith. sign in

REVIEW 2 major objections 1 minor 35 references

Comparison of Deep Learning and Particle Smoother EM Methods for Estimation of Rb-82 Myocardial Perfusion PET Kinetic Parameters

T0 review · 2 major / 1 minor · reviewed 2026-05-23 · grok-4.3

Challenge this review Download PDF Read on arXiv ↗

Pith's one-line read Convolutional neural network achieves lowest errors for Rb-82 PET kinetic parameters across frame durations

desk verdict CNN shows lower errors than NLLS/KEM/PSEM on simulated Rb-82 data, but the result stands or falls on whether the simulations capture real scanner and patient variability. read the letter →

arxiv 2412.04706 v2 submitted 2024-12-06 physics.med-ph

Myungheon Chin , Sarah J Zou , Garry Chinn , Craig S. Levin This is my paper
classification physics.med-ph
keywords Rb-82PETkineticmodelingCNNparticlesmootherEMmyocardialperfusioncompartmentdynamic
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper develops a particle smoother EM method and a convolutional neural network to estimate kinetic parameters from Rb-82 myocardial perfusion PET data and compares them to nonlinear least squares and Kalman EM approaches. It shows the CNN produces the smallest relative errors for flow F, k3, and k4 over 2-10 second frames at multiple noise levels when input functions match the training distribution. This matters because accurate kinetic estimates enable reliable myocardial blood flow quantification without fixing parameters to population averages.

What carries the argument

Convolutional neural network trained to map time-activity curves directly to kinetic parameters F, k3, k4 from simulated dynamic PET frames.

What would settle it

Apply the trained CNN to a set of real patient Rb-82 PET scans and measure whether the reported error reductions versus NLLS hold when compared against independent reference measurements.

Watch

Extended reading notes

Core claim

Across simulated Rb-82 dynamic studies, the CNN achieved the lowest relative errors for all parameters (F: 8.78-4.98%, k3: 26.05-25.50%, k4: 34.34-22.76%), significantly outperforming NLLS, KEM, and PSEM, while PSEM improved k3 estimation but underperformed for F.

Load-bearing premise

The simulated Rb-82 dynamic studies accurately reproduce the noise statistics, input-function variability, and physiological range of real clinical acquisitions.

Editorial extensions

If this is right

  • CNN estimates could reduce the need to fix parameters using population averages in clinical Rb-82 analysis.
  • PSEM shows parameter-dependent gains, improving k3 but not F, indicating targeted refinement may be required.
  • Both new methods degrade when input functions fall outside the training distribution.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The CNN approach may extend to other dynamic PET tracers if retrained on matching simulations.
  • Hybrid methods that embed compartment-model constraints inside the network could mitigate out-of-distribution failures.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 1 minor

Summary. The manuscript compares a convolutional neural network (CNN) and particle smoother EM (PSEM) method to nonlinear least squares (NLLS) and Kalman EM (KEM) for estimating kinetic parameters F, k3, and k4 from simulated Rb-82 myocardial perfusion PET dynamic studies. It reports that the CNN yields the lowest relative errors across 2-10 s frames and multiple noise levels (F: 8.78-4.98%, k3: 26.05-25.50%, k4: 34.34-22.76%), with Holm-adjusted p < 1e-15 superiority at 1.0x noise/2 s frames, while noting degradation under out-of-distribution input functions and mixed PSEM performance.

Significance. If the simulations faithfully capture clinical noise statistics, input-function variability, and physiological ranges, the results would indicate that CNN-based estimation can deliver more accurate and robust kinetic parameters than conventional NLLS or EM smoothers for Rb-82 PET, potentially improving myocardial blood flow quantification. The multi-frame-duration, multi-noise-level design with quantitative errors and adjusted p-values provides a clear empirical comparison.

major comments (2)
  1. [Abstract] Abstract: The headline claim of CNN superiority (lowest relative errors and p < 1e-15) rests entirely on simulated data, yet no quantitative verification is supplied that the simulated noise statistics, arterial input-function dispersion, or parameter ranges match those of real clinical Rb-82 acquisitions. The abstract itself notes performance degradation under out-of-distribution input functions, making simulation fidelity load-bearing for any claim of clinical relevance.
  2. [Results] Results (performance tables/figures): While relative errors and Holm-adjusted p-values are reported for held-out simulated cases, the absence of any real-patient or phantom validation means the reported error reductions cannot be assumed to translate when scanner-specific effects, patient motion, or actual input-function variability are present.
minor comments (1)
  1. [Abstract] Abstract: The CNN architecture, training details, and number of simulated realizations per condition are not summarized, which would help readers assess reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our simulation-based comparison of CNN, PSEM, NLLS, and KEM for Rb-82 myocardial perfusion PET kinetic parameter estimation. Our work focuses on controlled in silico evaluation with known ground truth, and we address the concerns regarding simulation fidelity and real-data validation below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The headline claim of CNN superiority (lowest relative errors and p < 1e-15) rests entirely on simulated data, yet no quantitative verification is supplied that the simulated noise statistics, arterial input-function dispersion, or parameter ranges match those of real clinical Rb-82 acquisitions. The abstract itself notes performance degradation under out-of-distribution input functions, making simulation fidelity load-bearing for any claim of clinical relevance.

    Authors: We agree that quantitative verification of simulation fidelity against specific clinical datasets is not provided and that this limits the strength of any clinical translation claims. The simulations follow standard noise and parameter models from the Rb-82 PET literature, but we did not include direct matching statistics. We will revise the abstract to explicitly qualify the superiority results as applying to in-distribution simulated cases only and to reinforce the implications of the noted OOD degradation. This is a partial revision focused on clearer scoping rather than new experiments. revision: partial

  2. Referee: [Results] Results (performance tables/figures): While relative errors and Holm-adjusted p-values are reported for held-out simulated cases, the absence of any real-patient or phantom validation means the reported error reductions cannot be assumed to translate when scanner-specific effects, patient motion, or actual input-function variability are present.

    Authors: We acknowledge that the absence of real-patient or phantom data means the error reductions cannot be assumed to hold under clinical conditions such as motion or scanner effects. The manuscript is designed as a simulation study to enable rigorous ground-truth comparison, which is a standard initial step in kinetic modeling method development. We will add an explicit limitations paragraph in the Discussion to state this scope limitation and the need for future real-data studies. No real-data validation will be added, as it lies outside the current work. revision: partial

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical comparisons on held-out simulations

full rationale

The paper reports direct numerical comparisons of estimation errors for CNN, NLLS, KEM, and PSEM on simulated Rb-82 dynamic PET data across frame durations and noise levels. No derivation chain, fitted parameters renamed as predictions, self-citation load-bearing steps, or ansatz smuggling appear in the abstract or described methods. The central claims (lowest relative errors for CNN, statistical significance) are computed from independent test simulations and do not reduce to the inputs by construction. This is a standard empirical benchmark study whose validity hinges on simulation fidelity rather than any definitional or self-referential reduction.

Assumptions & free parameters 0 free parameters · 0 assumptions · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the methods are described at the level of algorithmic approach and simulation protocol without listing fitted constants or unproven assumptions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Comparison of Deep Learning and Particle Smoother EM Methods for Estimation of Rb-82 Myocardial Perfusion PET Kinetic Parameters." pith.science (2026). https://pith.science/paper/2412.04706

@misc{pith2026241204706,
  author       = {Pith},
  title        = {Pith review of: Comparison of Deep Learning and Particle Smoother EM Methods for Estimation of Rb-82 Myocardial Perfusion PET Kinetic Parameters},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2412.04706}},
  note         = {Machine review of arXiv:2412.04706}
}
read the original abstract

Positron emission tomography (PET) enables quantification of dynamic physiological processes through time-resolved imaging. In Rb-82 myocardial perfusion PET, kinetic compartment modeling is used to estimate physiological parameters and derive myocardial blood flow. However, conventional nonlinear least squares (NLLS) estimation is sensitive to model misspecification when not all parameters can be reliably estimated and must instead be fixed or initialized using population averages, which can degrade accuracy. This work develops and evaluates two alternative kinetic analysis approaches for Rb-82 PET: a particle smoother-based Expectation-Maximization method (PSEM) and a convolutional neural network (CNN). Both methods were evaluated using simulated Rb-82 dynamic myocardial perfusion studies and compared against NLLS and a Kalman smoother-based Expectation-Maximization (KEM) algorithm across multiple frame durations and noise levels. Across 2-10 s frames, the CNN achieved the lowest relative errors for all parameters (F: 8.78-4.98%, k3: 26.05-25.50%, k4: 34.34-22.76%), significantly outperforming NLLS, KEM, and PSEM (Holm-adjusted p < 1e-15 at 1.0x noise, 2 s frames), although performance degraded under out-of-distribution input-function conditions. Overall, the CNN provided the most accurate and robust in-distribution kinetic parameter estimates across frame durations. In contrast, PSEM exhibited parameter-dependent behavior, improving k3 estimation while underperforming for F, suggesting that further methodological refinement is needed.

Figures

Figures reproduced from arXiv: 2412.04706 by the authors.

Figure 1
Figure 1. 2TC Model for 82Rb: The inflow and outflow rates between the plasma and the fast exchangeable compartment are assumed to be equal (F), but the flow rate from the first state to the plasma is divided by V 1 since q1(t) is not a concentration but the total amount of the activity. Many studies have demonstrated that regional myocardial perfusion PET studies with 82Rb are sufficiently modeled by a two-compartment model.… view at source ↗
Figure 2
Figure 2. The CNN model consists of seven 1-D convolutional layers, followed by a global [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Sample plot showing the particle smoother’s efficacy in estimating the two states [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Simulation result: The yellow dotted line and the green dotted line indicate the [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: Interpolation result: The red line in each subfigure represents the ground truth [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]
Figure 6
Figure 6. Figure 6: A plot of PSEM estimation across iterations for a sample simulation is shown: In [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]
Figure 7
Figure 7. Figure 7: Scatter plots showing the estimation errors compared to those of NLLS. The text [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    D. L. Bailey, D. W. Townsend, P. E. Valk, and M. N. Maisey, editors, Positron Emission Tomography : Basic Sciences , Springer, London, 2005

    work page 2005

  2. [2]

    Naghavi et al., From Vulnerable Plaque to Vulnerable Patient , Circulation 108 , 1664--1672 (2003)

    M. Naghavi et al., From Vulnerable Plaque to Vulnerable Patient , Circulation 108 , 1664--1672 (2003)

    work page 2003

  3. [3]

    S. C. Huang, B. A. Williams, J. Krivokapich, L. Araujo, M. E. Phelps, and H. R. Schelbert, Rabbit myocardial 82Rb kinetics and a compartmental model for blood flow estimation, American Journal of Physiology-Heart and Circulatory Physiology 256 , H1156--H1164 (1989)

    work page 1989

  4. [4]

    Schwaiger and O

    M. Schwaiger and O. Muzik, Assessment of myocardial perfusion by positron emission tomography, The American Journal of Cardiology 67 , 35D--43D (1991)

    work page 1991

  5. [5]

    Herrero, J

    P. Herrero, J. Markham, M. E. Shelton, and S. R. Bergmann, Implementation and evaluation of a two-compartment model for quantification of myocardial perfusion with rubidium-82 and positron emission tomography., Circulation Research 70 , 496--507 (1992)

    work page 1992

  6. [6]

    Lortie, R

    M. Lortie, R. S. B. Beanlands, K. Yoshinaga, R. Klein, J. N. DaSilva, and R. A. deKemp, Quantification of myocardial blood flow with 82Rb dynamic PET imaging, European Journal of Nuclear Medicine and Molecular Imaging 34 , 1765--1774 (2007)

    work page 2007

  7. [7]

    Yoshinaga, R

    K. Yoshinaga, R. Klein, and N. Tamaki, Generator-produced rubidium-82 positron emission tomography myocardial perfusion imaging— From basic aspects to clinical applications, Journal of Cardiology 55 , 163--173 (2010)

    work page 2010

  8. [8]

    J. H. Sohn, S. C. Behr, M. H. Pampaloni, and Y. Seo, Quantitative Assessment of Myocardial Ischemia With Positron Emission Tomography , Journal of Thoracic Imaging (2022), Publisher: LWW

    work page 2022

Show all 35 references
  1. [9]

    R. N. Gunn, S. R. Gunn, and V. J. Cunningham, Positron emission tomography compartmental models, Journal of Cerebral Blood Flow & Metabolism 21 , 635--652 (2001), Publisher: SAGE Publications Sage UK: London, England

    2001

  2. [10]

    E. D. Morris, C. J. Endres, K. C. Schmidt, B. T. Christian, R. F. Muzic, and R. E. Fisher, Kinetic modeling in positron emission tomography, Emission tomography 46 , 499--540 (2004), Publisher: Elsevier Amsterdam, The Netherlands

    2004

  3. [11]

    Watabe, Y

    H. Watabe, Y. Ikoma, Y. Kimura, M. Naganawa, and M. Shidahara, PET kinetic analysis—compartmental model, Annals of Nuclear Medicine 20 , 583--588 (2006)

    2006

  4. [12]

    C. S. Patlak, R. G. Blasberg, and J. D. Fenstermacher, Graphical Evaluation of Blood -to- Brain Transfer Constants from Multiple - Time Uptake Data , Journal of Cerebral Blood Flow & Metabolism 3 , 1--7 (1983)

    1983

  5. [13]

    Logan, J

    J. Logan, J. S. Fowler, N. D. Volkow, A. P. Wolf, S. L. Dewey, D. J. Schlyer, R. R. MacGregor, R. Hitzemann, B. Bendriem, S. J. Gatley, and D. R. Christman, Graphical Analysis of Reversible Radioligand Binding from Time — Activity Measurements Applied to [ N - 11C - Methyl ]-(...

    1990

  6. [14]

    Gibson and B

    S. Gibson and B. Ninness, Robust maximum-likelihood estimation of multivariable dynamic systems, Automatica 41 , 1667--1682 (2005)

    2005

  7. [15]

    R. E. Kalman, A New Approach to Linear Filtering and Prediction Problems , Journal of Basic Engineering 82 , 35--45 (1960)

    1960

  8. [16]

    H. E. RAUCH, F. TUNG, and C. T. STRIEBEL, Maximum likelihood estimates of linear dynamic systems, AIAA Journal 3 , 1445--1450 (1965)

    1965

  9. [17]

    Roweis and Z

    S. Roweis and Z. Ghahramani, A Unifying Review of Linear Gaussian Models , Neural Computation 11 , 305--345 (1999)

    1999

  10. [18]

    N. J. Gordon, D. J. Salmond, and A. F. M. Smith, Novel approach to nonlinear/non- Gaussian Bayesian state estimation, IEE Proceedings F (Radar and Signal Processing) 140 , 107--113(6) (1993)

    1993

  11. [19]

    T. B. Schön, A. Wills, and B. Ninness, System identification of nonlinear state-space models, Automatica 47 , 39--49 (2011)

    2011

  12. [20]

    LeCun, B

    Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel, Backpropagation Applied to Handwritten Zip Code Recognition , Neural Computation 1 , 541--551 (1989)

    1989

  13. [21]

    Alzubaidi, J

    L. Alzubaidi, J. Zhang, A. J. Humaidi, A. Al-Dujaili, Y. Duan, O. Al-Shamma, J. Santamaría, M. A. Fadhel, M. Al-Amidie, and L. Farhan, Review of deep learning: concepts, CNN architectures, challenges, applications, future directions, Journal of Big Data 8 , 53 (2021)

    2021

  14. [22]

    J. C. B. Gamboa, Deep Learning for Time - Series Analysis , 2017

    2017

  15. [23]

    R. Wang, H. Liu, T. Toyonaga, L. Shi, J. Wu, J. A. Onofrey, Y.-J. Tsai, M. Naganawa, T. Ma, Y. Liu, M.-K. Chen, A. P. Mecca, R. S. O’Dell, C. H. van Dyck, R. E. Carson, and C. Liu, Generation of synthetic PET images of synaptic density and amyloid from 18F - FDG images using d...

    2021

  16. [24]

    Y. Hu, G. I. Angelis, P. L. Kench, O. K. Fuller, Y. Liu, T. Ma, and S. R. Meikle, Direct Estimation of Neurotransmitter Activation Parameters in Dynamic PET Using Regression Neural Networks , in 2019 IEEE Nuclear Science Symposium and Medical Imaging Conference ( NSS / MIC ) ,...

    2019

  17. [25]

    S. M. Kazemi, R. Goel, S. Eghbali, J. Ramanan, J. Sahota, S. Thakur, S. Wu, C. Smyth, P. Poupart, and M. Brubaker, Time2Vec : Learning a Vector Representation of Time , 2019

    2019

  18. [26]

    N. A. Mullani, R. A. Goldstein, K. L. Gould, S. K. Marani, D. J. Fisher, H. A. O'Brien, and M. D. Loberg, Myocardial perfusion with rubidium-82. I . Measurement of extraction fraction and flow with external detectors, Journal of Nuclear Medicine 24 , 898--906 (1983), Publisher...

    1983

  19. [27]

    Jones, T

    E. Jones, T. Oliphant, P. Peterson, and others , SciPy : Open source scientific tools for Python , 2001

    2001

  20. [28]

    R. H. Shumway, D. S. Stoffer, and D. S. Stoffer, Time series analysis and its applications , volume 3, Springer, 2000

    2000

  21. [29]

    Elfring, E

    J. Elfring, E. Torta, and R. van de Molengraft, Particle Filters : A Hands - On Tutorial , Sensors 21 (2021)

    2021

  22. [30]

    Arulampalam, S

    M. Arulampalam, S. Maskell, N. Gordon, and T. Clapp, A tutorial on particle filters for online nonlinear/non- Gaussian Bayesian tracking, IEEE Transactions on Signal Processing 50 , 174--188 (2002)

    2002

  23. [31]

    Gustafsson, Particle filter theory and practice with positioning applications, IEEE Aerospace and Electronic Systems Magazine 25 , 53--82 (2010)

    F. Gustafsson, Particle filter theory and practice with positioning applications, IEEE Aerospace and Electronic Systems Magazine 25 , 53--82 (2010)

    2010

  24. [32]

    Nordh, pyParticleEst : A Python Framework for Particle - Based Estimation Methods , Journal of Statistical Software 78 , 1--25 (2017)

    J. Nordh, pyParticleEst : A Python Framework for Particle - Based Estimation Methods , Journal of Statistical Software 78 , 1--25 (2017)

    2017

  25. [33]

    Lindsten, P

    F. Lindsten, P. Bunch, S. J. Godsill, and T. B. Schön, Rao- Blackwellized particle smoothers for mixed linear/nonlinear state-space models, in 2013 IEEE International Conference on Acoustics , Speech and Signal Processing , pages 6288--6292, 2013

    2013

  26. [34]

    , " * write output.state after.block =

    ") INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sentence := #2 'after.sentence := #3 'after.block := STRINGS s t FUNCTION output.nonnull 's := output.state mid.sentence = ", " * write output.state...

  27. [35]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence skip FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTION or pop #1 'skip ...

Pith tools

Reviewed May 23, 2026 · model on record in the stance chip above.