Compressibility Analysis for the differentiable shift-variant Filtered Backprojection Model

Andreas Maier; Chengze Ye; Linda-Sophie Schneider; Mareike Thies; Yipeng Sun

REVIEW 4 major objections 6 minor 9 references

Compressibility Analysis for the differentiable shift-variant Filtered Backprojection Model

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The redundancy weights of differentiable shift-variant FBP can be replaced by a PCA decomposition, cutting trainable parameters 97.25% with no loss in reconstruction quality.

desk verdict Niche but plausible compression result for a specialized CT model; the speed and accuracy claims overreach, especially the ignored pretraining cost. read the letter →

arxiv 2501.11586 v1 pith:3RQFYD3B submitted 2025-01-20 cs.CV eess.IV

Chengze Ye , Linda-Sophie Schneider , Yipeng Sun , Mareike Thies , Andreas Maier This is my paper · ORCID

classification cs.CVeess.IV

keywords CTreconstructionnetworkcompressiondeeplearningknownoperatorPCAarbitrarytrajectoryshift-variantFBPcone-beamcomputedtomography

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

This paper sets out to show that the redundancy-weight layer of a differentiable shift-variant filtered backprojection (FBP) reconstruction network is highly compressible, and that replacing it with a principal-component decomposition preserves reconstruction quality. The authors integrate PCA directly into the pipeline, representing the redundancy weights as a low-dimensional coefficient vector mapped back through a principal-component basis plus the mean (Eq. 7). On a sinusoidal C-arm trajectory with 24 training samples, the compressed model cuts trainable parameters from 113,184,000 to 3,116,560 (a 97.25% reduction) while keeping MSE, PSNR, and SSIM comparable to the uncompressed model, and converges in about 140 epochs instead of 430. If the result holds beyond this single trajectory, it would make non-circular cone-beam CT reconstruction practical on resource-constrained hardware.

What carries the argument

The central object is the PCA decomposition of the redundancy-weight layer. Starting from the principal components $V_k$ and mean $\mu$ of the redundancy weights of an already-trained model, the pipeline replaces the high-dimensional weight layer with a trainable low-dimensional code $w'_{red}$ and the linear mapping $w'_{red} V_k^T + \mu$, inserted at the position of $w_{red}$ in the differentiable FBP chain (Eq. 7). The number of principal components $k$ controls the trade-off: $k = 50$, $30$, and $10$ yield 14,450,960, 8,783,760, and 3,116,560 trainable parameters respectively, with negligible changes in image quality.

What would settle it

Train the compressed model on a different non-circular trajectory, such as an elliptical or saddle orbit, using a PCA basis estimated on that trajectory, and compare reconstruction quality with the uncompressed model. If MSE, PSNR, or SSIM degrades substantially more than in Table 1, the claim that redundancy weights are generally low-rank is falsified for that setting; re-running the same sinusoidal setup from different random initializations would also test how much of the 97.25% figure depends on the specific trained model used to estimate the PCA basis.

Watch

Extended reading notes

Core claim

The central claim is that the redundancy weights in the differentiable shift-variant FBP model live on a low-dimensional linear subspace, so the learned weight tensor can be replaced by the reconstruction $w'_{red} V_k^T + \mu$ without sacrificing reconstruction accuracy. With only ten principal components the method retains image quality relative to the full model (MSE 0.0855 vs 0.0904, PSNR 37.67 dB vs 37.20 dB, SSIM 0.9481 vs 0.9591), while the number of trainable parameters falls from 113,184,000 to 3,116,560. The paper further reports that the compressed network reaches convergence in 140 epochs, compared with 430 for the uncompressed network, and that the redundancy weights recovered from the low-dimensional representation closely match the weights learned without compression.

Load-bearing premise

The load-bearing premise is that the PCA basis and mean computed from the redundancy weights of an already-trained uncompressed model remain a sufficient coordinate system for the redundancy weights during training of the compressed model, so the network does not need to leave that low-dimensional subspace; this is tested on only one sinusoidal trajectory with 24 training samples.

Editorial extensions

If this is right

With PCA(10), the differentiable shift-variant FBP model trains to convergence in roughly one-third of the epochs (140 vs 430), directly lowering the compute cost of learning redundancy weights for a given trajectory.
The parameter reduction from 113,184,000 to 3,116,560 makes the reconstruction network small enough for memory-constrained deployment, where the uncompressed model would be impractical.
Reconstruction quality stays within the same range as the uncompressed model (SSIM 0.9481 vs 0.9591), so the compressed pipeline can be substituted without a visible quality trade-off on this trajectory.
The same PCA insertion applies to any trajectory for which a redundancy-weight model has been trained, because the compression acts on the learned weights rather than on the analytic formula.
Larger $k$ values (50, 30) give a safety margin: even a 30-component model cuts parameters from 113,184,000 to 8,783,760 with SSIM 0.9409, leaving room to trade size against fidelity.

Reading between the lines

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

A natural extension the paper leaves implicit is cross-trajectory transfer: a PCA basis estimated from one orbit might serve as an initialization for another orbit, which would amortize the cost of computing the basis; this is testable by fitting $V_k$ on a sinusoidal trajectory and evaluating on an elliptical one.
Because PCA is linear, the success of this compression suggests that a spectral analysis of learned reconstruction weights could predict compressibility before full training, potentially selecting $k$ automatically from the eigenvalue decay.
The same low-rank substitution could be applied to the other learned weighting layers in the pipeline ($w_{cos}$, $w_{sino}$, $w_d$), which may yield additional parameter reductions beyond the 97.25% reported for the redundancy weights alone.
The evaluation on a single sinusoidal trajectory with 24 training samples means the numerical claims are best read as a proof of concept; testing on more trajectories and more samples would establish whether 97.25% is a typical compression rate or a favorable one.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Short version: this is a niche but legitimate compression result for a specialized CT reconstruction network. The 97% parameter reduction is real, but the two headline claims—no accuracy loss and much faster training—are both overstated. The speed claim is the bigger problem: the PCA basis comes from an already-trained uncompressed model, so the end-to-end training cost is not less than the uncompressed baseline; you pay for the uncompressed training first. The paper compares epochs (140 vs 430) and never reports wall-clock time or the pretraining cost. That is a load-bearing flaw in the main practical selling point.

What is genuinely new: applying PCA to the redundancy-weight layer of the differentiable shift-variant FBP, rather than to generic network weights. The paper shows that the learned redundancy weights live in a low-dimensional subspace—that is a real empirical finding on a non-trivial model, and Figure 1 is a useful diagnostic. The math is simple and correctly stated, and the paper does not hide the fact that quality degrades with smaller k.

The soft spots are mostly presentation and scope. Table 1 shows SSIM dropping from 0.959 to 0.943/0.948, so "without compromising accuracy" is too strong. The validation set is six volumes, no significance test, and only one sinusoidal trajectory is tested. The paper does not cite or compare against standard network-compression literature (pruning, low-rank factorization, distillation), so the novelty is genuinely incremental. There is also a technical ambiguity: the abstract calls V_k a "trainable eigenvector matrix," but Eq. 7 and the parameter counts suggest V_k is either fixed or shared; if it is trainable as a full matrix, the parameter savings do not add up. The reader should check that.

Who should read it: researchers working on learned non-circular trajectory CBCT reconstruction. For them, the compressibility observation and the PCA reparametrization are useful. For the broader ML community, this is a small case study.

Recommendation: I would send it to peer review, but with a clear request for major revision: reframe the speed claim, add wall-clock and end-to-end cost, include at least one standard compression baseline, fix the V_k ambiguity, and report significance or at least more validation samples. The core observation is sound enough to deserve referee time.

Referee Report

4 major / 6 minor

Summary. The paper proposes a PCA-based compression scheme for the redundancy-weight layer of a differentiable shift-variant filtered backprojection (FBP) reconstruction network. The authors train an uncompressed model on simulated cone-beam CT data acquired along a sinusoidal trajectory, compute a PCA basis from the trained redundancy weights, and then replace the high-dimensional redundancy weight layer with a low-dimensional trainable coefficient vector mapped back through the fixed PCA basis (Eq. 7). They report a 97.25% reduction in trainable parameters and claim that reconstruction accuracy is preserved and training is greatly accelerated (140 epochs vs. 430 epochs). Experimental results are presented for one sinusoidal trajectory with 24 training and 6 validation samples.

Significance. If the central claims hold, the work is a practical contribution: it offers a simple, operator-level compression for a reconstruction network that otherwise stores a large redundancy weight tensor, and it suggests that this layer has low intrinsic dimensionality. The authors provide quantitative metrics in Table 1, visual comparisons in Figures 3-4, and a clear parameter-count reduction. However, the evidence currently does not fully support the two headline claims: the training-time comparison is not end-to-end, because obtaining the PCA basis requires a fully trained uncompressed model, and the accuracy comparison shows a small but consistent SSIM drop with no statistical assessment. The generalizability to other trajectories is also untested, which limits the practical claim. These issues are addressable and do not invalidate the core idea, but they need to be fixed before publication.

major comments (4)

[Section 4 and Section 2.3, Eq. (7)] The claim that the method 'significantly accelerates the training process' is not supported end-to-end. The PCA basis V_k and mean mu are computed from the redundancy weights of an already-trained uncompressed model (Section 2.3), so the total cost of the proposed pipeline is at least the cost of training the uncompressed model plus training the compressed model. Comparing 140 epochs versus 430 epochs only measures the compressed training phase, not the whole procedure. Please report wall-clock training times, including the PCA estimation and any overhead, or clearly state the amortization scenario under which the speedup is realized (e.g., reusing the basis across many trajectories).
[Table 1] The statement 'without compromising reconstruction accuracy' is not rigorously established. SSIM drops from 0.9591 +/- 0.0051 (No PCA) to 0.9427 +/- 0.0088 for PCA(50) and 0.9481 +/- 0.0074 for PCA(10), while MSE and PSNR changes are small and in the opposite direction. With only six validation samples and no paired significance test, it is unclear whether these differences are statistically meaningful or reflect a tradeoff among metrics. Please report per-sample results, confidence intervals, or a paired test, and temper the accuracy claim accordingly.
[Abstract and Section 2.3, Eq. (7)] It is ambiguous whether V_k is trainable. The abstract says 'trainable eigenvector matrix,' but Eq. (7) shows only w'_red as the low-dimensional variable and treats V_k and mu as fixed quantities from PCA. This distinction is essential for the parameter-count claim: if V_k is frozen, the reported 3,116,560 parameters for PCA(10) make sense, but if V_k is trained, its entries must be included in the trainable parameter count and the 97.25% reduction may change. Please state explicitly which quantities are frozen and which are trained, and reconcile the parameter counts with that statement.
[Section 3] The generalizability claim is not supported by the experimental design. The PCA basis is estimated from the redundancy weights learned on a single sinusoidal trajectory, and the compressed model is tested on that same trajectory only. Since the paper motivates the method for non-circular and arbitrary trajectories, at least one additional trajectory (or a clear statement that cross-trajectory transfer is out of scope) is needed. The current 24-sample training set and 6-sample validation set also make it difficult to draw strong conclusions about reconstruction quality.

minor comments (6)

[Section 2.3, Eqs. (3)-(5)] The text describes standardization, but Eq. (3) only subtracts the mean and does not divide by the standard deviation. Please align the description with the equations or clarify that no scaling is applied.
[Section 2.2, Eq. (2)] The symbols w_cos, w_sino, w_red, and w_d are not defined. Please define each weight matrix and indicate its role in the reconstruction pipeline relative to Eq. (1).
[Section 2.1, Eq. (1)] The integration domain 'S^2/2' is unclear. Please specify the half-sphere notation or give a reference for this representation.
[Section 3] The paper does not report the parameters of the sinusoidal trajectory (e.g., amplitude, frequency, number of projections, detector geometry) or the forward projector settings. This information is necessary for reproducibility.
[Section 3] The sentence 'while the remaining samples were employed for validation' leaves the validation count implicit. Please state that only 6 samples were used for validation.
[Figure 1] Figure 1 is referenced in Section 2.3 but no visible figure is provided in the text; please ensure the figure is included and that its axis labels and legend are readable.

Circularity Check

1 steps flagged · score 2.0 of 10

PCA basis fit to the same trained weights makes Figure 4's recovered-vs-learned redundancy-weight comparison in-sample; the held-out reconstruction accuracy in Table 1 keeps the central claim independent.

fitted input called prediction [Section 2.3, Eq. (6); Section 4, Figure 4]
"Project the dimensionally reduced data Y back to the higher-dimensional space, then add the original mean to the centered data to obtain the reconstructed data matrix X ′. X ′ = Y VT k + µ (6) ... the figure presents a comparison between the redundancy weights recovered from the low-dimensional representation and the learned redundancy weights without dimensionality reduction. The results of this comparison demonstrate a remarkable similarity, thereby further validating the effectiveness of the dimensionality reduction in reconstruction process."

The PCA basis V_k and mean µ are estimated from the redundancy weights learned by the uncompressed network on the same sinusoidal-trajectory data. By Eqs. (5)-(6), the 'recovered' weights are X′ = (X − µ)V_k V_k^T + µ, i.e., the least-squares rank-k projection of X onto the subspace fitted from X itself. Matching on the fitting set is guaranteed by PCA's optimal reconstruction property, so Figure 4's similarity is not an independent prediction. The central accuracy claim does not reduce to this plot, because the compressed model is retrained and evaluated on held-out volumes in Table 1; the circularity is therefore confined to this in-sample comparison and the associated 'further validating' sentence.

full rationale

The paper's main claim is empirical: a PCA-parameterized redundancy-weight layer (Eq. 7) reaches comparable validation image quality with far fewer parameters. That claim is not forced by construction: the compressed model is retrained, the PCA basis is only an initialization (the abstract describes a trainable eigenvector matrix), and Table 1 reports metrics on held-out volumes (6 validation samples). The self-citations [3,4] supply the base differentiable FBP network (Eq. 2), but they do not by themselves establish compressibility. The only genuinely self-referential element is Figure 4: V_k and µ in Eq. (6) are computed from the very redundancy weights that Figure 4 then 'recovers,' so the similarity shown is the optimal rank-k reconstruction of the fitting data—an identity property of PCA, not independent evidence. This is a minor in-sample validation, not the paper's central result. Separately, the speed comparison in Section 4 (140 vs 430 epochs) omits the one-time cost of training the uncompressed model to obtain the PCA basis; that is a reporting gap affecting the speed claim, but it is not a circular reduction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the validity of the Defrise-Clack FBP formula, the faithfulness of the differentiable network to that formula, and the linear-subspace compressibility of the learned redundancy weights. The latter is the key new assumption and is tested only on a single trajectory with a small simulated dataset.

free parameters (3)

Number of principal components k = 10, 30, 50
Selected by inspecting Figure 1 and Table 1; determines the compression ratio and reconstruction quality.
Loss weighting lambda for SSIM = not reported
Ltotal = Lmse + lambda * Lssim in Eq. 8; the value is not reported and influences training behavior.
PCA eigenbasis V_k and mean mu = estimated from trained redundancy weights
Data-dependent projection operators estimated from the uncompressed model's weights; central to the compression and dependent on only 24 training samples.

assumptions (5)

standard math The Defrise-Clack shift-variant FBP formula (Eq. 1) correctly reconstructs CBCT data for non-circular trajectories.
Invoked in Section 2.1 as the starting point; the paper does not re-derive it.
domain assumption The differentiable network in Eq. 2 faithfully represents Eq. 1 via known operator learning.
Basis of the pipeline; relies on prior work by the same authors (refs. 3 and 4).
ad hoc to paper Redundancy weights learned by the uncompressed model lie near a low-dimensional linear subspace, so PCA with k=10 preserves enough information.
Core compressibility premise; tested only for one sinusoidal trajectory and 24 training samples.
domain assumption Simulated cone-beam forward projections with PyroNN accurately model the Artis zeego system for training and validation.
Used to generate all 30 data samples in Section 3.
domain assumption The Pancreatic-CT-CBCT-SEG dataset provides realistic volumes for evaluating the method.
Used in Section 3 for sinogram generation; no details on preprocessing or split are given.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Compressibility Analysis for the differentiable shift-variant Filtered Backprojection Model." pith.science (2026). https://pith.science/paper/3RQFYD3B

@misc{pith2026250111586,
  author       = {Pith},
  title        = {Pith review of: Compressibility Analysis for the differentiable shift-variant Filtered Backprojection Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3RQFYD3B}},
  note         = {Machine review of arXiv:2501.11586}
}

read the original abstract

The differentiable shift-variant filtered backprojection (FBP) model enables the reconstruction of cone-beam computed tomography (CBCT) data for any non-circular trajectories. This method employs deep learning technique to estimate the redundancy weights required for reconstruction, given knowledge of the specific trajectory at optimization time. However, computing the redundancy weight for each projection remains computationally intensive. This paper presents a novel approach to compress and optimize the differentiable shift-variant FBP model based on Principal Component Analysis (PCA). We apply PCA to the redundancy weights learned from sinusoidal trajectory projection data, revealing significant parameter redundancy in the original model. By integrating PCA directly into the differentiable shift-variant FBP reconstruction pipeline, we develop a method that decomposes the redundancy weight layer parameters into a trainable eigenvector matrix, compressed weights, and a mean vector. This innovative technique achieves a remarkable 97.25% reduction in trainable parameters without compromising reconstruction accuracy. As a result, our algorithm significantly decreases the complexity of the differentiable shift-variant FBP model and greatly improves training speed. These improvements make the model substantially more practical for real-world applications.

Figures

Figures reproduced from arXiv: 2501.11586 by the authors.

**Figure 2.** Compressed differentiable shift-variant FBP model: Low-dimensional redundancy weights are initially mapped [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Reconstruction result with PCA-based dimensionality reduction. (a) Reconstruction without PCA-based di [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Redundancy weights comparison. (a) Redundancy weights recovered from the low-dimensional representation. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

9 extracted references · 6 canonical work pages

[1]

Herl, G., Hiller, J., and Maier, A., ``Scanning trajectory optimisation using a quantitative tuybased local quality estimation for robot-based x-ray computed tomography,'' Nondestructive Testing and Evaluation 35 (3), 287--303 (2020)

work page 2020
[2]

and Clack, R., ``A cone-beam reconstruction algorithm using shift-variant filtering and cone-beam backprojection,'' IEEE transactions on medical imaging 13 (1), 186--195 (1994)

Defrise, M. and Clack, R., ``A cone-beam reconstruction algorithm using shift-variant filtering and cone-beam backprojection,'' IEEE transactions on medical imaging 13 (1), 186--195 (1994)

work page 1994
[3]

Ye, C., Schneider, L.-S., Sun, Y., and Maier, A., ``Deep learning computed tomography based on the defrise and clack algorithm,'' arXiv preprint arXiv:2403.00426 (2024)

work page Pith review arXiv 2024
[4]

Ye, C., Schneider, L.-S., Sun, Y., Thies, M., Mei, S., and Maier, A., ``Draco: Differentiable reconstruction for arbitrary cbct orbits,'' arXiv preprint arXiv:2410.14900 (2024)

work page Pith review arXiv 2024
[5]

and Ratajczak, W., ``Principal components analysis (pca),'' Computers & Geosciences 19 (3), 303--342 (1993)

Ma \'c kiewicz, A. and Ratajczak, W., ``Principal components analysis (pca),'' Computers & Geosciences 19 (3), 303--342 (1993)

work page 1993
[6]

N., ``A disciplined approach to neural network hyper-parameters: Part 1--learning rate, batch size, momentum, and weight decay,'' arXiv preprint arXiv:1803.09820 (2018)

Smith, L. N., ``A disciplined approach to neural network hyper-parameters: Part 1--learning rate, batch size, momentum, and weight decay,'' arXiv preprint arXiv:1803.09820 (2018)

arXiv 2018
[7]

K., ``Pyro-nn: Python reconstruction operators in neural networks,'' Medical physics 46 (11), 5110--5115 (2019)

Syben, C., Michen, M., Stimpel, B., Seitz, S., Ploner, S., and Maier, A. K., ``Pyro-nn: Python reconstruction operators in neural networks,'' Medical physics 46 (11), 5110--5115 (2019)

work page 2019
[8]

the cancer imaging archive,'' The Cancer Imaging Archive https://doi

Hong, J., Reyngold, M., Crane, C., Cuaron, J., Hajj, C., Mann, J., Zinovoy, M., Yorke, E., LoCastro, E., Apte, A., et al., ``Breath-hold ct and cone-beam ct images with expert manual organ-at-risk segmentations from radiation treatments of locally advanced pancreatic cancer [data set]. the cancer imaging archive,'' The Cancer Imaging Archive https://doi. ...

doi:10.7937/tcia 2021

Show all 9 references

[9]

write newline

" write newline "" before.all 'output.state := FUNCTION blank.sep after.quote 'output.state := FUNCTION fin.entry output.state after.quoted.block = 'skip 'add.period if write newline FUNCTION new.block output.state before.all = 'skip output.state after.quote = after.quoted.blo...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

Herl, G., Hiller, J., and Maier, A., ``Scanning trajectory optimisation using a quantitative tuybased local quality estimation for robot-based x-ray computed tomography,'' Nondestructive Testing and Evaluation 35 (3), 287--303 (2020)

work page 2020

[2] [2]

and Clack, R., ``A cone-beam reconstruction algorithm using shift-variant filtering and cone-beam backprojection,'' IEEE transactions on medical imaging 13 (1), 186--195 (1994)

Defrise, M. and Clack, R., ``A cone-beam reconstruction algorithm using shift-variant filtering and cone-beam backprojection,'' IEEE transactions on medical imaging 13 (1), 186--195 (1994)

work page 1994

[3] [3]

Ye, C., Schneider, L.-S., Sun, Y., and Maier, A., ``Deep learning computed tomography based on the defrise and clack algorithm,'' arXiv preprint arXiv:2403.00426 (2024)

work page Pith review arXiv 2024

[4] [4]

Ye, C., Schneider, L.-S., Sun, Y., Thies, M., Mei, S., and Maier, A., ``Draco: Differentiable reconstruction for arbitrary cbct orbits,'' arXiv preprint arXiv:2410.14900 (2024)

work page Pith review arXiv 2024

[5] [5]

and Ratajczak, W., ``Principal components analysis (pca),'' Computers & Geosciences 19 (3), 303--342 (1993)

Ma \'c kiewicz, A. and Ratajczak, W., ``Principal components analysis (pca),'' Computers & Geosciences 19 (3), 303--342 (1993)

work page 1993

[6] [6]

N., ``A disciplined approach to neural network hyper-parameters: Part 1--learning rate, batch size, momentum, and weight decay,'' arXiv preprint arXiv:1803.09820 (2018)

Smith, L. N., ``A disciplined approach to neural network hyper-parameters: Part 1--learning rate, batch size, momentum, and weight decay,'' arXiv preprint arXiv:1803.09820 (2018)

arXiv 2018

[7] [7]

K., ``Pyro-nn: Python reconstruction operators in neural networks,'' Medical physics 46 (11), 5110--5115 (2019)

Syben, C., Michen, M., Stimpel, B., Seitz, S., Ploner, S., and Maier, A. K., ``Pyro-nn: Python reconstruction operators in neural networks,'' Medical physics 46 (11), 5110--5115 (2019)

work page 2019

[8] [8]

the cancer imaging archive,'' The Cancer Imaging Archive https://doi

Hong, J., Reyngold, M., Crane, C., Cuaron, J., Hajj, C., Mann, J., Zinovoy, M., Yorke, E., LoCastro, E., Apte, A., et al., ``Breath-hold ct and cone-beam ct images with expert manual organ-at-risk segmentations from radiation treatments of locally advanced pancreatic cancer [data set]. the cancer imaging archive,'' The Cancer Imaging Archive https://doi. ...

doi:10.7937/tcia 2021