pith. sign in

arxiv: 2606.09953 · v1 · pith:AOQTI6ZKnew · submitted 2026-06-08 · 📡 eess.IV · cs.AI· cs.LG

Deep Slice Interpolation for Reducing Through-Plane Anisotropy and Noise in Head CT

Luis Cort\'es Ferre , Miguel A. Guti\'errez-Naranjo , Marcin Balcerzyk This is my paper

Pith reviewed 2026-06-27 14:56 UTC · model grok-4.3

classification 📡 eess.IV cs.AIcs.LG
keywords deep learningCT slice interpolationanisotropy reductionimplicit denoisinghead CTstructural similaritymultiplanar reconstruction
0
0 comments X

The pith

A deep learning system synthesizes intermediate CT slices to reduce through-plane anisotropy while adding implicit denoising.

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

The paper describes a neural network trained to generate new slices between existing axial CT images, effectively halving the spacing in the through-plane direction. This targets the common mismatch in head CT where in-plane resolution is high but slice thickness creates anisotropy that harms 3D views and volume estimates. Multiple loss combinations were tested, with all converged models beating classical interpolation and video methods on structural metrics. The same forward pass also produces denoised outputs, an effect that appeared on an external out-of-distribution scan.

Core claim

The system takes pairs of neighboring axial slices and outputs synthesized intermediate slices that halve effective through-plane spacing. All trained models surpass classical baselines and pretrained video interpolation networks on structural measures, with the MS-SSIM plus L1 combination giving the strongest overall profile. On an external head CT series the model reproduces the implicit denoising behavior predicted by the referenced theoretical analysis, indicating that both interpolation quality and the denoising side-effect are not limited to the training distribution.

What carries the argument

A convolutional network trained end-to-end with hybrid pixel-wise and multi-scale structural similarity losses to perform slice interpolation on anisotropic CT volumes.

If this is right

  • Multiplanar reformats and 3D visualizations gain resolution without additional scanning.
  • Volumetric measurements such as hematoma volume become more precise because voxel isotropy improves.
  • Downstream algorithms that expect near-isotropic input receive higher-quality data from the same acquisition.
  • Denoising occurs automatically during interpolation, removing the need for a separate noise-reduction step.
  • Patient-level bootstrap intervals and paired tests confirm the measured gains over baselines.

Where Pith is reading between the lines

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

  • The same architecture could be retrained on other anisotropic modalities such as certain MRI protocols to reduce slice spacing.
  • Routine clinical pipelines might adopt the method to improve existing low-dose or thick-slice acquisitions rather than increasing radiation exposure.
  • Larger external validation sets would be needed to confirm whether the single-case denoising observation scales to routine practice.
  • The documented instability of SSIM-family losses at small batch sizes points to a practical constraint for deployment on limited hardware.

Load-bearing premise

The implicit denoising and generalization seen on one external case will hold across wider clinical populations and scanner types.

What would settle it

Quantitative comparison of hematoma volume estimates computed from original versus interpolated volumes on a multi-center test set of head CT scans with known ground-truth volumes.

Figures

Figures reproduced from arXiv: 2606.09953 by Luis Cort\'es Ferre, Marcin Balcerzyk, Miguel A. Guti\'errez-Naranjo.

Figure 1
Figure 1. Figure 1: U-Net + EfficientNetV2-S architecture for CT slice interpolation. Top row: encoder (blue), left-to-right in decreasing spatial resolution. Bottom row: decoder (orange), right-to-left in increasing spatial resolution. Columns are aligned by spatial resolution: the leftmost column carries the 2-channel input [Ik, Ik+2] and the final decoder block fused with its 3 × 3 conv output projection (producing the sin… view at source ↗
Figure 2
Figure 2. Figure 2: Patch geometry shared by training-time sampling and test-time reconstruction. (a) Nine overlapping 256 × 256 crops whose top-left corners form a 3 × 3 grid at 64-pixel stride in the 512 × 512 slice; their union covers the central 384 × 384 band (solid blue outline). Translucent fills overlap so deeper color indicates regions covered by more crops; the dashed outline marks one representative crop (top-left … view at source ↗
Figure 3
Figure 3. Figure 3: Radially-averaged 1D noise power spectrum in uniform white-matter ROIs, averaged over 28 test patients (median pixel spacing 0.488 mm/px). Curves compare the acquired-slice NPS to the prediction NPS of each model on the same ROIs. Pixel-wise regression losses (L1, MSE, MS-SSIM+L1) suppress NPS across essentially the full frequency band; SSIMs shows a smaller but qualitatively similar reduction. The linear_… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative interpolation result at the mid-brain ventricular / basal-ganglia level. Patient ID_615f69e3, target triplet index 14; RSNA label any= 0. Model: SSIM loss (lr = 3×10−3 ). Detail cell: 2× zoom of a 160 × 160 basal-ganglia / ventricular ROI; the yellow rectangle marks the zoom origin. Brain window and radiological convention as described in the text. The posterior-right low-intensity halo is the … view at source ↗
Figure 5
Figure 5. Figure 5: Additional qualitative examples at mid-brain axial levels: hemorrhage case (top) and consecutive normal slice (bottom). Model, display window, zoom layout, and orientation follow [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Cross-method qualitative comparison on two mid-brain test slices. Top row of each panel: left input slice, ground truth, right input slice. Bottom row: our SSIMs model, RIFE, FILM, and mean baseline. RIFE and FILM produce plausible but structurally less accurate outputs, particularly near bone boundaries and hemorrhage margins; the mean baseline is visibly blurred. Brain window; radiological convention. 18… view at source ↗
Figure 7
Figure 7. Figure 7: Predictions and pixel-wise absolute error maps for four loss functions on the same test slice as [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Implicit denoising across 10 consecutive slices (hemorrhage patient). Column 1: original noisy acquired slice. Column 2: model prediction from neighboring slices. Column 3: absolute difference, showing predominantly noise-like texture removal. Gross hemorrhagic and anatomical structures remain visible, but diagnostic preservation requires reader validation. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Implicit denoising across 10 consecutive slices at normal (non-hemorrhage) axial levels from the same patient as [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Four consecutive reconstructed slices from HUVR-1 in the coarse-spacing region (k ≥ 18; ≈1.8 mm gap between non-overlapping 3.70 mm-thick slabs) and their model residuals. Leftmost column: acquired slice (brain window, center 44 HU, width 128 HU). Remaining columns: signed diff Iˆk −realk for each trained loss; red> 0, blue< 0; saturation ± the 99th-percentile absolute diff across all models and coarse-re… view at source ↗
Figure 11
Figure 11. Figure 11: Quantitative decomposition of the residual on HUVR-1. Top-left: per-slice mean signed residual in HU for each predictor, across the middle-slice range k ∈ [2, N−1]; the dashed vertical line marks the transition from the overlap region to the coarse-spacing region. Top-right: normalised spectrum |FFT|/total of the mean-centred residual restricted to the coarse-spacing region (k ≥ 18, L = 14), with the righ… view at source ↗
read the original abstract

Head computed tomography (CT) typically uses sub-millimeter in-plane resolution but 2-5 mm through-plane spacing, creating substantial anisotropy that degrades multiplanar reconstructions, volumetric measurements such as hematoma volume estimation, and downstream algorithms that assume near-isotropic voxels. We present a deep learning system that synthesizes intermediate CT slices from pairs of neighboring axial slices, halving the effective through-plane spacing. The system improves three-dimensional visualization while simultaneously producing inherently denoised outputs, yielding two complementary benefits from a single inference pass. To build a reliable system, we systematically evaluate pixel-wise losses, namely mean squared error (MSE) and mean absolute error (L1); structural-similarity losses, namely the structural similarity index (SSIM) and its multi-scale variant (MS-SSIM); and hybrid combinations. On a held-out test set, all converged models outperform classical interpolation baselines and pretrained video frame interpolation methods (RIFE, FILM) on all structural measures, with MS-SSIM+L1 offering the strongest balanced profile. We also document training instability in SSIM-family losses and identify partial remedies: the standard numerical fixes eliminate the dominant failure mode but leave residual divergence at smaller batch sizes. All results are reported with patient-level bootstrap confidence intervals and paired statistical tests. As an illustration, we apply the system to an out-of-distribution head CT series from Hospital Universitario Virgen del Roc\'io: the model synthesizes intermediate slices and exhibits on the real slices the implicit-denoising signature predicted by our theoretical analysis, supporting in a single external case that interpolation quality and implicit denoising are not confined to the training distribution.

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

1 major / 2 minor

Summary. The paper introduces a deep learning system to synthesize intermediate axial slices in head CT scans, halving through-plane spacing to mitigate anisotropy. It systematically compares pixel-wise (MSE, L1) and structural (SSIM, MS-SSIM) losses plus hybrids, reporting that all converged models outperform classical interpolation and pretrained video methods (RIFE, FILM) on held-out test data across structural metrics, with MS-SSIM+L1 strongest. Results include patient-level bootstrap CIs and paired tests. An out-of-distribution external case from Hospital Universitario Virgen del Rocío is used to illustrate implicit denoising consistent with a referenced theoretical analysis.

Significance. If the held-out performance claims hold, the work targets a practical clinical limitation in CT by improving multiplanar reformats, volumetric estimates, and downstream isotropic-assuming algorithms, while offering incidental denoising as a byproduct. The systematic loss-function ablation and use of patient-level bootstrap confidence intervals with paired statistical tests are explicit strengths that enhance credibility of the outperformance results over baselines.

major comments (1)
  1. [Abstract] Abstract (final paragraph): the assertion that results on the single external OOD series 'support... that interpolation quality and implicit denoising are not confined to the training distribution' rests on one case exhibiting the predicted signature; this is insufficient to underwrite the generalization claim without additional OOD series or quantitative metrics comparing synthesized vs. real slices on that data.
minor comments (2)
  1. [Abstract] Abstract: model architecture, training protocol, and dataset sizes (train/validation/test split) are omitted, which limits immediate verification of the central performance claims even though statistical reporting is present.
  2. [Abstract] Abstract: the phrase 'Roc\'io' contains a LaTeX escape artifact and should render as 'Rocío'.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation of minor revision. We address the major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract (final paragraph): the assertion that results on the single external OOD series 'support... that interpolation quality and implicit denoising are not confined to the training distribution' rests on one case exhibiting the predicted signature; this is insufficient to underwrite the generalization claim without additional OOD series or quantitative metrics comparing synthesized vs. real slices on that data.

    Authors: We agree that a single external case constitutes an illustration rather than robust evidence for generalization. We will revise the abstract to describe the OOD example as an illustration of the predicted implicit-denoising signature on out-of-distribution data, removing the phrasing that it 'supports' the claim that these properties are not confined to the training distribution. The revised text will emphasize the illustrative purpose without overstating generalizability. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical results on held-out and external data are independent of training inputs

full rationale

The paper reports model performance via standard held-out test-set metrics and one external OOD series, with no equations, fitted parameters, or self-citations that reduce any claimed prediction or denoising signature to the training data by construction. The referenced theoretical analysis is invoked only to interpret the single external observation and does not serve as a load-bearing derivation for the quantitative outperformance results.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on standard deep-learning assumptions for image synthesis tasks without introducing new physical entities or many explicitly fitted parameters beyond typical training choices.

free parameters (1)
  • hybrid loss weights
    Combinations such as MS-SSIM+L1 require weighting coefficients whose values are not stated in the abstract.
axioms (1)
  • domain assumption Convolutional networks trained on paired neighboring slices can synthesize plausible intermediate medical images.
    Central premise of the interpolation system.

pith-pipeline@v0.9.1-grok · 5843 in / 1228 out tokens · 33239 ms · 2026-06-27T14:56:20.770241+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

40 extracted references · 21 canonical work pages

  1. [1]

    Flanders, Luciano M

    Adam E. Flanders, Luciano M. Prevedello, George Shih, Safwan S. Halabi, Jayashree Kalpathy-Cramer, Robyn Ball, John T. Mongan, Anouk Stein, Felipe C. Kitamura, Matthew P. Lungren, Gagandeep Choudhary, Lesley Cala, Luiz Coelho, Monique Mogensen, Fanny Morón, Elka Miller, Ichiro Ikuta, Vahe Zohrabian, Olivia McDonnell, Christie Lincoln, Lubdha Shah, David J...

    work page doi:10.1148/ryai.2020190211 2019

  2. [2]

    Deep learning applied to intracranial hemorrhage detection.Journal of Imaging, 9(2),

    Luis Cortés-Ferre, Miguel Angel Gutiérrez-Naranjo, Juan José Egea-Guerrero, Soledad Pérez-Sánchez, and Marcin Balcerzyk. Deep learning applied to intracranial hemorrhage detection.Journal of Imaging, 9(2),

  3. [3]

    doi:10.3390/jimaging9020037

    ISSN 2313-433X. doi:10.3390/jimaging9020037. URLhttps://www.mdpi.com/2313-433X/9/2/37

    work page doi:10.3390/jimaging9020037

  4. [4]

    Real-time intermediate flow estimation for video frame interpolation

    Zhewei Huang, Tianyuan Zhang, Wen Heng, Boxin Shi, and Shuchang Zhou. Real-time intermediate flow estimation for video frame interpolation. In Shai Avidan, Gabriel Brostow, Moustapha Cissé, Giovanni Maria Farinella, and Tal Hassner, editors,Computer Vision – ECCV 2022, pages 624–642, Cham, 2022. Springer Nature Switzerland. ISBN 978-3-031-19781-9

    2022

  5. [5]

    Film: Frame interpolation for large motion

    Fitsum Reda, Janne Kontkanen, Eric Tabellion, Deqing Sun, Caroline Pantofaru, and Brian Curless. Film: Frame interpolation for large motion. InEuropean Conference on Computer Vision (ECCV), 2022

    2022

  6. [6]

    Bishop.Pattern Recognition and Machine Learning by Christopher M

    C.M. Bishop.Pattern Recognition and Machine Learning by Christopher M. Bishop. Springer Sci- ence+Business Media, LLC, 2006. URLhttps://books.google.es/books?id=Y44SyAEACAAJ

    2006

  7. [7]

    Hastie, R

    T. Hastie, R. Tibshirani, and J. Friedman.The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Second Edition. Springer Series in Statistics. Springer New York, 2009. ISBN 9780387848587. URLhttps://books.google.es/books?id=tVIjmNS3Ob8C

    2009

  8. [8]

    Noise2Noise: Learning image restoration without clean data

    Jaakko Lehtinen, Jacob Munkberg, Jon Hasselgren, Samuli Laine, Tero Karras, Miika Aittala, and Timo Aila. Noise2Noise: Learning image restoration without clean data. In Jennifer Dy and Andreas Krause, editors,Proceedings of the 35th International Conference on Machine Learning, volume 80 ofProceedings of Machine Learning Research, pages 2965–2974. PMLR, 1...

    2018

  9. [9]

    Lehmann, C

    T.M. Lehmann, C. Gonner, and K. Spitzer. Survey: interpolation methods in medical image processing. IEEE Transactions on Medical Imaging, 18(11):1049–1075, 1999. doi:10.1109/42.816070

    work page doi:10.1109/42.816070 1999

  10. [10]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Cheng Peng, Wei-An Lin, Haofu Liao, Rama Chellappa, and S. Kevin Zhou. Saint: Spatially aware interpolation network for medical slice synthesis. In2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7747–7756, 2020. doi:10.1109/CVPR42600.2020.00777

    work page doi:10.1109/cvpr42600.2020.00777 2020

  11. [11]

    Christodoulou, and Debiao Li

    Yuhua Chen, Yibin Xie, Zhengwei Zhou, Feng Shi, Anthony G. Christodoulou, and Debiao Li. Brain mri super resolution using 3d deep densely connected neural networks. In2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018), pages 739–742, 2018. doi:10.1109/ISBI.2018.8363679

    work page doi:10.1109/isbi.2018.8363679 2018

  12. [12]

    Dewey, Dzung L

    Can Zhao, Blake E. Dewey, Dzung L. Pham, Peter A. Calabresi, Daniel S. Reich, and Jerry L. Prince. Smore: A self-supervised anti-aliasing and super-resolution algorithm for mri using deep learning.IEEE Transactions on Medical Imaging, 40(3):805–817, 2021. doi:10.1109/TMI.2020.3037187

    work page doi:10.1109/tmi.2020.3037187 2021

  13. [13]

    I³net: Inter-intra-slice interpolation network for medical slice synthesis.IEEE Transactions on Medical Imaging, 43:3306–3318, 2024

    Haofei Song, Xintian Mao, Jing Yu, Qingli Li, and Yan Wang. I³net: Inter-intra-slice interpolation network for medical slice synthesis.IEEE Transactions on Medical Imaging, 43:3306–3318, 2024. URL https://api.semanticscholar.org/CorpusID:269407969

    2024

  14. [14]

    Adacof: Adaptive collaboration of flows for video frame interpolation

    Hyeongmin Lee, Taeoh Kim, Tae young Chung, Daehyun Pak, Yuseok Ban, and Sangyoun Lee. Adacof: Adaptive collaboration of flows for video frame interpolation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020

    2020

  15. [15]

    Coleman, Paddy Gilligan, and Stefano Sanvito

    Laura Gambini, Cian Gabbett, Luke Doolan, Lewys Jones, Jonathan N. Coleman, Paddy Gilligan, and Stefano Sanvito. Video frame interpolation neural network for 3d tomography across different length scales.Nature Communications, 15(1):7962, Sep 2024. ISSN 2041-1723. doi:10.1038/s41467-024-52260-2. URLhttps://doi.org/10.1038/s41467-024-52260-2

    work page doi:10.1038/s41467-024-52260-2 2024

  16. [16]

    Data-efficient unsupervised interpolation without any intermediate frame for 4d medical images, 2024

    JungEun Kim, Hangyul Yoon, Geondo Park, Kyungsu Kim, and Eunho Yang. Data-efficient unsupervised interpolation without any intermediate frame for 4d medical images, 2024. URLhttps://arxiv.org/ abs/2404.01464. 34

    arXiv 2024

  17. [17]

    Sit-sr 3d: Self-supervised slice interpolation via transfer learning for 3d volume super-resolution.Pattern Recognition Letters, 166: 97–104, 2023

    Muhammad Sarmad, Leonardo Carlos Ruspini, and Frank Lindseth. Sit-sr 3d: Self-supervised slice interpolation via transfer learning for 3d volume super-resolution.Pattern Recognition Letters, 166: 97–104, 2023. ISSN 0167-8655. doi:https://doi.org/10.1016/j.patrec.2023.01.008. URLhttps://www. sciencedirect.com/science/article/pii/S0167865523000144

    work page doi:10.1016/j.patrec.2023.01.008 2023

  18. [18]

    Review of the Finite Bandwidth Concept

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InMedical Image Computing and Computer-Assisted Intervention – MICCAI 2015, volume 9351 ofLecture Notes in Computer Science, pages 234–241. Springer, 2015. doi:10.1007/978-3- 319-24574-4_28

    work page doi:10.1007/978-3- 2015

  19. [19]

    Segmentation models pytorch.https://github.com/qubvel/segmentation_models

    Pavel Iakubovskii. Segmentation models pytorch.https://github.com/qubvel/segmentation_models. pytorch, 2019

    2019

  20. [20]

    Mingxing Tan and Quoc V. Le. Efficientnetv2: Smaller models and faster training. InInternational Con- ference on Machine Learning, 2021. URLhttps://api.semanticscholar.org/CorpusID:232478903

    2021

  21. [21]

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A. Efros. Image-to-image translation with conditional adversarial networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5967–5976, 2017. doi:10.1109/CVPR.2017.632

    work page doi:10.1109/cvpr.2017.632 2017

  22. [22]

    Loss functions for image restoration with neural networks.IEEE Transactions on Computational Imaging, 3:47–57, 2017

    Hang Zhao, Orazio Gallo, Iuri Frosio, and Jan Kautz. Loss functions for image restoration with neural networks.IEEE Transactions on Computational Imaging, 3:47–57, 2017. URLhttps://api. semanticscholar.org/CorpusID:5334482

    2017

  23. [23]

    IEEE Transactions on Image Processing 13(4), 600–612 (Apr 2004)

    Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from er- ror visibility to structural similarity.IEEE Transactions on Image Processing, 13(4):600–612, 2004. doi:10.1109/TIP.2003.819861

    work page doi:10.1109/tip.2003.819861 2004

  24. [24]

    Wang, E.P

    Z. Wang, E.P. Simoncelli, and A.C. Bovik. Multiscale structural similarity for image quality assessment. InThe Thrity-Seventh Asilomar Conference on Signals, Systems & Computers, 2003, volume 2, pages 1398–1402 Vol.2, 2003. doi:10.1109/ACSSC.2003.1292216

    work page doi:10.1109/acssc.2003.1292216 2003

  25. [25]

    Sergey Kastryulin, Jamil Zakirov, Denis Prokopenko, and Dmitry V. Dylov. Pytorch image quality: Metrics for image quality assessment, 2022. URLhttps://arxiv.org/abs/2208.14818

    arXiv 2022

  26. [26]

    Tizabi, Michael Baumgartner, Matthias Eisenmann, Doreen Heckmann- Nötzel, A

    Annika Reinke, Minu D. Tizabi, Michael Baumgartner, Matthias Eisenmann, Doreen Heckmann- Nötzel, A. Emre Kavur, Tim Rädsch, Carole H. Sudre, Laura Acion, Michela Antonelli, Tal Arbel, Spyridon Bakas, Arriel Benis, Florian Buettner, M. Jorge Cardoso, Veronika Cheplygina, Jianxu Chen, Evangelia Christodoulou, Beth A. Cimini, Keyvan Farahani, Luciana Ferrer,...

    work page doi:10.1038/s41592-023-02150-0 2024

  27. [27]

    Deep learning-based algorithms for low- dose ct imaging: A review.European Journal of Radiology, 172:111355, 2024

    Hongchi Chen, Qiuxia Li, Lazhen Zhou, and Fangzuo Li. Deep learning-based algorithms for low- dose ct imaging: A review.European Journal of Radiology, 172:111355, 2024. ISSN 0720-048X. doi:https://doi.org/10.1016/j.ejrad.2024.111355. URL https://www.sciencedirect.com/science/ article/pii/S0720048X24000718

    work page doi:10.1016/j.ejrad.2024.111355 2024

  28. [28]

    Trzasko, Natalia Khaylova, James M

    Armando Manduca, Lifeng Yu, Joshua D. Trzasko, Natalia Khaylova, James M. Kofler, Cynthia M. McCollough, and Joel G. Fletcher. Projection space denoising with bilateral filtering and ct noise modeling 35 for dose reduction in ct.Medical Physics, 36(11):4911–4919, 2009. doi:https://doi.org/10.1118/1.3232004. URLhttps://aapm.onlinelibrary.wiley.com/doi/abs/...

    work page doi:10.1118/1.3232004 2009

  29. [29]

    Reducing CT radiation dose with iterative reconstruction algorithms: the influence of scan and reconstruction parameters on image quality and CTDIvol.Eur

    Thorsten Klink, Verena Obmann, Johannes Heverhagen, Alexander Stork, Gerhard Adam, and Philipp Begemann. Reducing CT radiation dose with iterative reconstruction algorithms: the influence of scan and reconstruction parameters on image quality and CTDIvol.Eur. J. Radiol., 83(9):1645–1654, September 2014

    2014

  30. [30]

    Low-dose CT with a residual encoder-decoder convolutional neural network.IEEE Trans

    Hu Chen, Yi Zhang, Mannudeep K Kalra, Feng Lin, Yang Chen, Peixi Liao, Jiliu Zhou, and Ge Wang. Low-dose CT with a residual encoder-decoder convolutional neural network.IEEE Trans. Med. Imaging, 36(12):2524–2535, December 2017

    2017

  31. [31]

    2015, in 2015 IEEE International Conference on Computer Vision (ICCV), 1026–1034, doi: 10.1109/ICCV.2015.123

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 1026–1034, 2015. doi:10.1109/ICCV.2015.123

    work page doi:10.1109/iccv.2015.123 2015

  32. [32]

    Understanding the difficulty of training deep feedforward neural networks

    Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. InProceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics (AISTATS), volume 9, pages 249–256, 2010

    2010

  33. [33]

    Claude E. Duchon. Lanczos filtering in one and two dimensions.Journal of Applied Meteorology, 18(8): 1016–1022, 1979. doi:10.1175/1520-0450(1979)018<1016:LFIOAT>2.0.CO;2

    work page doi:10.1175/1520-0450(1979)018 1979

  34. [34]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations, 2019. URLhttps://openreview.net/forum?id=Bkg6RiCqY7

    2019

  35. [35]

    B. Efron. Bootstrap methods: Another look at the jackknife.The Annals of Statistics, 7(1):1–26, 1979. URLhttp://www.jstor.org/stable/2958830. Full publication date: Jan., 1979

    arXiv 1979

  36. [36]

    Individual comparisons by ranking methods.Biometrics Bulletin, 1(6):80–83, 1945

    Frank Wilcoxon. Individual comparisons by ranking methods.Biometrics Bulletin, 1(6):80–83, 1945. URLhttps://api.semanticscholar.org/CorpusID:53662922

    1945

  37. [37]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , author =

    Yoav Benjamini and Yosef Hochberg. Controlling the false discovery rate: A practical and powerful approachtomultipletesting.Journal of the Royal Statistical Society: Series B (Methodological), 57(1):289– 300, 1995. doi:https://doi.org/10.1111/j.2517-6161.1995.tb02031.x. URLhttps://rss.onlinelibrary. wiley.com/doi/abs/10.1111/j.2517-6161.1995.tb02031.x

    work page doi:10.1111/j.2517-6161.1995.tb02031.x 1995

  38. [38]

    H. B. Mann and D. R. Whitney. On a Test of Whether one of Two Random Variables is Stochastically Larger than the Other.The Annals of Mathematical Statistics, 18(1):50 – 60, 1947. doi:10.1214/aoms/1177730491. URLhttps://doi.org/10.1214/aoms/1177730491

    work page doi:10.1214/aoms/1177730491 1947

  39. [39]

    J. L. Hodges Jr. and E. L. Lehmann. Estimates of Location Based on Rank Tests.The Annals of Mathematical Statistics, 34(2):598 – 611, 1963. doi:10.1214/aoms/1177704172. URLhttps://doi.org/ 10.1214/aoms/1177704172

    work page doi:10.1214/aoms/1177704172 1963

  40. [40]

    The simple difference formula: An approach to teaching nonparametric correlation

    Dave Kerby. The simple difference formula: An approach to teaching nonparametric correlation. Comprehensive Psychology, 3, 01 2014. doi:10.2466/11.IT.3.1. 36

    work page doi:10.2466/11.it.3.1 2014