arxiv: 2604.16161 · v2 · submitted 2026-04-17 · ⚛️ physics.med-ph

Recognition: unknown

VQ-Wave: A physics-driven spatio-temporal deep learning approach for non-contrast-enhanced lung ventilation and perfusion MRI

Grzegorz Bauman , Pavlos Panos , Philipp Latzin , Oliver Bieri

Authors on Pith no claims yet

Pith reviewed 2026-05-10 06:57 UTC · model grok-4.3

classification ⚛️ physics.med-ph

keywords lung MRIventilation perfusion imagingdeep learningnon-contrast enhancedcystic fibrosisspatio-temporal networkspectral decompositionfunctional lung imaging

0 comments

The pith

A physics-driven neural network extracts stable lung ventilation and perfusion maps from short non-contrast MRI scans even when breathing patterns are irregular.

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

The paper presents VQ-Wave, a spatio-temporal deep learning model trained on synthetic signals that include amplitude modulations, frequency drifts, and noise. The network processes both local spatial context and temporal evolution to separate physiological ventilation and perfusion signals from noise. In numerical tests it maintains low error rates where matrix pencil decomposition shows instability from spectral leakage. In vivo scans of cystic fibrosis patients it produces consistent maps with under 12 percent mean variation even when total acquisition time drops from 45 to 15 seconds, while the conventional method suffers amplitude bias and regional dropouts.

Core claim

VQ-Wave is a physics-driven spatio-temporal inception neural network trained on synthetic signal models to estimate ventilation and perfusion parameters. By learning to decouple physiological signals from noise in the presence of non-stationary dynamics, the method yields ventilation and perfusion maps that remain quantitatively stable in short acquisitions and accurately reflect functional defects in cystic fibrosis patients, in contrast to matrix pencil decomposition which degrades under irregular physiology and reduced scan lengths.

What carries the argument

VQ-Wave, a physics-driven spatio-temporal inception neural network trained on synthetic signal models that simulate amplitude modulations, frequency drifts, and noise, to learn decoupling of ventilation and perfusion signals from noise.

If this is right

Functional lung MRI becomes feasible with acquisition protocols shortened to 15 seconds while preserving quantitative stability.
Ventilation and perfusion maps remain accurate in patients with irregular breathing such as those with cystic fibrosis.
The method avoids the systematic amplitude instability and regional signal dropouts that appear in matrix pencil results under short scans and non-stationary physiology.
Non-contrast functional imaging can be performed reliably at 1.5 T without requiring long breath-hold or stationary conditions.

Where Pith is reading between the lines

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

The same training strategy could be adapted to other dynamic MRI applications where physiological non-stationarity limits spectral methods.
Shorter scans would lower motion artifacts and improve patient compliance in pediatric or respiratory-compromised populations.
If the network generalizes, it could support repeated measurements during a single session to track treatment response without added radiation or contrast.

Load-bearing premise

The synthetic training signals that include amplitude modulations, frequency drifts, and noise are representative enough of the non-stationary physiological dynamics found in real in-vivo lung MRI acquisitions.

What would settle it

Acquire additional 15-second lung MRI scans in cystic fibrosis patients with independently verified ventilation defects and check whether VQ-Wave maps show mean quantitative variation above 12 percent or systematically miss the known defects while matrix pencil decomposition succeeds.

Figures

Figures reproduced from arXiv: 2604.16161 by Grzegorz Bauman, Oliver Bieri, Pavlos Panos, Philipp Latzin.

**Figure 1.** Figure 1: Generation of non-stationary physiological signal time-courses using the synthetic physiological simulator. (A) Evolution of instantaneous frequencies for ventilation (𝑓𝑣 , blue) and perfusion (𝑓𝑞, red). Unlike standard spectral models that assume constant periodicity, the simulator introduces stochastic frequency drifts and heart rate variability to mimic natural physiological instability. (B) The respira… view at source ↗

**Figure 3.** Figure 3: Quantitative validation of physiological parameter recovery. Performance of the proposed VQ-Wave network (blue) is compared to the reference global matrix pencil (MP, red) method using stationary physiological signal simulations at a uniform noise level of σ=5.0. (A, B) Scatter plots demonstrating the linearity of recovered amplitude for ventilation and perfusion channels. Pearson correlation coefficients … view at source ↗

**Figure 4.** Figure 4: Qualitative assessment of spatial reconstruction fidelity on the numerical lung phantom. Representative amplitude maps generated from ground truth signals under high noise conditions (σ=11.0). Columns compare the (left) ground truth reference, (center) matrix pencil reconstruction, and (right) the proposed VQ-Wave network. Zoomed insets focus on fine anatomical structures of pulmonary vessels. While both m… view at source ↗

**Figure 6.** Figure 6: Spatial fidelity of ventilation reconstruction. Comparison of reconstruction accuracy between the conventional matrix pencil algorithm (MP, left block) and the proposed VQ-Wave network (right block) against numerical ground truth. Spatial error maps (reconstructed minus ground truth amplitude) and voxel-wise scatter plots are evaluated under three simulated scenarios: an ideal baseline steady-state acquisi… view at source ↗

**Figure 8.** Figure 8: In-vivo performance and quantitative stability under physiological non-stationarity. Evaluation of a 14-year-old male with cystic fibrosis exhibiting irregular cardiopulmonary dynamics. Clinical lung function tests indicated a forced vital capacity (FVC) of 100%, forced expiratory volume in 1 second (FEV1) of 85%, and a lung clearance index at 2.5% (LCI 2.5%) of 13.02. (Top rows) Qualitative comparison of … view at source ↗

**Figure 9.** Figure 9: Temporal reproducibility of functional reconstructions from truncated acquisitions. Comparison of fractional ventilation and perfusion maps generated using three distinct 40-frame windows extracted from beginning, middle and end of the full-length free-breathing cystic fibrosis patient acquisition. While VQ-Wave maintains consistent structural and quantitative fidelity across all temporal windows, the matr… view at source ↗

**Figure 10.** Figure 10: Diagnostic defect mapping under stable physiological conditions. Evaluation of a 12- year-old female with cystic fibrosis exhibiting stable cardiopulmonary dynamics. Clinical lung function tests indicated a forced vital capacity (FVC) of 82%, forced expiratory volume in 1 second (FEV1) of 82%, and a lung clearance index at 2.5% (LCI 2.5%) of 7.68. (Top rows) Comparison of fractional ventilation and perfus… view at source ↗

read the original abstract

Purpose: To develop a robust deep learning framework for non-contrast-enhanced functional lung MRI, overcoming the limitations of spectral decomposition in the presence of physiological non-stationarity. Methods: We introduce VQ-Wave (Ventilation/Q-perfusion Waveform-based Assessment of Variable Evolutions), a physics-driven spatio-temporal inception neural network trained on synthetic signal models to estimate ventilation and perfusion parameters. By processing local spatial context alongside temporal evolution, the network learns to decouple physiological signals from noise. The training generator simulated non-stationary dynamics, including amplitude modulations, frequency drifts, and noise. Performance was validated against matrix pencil (MP) decomposition using numerical phantoms and in-vivo lung MRI acquired in four healthy volunteers and two children with cystic fibrosis (CF) at 1.5T. Results: In numerical benchmarks, VQ-Wave demonstrated superior robustness to non-stationarity, maintaining low global and regional error rates where MP exhibited stochastic instability due to spectral leakage. In-vivo, VQ-Wave accurately captured functional defects in patients with CF yielding ventilation and perfusion maps with high quantitative stability (mean variation < 12%) even when scan time was reduced from 45s to 15s. Conversely, under irregular physiology and short scan lengths, MP decomposition severely degraded, exhibiting systematic amplitude instability, overestimation bias, and regional signal dropouts. Conclusion: VQ-Wave offers a robust, physics-driven neural network-based alternative to spectral decomposition. By effectively handling physiological irregularity and noise, it enables reliable functional lung imaging with substantially shortened acquisition protocols.

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.

Desk Editor's Note private letter to a colleague

VQ-Wave beats matrix pencil on phantoms with a new inception network but in-vivo claims need more validation.

read the letter

The punchline is that this paper presents a physics-driven deep learning method called VQ-Wave that uses a spatio-temporal inception network trained on synthetic non-stationary signals to separate ventilation and perfusion in lung MRI without contrast. It shows clear improvements over the matrix pencil approach in controlled phantom tests. What the paper does well is address the problem of physiological non-stationarity by including amplitude modulations, frequency drifts, and noise in the training data. The results on numerical phantoms indicate lower global and regional errors, and the in-vivo tests suggest the maps remain stable even with shorter 15-second scans in a small group of subjects including children with cystic fibrosis. The soft spots are the very limited in-vivo sample of only six subjects total, with just two CF patients, and no independent ground truth reference for the functional maps. The stability metric of less than 12% variation is reported, but without details on statistical testing or controls for domain shift from synthetic to real data, the generalization to irregular physiology remains uncertain. The comparison is mainly against matrix pencil, which degrades as expected, but stronger baselines or real-world references would strengthen the case. This paper is for MRI researchers and pulmonologists focused on non-contrast functional imaging, particularly for pediatric applications where shorter scans matter. Readers looking for alternatives to spectral decomposition in noisy or variable signals will find the training approach useful. It deserves a serious referee because the core idea and phantom results are concrete and reproducible in principle, even if the clinical evidence is early stage. I recommend sending it for peer review, with the expectation that revisions will add more subjects and validation data.

Referee Report

3 major / 2 minor

Summary. The paper introduces VQ-Wave, a physics-driven spatio-temporal inception neural network trained exclusively on synthetic signal models incorporating amplitude modulations, frequency drifts, and noise, to estimate ventilation and perfusion maps from non-contrast-enhanced lung MRI. It claims superior robustness to physiological non-stationarity compared to matrix pencil (MP) decomposition, with lower errors on numerical phantoms and stable in-vivo maps (mean variation <12%) on a cohort of 4 healthy volunteers and 2 CF patients at 1.5T, even under reduced scan times from 45s to 15s, enabling shorter acquisition protocols.

Significance. If the generalization from synthetic training to real non-stationary lung signals holds, VQ-Wave could provide a practical alternative to spectral methods for functional lung imaging without contrast or hyperpolarized gases, particularly benefiting pediatric CF patients by allowing reliable defect mapping with abbreviated scans. The physics-informed training approach and direct comparison to MP on both phantoms and in-vivo data represent a constructive step toward handling irregular physiology.

major comments (3)

[In-vivo validation and results] The in-vivo results section asserts that VQ-Wave 'accurately captured functional defects in patients with CF' based on visual inspection and quantitative stability (<12% mean variation under scan-time reduction), but provides no independent reference standard (e.g., hyperpolarized gas MRI or contrast-enhanced perfusion) for the 2 CF subjects; this leaves the accuracy claim unanchored and load-bearing for the central generalization argument.
[Methods: synthetic signal models and training] Methods description of the synthetic training generator (amplitude modulations, frequency drifts, additive noise) does not include quantitative validation that these statistics match the spectral or temporal characteristics of real CF lung MRI (e.g., cardiac-respiratory coupling or perfusion heterogeneity), creating a domain-shift risk that directly affects the reported robustness advantage over MP.
[Methods: VQ-Wave network and training] No details are provided on network architecture specifics (number of inception modules, kernel sizes, or spatio-temporal fusion layers), exact training protocol (loss function, optimizer, epoch count, or regularization), or statistical testing (e.g., paired t-tests or regional error distributions) for the phantom comparisons, undermining reproducibility and the claim of 'superior robustness'.

minor comments (2)

[Results: in-vivo quantitative stability] Clarify the precise definition and computation of 'mean variation <12%' (e.g., across subjects, regions, or voxels) and whether it is reported with standard deviation or confidence intervals.
[Methods] The abstract and results would benefit from explicit mention of the total number of synthetic training samples and any data augmentation strategies used to mitigate overfitting to the generator.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the thoughtful and constructive review. We address each major comment point-by-point below. Revisions have been made to improve clarity, reproducibility, and to acknowledge limitations where appropriate.

read point-by-point responses

Referee: The in-vivo results section asserts that VQ-Wave 'accurately captured functional defects in patients with CF' based on visual inspection and quantitative stability (<12% mean variation under scan-time reduction), but provides no independent reference standard (e.g., hyperpolarized gas MRI or contrast-enhanced perfusion) for the 2 CF subjects; this leaves the accuracy claim unanchored and load-bearing for the central generalization argument.

Authors: We agree that the lack of an independent reference standard (such as hyperpolarized 129Xe MRI) for the two CF patients means the accuracy claim rests primarily on expert visual assessment and quantitative stability metrics. In the revised manuscript we have softened the language from 'accurately captured' to 'captured functional defects consistent with clinical expectations' and added an explicit limitations paragraph discussing the need for future multi-modal validation. The core robustness argument, however, is still supported by the phantom experiments and the direct head-to-head comparison with matrix-pencil decomposition under controlled non-stationarity. revision: partial
Referee: Methods description of the synthetic training generator (amplitude modulations, frequency drifts, additive noise) does not include quantitative validation that these statistics match the spectral or temporal characteristics of real CF lung MRI (e.g., cardiac-respiratory coupling or perfusion heterogeneity), creating a domain-shift risk that directly affects the reported robustness advantage over MP.

Authors: The synthetic generator was parameterized from literature values and from qualitative inspection of our own in-vivo time series, but no quantitative spectral matching was reported. We have now added a supplementary figure that overlays power-spectral-density and autocorrelation functions computed on real healthy and CF lung MRI against the synthetic ensemble; the dominant respiratory and cardiac frequency bands show good overlap. This addition directly addresses the domain-shift concern while preserving the physics-driven training strategy. revision: yes
Referee: No details are provided on network architecture specifics (number of inception modules, kernel sizes, or spatio-temporal fusion layers), exact training protocol (loss function, optimizer, epoch count, or regularization), or statistical testing (e.g., paired t-tests or regional error distributions) for the phantom comparisons, undermining reproducibility and the claim of 'superior robustness'.

Authors: We apologize for these omissions. The revised Methods section now specifies: three inception modules with parallel 1-D kernels of size 3, 5 and 7 (temporal) and 3×3 (spatial), followed by concatenation and a 1×1 fusion convolution; training used mean-squared-error loss, Adam optimizer (learning rate 1e-4), 200 epochs with early stopping (patience 20), and L2 regularization (weight 1e-5). Phantom comparisons now include paired t-tests (p < 0.01) and a supplementary table of regional error distributions. These additions restore full reproducibility. revision: yes

standing simulated objections not resolved

Independent reference-standard validation (e.g., hyperpolarized gas MRI) for the two CF patients cannot be provided within the current dataset and would require new acquisitions.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper trains VQ-Wave on independently generated synthetic signals that incorporate amplitude modulations, frequency drifts and noise, then evaluates performance on separate numerical phantoms and in-vivo scans from distinct subjects. Quantitative stability and defect-capture claims are measured relative to matrix-pencil decomposition on held-out data rather than being algebraically forced by the training generator or any fitted parameter within the same dataset. No self-definitional, fitted-input-called-prediction, or self-citation-load-bearing steps appear in the provided abstract or method description.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach depends on the domain assumption that synthetic non-stationary signal models capture essential real-world physiology and that the learned network generalizes without introducing systematic bias from the training distribution.

free parameters (1)

Neural network weights and hyperparameters
Large number of trainable parameters learned from synthetic data; specific values and selection criteria not provided in abstract.

axioms (1)

domain assumption Synthetic signals with amplitude modulations, frequency drifts, and noise adequately model physiological non-stationarity in lung MRI.
Invoked to justify training the network on simulated data before applying to in-vivo scans.

pith-pipeline@v0.9.0 · 5599 in / 1323 out tokens · 51517 ms · 2026-05-10T06:57:06.783370+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

34 extracted references · 28 canonical work pages · 2 internal anchors

[1]

MRI of the Lung, Springer International Publishing, Cham,

Kauczor HU, Wielpütz MO, eds. MRI of the Lung, Springer International Publishing, Cham,
[2]

Doi: 10.1007/978-3-319-42617-4

work page doi:10.1007/978-3-319-42617-4
[3]

MRI of the lung (1/3): methods

Wild JM, Marshall H, Bock M, Schad LR, Jakob PM, Puderbach M, Molinari F, Van Beek EJ, Biederer J. MRI of the lung (1/3): methods. Insights Imaging. 2012 Aug;3(4):345 -53. doi: 10.1007/s13244-012-0176-x

work page doi:10.1007/s13244-012-0176-x 2012
[4]

Lung MRI: Indications, Capabilities, and Techniques-AJR Expert Panel Narrative Review

Azour L, Ohno Y, Biederer J, Hochhegger B, Bauman G, Hatabu H, Schiebler ML, Ackman JB. Lung MRI: Indications, Capabilities, and Techniques-AJR Expert Panel Narrative Review. AJR Am J Roentgenol. 2025 Oct;225(4):e2532637. doi: 10.2214/AJR.25.32637

work page doi:10.2214/ajr.25.32637 2025
[5]

Pulmonary perfusion imaging using MRI: clinical application

Ley S, Ley-Zaporozhan J. Pulmonary perfusion imaging using MRI: clinical application. Insights Imaging. 2012 Feb;3(1):61-71. doi: 10.1007/s13244-011-0140-1

work page doi:10.1007/s13244-011-0140-1 2012
[6]

Gadolinium-Based Contrast Media Nephrotoxicity in Kidney Impairment: The Physio-Pathological Conditions for the Perfect Murder

Martino F, Amici G, Rosner M, Ronco C, Novara G. Gadolinium-Based Contrast Media Nephrotoxicity in Kidney Impairment: The Physio-Pathological Conditions for the Perfect Murder. J Clin Med. 2021 Jan 13;10(2):271. doi: 10.3390/jcm10020271

work page doi:10.3390/jcm10020271 2021
[7]

Gadolinium-Based Contrast Agents in Pregnant Women: A Literature Review of MRI Safety

Alghamdi SA Sr. Gadolinium-Based Contrast Agents in Pregnant Women: A Literature Review of MRI Safety. Cureus. 2023 May 3;15(5):e38493. doi: 10.7759/cureus.38493

work page doi:10.7759/cureus.38493 2023
[8]

Gadolinium deposition in the brain: summary of evidence and recommendations

Gulani V, Calamante F, Shellock FG, Kanal E, Reeder SB; International Society for Magnetic Resonance in Medicine. Gadolinium deposition in the brain: summary of evidence and recommendations. Lancet Neurol. 2017 Jul;16(7):564-570. doi: 10.1016/S1474- 4422(17)30158-8

work page doi:10.1016/s1474- 2017
[9]

Imaging of lung function using hyperpolarized helium-3 magnetic resonance imaging: Review of current and emerging translational methods and applications

Fain S, Schiebler ML, McCormack DG, Parraga G. Imaging of lung function using hyperpolarized helium-3 magnetic resonance imaging: Review of current and emerging translational methods and applications. J Magn Reson Imaging. 2010 Dec;32(6):1398-408. doi: 10.1002/jmri.22375

work page doi:10.1002/jmri.22375 2010
[10]

The role of hyperpolarized 129xenon in MR imaging of pulmonary function

Ebner L, Kammerman J, Driehuys B, Schiebler ML, Cadman RV, Fain SB. The role of hyperpolarized 129xenon in MR imaging of pulmonary function. Eur J Radiol. 2017 Jan;86:343-352. doi: 10.1016/j.ejrad.2016.09.015. Epub 2016 Sep 16

work page doi:10.1016/j.ejrad.2016.09.015 2017
[11]

Magnetic resonance lung function--a breakthrough for lung imaging and functional assessment? A phantom study and clinical trial

Zapke M, Topf HG, Zenker M, Kuth R, Deimling M, Kreisler P, Rauh M, Chefd'hotel C, Geiger B, Rupprecht T. Magnetic resonance lung function--a breakthrough for lung imaging and functional assessment? A phantom study and clinical trial. Respir Res. 2006 Aug 6;7(1):106. doi: 10.1186/1465-9921-7-106

work page doi:10.1186/1465-9921-7-106 2006
[12]

Flow-weighted MRI of the lungs with the ECG-gated half-Fourier FSE technique: evaluation of the effect of the cardiac cycle

Kawanami S, Nakamura K, Miyazaki M, Sugiura S, Yamamoto S, Nakata H. Flow-weighted MRI of the lungs with the ECG-gated half-Fourier FSE technique: evaluation of the effect of the cardiac cycle. Magn Reson Med Sci. 2002 Nov 1;1(3):137-47. doi: 10.2463/mrms.1.137

work page doi:10.2463/mrms.1.137 2002
[13]

Non-contrast-enhanced perfusion and ventilation assessment of the human lung by means of fourier decomposition in proton MRI

Bauman G, Puderbach M, Deimling M, Jellus V, Chefd'hotel C, Dinkel J, Hintze C, Kauczor HU, Schad LR. Non-contrast-enhanced perfusion and ventilation assessment of the human lung by means of fourier decomposition in proton MRI. Magn Reson Med. 2009 Sep;62(3):656-64. doi: 10.1002/mrm.22031

work page doi:10.1002/mrm.22031 2009
[14]

Matrix pencil decomposition of time-resolved proton MRI for robust and improved assessment of pulmonary ventilation and perfusion

Bauman G, Bieri O. Matrix pencil decomposition of time-resolved proton MRI for robust and improved assessment of pulmonary ventilation and perfusion. Magn Reson Med. 2017 Jan;77(1):336-342. doi: 10.1002/mrm.26096

work page doi:10.1002/mrm.26096 2017
[15]

Feasibility of quantitative regional ventilation and perfusion mapping with phase-resolved functional lung (PREFUL) MRI in healthy volunteers and COPD, CTEPH, and CF patients

Voskrebenzev A, Gutberlet M, Klimeš F, Kaireit TF, Schönfeld C, Rotärmel A, Wacker F, Vogel-Claussen J. Feasibility of quantitative regional ventilation and perfusion mapping with phase-resolved functional lung (PREFUL) MRI in healthy volunteers and COPD, CTEPH, and CF patients. Magn Reson Med. 2018 Apr;79(4):2306-2314. doi: 10.1002/mrm.26893

work page doi:10.1002/mrm.26893 2018
[16]

SElf-gated Non-Contrast-Enhanced FUnctional Lung imaging (SENCEFUL) using a quasi-random fast low-angle shot (FLASH) sequence and proton MRI

Fischer A, Weick S, Ritter CO, Beer M, Wirth C, Hebestreit H, Jakob PM, Hahn D, Bley T, Köstler H. SElf-gated Non-Contrast-Enhanced FUnctional Lung imaging (SENCEFUL) using a quasi-random fast low-angle shot (FLASH) sequence and proton MRI. NMR Biomed. 2014 Aug;27(8):907-17. doi: 10.1002/nbm.3134

work page doi:10.1002/nbm.3134 2014
[17]

Dynamic mode decomposition of dynamic MRI for assessment of pulmonary ventilation and perfusion

Ilicak E, Ozdemir S, Zapp J, Schad LR, Zöllner FG. Dynamic mode decomposition of dynamic MRI for assessment of pulmonary ventilation and perfusion. Magn Reson Med. 2023 Aug;90(2):761-769. doi: 10.1002/mrm.29656

work page doi:10.1002/mrm.29656 2023
[18]

Automated reference-free detection of motion artifacts in magnetic resonance images

Küstner T, Liebgott A, Mauch L, Martirosian P, Bamberg F, Nikolaou K, Yang B, Schick F, Gatidis S. Automated reference-free detection of motion artifacts in magnetic resonance images. MAGMA. 2018 Apr;31(2):243-256. doi: 10.1007/s10334-017-0650-z

work page doi:10.1007/s10334-017-0650-z 2018
[19]

End- systole and end-diastole detection in short axis cine MRI using a fully convolutional neural network with dilated convolutions

Pérez-Pelegrí M, Monmeneu JV, López-Lereu MP, Maceira AM, Bodi V, Moratal D. End- systole and end-diastole detection in short axis cine MRI using a fully convolutional neural network with dilated convolutions. Comput Med Imaging Graph. 2022 Jul;99:102085. doi: 10.1016/j.compmedimag.2022.102085

work page doi:10.1016/j.compmedimag.2022.102085 2022
[20]

Fast and accurate view classification of echocardiograms using deep learning

Madani A, Arnaout R, Mofrad M, Arnaout R. Fast and accurate view classification of echocardiograms using deep learning. NPJ Digit Med. 2018;1:6. doi: 10.1038/s41746-017- 0013-1

work page doi:10.1038/s41746-017- 2018
[21]

Voskrebenzev A, Gutberlet M, Klimeš F, Kaireit TF, Shin HO, Kauczor HU, Welte T, Wacker F, Vogel-Claussen J. A synthetic lung model (ASYLUM) for validation of functional lung imaging methods shows significant differences between signal-based and deformation-field- based ventilation measurements. Front Med (Lausanne). 2024 Sep 4;11:1418052. doi: 10.3389/fm...

work page doi:10.3389/fmed.2024.1418052 2024
[22]

A method for incorporating organ motion due to breathing into 3D dose calculations

Lujan AE, Larsen EW, Balter JM, Ten Haken RK. A method for incorporating organ motion due to breathing into 3D dose calculations. Med Phys 1999;26:715–720

1999
[23]

An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling

Bai S, Kolter JZ, Koltun V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271. 2018 Apr 19

work page internal anchor Pith review arXiv 2018
[24]

Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, Kaiser Ł, Polosukhin I,
[25]

Attention Is All You Need

Attention is all you need. In: Guyon I, Luxburg UV, Bengio S, Wallach H, Fergus R, Vishwanathan S, Garnett R (Eds.). Attention is All you Need. In Guyon I, Von Luxburg U, Bengio S, Wallach H, Fergus R, Vishwanathan S, Garnett R (ed.). 31st Conference on Neural Information Processing Systems (NIPS). Advances in Neural Information Processing Systems. Vol. 3...

work page internal anchor Pith review Pith/arXiv arXiv 2017
[26]

Speaker recognition from raw waveform with SincNet

Ravanelli M, Bengio Y. Speaker recognition from raw waveform with SincNet. 2018 IEEE Spoken Language Technology Workshop (SLT), 1021-1028

2018
[27]

Squeeze-and-Excitation Networks

Hu J, Shen L, Sun G. Squeeze-and-Excitation Networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2018:7132-7141

2018
[28]

InceptionTime: Finding AlexNet for Time Series Classification

Fawaz HI, Lucas B, Forestier G, Pelletier C, Schmidt DF, Weber J, Webb GI, Idoumghar L, Muller PA, Petitjean F. InceptionTime: Finding AlexNet for Time Series Classification. Data Mining and Knowledge Discovery. 2020; 34(6), 1936-1962. doi: 10.1007/s10618-020-00710

work page doi:10.1007/s10618-020-00710 2020
[29]

PyTorch: An imperative style, high-performance deep learning library

Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G, Killeen T, Lin Z, Gimelshein N, Antiga L, Desmaison A, Köpf A, Yang E, DeVito Z, Raison M, Tejani A, Chilamkurthy S, Steiner B, Fang L, Bai J, Chintala S. PyTorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems (32), 2019: 8024– 8035

2019
[30]

MRI lung lobe segmentation in pediatric cystic fibrosis patients using a recurrent neural network trained with publicly accessible CT datasets

Pusterla O, Heule R, Santini F, Weikert T, Willers C, Andermatt S, Sandkühler R, Nyilas S, Latzin P, Bieri O, Bauman G. MRI lung lobe segmentation in pediatric cystic fibrosis patients using a recurrent neural network trained with publicly accessible CT datasets. Magn Reson Med. 2022 Jul;88(1):391-405. doi: 10.1002/mrm.29184

work page doi:10.1002/mrm.29184 2022
[31]

Ultra-fast Steady-State Free Precession Pulse Sequence for Fourier Decomposition Pulmonary MRI

Bauman G, Pusterla O, Bieri O. Ultra-fast Steady-State Free Precession Pulse Sequence for Fourier Decomposition Pulmonary MRI. Magn Reson Med. 2016 Apr;75(4):1647-53. doi: 10.1002/mrm.25697

work page doi:10.1002/mrm.25697 2016
[32]

Adaptive Graph Diffusion Regularisation for Discontinuity Preserving Image Registration, in: Klein S, Staring M, Durrleman S, Sommer S (Eds.), Biomed

Sandkühler R, Jud C, Pezold S, Cattin PC. Adaptive Graph Diffusion Regularisation for Discontinuity Preserving Image Registration, in: Klein S, Staring M, Durrleman S, Sommer S (Eds.), Biomed. Image Regist., Springer International Publishing, Cham, 2018: pp. 24–34. https://doi.org/10.1007/978-3-319-92258-4_3

work page doi:10.1007/978-3-319-92258-4_3 2018
[33]

An automated pipeline for computation and analysis of functional ventilation and perfusion lung MRI with matrix pencil decomposition: TrueLung

Pusterla O, Willers C, Sandkühler R, Andermatt S, Nyilas S, Cattin PC, Latzin P, Bieri O, Bauman G. An automated pipeline for computation and analysis of functional ventilation and perfusion lung MRI with matrix pencil decomposition: TrueLung. Z Med Phys. 2024 Sep 19:S0939-3889(24)00084-9. doi: 10.1016/j.zemedi.2024.08.001

work page doi:10.1016/j.zemedi.2024.08.001 2024
[34]

stochastic instability

Kjørstad Å, Corteville DM, Fischer A, Henzler T, Schmid-Bindert G, Zöllner FG, Schad LR. Quantitative lung perfusion evaluation using Fourier decomposition perfusion MRI. Magn Reson Med. 2014 Aug;72(2):558-62. doi: 10.1002/mrm.24930. FIGURES Figure 1: Generation of non -stationary physiological signal time -courses using the synthetic physiological simula...

work page doi:10.1002/mrm.24930 2014