Pith. sign in

REVIEW 3 major objections 4 minor 79 references

A 2.5D spectral graph encoder outperforms 3D transformers and CNNs on chest CT abnormality classification.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 10:14 UTC pith:PMV7TRKU

load-bearing objection Solid 2.5D graph-based CT encoder with broad multi-dataset evaluation, but the statistical significance claims and NLP-derived labels are weaker than the presentation suggests. the 3 major comments →

arxiv 2510.10779 v5 pith:PMV7TRKU submitted 2025-10-12 cs.CV

Structured Spectral Graph Representation Learning for Multi-label Abnormality Analysis from 3D CT Scans

Theo Di Piazza , Carole Lazarus , Olivier Nempont , Loic Boussel This is my paper
classification cs.CV
keywords 3D CTmulti-label classificationgraph neural networksspectral convolutionChebyshev polynomials2.5D representationcross-dataset generalizationabnormality detection
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper argues that a 3D CT scan is better modeled as a sparse graph of axial slice triplets than as a dense volume of 3D patches. It introduces CT-SSG, a 2.5D encoder in which 80 triplet nodes exchange information through spectral (Chebyshev) graph convolutions weighted by physical slice spacing. Trained and tested with 5-fold cross-validation on a large public chest CT cohort, the model reports a macro-F1 of 57.18, with consistent margins over a 2.5D attention baseline, a 3D CNN, and a 3D video transformer, all initialized from the same natural-image pretrained weights. The same frozen representations improve report-generation metrics and, under linear probing, beat training-from-scratch on abdominal CT when labeled data are scarce. If correct, the paper shows that explicit inter-slice geometry can substitute for massive 3D pretraining.

Core claim

On the paper's own terms, the central discovery is that restricting inter-slice message passing to a local, physically weighted neighborhood—rather than letting every slice triplet attend to every other—is the right inductive bias for volumetric CT interpretation. CT-SSG partitions the volume into non-overlapping triplets of axial slices, encodes each independently with a shared 2D network, adds a learnable axial positional embedding, and then runs a single spectral block using Chebyshev polynomial filters on the graph Laplacian. In the paper's experiments this sparse spectral configuration outperforms fully connected graph and attention topologies and beats 3D CNN and 3D transformer baselin

What carries the argument

The load-bearing object is a structured graph whose nodes are axial slice triplets (80 nodes for a standardized 240-slice volume) and whose edges are undirected, weighted by an inverse function of inter-triplet distance along the z-axis. Messages are propagated by Chebyshev spectral convolution on the normalized graph Laplacian (filter size K=3, receptive field q=16), followed by layer normalization, a feedforward block, residual connections, mean pooling, and a linear classifier. This mechanism converts the 3D volume into a small, sparse graph so that long-range dependencies are handled by spectral filtering rather than by dense attention, keeping the model shallow (one block) and inexpensi

Load-bearing premise

The claim's load-bearing premise is that the NLP-derived abnormality labels (from a radiology language model, a rule-based labeler, and LLM inference) are accurate enough that the reported F1 differences reflect real visual findings rather than label noise, and if those labels are systematically biased the cross-dataset ranking could be measuring labeler artifacts.

What would settle it

Re-run the CT-RATE comparison on a held-out subset annotated by board-certified radiologists (rather than report-extracted labels), keeping the same train/test split and checkpoints; if CT-SSG's F1 margin over CT-Scroll and ViViT drops to statistical non-significance, the generalization claim is falsified.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If CT-SSG is right, a 2.5D graph representation can match or beat 3D transformers and 3D CNNs at the same ImageNet initialization, without domain-specific pretraining.
  • Sparse local connectivity is not a compromise but a benefit: across operators, restricting edges to q=16 neighbors along the z-axis beats fully connected graphs, implying structure beats expressivity for this task.
  • A single spectral block suffices; adding depth (3 or 5 blocks) does not improve F1, suggesting that the information needed for abnormality detection lives in nearby slices.
  • Pretrained CT-SSG representations transfer: a frozen encoder improves automated report generation and, via linear probing, outperforms from-scratch training on abdominal CT when fewer than about 3,750 labeled samples are available.
  • The reported robustness to z-axis translation and noise, if confirmed, supports deployment in settings where patient positioning and scanner calibration vary.

Where Pith is reading between the lines

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

  • I would test whether the margin survives expert-labeled ground truth: the paper's labels come from language-model extraction of radiology reports, so a re-evaluation on a radiologist-annotated subset is the cleanest check on whether the ranking reflects image content or label noise.
  • The graph formulation invites anatomically grounded edges (e.g., connecting nodes within the same lung lobe or mediastinum), which the paper deliberately avoids to stay segmentation-free; such edges could further close the gap to fully 3D models if anatomical priors are available.
  • Because all methods share the same 2D pretrained backbone as feature extractor, the observed gains are attributable to the aggregation mechanism; ablating the spectral module with a weaker backbone would quantify how much of the benefit is graph structure versus backbone capacity.
  • The 70 ms inference time and one-day single-GPU training suggest a practical path to real-time CT triage, but the paper does not study deployment constraints like memory under batch inference or latency on varied hardware; that is an extension, not a claim.

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

3 major / 4 minor

Summary. This paper introduces CT-SSG, a 2.5D graph-based encoder for multi-label abnormality classification in 3D chest CT. A CT volume is partitioned into 80 triplets of axial slices; each triplet is embedded by an ImageNet-pretrained 2D ResNet, the embeddings become nodes of a chain-like graph with distance-based edge weights, and a Chebyshev spectral convolution block (L=1) performs message passing. A pooled representation feeds an 18-way multi-label classifier. The authors report macro-F1 57.18±0.19 on CT-RATE, with consistent ranking on Rad-ChestCT and CT-HCL, ablations of components/topology/filter size, robustness to translation and noise, and transfer experiments to report generation and abdominal CT. The main claim is that this lightweight spectral graph encoder, with no 3D convolution or attention, outperforms 3D CNN/Transformer and 2.5D baselines initialized from ImageNet.

Significance. If the empirical claims are correct, the paper makes a useful contribution: it demonstrates that a carefully structured sparse graph over slice triplets can be a competitive and computationally light alternative to 3D transformers for volumetric CT classification. The evaluation is broad — three datasets, multiple baselines, ablations, robustness, and two downstream transfer tasks — and the pseudocode in Appendix C makes the implementation easy to reproduce. The main weaknesses concern the statistical foundation of the headline comparison: the significance test is based on five non-independent cross-validation folds, and the winning configuration is selected from ablations on the same test folds. These issues are fixable but currently leave the central superiority claim not fully established.

major comments (3)
  1. [§5.1, Table 2] The claim that CT-SSG outperforms CT-Scroll/CNN3D/ViViT with p<0.01 rests on paired t-tests over the five CV folds. These observations are not independent: training folds overlap substantially, so the t-test is anti-conservative; with only five differences the variance estimate is also unstable. The paper should use a valid test for repeated k-fold CV (e.g., Bouckaert-Frank corrected t-test, or a subject-level bootstrap/permutation), and should report confidence intervals. As written, the headline advantage over the closest baseline (+2.88 F1 over CT-Scroll) is not statistically established. Additionally, testing many baselines and metrics without multiplicity control inflates the Type I error rate.
  2. [§5.2, Tables 3–6] The hyperparameters L=1, q=16, K=3 in Table 1 and the component increments in Table 3 are selected/validated on the same CT-RATE test folds that produce the headline Table 2. No nested holdout or independent tuning set is described, so the reported 57.18 is an optimistically biased selection estimate. Moreover, several ablation differences are small (e.g., +0.42 F1 for sparse topology, +0.21 F1 for residual connection) and are reported without significance tests or multiplicity correction. Please provide a nested evaluation, or show that the chosen configuration, fixed a priori, also wins on a held-out dataset.
  3. [§4 and §5.4] The ground truth for the clinical metrics is NLP-derived: CT-RATE labels are extracted with RadBERT, Rad-ChestCT uses a SARLE labeler, CT-HCL labeling is described inconsistently (manual in §4 vs. BERT-based in Fig. 4a), and abdominal pseudo-labels come from LLM inference. In §5.4, the CE F1 is computed by extracting abnormalities from generated reports with RadBERT and comparing them to RadBERT-derived ground-truth labels; this circularity can inflate CE scores and makes absolute clinical claims dependent on the labeler's biases. This does not necessarily change the relative ranking of encoders under an identical protocol, but the paper should validate a subset against manual annotations and report CE with an alternative labeler.
minor comments (4)
  1. [§5.1] The text reports F1=57.06 while Table 2 reports 57.18; the relative gains are computed from 57.06. Please harmonize the reported value and the table.
  2. [Eq. (2), §4] The edge weights depend on s_z, but all experiments resample volumes to 0.75 mm z-spacing. The claimed adaptivity to varying slice spacing is therefore not tested; either add variable-spacing experiments or temper the claim.
  3. [Figure 4a / §4] Figure 4a says CT-HCL abnormalities are extracted with a BERT-based language model, while §4 says they are manually extracted by radiologists. Please reconcile this contradiction.
  4. [General] Typos and formatting issues: 'empiricaly', 'formaly', 'understability', a duplicated author name in the author block, and 'Nempont' appearing in the running header but not in the author list. Careful proofreading is needed.

Circularity Check

1 steps flagged

Classification claim is externally benchmarked and non-circular; the report-generation Clinical Efficacy metric is partially circular because predicted and ground-truth labels are produced by the same RadBERT extractor.

specific steps
  1. self definitional [Section 5.4 'Evaluation protocol'; Section 4 'Database']
    "Specifically, generated reports are given to a RadBERT (Yan et al., 2022) to extract predicted abnormalities as binary label vectors. These are compared against ground-truth annotations ... [CT-RATE] labels are automatically extracted from radiology reports using RadBERT (Yan et al., 2022)"

    The CE F1/CRG pipeline computes labels from the generated report with RadBERT, while the CT-RATE ground-truth vectors are also RadBERT outputs from the original reports. Thus the metric is RadBERT(generated_text) vs RadBERT(original_text): it measures whether the decoder produces text that triggers the same external extractor, not whether the encoder-detected pathology exists in the volume. A decoder that copies the original report (or its RadBERT-triggering patterns) achieves high CE independent of the visual representation, so the transfer-to-report-generation superiority claim is partly an artifact of the shared labeler. The main classification result is unaffected because there the model predicts directly from images against the same fixed external labels.

full rationale

The central claim (Tables 2 and 6) is an empirical benchmark comparison: CT-SSG is trained and evaluated on external CT-RATE folds and evaluated on Rad-ChestCT/CT-HCL with fixed labels. No architectural component is fitted to the test labels; the spectral convolution, edge weighting, and positional embeddings are standard components with stated formulas. Self-citations to CT-Scroll and the EMERGE workshop paper are contextual/baseline references, not invoked as proof of CT-SSG's superiority; the improvement over CT-Scroll is measured in the paper rather than assumed. The one genuine definitional circularity is in the report-generation Clinical Efficacy metric: predicted and ground-truth label vectors are both produced by the same RadBERT extractor, so the CE comparison reduces to RadBERT self-consistency across texts. This affects only the secondary report-generation transfer claim, not the classification core. The abdominal-transfer experiment uses LLM-derived pseudo-labels for both the linear probe and the supervised baseline, so the comparison remains fair despite label noise. Statistical concerns about paired t-tests on 5 non-independent folds and same-fold hyperparameter selection are correctness risks, not circularity, and are not scored here.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

CT-SSG is an empirical architecture paper. The central contribution depends on tuned architecture hyperparameters (L, K, q), a hand-designed edge-weighting function, and preprocessing choices, while the label quality and transfer targets are assumed from external datasets and NLP labelers.

free parameters (5)
  • Spectral filter size K = 3
    Tuned in Table 5; sparse q=16 with K=3 gives the best F1 (57.18) and is the headline configuration in Table 1.
  • Graph receptive field q = 16
    Ablated in Tables 5 and 6; sparse topology q=16 is chosen over fully connected q=80.
  • Number of spectral blocks L = 1
    Table 4 depth ablation; L=1 is best, deeper models do not improve.
  • Edge-weighting constants in Eq. 2 = 1 + 1/(1 + 3|i-j|s_z)
    The constants 3 and 1 in the hand-designed edge-weighting scheme are not learned and no sensitivity analysis is reported.
  • Triplet size C = 3 axial slices
    Fixed at 3 slices per node; inherited from the 2.5D design and not ablated.
axioms (6)
  • domain assumption Axial slice triplets processed independently by a 2D ResNet capture sufficient local context before graph propagation.
    Section 3.2 embeds each triplet separately; if useful information spans more than 3 slices or depends on true 3D texture, the feature initialization is lossy.
  • domain assumption Automatically extracted report labels are accurate enough to serve as ground truth.
    CT-RATE labels come from RadBERT, Rad-ChestCT from SARLE, and Merlin abdominal pseudo-labels from LLM inference (Sections 4 and 5.5); label noise is acknowledged but not quantified.
  • domain assumption Standardized crop/pad to 240x480x480 and resampling preserve all abnormalities.
    Section 4 Processing; any abnormality lost in resampling or cropping is invisible to every model and metric.
  • ad hoc to paper A spectral Chebyshev convolution on a hand-weighted chain graph is an appropriate inductive bias for inter-slice dependencies.
    Section 3.5 argues this qualitatively; the claim that spectral > spatial is supported only by selected ablation configurations (Table 6), not by a derivation.
  • domain assumption ImageNet-pretrained weights are a beneficial initialization for CT slices.
    All baselines use ImageNet initialization or weight inflation (Section 5.1); conclusions compare architectures conditional on this choice.
  • standard math Chebyshev polynomial approximation of the graph Laplacian provides valid localized spectral filtering on undirected graphs.
    Equations 6-7 and Section 3.5 rely on standard spectral graph theory from Defferrard et al.

pith-pipeline@v1.3.0-alltime-deepseek · 28424 in / 15676 out tokens · 142275 ms · 2026-08-04T10:14:38.601358+00:00 · methodology

0 comments
read the original abstract

With the growing volume of CT examinations, there is an increasing demand for automated tools such as organ segmentation, abnormality detection, and report generation to support radiologists in managing their clinical workload. Multi-label classification of 3D Chest CT scans remains a critical yet challenging problem due to the complex spatial relationships inherent in volumetric data and the wide variability of abnormalities. Existing methods based on 3D convolutional neural networks struggle to capture long-range dependencies, while Vision Transformers often require extensive pre-training on large-scale, domain-specific datasets to perform competitively. In this work, we propose a 2.5D alternative by introducing a new graph-based framework that represents 3D CT volumes as structured graphs, where axial slice triplets serve as nodes processed through spectral graph convolution, enabling the model to reason over inter-slice dependencies while maintaining complexity compatible with clinical deployment. Our method, trained and evaluated on 3 datasets from independent institutions, achieves strong cross-dataset generalization, and shows competitive performance compared to state-of-the-art visual encoders. We further conduct comprehensive ablation studies to evaluate the impact of various aggregation strategies, edge-weighting schemes, and graph connectivity patterns. Additionally, we demonstrate the broader applicability of our approach through transfer experiments on automated radiology report generation and abdominal CT data.

Figures

Figures reproduced from arXiv: 2510.10779 by Carole Lazarus, Loic Boussel, Olivier Nempont, Theo Di Piazza.

Figure 1
Figure 1. Figure 1: Axial slices from 3D CT Scans, with abnormali￾ties manually contoured in red, illustrating distinct visual characteristics. real-world clinical deployment (Aravazhi et al., 2025). Early approaches to multi-label abnormality classifica￾tion in CT imaging predominantly leveraged fully convolu￾tional networks. The recent release of CT-RATE, a large￾scale public dataset containing chest CT scans from over 21,0… view at source ↗
Figure 2
Figure 2. Figure 2: CT-SSG Architecture Overview. Adjacent axial slices are grouped into triplets, each representing a node in a graph. Edges between nodes are weighted according to their physical distance along the z-axis. Node features are enhanced with Triplet Axial Slices positional embeddings, and then processed by a Spectral Block that incorporates Chebyshev graph convolution for structured spectral modeling. The result… view at source ↗
Figure 3
Figure 3. Figure 3: Spectral Block with detailed notations. Input features are given to a first normalization layer, followed by spectral graph convolutions with a residual skip connection. These updated features are then fed to a feedforward neural network followed by a second normalization layer with a residual skip connection. 3.5 Spectral Domain Module A key challenge in this formulation is the variability in anatomical p… view at source ↗
Figure 4
Figure 4. Figure 4: Comprehensive analysis of the datasets. Metadata not available for the Rad-ChestCT dataset. a) Abnormalities from CT-HCL are extracted with a BERT-based language model trained on french radiology reports from manually extracted anotations. b) CT-HCL comprises data from 2,000 unique patients, with age randing from 20 to 100 years. c) CT-HCL volumes comes from Hospices Civil de Lyon, with scanners from four … view at source ↗
Figure 5
Figure 5. Figure 5: F1-Score per abnormality for the 18 abnormalities from the CT-RATE test set, comparing our proposed CT-SSG with representative 3D Convolutional and 3D Transformer baselines. For clarity, one representative model per family is reported. CT-SSG consistently improves over both baselines, with the largest absolute gains observed in Pericardial effusion (+∆8.96%), Calcification (+∆6.23%), and Pleural effusion (… view at source ↗
Figure 6
Figure 6. Figure 6: Average absolute F1-score improvements of CT-SSG over representative 3D convolutional (CNN) and 3D Transformer (ViViT) baselines. Abnormalities are grouped by anatomical region and pathophysiological type to highlight systematic patterns of gain. CT-SSG yields consistent improvements across groups. Model depth L F1-Score AUROC mAP Accuracy 1 57.18±0.19 83.69±0.31 40.81±0.19 81.03±0.38 3 56.46±0.20 83.04±0.… view at source ↗
Figure 7
Figure 7. Figure 7: Gradient-weighted class activation maps, extracted from the 2D ResNet from the triplet slices embeddings module, where darker regions indicate lower activations. For each input, we display the slice with the highest absolute activation value from the heatmap. 1347 [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Robustness evaluation. Left: macro-F1 under axial z-axis translations (−30 to +30 slices), where all meth￾ods remain invariant to volumetric shifts. Right: macro-F1 under Gaussian noise perturbations of increasing standard deviation, where performance is stable up to σ = 0.025 and CT-SSG maintains higher F1-Score than baselines even as noise increases. Sensitivity to patient body translation To emulate var… view at source ↗
Figure 9
Figure 9. Figure 9: Report generation framework overview. The frozen pretrained image encoder extract visual features that are given to a decoder which generates the report, in a auto-regressively manner. Evaluation protocol [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison between ground-truth reports and those generated with CT-SSG for 3D Chest CT volumes. Color-coded highlights indicate abnormalities correctly captured by the model, demonstrating alignment with ground-truth annotations. the robustness of the learned representations and suggest a promising avenue for future research in connecting abnormal￾ity representation learning with clinically f… view at source ↗
Figure 12
Figure 12. Figure 12: Linear probe framework overview. The frozen pretrained image encoder extract visual features that are given to a linear layer to predict abnormalities. Following established evaluation transfer protocols (Misra and Maaten, 2019; Bardes et al., 2021), we compare two configurations: a supervised baseline in which CT-SSG is trained from scratch with ImageNet-initialized ResNet weights, and a linear probing, … view at source ↗
Figure 13
Figure 13. Figure 13: Transfer to abdominal CT. Comparison of a linear probe trained on frozen chest-pretrained CT-SSG representations against a supervised CT-SSG trained from scratch on the Merlin Abdominal CT dataset (Blanke￾meier et al., 2024). Performance is reported in terms of macro-F1 and macro-mAP for different size of the train set. 5.6 Implementation Details Multi-label Abnormality Classification CT-SSG was trained u… view at source ↗
Figure 14
Figure 14. Figure 14: Comparison of F1-Score, AUROC and mAP across models with varying parameter counts. Despite comparable [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: t-SNE visualization, of the pooled features for the [PITH_FULL_IMAGE:figures/full_fig_p024_15.png] view at source ↗

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

79 extracted references · 44 linked inside Pith

  1. [1]

    Graph- Based Deep Learning for Medical Diagnosis and Analysis : Past , Present and Future

    David Ahmedt-Aristizabal, Mohammad Ali Armin, Simon Denman, Clinton Fookes, and Lars Petersson. Graph- Based Deep Learning for Medical Diagnosis and Analysis : Past , Present and Future . Sensors (Basel, Switzerland), 21 0 (14): 0 4758, July 2021. ISSN 1424-8220

    2021

  2. [2]

    An overview of deep learning in medical imaging

    Andrés Anaya-Isaza, Leonel Mera-Jiménez, and Martha Zequera-Diaz. An overview of deep learning in medical imaging. Informatics in Medicine Unlocked, 26: 0 100723, January 2021. ISSN 2352-9148

    2021

  3. [3]

    The integration of artificial intelligence into clinical medicine: Trends , challenges, and future directions

    Prasanna Sakthi Aravazhi, Praveen Gunasekaran, Neo Zhong Yi Benjamin, Andy Thai, Kiran Kishor Chandrasekar, Nikhil Deep Kolanu, Priyadarshi Prajjwal, Yogesh Tekuru, Lissette Villacreses Brito, and Pugazhendi Inban. The integration of artificial intelligence into clinical medicine: Trends , challenges, and future directions. Disease-a-month: DM, 71 0 (6): ...

    2025

  4. [4]

    ViViT : A Video Vision Transformer

    Anurag Arnab, Mostafa Dehghani, Georg Heigold, Chen Sun, Mario Lucic, and Cordelia Schmid. ViViT : A Video Vision Transformer . In 2021 IEEE / CVF International Conference on Computer Vision ( ICCV ) , pages 6816--6826, Montreal, QC, Canada, October 2021. IEEE. ISBN 978-1-6654-2812-5

    2021

  5. [5]

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer Normalization , July 2016. arXiv:1607.06450 [cs, stat]

    Pith/arXiv arXiv 2016

  6. [6]

    Neural Machine Translation by Jointly Learning to Align and Translate , May 2016

    Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural Machine Translation by Jointly Learning to Align and Translate , May 2016. arXiv:1409.0473 [cs]

    Pith/arXiv arXiv 2016

  7. [7]

    VICReg : Variance - Invariance - Covariance Regularization for Self - Supervised Learning

    Adrien Bardes, Jean Ponce, and Yann LeCun. VICReg : Variance - Invariance - Covariance Regularization for Self - Supervised Learning . October 2021

    2021

  8. [8]

    The Intelligible and Effective Graph Neural Additive Networks , December 2024

    Maya Bechler-Speicher, Amir Globerson, and Ran Gilad-Bachrach. The Intelligible and Effective Graph Neural Additive Networks , December 2024. arXiv:2406.01317 [cs]

    arXiv 2024

  9. [9]

    Laplacian Eigenmaps and Spectral Techniques for Embedding and Clustering

    Mikhail Belkin and Partha Niyogi. Laplacian Eigenmaps and Spectral Techniques for Embedding and Clustering . In Advances in Neural Information Processing Systems , volume 14. MIT Press, 2001

    2001

  10. [10]

    Mollie Bianchi and Timothy D. Barfoot. UAV Localization Using Autoencoded Satellite Images , February 2021. arXiv:2102.05692 [cs]

    Pith/arXiv arXiv 2021

  11. [11]

    Shah, Andrew Johnston, Robert D

    Louis Blankemeier, Joseph Paul Cohen, Ashwin Kumar, Dave Van Veen, Syed Jamal Safdar Gardezi, Magdalini Paschali, Zhihong Chen, Jean-Benoit Delbrouck, Eduardo Reis, Cesar Truyts, Christian Bluethgen, Malte Engmann Kjeldskov Jensen, Sophie Ostmeier, Maya Varma, Jeya Maria Jose Valanarasu, Zhongnan Fang, Zepeng Huo, Zaid Nabulsi, Diego Ardila, Wei-Hung Weng...

    arXiv 2024

  12. [12]

    On the Opportunities and Risks of Foundation Models , July 2022

    Rishi Bommasani. On the Opportunities and Risks of Foundation Models , July 2022. arXiv:2108.07258 [cs]

    Pith/arXiv arXiv 2022

  13. [13]

    Warshauer

    Joshua Broder and David M. Warshauer. Increasing utilization of computed tomography in the adult emergency department, 2000-2005. Emergency Radiology, 13 0 (1): 0 25--30, October 2006. ISSN 1070-3004

    2000

  14. [14]

    How Attentive are Graph Attention Networks ?, January 2022

    Shaked Brody, Uri Alon, and Eran Yahav. How Attentive are Graph Attention Networks ?, January 2022. arXiv:2105.14491 [cs]

    Pith/arXiv arXiv 2022

  15. [15]

    Spectral Networks and Locally Connected Networks on Graphs , May 2014

    Joan Bruna, Wojciech Zaremba, Arthur Szlam, and Yann LeCun. Spectral Networks and Locally Connected Networks on Graphs , May 2014. arXiv:1312.6203 [cs]

    Pith/arXiv arXiv 2014

  16. [16]

    Basketball Player Action Recognition and Tracking Using R (2+1) D CNN With Spatial -temporal Features

    Hao-Hsiang Chang, Yu-Hua Chang, Yi-Lung Shih, Cheng-Hsun Lin, and Huang-Chia Shih. Basketball Player Action Recognition and Tracking Using R (2+1) D CNN With Spatial -temporal Features . In 2024 IEEE 13th Global Conference on Consumer Electronics ( GCCE ) , pages 388--389, October 2024. ISSN: 2693-0854

    2024

  17. [17]

    Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering , February 2017

    Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering , February 2017. arXiv:1606.09375 [cs]

    Pith/arXiv arXiv 2017

  18. [18]

    DenOtter and Johanna Schubert

    Tami D. DenOtter and Johanna Schubert. Hounsfield Unit . In StatPearls . StatPearls Publishing, Treasure Island (FL), 2024

    2024

  19. [19]

    BERT : Pre -training of Deep Bidirectional Transformers for Language Understanding , October 2018

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT : Pre -training of Deep Bidirectional Transformers for Language Understanding , October 2018

    2018

  20. [20]

    Imitating Radiological Scrolling : A Global - Local Attention Model for 3D Chest CT Volumes Multi - Label Anomaly Classification

    Theo Di Piazza, Carole Lazarus, Olivier Nempont, and Loic Boussel. Imitating Radiological Scrolling : A Global - Local Attention Model for 3D Chest CT Volumes Multi - Label Anomaly Classification . January 2025 a

    2025

  21. [21]

    Structured Spectral Graph Learning for Anomaly Classification in 3D Chest CT Scans , August 2025 b

    Theo Di Piazza, Carole Lazarus, Olivier Nempont, and Loic Boussel. Structured Spectral Graph Learning for Anomaly Classification in 3D Chest CT Scans , August 2025 b . arXiv:2508.01045 [cs]

    Pith/arXiv arXiv 2025

  22. [22]

    An Image is Worth 16x16 Words : Transformers for Image Recognition at Scale , June 2021

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An Image is Worth 16x16 Words : Transformers for Image Recognition at Scale , June 2021. arXiv:2010.11929 [cs]

    Pith/arXiv arXiv 2021

  23. [23]

    Mazurowski, Joseph Y

    Rachel Lea Draelos, David Dov, Maciej A. Mazurowski, Joseph Y. Lo, Ricardo Henao, Geoffrey D. Rubin, and Lawrence Carin. Machine-learning-based multiple abnormality prediction with large-scale chest computed tomography volumes. Medical Image Analysis, 67: 0 101857, January 2021. ISSN 1361-8415

    2021

  24. [24]

    Fast Graph Representation Learning with PyTorch Geometric , April 2019

    Matthias Fey and Jan Eric Lenssen. Fast Graph Representation Learning with PyTorch Geometric , April 2019. arXiv:1903.02428 [cs]

    Pith/arXiv arXiv 2019

  25. [25]

    Beam Search Strategies for Neural Machine Translation

    Markus Freitag and Yaser Al-Onaizan. Beam Search Strategies for Neural Machine Translation . In Proceedings of the First Workshop on Neural Machine Translation , pages 56--60, 2017. arXiv:1702.01806 [cs]

    Pith/arXiv arXiv 2017

  26. [26]

    On Over - Squashing in Message Passing Neural Networks : The Impact of Width , Depth , and Topology , May 2023

    Francesco Di Giovanni, Lorenzo Giusti, Federico Barbero, Giulia Luise, Pietro Lio', and Michael Bronstein. On Over - Squashing in Message Passing Neural Networks : The Impact of Width , Depth , and Topology , May 2023. arXiv:2302.02941 [cs]

    Pith/arXiv arXiv 2023

  27. [27]

    HIGT : Hierarchical Interaction Graph - Transformer for Whole Slide Image Analysis , September 2023

    Ziyu Guo, Weiqin Zhao, Shujun Wang, and Lequan Yu. HIGT : Hierarchical Interaction Graph - Transformer for Whole Slide Image Analysis , September 2023. arXiv:2309.07400 [cs]

    Pith/arXiv arXiv 2023

  28. [28]

    GenerateCT : Text - Conditional Generation of 3D Chest CT Volumes , November 2023

    Ibrahim Ethem Hamamci, Sezgin Er, Enis Simsar, Anjany Sekuboyina, Chinmay Prabhakar, Alperen Tezcan, Ayse Gulnihan Simsek, Sevval Nil Esirgun, Furkan Almas, Irem Doğan, Muhammed Furkan Dasdelen, Hadrien Reynaud, Sarthak Pati, Christian Bluethgen, Mehmet Kemal Ozdemir, and Bjoern Menze. GenerateCT : Text - Conditional Generation of 3D Chest CT Volumes , No...

    Pith/arXiv arXiv 2023

  29. [29]

    Developing Generalist Foundation Models from a Multimodal Dataset for 3D Computed Tomography , October 2024 a

    Ibrahim Ethem Hamamci, Sezgin Er, Furkan Almas, Ayse Gulnihan Simsek, Sevval Nil Esirgun, Irem Dogan, Muhammed Furkan Dasdelen, Omer Faruk Durugol, Bastian Wittmann, Tamaz Amiranashvili, Enis Simsar, Mehmet Simsar, Emine Bensu Erdemir, Abdullah Alanbay, Anjany Sekuboyina, Berkan Lafci, Christian Bluethgen, Mehmet Kemal Ozdemir, and Bjoern Menze. Developin...

    arXiv 2024

  30. [30]

    CT2Rep : Automated Radiology Report Generation for 3D Medical Imaging , March 2024 b

    Ibrahim Ethem Hamamci, Sezgin Er, and Bjoern Menze. CT2Rep : Automated Radiology Report Generation for 3D Medical Imaging , March 2024 b . arXiv:2403.06801 [cs, eess]

    Pith/arXiv arXiv 2024

  31. [31]

    CRG Score : A Distribution - Aware Clinical Metric for Radiology Report Generation , May 2025

    Ibrahim Ethem Hamamci, Sezgin Er, Suprosanna Shit, Hadrien Reynaud, Bernhard Kainz, and Bjoern Menze. CRG Score : A Distribution - Aware Clinical Metric for Radiology Report Generation , May 2025. arXiv:2505.17167 [cs]

    Pith/arXiv arXiv 2025

  32. [32]

    Deep Residual Learning for Image Recognition , December 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep Residual Learning for Image Recognition , December 2015. arXiv:1512.03385 [cs]

    Pith/arXiv arXiv 2015

  33. [33]

    Ilesanmi, Taiwo O

    Ademola E. Ilesanmi, Taiwo O. Ilesanmi, and Babatunde O. Ajayi. Reviewing 3D convolutional neural network approaches for medical image segmentation. Heliyon, 10 0 (6): 0 e27398, March 2024. ISSN 2405-8440

    2024

  34. [34]

    Mong, Safwan S

    Jeremy Irvin, Pranav Rajpurkar, Michael Ko, Yifan Yu, Silviana Ciurea-Ilcus, Chris Chute, Henrik Marklund, Behzad Haghgoo, Robyn Ball, Katie Shpanskaya, Jayne Seekins, David A. Mong, Safwan S. Halabi, Jesse K. Sandberg, Ricky Jones, David B. Larson, Curtis P. Langlotz, Bhavik N. Patel, Matthew P. Lungren, and Andrew Y. Ng. CheXpert : A Large Chest Radiogr...

    Pith/arXiv arXiv 2019

  35. [35]

    Alistair E. W. Johnson, Tom Pollard, Roger Mark, Seth Berkowitz, and Steven Horng. The MIMIC - CXR Database , 2019

    2019

  36. [36]

    Chaitanya K. Joshi. Transformers are Graph Neural Networks , June 2025. arXiv:2506.22084 [cs]

    Pith/arXiv arXiv 2025

  37. [37]

    John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ronneberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin Žídek, Anna Potapenko, Alex Bridgland, Clemens Meyer, Simon A. A. Kohl, Andrew J. Ballard, Andrew Cowie, Bernardino Romera-Paredes, Stanislav Nikolov, Rishub Jain, Jonas Adler, Trevor Back, Stig Petersen, David Reiman...

    2021

  38. [38]

    BERT -based natural language processing analysis of French CT reports: Application to the measurement of the positivity rate for pulmonary embolism

    Emilien Jupin-Delevaux, Aissam Djahnine, François Talbot, Antoine Richard, Sylvain Gouttard, Adeline Mansuy, Philippe Douek, Salim Si-Mohamed, and Loïc Boussel. BERT -based natural language processing analysis of French CT reports: Application to the measurement of the positivity rate for pulmonary embolism. Research in Diagnostic and Interventional Imagi...

    2023

  39. [39]

    CT - GRAPH : Hierarchical Graph Attention Network for Anatomy - Guided CT Report Generation , August 2025

    Hamza Kalisch, Fabian Hörst, Jens Kleesiek, Ken Herrmann, and Constantin Seibold. CT - GRAPH : Hierarchical Graph Attention Network for Anatomy - Guided CT Report Generation , August 2025. arXiv:2508.05375 [cs]

    Pith/arXiv arXiv 2025

  40. [40]

    Deep Visual - Semantic Alignments for Generating Image Descriptions , April 2015

    Andrej Karpathy and Li Fei-Fei. Deep Visual - Semantic Alignments for Generating Image Descriptions , April 2015. arXiv:1412.2306 [cs]

    Pith/arXiv arXiv 2015

  41. [41]

    Lang, Stefan M

    Johannes Kiechle, Daniel M. Lang, Stefan M. Fischer, Lina Felsner, Jan C. Peeken, and Julia A. Schnabel. Graph Neural Networks : A suitable Alternative to MLPs in Latent 3D Medical Image Classification ?, July 2024. arXiv:2407.17219 [cs]

    Pith/arXiv arXiv 2024

  42. [42]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A Method for Stochastic Optimization , January 2017. arXiv:1412.6980 [cs]

    Pith/arXiv arXiv 2017

  43. [43]

    A Survey on Biomedical Image Captioning , May 2019

    Vasiliki Kougia, John Pavlopoulos, and Ion Androutsopoulos. A Survey on Biomedical Image Captioning , May 2019. arXiv:1905.13302 [cs] version: 1

    Pith/arXiv arXiv 2019

  44. [44]

    Denkowski

    Alon Lavie and Michael J. Denkowski. The Meteor metric for automatic evaluation of machine translation. Machine Translation, 23 0 (2): 0 105--115, September 2009. ISSN 1573-0573

    2009

  45. [45]

    Deep learning

    Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 521 0 (7553): 0 436--444, May 2015. ISSN 0028-0836, 1476-4687

    2015

  46. [46]

    µ2 Tokenizer : Differentiable Multi - Scale Multi - Modal Tokenizer for Radiology Report Generation

    Siyou Li, Pengyao Qin, Huanan Wu, Dong Nie, Arun J Thirunavukarasu, Juntao Yu, and Le Zhang. µ2 Tokenizer : Differentiable Multi - Scale Multi - Modal Tokenizer for Radiology Report Generation . 2025

    2025

  47. [47]

    Exploring and Distilling Posterior and Prior Knowledge for Radiology Report Generation , June 2021 a

    Fenglin Liu, Xian Wu, Shen Ge, Wei Fan, and Yuexian Zou. Exploring and Distilling Posterior and Prior Knowledge for Radiology Report Generation , June 2021 a . arXiv:2106.06963 [cs]

    Pith/arXiv arXiv 2021

  48. [48]

    Enhanced Contrastive Learning with Multi -view Longitudinal Data for Chest X -ray Report Generation , February 2025

    Kang Liu, Zhuoqi Ma, Xiaolu Kang, Yunan Li, Kun Xie, Zhicheng Jiao, and Qiguang Miao. Enhanced Contrastive Learning with Multi -view Longitudinal Data for Chest X -ray Report Generation , February 2025. arXiv:2502.20056 [cs]

    arXiv 2025

  49. [49]

    Swin Transformer : Hierarchical Vision Transformer using Shifted Windows , August 2021 b

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin Transformer : Hierarchical Vision Transformer using Shifted Windows , August 2021 b . arXiv:2103.14030 [cs]

    Pith/arXiv arXiv 2021

  50. [50]

    Video Swin Transformer , June 2021 c

    Ze Liu, Jia Ning, Yue Cao, Yixuan Wei, Zheng Zhang, Stephen Lin, and Han Hu. Video Swin Transformer , June 2021 c . arXiv:2106.13230 [cs]

    Pith/arXiv arXiv 2021

  51. [51]

    U- Mamba : Enhancing Long -range Dependency for Biomedical Image Segmentation , January 2024

    Jun Ma, Feifei Li, and Bo Wang. U- Mamba : Enhancing Long -range Dependency for Biomedical Image Segmentation , January 2024. arXiv:2401.04722 [eess]

    Pith/arXiv arXiv 2024

  52. [52]

    Graph neural network surrogate for strategic transport planning, August 2024

    Nikita Makarov, Santhanakrishnan Narayanan, and Constantinos Antoniou. Graph neural network surrogate for strategic transport planning, August 2024. arXiv:2408.07726 [cs]

    Pith/arXiv arXiv 2024

  53. [53]

    Cross- Entropy Loss Functions : Theoretical Analysis and Applications , June 2023

    Anqi Mao, Mehryar Mohri, and Yutao Zhong. Cross- Entropy Loss Functions : Theoretical Analysis and Applications , June 2023. arXiv:2304.07288 [cs]

    Pith/arXiv arXiv 2023

  54. [54]

    Self- Supervised Learning of Pretext - Invariant Representations , December 2019

    Ishan Misra and Laurens van der Maaten. Self- Supervised Learning of Pretext - Invariant Representations , December 2019. arXiv:1912.01991 [cs]

    Pith/arXiv arXiv 2019

  55. [55]

    Hamilton, Jan Eric Lenssen, Gaurav Rattan, and Martin Grohe

    Christopher Morris, Martin Ritzert, Matthias Fey, William L. Hamilton, Jan Eric Lenssen, Gaurav Rattan, and Martin Grohe. Weisfeiler and Leman Go Neural : Higher -order Graph Neural Networks , November 2021. arXiv:1810.02244 [cs]

    Pith/arXiv arXiv 2021

  56. [56]

    Redefining Radiology : A Review of Artificial Intelligence Integration in Medical Imaging

    Reabal Najjar. Redefining Radiology : A Review of Artificial Intelligence Integration in Medical Imaging . Diagnostics (Basel, Switzerland), 13 0 (17): 0 2760, August 2023. ISSN 2075-4418

    2023

  57. [57]

    Gupta, Mohammad Khubeb Siddiqui, Ruben Morales-Menendez, Prakhar Bhardwaj, and Vaishnavi Singh

    Harsh Panwar, P.K. Gupta, Mohammad Khubeb Siddiqui, Ruben Morales-Menendez, Prakhar Bhardwaj, and Vaishnavi Singh. A deep learning and grad- CAM based color visualization approach for fast detection of COVID -19 cases using chest X -ray and CT - Scan images. Chaos, Solitons, and Fractals, 140: 0 110190, November 2020. ISSN 0960-0779

    2020

  58. [58]

    Bleu: a Method for Automatic Evaluation of Machine Translation

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a Method for Automatic Evaluation of Machine Translation . In Pierre Isabelle, Eugene Charniak, and Dekang Lin, editors, Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics , pages 311--318, Philadelphia, Pennsylvania, USA, July 2002. Association for Com...

    2002

  59. [59]

    Patel and Orlando De Jesus

    Paula R. Patel and Orlando De Jesus. CT Scan . In StatPearls . StatPearls Publishing, Treasure Island (FL), 2024

    2024

  60. [60]

    A scoping review of large language model based approaches for information extraction from radiology reports npj Digital Medicine , 2025

    Daniel Reichenpfader, Henning Muller, and Kerstin Denecke. A scoping review of large language model based approaches for information extraction from radiology reports npj Digital Medicine , 2025

    2025

  61. [61]

    Graph neural networks for materials science and chemistry

    Patrick Reiser, Marlen Neubert, André Eberhard, Luca Torresi, Chen Zhou, Chen Shao, Houssam Metni, Clint van Hoesel, Henrik Schopmans, Timo Sommer, and Pascal Friederich. Graph neural networks for materials science and chemistry. Communications Materials, 3 0 (1): 0 1--18, November 2022. ISSN 2662-4443. Publisher: Nature Publishing Group

    2022

  62. [62]

    Amanda Ross and Victor L. Willson. Paired Samples T - Test . In Amanda Ross and Victor L. Willson, editors, Basic and Advanced Statistical Tests : Writing Results Sections and Creating Tables and Figures , pages 17--19. SensePublishers, Rotterdam, 2017. ISBN 978-94-6351-086-8. doi:10.1007/978-94-6351-086-8_4

    doi:10.1007/978-94-6351-086-8_4 2017

  63. [63]

    Berg, and Li Fei-Fei

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-Fei. ImageNet Large Scale Visual Recognition Challenge , January 2015. arXiv:1409.0575 [cs]

    Pith/arXiv arXiv 2015

  64. [64]

    Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra

    Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad- CAM : Visual Explanations from Deep Networks via Gradient -based Localization , December 2019. arXiv:1610.02391

    Pith/arXiv arXiv 2019

  65. [65]

    GLU Variants Improve Transformer , February 2020

    Noam Shazeer. GLU Variants Improve Transformer , February 2020. arXiv:2002.05202 [cs]

    Pith/arXiv arXiv 2020

  66. [66]

    M. Stone. Cross- Validatory Choice and Assessment of Statistical Predictions . Journal of the Royal Statistical Society. Series B (Methodological), 36 0 (2): 0 111--147, 1974. ISSN 0035-9246. Publisher: [Royal Statistical Society, Oxford University Press]

    1974

  67. [67]

    Sudre, M

    Carole H. Sudre, M. Jorge Cardoso, Sebastien Ourselin, and Alzheimer’s Disease Neuroimaging Initiative . Longitudinal segmentation of age-related white matter hyperintensities. Medical Image Analysis, 38: 0 50--64, May 2017. ISSN 1361-8423

    2017

  68. [68]

    Self- Supervised Pre - Training of Swin Transformers for 3D Medical Image Analysis , March 2022

    Yucheng Tang, Dong Yang, Wenqi Li, Holger Roth, Bennett Landman, Daguang Xu, Vishwesh Nath, and Ali Hatamizadeh. Self- Supervised Pre - Training of Swin Transformers for 3D Medical Image Analysis , March 2022. arXiv:2111.14791 [cs]

    Pith/arXiv arXiv 2022

  69. [69]

    Interactive and Explainable Region -guided Radiology Report Generation

    Tim Tanida, Philip Müller, Georgios Kaissis, and Daniel Rueckert. Interactive and Explainable Region -guided Radiology Report Generation . In 2023 IEEE / CVF Conference on Computer Vision and Pattern Recognition ( CVPR ) , pages 7433--7442, June 2023. arXiv:2304.08295 [cs]

    Pith/arXiv arXiv 2023

  70. [70]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention Is All You Need , August 2023. arXiv:1706.03762 [cs]

    Pith/arXiv arXiv 2023

  71. [71]

    Everything is Connected : Graph Neural Networks

    Petar Veličković. Everything is Connected : Graph Neural Networks . Current Opinion in Structural Biology, 79: 0 102538, April 2023. ISSN 0959440X. arXiv:2301.08210 [cs]

    Pith/arXiv arXiv 2023

  72. [72]

    Graph Attention Networks , February 2018

    Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph Attention Networks , February 2018. arXiv:1710.10903 [stat]

    Pith/arXiv arXiv 2018

  73. [73]

    Show and Tell : A Neural Image Caption Generator , April 2015

    Oriol Vinyals, Alexander Toshev, Samy Bengio, and Dumitru Erhan. Show and Tell : A Neural Image Caption Generator , April 2015. arXiv:1411.4555 [cs]

    Pith/arXiv arXiv 2015

  74. [74]

    A Review on 3D Convolutional Neural Network

    Chuqi Wang. A Review on 3D Convolutional Neural Network . In 2023 IEEE 3rd International Conference on Power , Electronics and Computer Applications ( ICPECA ) , pages 1204--1208, January 2023

    2023

  75. [75]

    Chang, Amilcare Gentili, and Chun-Nan Hsu

    An Yan, Julian McAuley, Xing Lu, Jiang Du, Eric Y. Chang, Amilcare Gentili, and Chun-Nan Hsu. RadBERT : Adapting Transformer -based Language Models to Radiology . Radiology: Artificial Intelligence, 4 0 (4): 0 e210258, July 2022. Publisher: Radiological Society of North America

    2022

  76. [76]

    Swin3D : A Pretrained Transformer Backbone for 3D Indoor Scene Understanding , August 2023

    Yu-Qi Yang, Yu-Xiao Guo, Jian-Yu Xiong, Yang Liu, Hao Pan, Peng-Shuai Wang, Xin Tong, and Baining Guo. Swin3D : A Pretrained Transformer Backbone for 3D Indoor Scene Understanding , August 2023. arXiv:2304.06906 [cs]

    Pith/arXiv arXiv 2023

  77. [77]

    MagNet : A Neural Network for Directed Graphs , June 2021

    Xitong Zhang, Yixuan He, Nathan Brugnone, Michael Perlmutter, and Matthew Hirn. MagNet : A Neural Network for Directed Graphs , June 2021. arXiv:2102.11391 [cs]

    Pith/arXiv arXiv 2021

  78. [78]

    Lungren, and Serena Yeung

    Yuhui Zhang, Shih-Cheng Huang, Zhengping Zhou, Matthew P. Lungren, and Serena Yeung. Adapting Pre -trained Vision Transformers from 2D to 3D through Weight Inflation Improves Medical Image Segmentation , February 2023. arXiv:2302.04303 [cs]

    Pith/arXiv arXiv 2023

  79. [79]

    Graph neural networks: A review of methods and applications

    Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang, Changcheng Li, and Maosong Sun. Graph neural networks: A review of methods and applications. AI Open, 1: 0 57--81, January 2020. ISSN 2666-6510

    2020