Pith. sign in

REVIEW 3 major objections 5 minor 38 references

GAN-based synthetic FDG PET images from T1 brain MRI can serve to improve performance of deep unsupervised anomaly detection models

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Synthetic FDG PET images generated from T1 MRI by GANs can replace real PET scans in the normative database used to train an unsupervised detector of epilepsy lesions.

desk verdict Useful proof-of-concept, but the headline comparison (synthetic vs real PET for UAD training) is confounded by switching normative datasets, so the 74% vs 42% result does not support the replacement claim as stated. read the letter →

arxiv 2505.07364 v1 pith:CCFF2FXU submitted 2025-05-12 eess.IV cs.AI

classification eess.IVcs.AI
keywords MedicalimagesynthesisCycle-GANPETMRIUnsupervisedanomalydetectionout-of-distribution(OOD)epileptogeniczone
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper's claim is that synthetic FDG PET images, generated from T1 MRI by GAN-based architectures, can replace true FDG PET images in a multimodal normative database used to train an unsupervised anomaly detection model. It reports that the best detector trained on synthetic PET from a 3D-patch Cycle-GAN finds 14 of 19 subtle epilepsy lesions (74% sensitivity, mean rank 2.1) among 17 surgical patients, versus 8 of 19 (42%, mean rank 3.9) for the same detector trained on real PET. The practical stake is direct: real FDG PET control databases are costly and hard to collect, while T1 MRI is routine, so synthesizing the missing modality could make PET-based anomaly detection feasible where it currently is not. The paper also introduces task-oriented metrics, reconstruction error and Mahalanobis distance in latent space, to test whether synthetic data behave as in-distribution for the detector.

What carries the argument

The machinery is a Cycle-GAN for T1-to-PET translation, augmented with a mean-squared-error loss between paired real and synthetic PET, and configured either as 2.5D slice triplets or 3D patches; the best version is the 3D-patch model, followed by Gaussian smoothing and histogram matching of the reconstructed volumes. The downstream UAD model is a siamese convolutional autoencoder whose latent vectors feed a per-voxel one-class SVM, producing ranked anomaly clusters. The paper's new quality metrics, global reconstruction MSE and Mahalanobis distance in the autoencoder latent space, check whether fake PET behaves as in-distribution for the detector before the detection experiment is run.

What would settle it

Take the same UAD pipeline and the same 17-patient test set, but train the detector on a larger real-PET normative database that is scanner-matched to the patients; if its sensitivity reaches or exceeds 74%, the synthetic advantage is not caused by better normative fidelity. Independently, recompute detection counts using only the top-1 or top-3 clusters instead of the top-10 rule described in Appendix C; if the synthetic-trained sensitivity drops below the real-trained 42%, the reported gain is an artifact of the cluster-counting protocol.

Watch

Extended reading notes

Core claim

The central claim, stated directly in the conclusion, is that synthetic FDG PET data generated from T1 MRI can efficiently replace true FDG PET images in normative databases for training unsupervised anomaly detection models. The paper supports this with three linked results: a Cycle-GAN with an added mean-squared-error loss, trained on paired T1 and PET data, produces synthetic PET with SSIM around 0.9 and PSNR around 23.8; the synthetic PET falls in-distribution relative to real PET under the proposed OOD metrics; and the UAD model trained on real T1 plus synthetic 3D-patch PET detects more subtle epileptogenic lesions than the same model trained on real PET. The authors caution that their explanation for the improvement, a denser synthetic latent distribution due to lower inter-individual variability, should be interpreted with caution.

Load-bearing premise

The load-bearing premise is that the synthetic-trained detector's higher sensitivity reflects a genuinely useful denser normative latent distribution, not an artifact of reduced inter-individual variability that happens to produce more or smaller candidate clusters on this particular 17-patient cohort.

Editorial extensions

If this is right

  • A center with only T1 MRI of healthy subjects could build a PET-trained unsupervised anomaly detector without collecting a real FDG PET control database.
  • The 3D-patch Cycle-GAN with MSE loss is the recommended synthesis configuration, giving the best visual metrics and the best downstream detection performance.
  • The proposed OOD metrics offer a fast pre-screening check for whether a new patient's multimodal images are in-distribution for a trained UAD model.
  • For localizing subtle epileptogenic zones, training on synthetic rather than real PET is reported to raise sensitivity from 42% to 74% in this 17-patient cohort.

Reading between the lines

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

  • Because the authors themselves flag the denser-latent-distribution explanation as tentative, a direct test would be to measure the covariance of latent vectors for synthetic versus real control PET: a substantially narrower synthetic distribution would predict that the detection gain shrinks on more heterogeneous patient data.
  • The top-ten cluster counting rule makes a hit easy, since a lesion counts as detected if it appears anywhere among the ten most suspicious clusters; recomputing the comparison with only the top-ranked cluster would show whether the synthetic-trained advantage reflects sharper localization or simply more candidate clusters.
  • If the effect generalizes, the same replacement strategy could apply to other radiotracers or modalities where normative databases are scarce, such as amyloid or dopamine-transporter PET, provided a paired MRI-based synthesis model can be trained.
  • A natural next experiment is mixing real and synthetic PET in the normative database; if detection improves monotonically with the amount of real data, the benefit is likely due to data scale rather than something unique to synthetic images.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript proposes to synthesize FDG PET images from T1-weighted MRI using GAN-based architectures, with the stated goal of using such synthetic PET data as normative training data for an unsupervised anomaly detection (UAD) model for epileptogenic zone screening. The authors compare several GAN configurations (2.5D vs 3D-patch, simple GAN vs Cycle-GAN, with and without an additional MSE loss), evaluate visual quality using PSNR/SSIM/LPIPS and task-oriented out-of-distribution metrics, and then train a siamese-autoencoder/OC-SVM UAD model on either 35 real paired T1/PET exams (UAD1) or on 35 T1 exams from a different control dataset paired with synthetic PET generated from those T1 scans (UAD2/UAD3). On 17 epilepsy patients with 19 subtle lesions, the model trained on 3D-patch synthetic PET achieves 74% sensitivity (14/19) versus 42% (8/19) for the model trained on real PET, with mean ranks of 2.1 and 3.9 respectively. The authors conclude that synthetic PET could efficiently replace true PET in multi-modality normative databases for training UAD models.

Significance. If the central claim were established, this would be a valuable contribution: it would relax a significant practical bottleneck for UAD training, namely the difficulty of acquiring normative PET databases, and it would extend prior work by evaluating synthetic data on a concrete downstream detection task rather than only with visual metrics. The paper has clear strengths: the code and normative image dataset are released, the visual-quality comparison includes formal statistical tests (Wilcoxon signed-rank), and the clinical evaluation targets a well-defined, difficult problem (MRI-negative epilepsy with subtle hypometabolism). The task-oriented OOD metrics are a reasonable attempt to bridge visual quality and utility, although their self-referential nature limits their probative value. However, the central claim—that synthetic PET can replace true PET—is supported by a comparison that is confounded by the training population, and the effect size is not statistically established on the small lesion cohort. Because these issues bear directly on the main conclusion, the current evidence supports a more cautious claim than the one stated in the Conclusion.

major comments (3)
  1. [Section 4.2 and Table 1] The comparison between UAD1 (trained on real PET) and UAD2/UAD3 (trained on synthetic PET) is confounded: UAD1 is trained on the 35 true T1/PET exams of DBC1, while UAD2 and UAD3 are trained on 35 T1 exams of DBC2 paired with synthetic PET generated from those T1 scans. The observed 74% vs 42% difference in Table 3 therefore conflates the effect of synthetic PET with differences in the normative T1 training population (DBC2 vs DBC1) and any differences in training-set composition. The paper calls this a 'fair comparison' in Section 4.2, but fairness would require true and synthetic PET for the same normative subjects, e.g., by generating synthetic PET from DBC1 T1 scans (the GANs are trained on DBC1) or by obtaining true PET for DBC2. This confound is load-bearing for the replacement claim and needs to be resolved with a within-dataset control.
  2. [Table 3 and Section 5.3] The key result, 14/19 vs 8/19 detections, is reported without confidence intervals or a significance test. A McNemar exact test on the discordant pairs from Table 3 yields p≈0.07, so even under the current (confounded) design the superiority of the synthetic-trained model is not statistically established. The authors should report an appropriate paired test or a bootstrap interval, and should frame the 74% vs 42% difference as an exploratory finding given the small lesion count.
  3. [Section 6 and Appendix C] The explanation for the improvement—that synthetic PET induces a 'denser latent distribution'—is acknowledged by the authors as speculative, but the evaluation protocol itself may contribute to the gap: clusters are thresholded per score map to yield at most 10 clusters (Appendix C), and a lesion counts as detected if it lies in any of the top-10 clusters. If the synthetic-trained models produce more or smaller clusters on this cohort, the sensitivity difference could be inflated. The paper should report the number of clusters per patient per model and evaluate sensitivity to the cluster threshold; without this, the 74% vs 42% difference cannot be confidently attributed to a genuinely more useful normative representation.
minor comments (5)
  1. [Section 3.1.3] The Gaussian smoothing FWHM (1.5 mm) and the histogram-matching reference image are selected on the validation set; a brief sensitivity analysis or justification would clarify how dependent the reported visual metrics are on these post-processing choices.
  2. [Section 5.3] In the comparison sentence, 'the real paired T1 MRI and FDG PET of D_B3Dpatch_C1' should read 'of DBC1'; the subscript appears to be a typo.
  3. [Figure 5] The scatter plots are dense and the three groups overlap; consider plotting marginal distributions or increasing transparency, and state the number of points per group in the caption.
  4. [Appendix C] The cluster-ranking criterion uses equal weights for cluster size and average score, and the OC-SVM RBF kernel is used without a reported hyperparameter search; a brief description of how these choices were made, or a sensitivity check, would improve reproducibility.
  5. [Introduction and Conclusion] The claim that GANs 'outperform transformer or diffusion models' rests on the cited review [6] and not on experiments in this paper; this limitation is acknowledged in the Discussion but could be stated more carefully in the Conclusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the UAD comparison is an external evaluation on patient data; the training-set difference is a validity confound, not a circular reduction.

full rationale

The central claim—that synthetic FDG PET generated from T1 MRI can replace true PET in a normative database for training a UAD model—is tested on the 17 epilepsy-patient exams of DBCep (plus patient R), which are not used to train either the Cycle-GAN or the UAD models. UAD1 is trained on true T1+PET pairs from DBC1; UAD2 and UAD3 are trained on DBC2 T1 images paired with synthetic PET produced by a GAN trained on DBC1. The Table 3 sensitivities (8/19, 11/19, 14/19) are empirical outcomes on held-out patients, not quantities reconstructed from the training data by construction. The task-oriented OOD metrics in Section 3.3 and Figure 5 are computed with the same UAD architecture that performs detection and are therefore model-relative; however, they are not fitted to the detection result and are presented as a proxy, with the patient-level detection comparison providing an independent check. References [1] and [37] are self-citations, but they are used as method sources (the siamese-AE plus OC-SVM architecture and prior synthesis pipeline) rather than as theorems that force the paper's conclusion, so they are not load-bearing in a circular sense. The paper itself cautions that the 'denser latent distribution' explanation is speculative and that the cohort is small. The main validity concern is experimental rather than circular: UAD1 trains on DBC1 while UAD2 and UAD3 train on DBC2, so the 'synthetic vs. real' comparison is entangled with the control-population difference; this affects interpretation and generalizability but does not make the reported prediction equivalent to its inputs.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim does not introduce new physical entities or new laws. It depends on standard deep learning training assumptions and on several explicitly chosen post-processing and evaluation hyperparameters. The most consequential domain assumption is that T1 MRI determines the FDG PET signal closely enough for the synthetic distribution to be a valid normative distribution for anomaly detection. Another is that 35 training subjects suffice for per-voxel OC-SVM support estimation in a 64-dimensional latent space. These assumptions, not any fitted physical constant, carry the paper's conclusions.

free parameters (6)
  • Gaussian smoothing FWHM for synthetic PET post-processing = 1.5 mm
    Chosen among 0-3 mm to produce the best SSIM on validation data (Section 4.3). Affects texture and border artifacts of synthetic PET, hence OOD metrics and downstream detection.
  • Histogram matching reference PET image = Randomly chosen PET image from GAN training database
    Section 3.1.3: intensity distribution of each fake PET is adjusted to a randomly chosen reference PET. The arbitrary reference influences absolute intensities and can affect the autoencoder and OC-SVM.
  • UAD cluster threshold = Set to yield at most n=10 detected clusters per patient
    Appendix C. Directly defines the sensitivity metric: any of the top 10 clusters overlapping ground truth counts as a detection. A more permissive threshold increases apparent sensitivity.
  • Cluster ranking criterion weights = Equal contribution of cluster size and score
    Appendix C states 'equal relative contribution of the cluster size and score in computation of the ranking criterion.' This weighting determines which clusters appear in the top 10 and thus which detections are counted.
  • OC-SVM RBF kernel hyperparameters
    The per-voxel OC-SVM models use an RBF kernel, but the kernel parameters (e.g., nu/gamma) are not reported in the manuscript. They control the decision boundary and hence the anomaly score maps.
  • Training/validation/test split for final GAN retraining = 29 training, 3 validation, 3 test controls
    Section 4.2. The 3 test control subjects provide a very small sample for assessing visual quality and OOD behavior of the selected GAN models.
assumptions (6)
  • domain assumption FDG PET patterns of healthy subjects are sufficiently determined by T1-weighted MRI anatomy for the purpose of training an anomaly detector.
    The entire synthesis approach assumes T1 carries the needed metabolic information. If PET reflects metabolic variability not visible in T1, the synthetic distribution will be artificially narrow, which is exactly the denser-distribution hypothesis discussed in Section 6.
  • domain assumption The GAN trained on DBC1 (35 paired exams) generalizes to T1 images of DBC2 and patient scans acquired on the same scanner with the same protocol.
    Section 4.1: all databases were acquired on the same 1.5T Sonata and mCT scanners. The homogeneous acquisition is acknowledged as a limitation in Section 6; across-scanner generalization is not established.
  • domain assumption Per-voxel OC-SVM support estimation in a 64-dimensional latent space is reliable with 35 training subjects.
    Appendix C trains one OC-SVM per voxel on latent vectors from all training subjects. High-dimensional support estimation with 35 points is statistically fragile.
  • domain assumption The Mahalanobis distance and reconstruction error computed with the UAD model trained on synthetic data are valid proxies for whether a new sample is in-distribution for the detection task.
    Sections 3.3 and 4.5. This assumption underlies the OOD metrics; it uses the same model that is later evaluated for detection, making the check self-referential rather than an independent benchmark.
  • domain assumption The Cycle-GAN with auxiliary MSE loss converges to a useful mapping with the described architectures and hyperparameters.
    GAN training is not guaranteed to converge; the paper relies on validation SSIM for model selection, which may not reflect downstream task performance.
  • standard math The empirical covariance matrix in the Mahalanobis distance is invertible and well-estimated from the training data.
    Equations (4)-(5) use the empirical covariance of latent vectors. With 35 samples and a 64-dimensional latent space, the covariance may be ill-conditioned; the paper does not report regularization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GAN-based synthetic FDG PET images from T1 brain MRI can serve to improve performance of deep unsupervised anomaly detection models." pith.science (2026). https://pith.science/paper/CCFF2FXU

@misc{pith2026250507364,
  author       = {Pith},
  title        = {Pith review of: GAN-based synthetic FDG PET images from T1 brain MRI can serve to improve performance of deep unsupervised anomaly detection models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CCFF2FXU}},
  note         = {Machine review of arXiv:2505.07364}
}
read the original abstract

Background and Objective. Research in the cross-modal medical image translation domain has been very productive over the past few years in tackling the scarce availability of large curated multimodality datasets with the promising performance of GAN-based architectures. However, only a few of these studies assessed task-based related performance of these synthetic data, especially for the training of deep models. Method. We design and compare different GAN-based frameworks for generating synthetic brain [18F]fluorodeoxyglucose (FDG) PET images from T1 weighted MRI data. We first perform standard qualitative and quantitative visual quality evaluation. Then, we explore further impact of using these fake PET data in the training of a deep unsupervised anomaly detection (UAD) model designed to detect subtle epilepsy lesions in T1 MRI and FDG PET images. We introduce novel diagnostic task-oriented quality metrics of the synthetic FDG PET data tailored to our unsupervised detection task, then use these fake data to train a use case UAD model combining a deep representation learning based on siamese autoencoders with a OC-SVM density support estimation model. This model is trained on normal subjects only and allows the detection of any variation from the pattern of the normal population. We compare the detection performance of models trained on 35 paired real MR T1 of normal subjects paired either on 35 true PET images or on 35 synthetic PET images generated from the best performing generative models. Performance analysis is conducted on 17 exams of epilepsy patients undergoing surgery. Results. The best performing GAN-based models allow generating realistic fake PET images of control subject with SSIM and PSNR values around 0.9 and 23.8, respectively and in distribution (ID) with regard to the true control dataset. The best UAD model trained on these synthetic normative PET data allows reaching 74% sensitivity. Conclusion. Our results confirm that GAN-based models are the best suited for MR T1 to FDG PET translation, outperforming transformer or diffusion models. We also demonstrate the diagnostic value of these synthetic data for the training of UAD models and evaluation on clinical exams of epilepsy patients. Our code and the normative image dataset are available.

Figures

Figures reproduced from arXiv: 2505.07364 by the authors.

Figure 1
Figure 1. Graphical abstract depicting the main steps of the conducted analysis. The main contributions of this paper, depicted on [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Cycle-GAN architecture based on two baseline GANs translating images from domain A to domain B (upper GAN) and vice versa (lower GAN). where 𝑦 ′ 𝑎 is the fake image of domain A generated by generator 𝐺𝐴 from the fake 𝑥𝑏 , that is 𝑦 ′ 𝑎 = 𝐺𝐴(𝑥𝑏 ) with 𝑥𝑏 = 𝐺𝐵(𝑦𝑎 ). 3.1.2. Adapted loss term for supervised image translation CycleGAN-models have been proposed and shown effi￾cient when trained on unpaired modalities, e.g… view at source ↗
Figure 3
Figure 3. Scheme of brain anomaly detection model proposed in [1]. The original model was trained to detect epilepsy lesion from T1 and FLAIR images. In this study, it is adapted to perform the detection task based on T1 and [18F]FDG PET images. vectors that are then fed into OC-SVM models [23]. We train one OC-SVM model per voxel on the matrix composed of the latent representations of this voxel extracted from all subjects f… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Qualitative result on one validation control subject of 𝐷𝐵𝐶1 . First left column: original T1 MRI, then, from left to right: original FDG PET image, synthetic PET images generated with 2.5D and 3D-patch Cycle-GAN models with MSE loss. was achieved with the 𝑈𝐴𝐷3 model t…
Figure 5
Figure 5. Figure 5: Mahalanobis distance 𝐷𝑚 and reconstruction error 𝑀𝑆𝐸 on test subjects inputted to the 𝑈𝐴𝐷 model of [1]. Blue points correspond to the 35 real controls from 𝐷𝐵𝐶1 , purple squares to the 18 patients of 𝐷𝐵𝑒𝑝. a) Results for 𝑈𝐴𝐷2 trained on 𝐷𝐵2.5𝐷 𝐶2 , red triangles corres…
Figure 6
Figure 6. Figure 6: Example cluster maps overlaid on T1 MRI by the detection models, from top to bottom: 𝑈𝐴𝐷1 , 𝑈𝐴𝐷2 and 𝑈𝐴𝐷3 , respectively. Selected transverse or coronal slices of patients 𝐻, 𝑀, and 𝑂 are centered on confirmed EZ localisations in various areas of the brain, namely the …
Figure 7
Figure 7. Figure 7: ResNet architecture for 2.5D and 3D-patch generators 4x4 Conv, stride 2 Leaky ReLU Batch Norm Leaky ReLU Batch Norm Leaky ReLU Batch Norm Leaky ReLU 3x128x128 64x64x64 128x32x32 256x16x16 512x15x15 1x14x14 2.5D 3D-patch 1x32x32x32 64x16x16x16 128x8x8x8 256x4x4x4 512x3x…
Figure 8
Figure 8. Figure 8: PatchGAN architecture for 2.5D and 3D-patch discriminators is assigned the signed score output by the classifier. This yields a distance map for the given patient, which is then thresholded at some pre-chosen value and a 26-connectivity rule is applied to identify the …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 24 canonical work pages

  1. [1]

    Regularized siamese neural network for unsupervised outlier detection on brain multiparametricmagneticresonanceimaging:applicationtoepilepsy lesion screening

    Alaverdyan, Z., Jung, J., Bouet, R., Lartizien, C., 2020. Regularized siamese neural network for unsupervised outlier detection on brain multiparametricmagneticresonanceimaging:applicationtoepilepsy lesion screening. Medical Image Analysis 60, 101618

  2. [2]

    Unsupervised medical image translation using cycle-medgan,in:201927thEuropeanSignalProcessingConference (EUSIPCO), IEEE

    Armanious, K., Jiang, C., Abdulatif, S., Küstner, T., Gatidis, S., Yang, B., 2019. Unsupervised medical image translation using cycle-medgan,in:201927thEuropeanSignalProcessingConference (EUSIPCO), IEEE. pp. 1–5

  3. [3]

    Autoencoders for unsupervised anomaly segmentation in brain MR images:Acomparativestudy

    Baur, C., Denner, S., Wiestler, B., Navab, N., Albarqouni, S., 2021. Autoencoders for unsupervised anomaly segmentation in brain MR images:Acomparativestudy. MedicalImageAnalysis69. doi: https: //doi.org/10.1016/j.media.2020.101952

  4. [4]

    Journal of Medical Imaging 8

    Burgos, N., Cardoso, M., Samper-González, J., Habert, M., Dur- rleman, S., Ourselin, S., Colliot, O., Initiative, A.D.N., Initiative, F.L.D.N.,2021.Anomalydetectionfortheindividualanalysisofbrain pet images. Journal of Medical Imaging 8. doi:10.1117/1.JMI.8.2. 024003

  5. [5]

    Resvit:Residualvisiontrans- formers for multimodal medical image synthesis

    Dalmaz,O.,Yurt,M.,Çukur,T.,2022. Resvit:Residualvisiontrans- formers for multimodal medical image synthesis. IEEE Transactions on Medical Imaging 41, 2598–2614. doi:10.1109/TMI.2022.3167808

  6. [6]

    Deep learning based synthesis of mri, ct and pet: Review and analysis

    Dayarathna, S., Islam, K.T., Uribe, S., Yang, G., Hayat, M., Chen, Z., 2024. Deep learning based synthesis of mri, ct and pet: Review and analysis. Medical Image Analysis 92, 103046. URL: https: //www.sciencedirect.com/science/article/pii/S1361841523003067, doi:https://doi.org/10.1016/j.media.2023.103046

  7. [7]

    Deep-learning predicted pet can be subtracted from the true clinical fluorodeoxyglucose pet co-registered to mri to identify the epileptogenic focus in focal epilepsy

    Flaus, A., Jung, J., Ostrowky-Coste, K., Sylvain, R., Guénot, M., Bouvard, S., Janier, M., Yaakub, S.N., Lartizien, C., Costes, N., Hammers, A., 2023. Deep-learning predicted pet can be subtracted from the true clinical fluorodeoxyglucose pet co-registered to mri to identify the epileptogenic focus in focal epilepsy. Epilepsia Open 00, 1–12. doi:10.1002/e...

  8. [8]

    Distance-based detection of out-of-distribution silent failures for covid-19 lung lesion segmentation

    González, C., Gotkowski, K., Fuchs, M., Bucher, A., Dadras, A., Fischbach, R., Kaltenborn, I.J., Mukhopadhyay, A., 2022. Distance-based detection of out-of-distribution silent failures for covid-19 lung lesion segmentation. Medical Image Analysis 82, 102596. URL: https://www.sciencedirect.com/science/article/ pii/S1361841522002298, doi: https://doi.org/10...

Show all 38 references
  1. [9]

    Generative adversarial nets, in: Advances in neural information processing systems, pp

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y., 2014. Generative adversarial nets, in: Advances in neural information processing systems, pp. 2672–2680

  2. [10]

    Three-dimensional maximum probability atlas of the human brain, with particular reference to the temporal lobe

    Hammers, A., Allom, R., Koepp, M.J., Free, S.L., Myers, R., Lemieux, L., Mitchell, T.N., Brooks, D.J., Duncan, J.S., 2003. Three-dimensional maximum probability atlas of the human brain, with particular reference to the temporal lobe. Human Brain Mapping 19, 224–247. URL: http...

  3. [11]

    Tiling and stitching segmentation output for remote sensing: Basic challenges and recommendations

    Huang, B., Reichman, D., Collins, L.M., Bradbury, K., Malof, J.M., 2018. Tiling and stitching segmentation output for remote sensing: Basic challenges and recommendations. arXiv preprint arXiv:1805.12219

  4. [12]

    One model to synthesize them all: Multi-contrast multi- scale transformer for missing data imputation

    Liu, J., Pasumarthi, S., Duffy, B., Gong, E., Datta, K., Zaharchuk, G., 2023. One model to synthesize them all: Multi-contrast multi- scale transformer for missing data imputation. IEEE Transactions on Medical Imaging 42, 2577–2591. doi:10.1109/TMI.2023.3261707

  5. [13]

    Least squares generative adversarial networks, in: Proceedings of the IEEE conference on computer vision (ICCV), pp

    Mao, X., Li, Q., Xie, H., Lau, R.Y.K., Wang, Z., Smolley, S.P., 2017. Least squares generative adversarial networks, in: Proceedings of the IEEE conference on computer vision (ICCV), pp. 2813–2821

  6. [14]

    On the pitfalls of using the residual as anomaly score, in: Medical Imaging with Deep Learning (MIDL), 2022 International Conference on

    Meissen, F., Wiestler, B., Kaissis, G., Rueckert, D., 2022. On the pitfalls of using the residual as anomaly score, in: Medical Imaging with Deep Learning (MIDL), 2022 International Conference on. URL: https://openreview.net/forum?id=ZsoHLeupa1D

  7. [15]

    Cermep- idb-mrxfdg: a database of 37 normal adult human brain [18f]fdg pet, t1 and flair mri, and ct images available for research

    Mérida, I., Jung, J., Bouvard, S., Le Bars, D., Lancelot, S., Lavenne, F.,Bouillot,C.,Redouté,J.,Hammers,A.,Costes,N.,2021. Cermep- idb-mrxfdg: a database of 37 normal adult human brain [18f]fdg pet, t1 and flair mri, and ct images available for research. EJNMMI Research 11. d...

  8. [16]

    Muñoz-Ramírez, V., Pinon, N., Forbes, F., Lartizen, C., Dojat, M.,

  9. [17]

    Pan, Y., Chen, Y., Shen, D., Xia, Y., 2021. Collaborative image synthesisanddiseasediagnosisforclassificationofneurodegenerative disorders with incomplete multi-modal neuroimages, in: de Bruijne, M.,Cattin,P.C.,Cotin,S.,Padoy,N.,Speidel,S.,Zheng,Y.,Essert,C. (Eds.), Medical Im...

  10. [18]

    Spatially- constrained fisher representation for brain disease identification with incompletemulti-modalneuroimages

    Pan, Y., Liu, M., Lian, C., Xia, Y., Shen, D., 2020. Spatially- constrained fisher representation for brain disease identification with incompletemulti-modalneuroimages. IEEETransactionsonMedical Imaging 39, 2965–2975. doi:10.1109/TMI.2020.2983085

  11. [19]

    Disease-image-specific learningfordiagnosis-orientedneuroimagesynthesiswithincomplete multi-modalitydata

    Pan, Y., Liu, M., Xia, Y., Shen, D., 2022. Disease-image-specific learningfordiagnosis-orientedneuroimagesynthesiswithincomplete multi-modalitydata. IEEETransactionsonPatternAnalysisandMa- chine Intelligence 44, 6839–6853. doi:10.1109/TPAMI.2021.3091214

  12. [20]

    Unsupervisedbrainimaging3danomalyde- tectionandsegmentationwithtransformers

    Pinaya,W.H.,Tudosiu,P.D.,Gray,R.,Rees,G.,Nachev,P.,Ourselin, S.,Cardoso,M.J.,2022. Unsupervisedbrainimaging3danomalyde- tectionandsegmentationwithtransformers. MedicalImageAnalysis 79

  13. [21]

    Pinon, N., Oudoumanessah, G., Trombetta, R., Dojat, M., Forbes, F., Lartizien, C., 2023a. Brain subtle anomaly detection based on auto-encoderslatentspaceanalysis:applicationtodenovoparkinson patients, in: ISBI 2023 - IEEE 20th International Symposium on Biomedical Imaging, IE...

  14. [22]

    Pinon, N., Trombetta, R., Lartizien, C., 2023b. One-class svm on siamese neural network latent space for unsupervised anomaly detection on brain mri white matter hyperintensities, in: Medical ImagingwithDeepLearning(MIDL),2023InternationalConference on

  15. [23]

    Neural computation 13, 1443–1471

    Schölkopf, B., Platt, J.C., Shawe-Taylor, J., Smola, A.J., Williamson, R.C.,2001.Estimatingthesupportofahigh-dimensionaldistribution. Neural computation 13, 1443–1471

  16. [24]

    Sikka,A.,Peri,S.V.,Bathula,D.R.,2018.Mritofdg-pet:cross-modal synthesis using 3d u-net for multi-modal alzheimer’s classification, in: International Workshop on Simulation and Synthesis in Medical Imaging, Springer. pp. 80–89

  17. [25]

    Mri to pet cross-modality translation using globally and locally aware gan (gla- gan) for multi-modal diagnosis of alzheimer’s disease

    Sikka, A., Skand, Virk, J.S., Bathula, D.R., 2021. Mri to pet cross-modality translation using globally and locally aware gan (gla- gan) for multi-modal diagnosis of alzheimer’s disease. ArXiv abs/2108.02160. URL: https://api.semanticscholar.org/CorpusID: 236912958

  18. [26]

    Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation

    Tajbakhsh, N., Jeyaseelan, L., Li, Q., Chiang, J.N., Wu, Z., Ding, X., 2020. Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation. Medical Image Analysis 63, 101693. URL: https://www.sciencedirect. com/science/article/pii/S1361841520...

  19. [27]

    Image qualityassessment:fromerrorvisibilitytostructuralsimilarity

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P., 2004. Image qualityassessment:fromerrorvisibilitytostructuralsimilarity. IEEE transactions on image processing 13, 600–612

  20. [28]

    Predictingpet-deriveddemyelinationfrommul- timodal mri using sketcher-refiner adversarial training for multiple sclerosis

    Wei,W.,Poirion,E.,Bodini,B.,Durrleman,S.,Ayache,N.,Stankoff, B.,Colliot,O.,2019. Predictingpet-deriveddemyelinationfrommul- timodal mri using sketcher-refiner adversarial training for multiple sclerosis. Medical image analysis 58, 101546

  21. [29]

    Pro- posal for a new classification of outcome with respect to epileptic seizures following epilepsy surgery

    Wieser, H.G., Blume, W.T., Fish, D.R., Goldensohn, E.S., Hufnagel, A., King, D., Sperling, M.R., Lüders, H., Pedley, T.A., 2001. Pro- posal for a new classification of outcome with respect to epileptic seizures following epilepsy surgery. Epilepsia 42. URL: https: Zotova, Pino...

  22. [30]

    Gener- ative adversarial networks for noise reduction in low-dose ct

    Wolterink, J.M., Leiner, T., Viergever, M.A., Išgum, I., 2017. Gener- ative adversarial networks for noise reduction in low-dose ct. IEEE transactions on medical imaging 36, 2536–2545

  23. [31]

    Generative adversarial networks: A primer for radiologists

    Wolterink, J.M., Mukhopadhyay, A., Leiner, T., Vogl, T.J., Bucher, A.M., Išgum, I., 2021. Generative adversarial networks: A primer for radiologists. RadioGraphics 41, 840–857. URL: https://doi.org/10.1148/rg.2021200151, doi: 10.1148/rg.2021200151, arXiv:https://doi.org/10.114...

  24. [32]

    Synthesizing pet images from high-field and ultra-high-field mr images using joint diffusion attention model

    Xie, T., Cao, C., Cui, Z., Guo, Y., Wu, C., Wang, X., Li, Q., Hu, Z., Sun, T., Sang, Z., Zhou, Y., Zhu, Y., Liang, D., Jin, Q., Zeng, H., Chen, G., Wang, H., 2024. Synthesizing pet images from high-field and ultra-high-field mr images using joint diffusion attention model. URL...

  25. [33]

    Yaakub, S.N., McGinnity, C.J., Clough, J.R., Kerfoot, E., Girard, N., Guedj, E., Hammers, A., 2019. Pseudo-normal pet synthesis with generative adversarial networks for localising hypometabolism inepilepsies,in:InternationalWorkshoponSimulationandSynthesis in Medical Imaging, ...

  26. [34]

    Bpgan: Brain pet synthesis from mri using generative adversarial network for multi-modal alzheimer’s disease diagnosis

    Zhang, J., He, X., Qing, L., Gao, F., Wang, B., 2022. Bpgan: Brain pet synthesis from mri using generative adversarial network for multi-modal alzheimer’s disease diagnosis. Computer Methods and Programs in Biomedicine 217, 106676. URL: https: //www.sciencedirect.com/science/a...

  27. [35]

    The unreasonableeffectivenessofdeepfeaturesasaperceptualmetric,in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Zhang,R.,Isola,P.,Efros,A.A.,Shechtman,E.,Wang,O.,2018. The unreasonableeffectivenessofdeepfeaturesasaperceptualmetric,in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 586–595

  28. [36]

    Unpaired image- to-image translation using cycle-consistent adversarial networks, in: Computer Vision (ICCV), 2017 IEEE International Conference on

    Zhu, J.Y., Park, T., Isola, P., Efros, A.A., 2017. Unpaired image- to-image translation using cycle-consistent adversarial networks, in: Computer Vision (ICCV), 2017 IEEE International Conference on

  29. [37]

    Zotova, D., Jung, J., Lartizien, C., 2021. Gan-based synthetic fdg pet images from t1 brain mri can serve to improve performance of deep unsupervisedanomalydetectionmodels,in:Svoboda,D.,Burgos,N., Wolterink,J.M.,Zhao,C.(Eds.),SimulationandSynthesisinMedical Imaging, Springer I...

  30. [2021]

    Global Image-Based Unsupervised Anomaly Detec- tion in MR Brain Scans of Early Parkinsonian Patients, in: Machine Learning in Clinical Neuroimaging, Cham

    Patch vs. Global Image-Based Unsupervised Anomaly Detec- tion in MR Brain Scans of Early Parkinsonian Patients, in: Machine Learning in Clinical Neuroimaging, Cham. pp. 34–43

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.