AbstractThe success of deep learning (DL) relies heavily on training data from which DL models encapsulate information. Consequently, the development and deployment of DL models expose data to potential privacy breaches, which are particularly critical in data-sensitive contexts like medicine. We propose a new technique named DiffGuard that generates realistic and diverse synthetic medical images with annotations, even indistinguishable for experts, to replace real data for DL model training, which cuts off their direct connection and enhances privacy safety.

摘要深度学习（DL）的成功在很大程度上依赖于DL模型封装信息的训练数据。因此，DL模型的开发和部署将数据暴露于潜在的隐私侵犯，这在医学等数据敏感的环境中尤为关键。我们提出了一种名为DiffGuard的新技术，该技术可以生成具有注释的真实且多样的合成医学图像，甚至专家无法区分，以替代DL模型训练的真实数据，从而切断了它们的直接连接并增强了隐私安全。

We demonstrate that DiffGuard enhances privacy safety with much less data leakage and better resistance against privacy attacks on data and model. It also improves the accuracy and generalizability of DL models for segmentation and classification of mediastinal neoplasms in multi-center evaluation. We expect that our solution would enlighten the road to privacy-preserving DL for precision medicine, promote data and model sharing, and inspire more innovation on artificial-intelligence-generated-content technologies for medicine..

。它还提高了DL模型在多中心评估中用于纵隔肿瘤分割和分类的准确性和通用性。我们希望我们的解决方案能够为精准医学的隐私保护DL之路提供启示，促进数据和模型共享，并激发人工智能生成的医学内容技术的更多创新。。

IntroductionDeep learning (DL) has served as a fundamental technology and empowered many practical applications in medical image-based diagnosis and treatment1,2,3. The fascinating performance of DL models comes from the carefully designed algorithms, high-performance computing devices, and most fundamentally, the large-scale training data of high quality.

。DL模型的迷人性能来自精心设计的算法，高性能的计算设备，最根本的是，高质量的大规模训练数据。

Despite the promising performance and wide deployment of DL models in clinical applications, it is necessary to rethink a question: would these DL models leak the training images and more importantly, the privacy of numerous patients that provided these images? The answer is pessimistic because with the rapid development of privacy attack techniques, privacy leakage risks may be throughout the whole pipeline of DL-based medical image diagnosis, including data leakage in data inspection and data sharing, data memorization attacks during model training4,5, data leakage from gradients during federated learning6,7, white-box model inversion attacks8,9,10 and black-box membership inference attacks11,12,13 during and after model deployment (Fig.

尽管DL模型在临床应用中具有良好的性能和广泛的部署，但有必要重新思考一个问题：这些DL模型是否会泄漏训练图像，更重要的是，会泄漏提供这些图像的众多患者的隐私？答案是悲观的，因为随着隐私攻击技术的快速发展，隐私泄漏风险可能贯穿于基于DL的医学图像诊断的整个流程，包括数据检查和数据共享中的数据泄漏，模型训练期间的数据存储攻击4,5，联邦学习期间梯度的数据泄漏6,7，模型部署期间和之后的白盒模型反转攻击8,9,10和黑盒成员推理攻击11,12,13（图）。

1a). With these risks on medical images, two types of privacy may be violated: membership privacy14 and identity privacy15. When membership privacy is violated, attackers may leverage membership information to infer sensitive attributes related to the data, such as health conditions and treatments. The violation of identity privacy is more serious, where attackers can directly recover the images and use them as identifiers of patients.

1a）。由于医学图像存在这些风险，可能会侵犯两种类型的隐私：会员隐私14和身份隐私15。当会员隐私受到侵犯时，攻击者可能会利用会员信息推断与数据相关的敏感属性，例如健康状况和治疗方法。侵犯身份隐私的情况更为严重，攻击者可以直接恢复图像并将其用作患者的标识符。

This means that attacks can precisely recognize every single patient in the dataset and obtain their sensitive attributes. Although there are some attempts to address these issues during model training, including knowledge distillation16,17, adversari.

这意味着攻击可以精确识别数据集中的每个患者并获得其敏感属性。虽然在模型训练期间有一些尝试来解决这些问题，包括知识提取16,17，对抗。

(1)

(1)

$$p({y}_{1},{y}_{2},\cdots ,{y}_{T}{{|}}{y}_{0})=\mathop{\prod }\limits_{t=1}^{T}p({y}_{t}{\rm{|}}{y}_{t-1})$$

$$p（{y}_1、{y}_｛2｝，\cdots，{y}_｛t｝｛|｝｝{y}_｛0｝）=\mathop｛prod｝\limits_{t=1}^{T}p（{y}_｛t｝｛rm｛|｝｝{y}_{T-1}）$$

(2)

(2)

where \(\{{\alpha }_{t}{\}}_{t=1}^{T}\) are the hyper-parameters of the noise schedule. From Eq. (2), we can marginalize the forward process:$$p({y}_{t}{\rm{|}}{y}_{0}{\mathscr{)}}{\mathscr{=}}{\mathscr{N}}{\mathscr{(}}{y}_{t}{\rm{;}}\sqrt{{\gamma }_{t}}{y}_{0},(1-{\gamma }_{t})I),t=1,2,\cdots ,T$$.

其中\（{{\ alpha}}ut}}ut=1}^{t}）是噪声时间表的超参数。从等式（2）中，我们可以将前进过程边缘化：$$p({y}_{t} {\rm{|}{y}_{0}{\mathscr{}}{\mathscr{=}}{\mathscr{N}}{\mathscr{（}}{y}_{t} {\rm{；}}\sqrt{\gamma}\uu{t}}{y}_{0}，（1-{\ gamma}{t}）I），t=1,2，\cdots，t$$。

(3)

(3)

where \({\gamma }_{t}={\prod }_{s=0}^{t}(1-{\alpha }_{s})\). The above equations lay the foundation for the forward diffusion process. Our goal is to recover the target image \({y}_{0}\) given a noisy image \({y}_{T}\):$${y}_{T}=\sqrt{{\gamma }_{T}}{y}_{0}+\sqrt{1-{\gamma }_{T}}\epsilon ,\epsilon \sim (0,I)$$.

其中\（{\ gamma}u{t}={\ prod}u{s=0}^{t}（1-{\ alpha}u{s}）\）。上述方程为前向扩散过程奠定了基础。我们的目标是恢复目标图像\({y}_{0}\）给定一个有噪声的图像\({y}_{T} \）：$${y}_{T} =\sqrt{{\gamma}\ut}{y}_{0}+\sqrt{1-{\ gamma}}\ut}}\ epsilon \ epsilon \ sim（0，I）$$。

(4)

(4)

Therefore, if we can estimate \(\epsilon\), then we can calculate \({y}_{0}\):$${y}_{0}=\frac{1}{\sqrt{{\gamma }_{t}}}({y}_{t}-\sqrt{1-{\gamma }_{t}}\epsilon )$$

因此，如果我们可以估计\（\ epsilon \），那么我们可以计算\({y}_{0}\）：$${y}_{0}=\frac{1}{\sqrt{{\gamma}\uu{t}}({y}_{t}-\sqrt{1-{\ gamma}{t}}\ epsilon）$$

(5)

(5)

To estimate \(\epsilon\), we train a neural network \(f\) parameterized by \(\theta\) with the following objective:$${{\mathbb{E}}}_{y}{{\mathbb{E}}}_{(\epsilon ,\gamma )}{\rm{||}}{f}_{\theta }(y,\gamma )-\epsilon {\rm{|}}{{\rm{|}}}_{2}^{2}$$

为了估计\（\ epsilon \），我们训练了一个由\（\ theta \）参数化的神经网络\（f \），其目标如下：$${{\ mathbb{E}}}{y}{\ mathbb{E}}{（\ epsilon，\ gamma）}{\ rm{| |}}{f}_{\θ}（y，\ gamma）-\ε{\ rm{|}}{{\ rm{|}}}{{2}^{2}$$

(6)

(6)

We adopted the same network architecture and hyper-parameters as the method proposed by Ho et al.58 and trained with AdamW optimizer114 at a learning rate of 0.0001 for 1,000,000 iterations of batch size 12.During the inference stage, instead of directly estimating \({y}_{0}\) using \(f\), we iteratively perform the denoising for \(T\) steps.

我们采用了与Ho等人58提出的方法相同的网络体系结构和超参数，并使用AdamW optimizer114以0.0001的学习率对批量大小为12的1000000次迭代进行了训练。在推理阶段，而不是直接估计\({y}_。

The posterior distribution of \({y}_{t-1}\) given \({y}_{0}\) and \({y}_{t}\) can be formulated as:$$p({y}_{t-1}{\rm{|}}{y}_{0},{y}_{t}{\mathscr{)}}{\mathscr{=}}{\mathscr{N}}{\mathscr{(}}{y}_{t-1}{\rm{|}}{\mu }_{t},{{\sigma }_{t}}^{2}I)$$.

的后验分布\({y}_\({y}_{0}\）和\({y}_{t} \）可以表示为：$$p({y}_{t-1}{\rm{|}}{y}_{0}，{y}_{t} {\mathscr{}}{\mathscr{=}}{\mathscr{N}}{\mathscr{（}}{y}_{t-1}{\rm{|}}{\mu}{t}，{{\sigma}{t}^{2}I)$$。

(7)

(7)

where \({\mu }_{t}=\frac{\sqrt{{\gamma }_{t-1}}(1-{\alpha }_{t})}{1-{\gamma }_{t}}{y}_{0}+\frac{\sqrt{{\alpha }_{t}}(1-{\gamma }_{t-1})}{1-{\gamma }_{t}}{y}_{t}\) and \({\sigma }_{t}=\sqrt{\frac{(1-{\gamma }_{t-1})(1-{\alpha }_{t})}{1-{\gamma }_{t}}}\). Given Eq. (5) and Eq. (7), we can obtain the estimation:$$\hat{{\mu }_{t}}=\frac{1}{\sqrt{{\alpha }_{t}}}({y}_{t}-\frac{1-{\alpha }_{t}}{\sqrt{1-{\gamma }_{t}}}{f}_{\theta }({y}_{t},{\gamma }_{t}))$$.

其中\（{\ mu}{t}=\ frac{\ sqrt{\ gamma}{t-1}（1-{\ alpha}{t}）}{1-{\ gamma}{t}{y}_{0}+\frac{\sqrt{{\alpha}\ut}（1-{\gamma}\ut-1}}{1-{\gamma}\ut}{y}_{t} \）和\（{\ sigma}ut}=\ sqrt{\ frac{（1-{\ gamma}ut-1}）（1-{\ alpha}ut}}{1-{\ gamma}ut}}）。给定等式（5）和等式（7），我们可以得到估计：$$$\ hat{{\ mu}}UUT}=\ frac{1}{\ sqrt{\ alpha}UT}}({y}_{t}-\分形{1-{\ alpha}{ut}}{\ sqrt{1-{\ gamma}{ut}}{f}_{\θ}({y}_{t} ，{\ gamma}}{t}））$$。

(8)

(8)

Following Ho et al.58, we use \(\hat{{\sigma }_{t}}=\sqrt{1-{\alpha }_{t}}\). As a result, each iteration of the denoising process can be computed as:$${\widehat{{y}_{t-1}}}=\frac{1}{\sqrt{{\alpha }_{t}}}\left({y}_{t}-\frac{1-{\alpha }_{t}}{\sqrt{1-{\gamma }_{t}}}{f}_{\theta }({y}_{t},{\gamma }_{t})\right)+\sqrt{1-{\alpha }_{t}}{\epsilon }_{t},t=T,T-1,\cdots ,1$$.

根据Ho等人的58，我们使用\（\ hat{\ sigma}ut}=\ sqrt{1-{\ alpha}ut}\）。因此，降噪过程的每次迭代都可以计算为：$${\widehat{{y}_{t-1}}=\frac{1}{\sqrt{{\alpha}}\uu{t}}\left({y}_{t}-\分形{1-{\ alpha}{ut}}{\ sqrt{1-{\ gamma}{ut}}{f}_{\θ}({y}_{t} ，{\伽玛}{t}\右）+\ sqrt{1-{\阿尔法}{t}{\ε}{t}，t=t，t-1，cdots，1美元。

(9)

(9)

where \({\epsilon }_{t}\) is randomly sampled from \({\mathscr{N}}(0,I)\). In practice, we set \(T=1000\).Finally, each sampled \(\hat{y}\) is decoded into a synthetic image and its corresponding annotation mask. The former channels of \(\hat{y}\) represents the synthetic image and are linearly rescaled to the range \((\mathrm{0,1})\).

其中\（{\ epsilon}\{t}\）是从\（{\ mathscr{N}}（0，I）\）中随机采样的。在实践中，我们设置了\（T=1000 \）。最后，将每个采样的“hat{y}”解码为合成图像及其相应的注释掩码。前一个通道（hat{y}）代表合成图像，并线性重新缩放到范围（mathrm{0,1}）。

The last channel of \(\hat{y}\) represents the annotation mask and is linearly rescaled to the range \((0,C)\). For each pixel, the floating-point value is converted to the class label with the smallest absolute difference.To ensure the quality of synthetic datasets, we filtered out low-quality synthetic images with an autoencoder model.

\（\ hat{y}\）的最后一个通道表示注释掩码，并线性重新缩放到范围\（（0，C）\）。对于每个像素，浮点值将转换为绝对差值最小的类标签。为了确保合成数据集的质量，我们使用自动编码器模型过滤掉了低质量的合成图像。

We also filtered out the synthetic samples with more than one type of mediastinal neoplasm.Baseline methodsIn this study, we compared DiffGuard with several baseline methods. For the 2D augmentation method, each axial CT slice in the real CT scans was randomly rotated by an angle uniformly sampled from \(-2^{\circ}\) to \(2^{\circ}\), then randomly cropped and resized to a height and weight of 256.

我们还筛选出具有多种纵隔肿瘤的合成样品。基线方法在这项研究中，我们将DiffGuard与几种基线方法进行了比较。对于2D增强方法，将真实CT扫描中的每个轴向CT切片随机旋转一个从（-2 ^{\ circ}）到（2 ^{\ circ}）均匀采样的角度，然后随机裁剪并调整大小至256的高度和重量。

For the 3D augmentation method, each CT scan was randomly rotated by an angle uniformly sampled from \(-5^{\circ}\) to \(5^{\circ}\), then randomly resized in three dimensions whose scales were sampled independently between 0.9 and 1.1. To make a fair comparison with AsynDGAN, we followed the original training hyperparameters except for the multi-institutional federated learning setting, and we trained the models on the gathered training images.

对于3D增强方法，将每个CT扫描随机旋转一个从（-5 ^{\ circ}）到（5 ^{\ circ}）均匀采样的角度，然后随机调整三维尺寸，其尺度在0.9和1.1之间独立采样。为了与AsynDGAN进行公平比较，我们遵循了除多机构联合学习设置之外的原始训练超参数，并在收集的训练图像上训练了模型。

We used the original training hyperparameters and adopted the same data augmentation strategies as DiffGuard, i.e., random horizontal flipping, rotation, and resizing. After training, we generated images with randomly rotated and resized ground-truth lab.

我们使用了原始的训练超参数，并采用了与DiffGuard相同的数据增强策略，即随机水平翻转，旋转和调整大小。训练后，我们使用随机旋转和调整大小的地面真相实验室生成图像。

Data availability

数据可用性

The data that support the findings of this study are divided into two groups: shared data and restricted data. Shared data are available from the manuscript, references, and supplementary materials. Restricted data relating to individuals in this study are subject to a license that allows for the use of the data only for analysis.

支持本研究结果的数据分为两组：共享数据和受限数据。共享数据可从手稿，参考文献和补充材料中获得。本研究中与个人相关的受限数据需要获得许可证，该许可证仅允许将数据用于分析。

The internal test datasets, external test datasets, and DiffGuard-generated data will be released upon publication..

内部测试数据集、外部测试数据集和DiffGuard生成的数据将在发布时发布。。

Code availability

代码可用性

We have uploaded our code, trained models, and part of the DiffGuard-generated data at https://github.com/ZhanpZhou/DiffGuard (https://doi.org/10.5281/zenodo.13946208). For experiments on nnU-Net, we used the public code at https://github.com/MIC-DKFZ/nnUNet/tree/nnunetv1. For experiments on AsynDGAN, we used the public code at https://github.com/tommy-qichang/AsynDGAN.

我们已经上传了我们的代码、训练过的模型和DiffGuard生成的部分数据https://github.com/ZhanpZhou/DiffGuard（笑声）(https://doi.org/10.5281/zenodo.13946208)。为了在nnU Net上进行实验，我们使用了https://github.com/MIC-DKFZ/nnUNet/tree/nnunetv1.对于AsynDGAN的实验，我们使用了https://github.com/tommy-qichang/AsynDGAN.

For DP-SGD, we used the python package pyvacy (version 0.0.23)..

对于DP-SGD，我们使用了python包pyvacy（版本0.0.23）。。

ReferencesEsteva, A. et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature 542, 115–118 (2017).Article

参考文献Esteva，A。等人。皮肤科医生用深度神经网络对皮肤癌进行分类。自然542115-118（2017）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

McKinney, S. M. et al. International evaluation of an AI system for breast cancer screening. Nature 577, 89–94 (2020).Article

McKinney，S.M.等人，《人工智能乳腺癌筛查系统的国际评估》。《自然》577,89-94（2020）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Lu, M. Y. et al. AI-based pathology predicts origins for cancers of unknown primary. Nature 594, 106–110 (2021).Article

Lu，M.Y.等人，基于AI的病理学预测未知原发性癌症的起源。自然594106-110（2021）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Song, C., Ristenpart, T. & Shmatikov, V. Machine Learning Models that Remember Too Much. in Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security 587–601 (ACM, Dallas Texas USA, 2017). https://doi.org/10.1145/3133956.3134077.Li, H., Ayache, N. & Delingette, H.

Song，C.，Ristenpart，T。＆Shmatikov，V。记住太多的机器学习模型。2017年ACM SIGSAC计算机与通信安全会议论文集587-601（ACM，美国德克萨斯州达拉斯，2017）。https://doi.org/10.1145/3133956.3134077.Li，H.，Ayache，N。和Delingette，H。

Data Stealing Attack on Medical Images: Is It Safe to Export Networks from Data Lakes? in Distributed, Collaborative, and Federated Learning, and Affordable AI and Healthcare for Resource Diverse Global Health (eds. Albarqouni, S. et al.) vol. 13573 28–36 (Springer Nature Switzerland, Cham, 2022).Zhu, L., Liu, Z.

医学图像数据窃取攻击：从数据湖导出网络安全吗？《分布式、协作式和联合式学习，以及可负担得起的人工智能和医疗保健，以实现资源多样化的全球健康》（编辑Albarqouni，S.等人）第13573卷28-36（瑞士施普林格自然杂志，查姆，2022年）。朱，刘，Z。

& Han, S. Deep leakage from gradients. in Advances in neural information processing systems (eds. Wallach, H. et al.) vol. 32 (Curran Associates, Inc., 2019).Geiping, J., Bauermeister, H., Dröge, H. & Moeller, M. Inverting Gradients - How easy is it to break privacy in federated learning? in Advances in neural information processing systems (eds.

&Han，S。梯度深度渗漏。《神经信息处理系统的进展》（eds.Wallach，H.et al。）第32卷（Curran Associates，Inc.，2019）。Geiping，J.，Bauermeister，H.，Dröge，H.＆Moeller，M.反转梯度-在联邦学习中打破隐私有多容易？神经信息处理系统的进展（eds）。

Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M. F. & Lin, H.) vol. 33 16937–16947 (Curran Associates, Inc., 2020).Fredrikson, M., Jha, S. & Ristenpart, T. Model inversion attacks that exploit confidence information and basic countermeasures. in Proceedings of the 22nd ACM SIGSAC conference on computer and communications security, denver, CO, USA, october 12-16, 2015 (eds.

拉罗谢尔（H.Larochelle），兰萨托（M.Ranzato），哈塞尔（R.Hadsell），巴尔坎（Balcan），M.F。和林（Lin），H。）第33卷，16937–16947（Curran Associates，Inc.，2020）。Fredrikson，M.，Jha，S。＆Ristenpart，T。利用置信信息和基本对策的模型反转攻击。2015年10月12日至16日在美国科罗拉多州丹佛举行的第22届ACM SIGSAC计算机与通信安全会议论文集（编辑：。

Ray, I., Li, N. & Kruegel, C.) 1322–1333 (ACM, 2015). https://doi.org/10.1145/2810103.2813677.Zhang, Y. et al. The Secret Revealer: Generative Model-Inversion Attacks Against Deep Neural Networks. in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 250–258 (IEEE, Seattle, WA, USA, 2020).

Ray，I.，Li，N。和Kruegel，C。）1322-1333（ACM，2015）。https://doi.org/10.1145/2810103.2813677.Zhang，Y.等人，《秘密揭秘者：针对深度神经网络的生成模型反转攻击》。2020年IEEE/CVF计算机视觉和模式识别会议（CVPR）250–258（IEEE，西雅图，华盛顿，美国，2020）。

https://doi.org/10.1109/CVPR42600.2020.00033.Struppek, L. et al. Plug & play attacks: Towards robust and flexible model inversion at.

https://doi.org/10.1109/CVPR42600.2020.00033.Struppek，L。等人。即插即用攻击：在。

Google Scholar

谷歌学者

Shejwalkar, V. & Houmansadr, A. Membership Privacy for Machine Learning Models Through Knowledge Transfer. Proc. AAAI Conf. Artif. Intell. 35, 9549–9557 (2021).

Shejwalkar，V。＆Houmansadr，A。通过知识转移的机器学习模型的成员隐私。程序。AAAI配置文件。因特尔。359549-9557（2021）。

Google Scholar

谷歌学者

Tang, X. et al. Mitigating membership inference attacks by {Self-Distillation} through a novel ensemble architecture. In 31st USENIX security symposium (USENIX security 22). 1433–1450 (2022).Nasr, M., Shokri, R. & Houmansadr, A. Machine Learning with Membership Privacy using Adversarial Regularization.

Tang，X.等人。通过一种新的集成架构，通过{自蒸馏}减轻成员推理攻击。在第31届USENIX安全研讨会（USENIX security 22）上。1433年至1450年（2022年）。Nasr，M.，Shokri，R。＆Houmansadr，A。使用对抗性正则化的具有成员隐私的机器学习。

in Proceedings of the 2018 ACM SIGSAC Conference on Computer and Communications Security 634–646 (ACM, Toronto Canada, 2018). https://doi.org/10.1145/3243734.3243855.Hu, H., Salcic, Z., Dobbie, G., Chen, Y. & Zhang, X. EAR: An Enhanced Adversarial Regularization Approach against Membership Inference Attacks.

在2018年ACM SIGSAC计算机与通信安全会议论文集634-646（ACM，加拿大多伦多，2018）中。https://doi.org/10.1145/3243734.3243855.Hu，H.，Salcic，Z.，Dobbie，G.，Chen，Y。＆Zhang，X。EAR：一种针对成员推理攻击的增强对抗性正则化方法。

in 2021 International Joint Conference on Neural Networks (IJCNN) 1–8 (IEEE, Shenzhen, China, 2021). https://doi.org/10.1109/IJCNN52387.2021.9534381.Abadi, M. et al. Deep Learning with Differential Privacy. in Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security 308–318 (ACM, Vienna Austria, 2016).

2021年国际神经网络联合会议（IJCNN）1-8（IEEE，中国深圳，2021）。https://doi.org/10.1109/IJCNN52387.2021.9534381.Abadi，M.等人，《差异隐私的深度学习》。在2016年ACM SIGSAC计算机与通信安全会议论文集308-318（ACM，维也纳-奥地利，2016）中。

https://doi.org/10.1145/2976749.2978318.Shin, H.-C. et al. Medical Image Synthesis for Data Augmentation and Anonymization Using Generative Adversarial Networks. in Simulation and Synthesis in Medical Imaging (eds. Gooya, A., Goksel, O., Oguz, I. & Burgos, N.) vol. 11037 1–11 (Springer International Publishing, Cham, 2018).Han, T.

https://doi.org/10.1145/2976749.2978318.Shin，H.-C.等人。使用生成对抗网络进行数据增强和匿名化的医学图像合成。在医学成像中的模拟和合成（编辑：Gooya，A.，Goksel，O.，Oguz，I。＆Burgos，N。）第11037卷1-11（Springer International Publishing，Cham，2018）。。

et al. Breaking medical data sharing boundaries by using synthesized radiographs. Sci. Adv. 6, eabb7973 (2020).Article .

等人。通过使用合成射线照片打破医学数据共享边界。科学。广告6，eabb7973（2020）。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

DuMont Schütte, A. et al. Overcoming barriers to data sharing with medical image generation: a comprehensive evaluation. Npj Digit. Med. 4, 141 (2021).Article

DuMont Schütte，A.等人。克服医学图像生成数据共享的障碍：综合评估。。医学杂志4141（2021）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Henschke, C. I. et al. CT Screening for Lung Cancer:Prevalence and Incidence of Mediastinal Masses. Radiology 239, 586–590 (2006).Article

Henschke，C.I。等。肺癌的CT筛查：纵隔肿块的患病率和发病率。放射学239586-590（2006）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Yoon et al. Incidental Anterior Mediastinal Nodular Lesions on Chest CT in Asymptomatic Subjects. J. Thorac. Oncol. Publ. Int. Assoc. Study Lung Cancer 13, 359–366 (2017).

Yoon等人。无症状受试者胸部CT偶发前纵隔结节性病变。J、 。Oncol公司。公共。国际协会肺癌研究13359-366（2017）。

Google Scholar

谷歌学者

Miyazawa, R. et al. Incidental mediastinal masses detected at low-dose CT screening: prevalence and radiological characteristics. Jpn. J. Radio. 38, 1150–1157 (2020).Article

Miyazawa，R.等人。在低剂量CT筛查中发现的偶发纵隔肿块：患病率和放射学特征。日本。J、 收音机。381150-1157（2020）。文章

Google Scholar

谷歌学者

Strollo, D. C., de, C., Melissa, L., Rosado, J. & James, R. Primary Mediastinal Tumors. Part 1*: Tumors of the Anterior Mediastinum. Chest 112, 511–522 (1997).Article

D.C.，D.C.，Melissa，L.，Rosado，J。＆James，R。原发性纵隔肿瘤。第1部分*：前纵隔肿瘤。胸部112511-522（1997）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Juanpere, S. et al. A diagnostic approach to the mediastinal masses. Insights Imaging 4, 29–52 (2012).Article

Juanpere，S.等人。纵隔肿块的诊断方法。。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Somepalli, G., Singla, V., Goldblum, M., Geiping, J. & Goldstein, T. Diffusion art or digital forgery? investigating data replication in diffusion models. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 6048–6058 (2023).Yeom, S., Giacomelli, I., Fredrikson, M.

Somepalli，G.，Singla，V.，Goldblum，M.，Geiping，J。＆Goldstein，T。扩散艺术还是数字伪造？研究扩散模型中的数据复制。在IEEE/CVF计算机视觉和模式识别会议论文集6048-6058（2023）中。。

& Jha, S. Privacy Risk in Machine Learning: Analyzing the Connection to Overfitting. in 2018 IEEE 31st Computer Security Foundations Symposium (CSF) 268–282 (IEEE, Oxford, 2018). https://doi.org/10.1109/CSF.2018.00027.Aberle, D. R. et al. Reduced lung-cancer mortality with low-dose computed tomographic screening.

&Jha，S。机器学习中的隐私风险：分析与过度拟合的关系。2018年IEEE第31届计算机安全基础研讨会（CSF）268–282（IEEE，牛津，2018）。https://doi.org/10.1109/CSF.2018.00027.Aberle，D.R.等人通过低剂量计算机断层扫描筛查降低了肺癌死亡率。

N. Engl. J. Med. 365, 395–409 (2011).Article .

N、 英语。J、 医学365395-409（2011）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B. & Hochreiter, S. GANs trained by a two time-scale update rule converge to a local nash equilibrium. in Advances in neural information processing systems 30: Annual conference on neural information processing systems 2017, december 4–9, 2017, long beach, CA, USA (eds.

Heusel，M.，Ramsauer，H.，Unterthiner，T.，Nessler，B。＆Hochreiter，S.GANs通过两时间尺度更新规则训练，收敛于局部纳什均衡。神经信息处理系统进展30:2017年神经信息处理系统年会，2017年12月4日至9日，美国加利福尼亚州长滩（eds。

Guyon, I. et al.) 6626–6637 (2017).Binkowski, M., Sutherland, D. J., Arbel, M. & Gretton, A. Demystifying MMD gans. in 6th international conference on learning representations, ICLR 2018, vancouver, BC, canada, april 30 - may 3, 2018, conference track proceedings (OpenReview.net, 2018).Salimans, T. et al.

。Binkowski，M.，Sutherland，D.J.，Arbel，M。和Gretton，A。揭开MMD gans的神秘面纱。在2018年4月30日至5月3日于加拿大不列颠哥伦比亚省温哥华举行的第六届国际学习表征会议（ICLR 2018）上，会议记录（OpenReview.net，2018）。Salimans，T。等人。

Improved techniques for training gans. in Advances in neural information processing systems 29: Annual conference on neural information processing systems 2016, december 5-10, 2016, barcelona, spain (eds. Lee, D. D., Sugiyama, M., von Luxburg, U., Guyon, I. & Garnett, R.) 2226–2234 (2016).Chang, Q. et al.

改进了训练gans的技术。《神经信息处理系统的进展》29：《2016年神经信息处理系统年度会议》，2016年12月5日至10日，西班牙巴塞罗那（编辑：Lee，D.D.，Sugiyama，M.，von Luxburg，U.，Guyon，I.＆Garnett，R。）2226-2234（2016）。Chang，Q。等人。

Synthetic Learning: Learn From Distributed Asynchronized Discriminator GAN Without Sharing Medical Image Data. in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 13853–13863 (IEEE, Seattle, WA, USA, 2020). https://doi.org/10.1109/CVPR42600.2020.01387.Isensee, F., Jaeger, P.

综合学习：在不共享医学图像数据的情况下，从分布式异步鉴别器GAN学习。2020年IEEE/CVF计算机视觉和模式识别会议（CVPR）13853–13863（IEEE，西雅图，华盛顿，美国，2020）。https://doi.org/10.1109/CVPR42600.2020.01387.Isensee，F.，杰格，P。

F., Kohl, S. A. A., Petersen, J. & Maier-Hein, K. H. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat. Methods 18, 203–211 (2021).Article .

F、 ，Kohl，S.A.A.，Petersen，J.＆Maier-Hein，K.H.nnU-Net：一种用于基于深度学习的生物医学图像分割的自配置方法。自然方法18203-211（2021）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Ronneberger, O., Fischer, P. & Brox, T. U-net: Convolutional networks for biomedical image segmentation. in Medical image computing and computer-assisted intervention - MICCAI 2015 - 18th international conference munich, germany, october 5 - 9, 2015, proceedings, part III (eds. Navab, N., Hornegger, J., I.

Ronneberger，O.，Fischer，P。＆Brox，T.U-net：用于生物医学图像分割的卷积网络。在医学图像计算和计算机辅助干预-MICCAI 2015-第18届国际会议慕尼黑，德国，2015年10月5日至9日，会议记录，第三部分（编辑Navab，N.，Hornegger，J.，I。

I. I., W. M. W. & Frangi, A. F.) vol. 9351 234–241 (Springer, 2015).Chen, J. et al. TransUNet: Rethinking the U-Net architecture design for medical image segmentation through the lens of transformers. Med. Image Anal. 97, 103280 (2024).Dwork, C. Differential Privacy. in Automata, Languages and Programming (eds.

一、 I.，W.M.W.＆Frangi，A.F。）第9351卷234-241（Springer，2015）。Chen，J。et al。TransUNet：重新思考通过变压器镜头进行医学图像分割的U-Net架构设计。医学图像肛门。97103280（2024）。Dwork，C。差异隐私。在自动机，语言和编程（eds。

Bugliesi, M., Preneel, B., Sassone, V. & Wegener, I.) vol. 4052 1–12 (Springer Berlin Heidelberg, Berlin, Heidelberg, 2006).Dwork, C. & Roth, A. The Algorithmic Foundations of Differential Privacy. (now Publishers Inc, 2013). https://doi.org/10.1561/9781601988195.Gadotti, A., Rocher, L., Houssiau, F., Creţu, A.-M.

。Dwork，C。＆Roth，A。差异隐私的算法基础。（现为Publishers Inc，2013）。https://doi.org/10.1561/9781601988195.Gadotti，A.，Rocher，L.，Houssiau，F.，Creţu，A.-M。

& De Montjoye, Y.-A. Anonymization: The imperfect science of using data while preserving privacy. Sci. Adv. 10, eadn7053 (2024).Article .

&匿名化：在保护隐私的同时使用数据的不完善科学。科学。Adv.10，eadn7053（2024）。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Jayaraman, B. & Evans, D. Evaluating differentially private machine learning in practice. in 28th USENIX security symposium, USENIX security 2019, santa clara, CA, USA, august 14-16, 2019 (eds. Heninger, N. & Traynor, P.) 1895–1912 (USENIX Association, 2019).Blanco-Justicia, A., Sánchez, D., Domingo-Ferrer, J.

Jayaraman，B。＆Evans，D。在实践中评估差异私有机器学习。。布兰科·贾斯蒂西亚（BlancoJusticia），桑切斯（Sánchez），多明戈·费雷尔（DomingoFerrer），J。

& Muralidhar, K. A Critical Review on the Use (and Misuse) of Differential Privacy in Machine Learning. ACM Comput. Surv. 55, 1–16 (2023).Article .

&Muralidhar，K。关于机器学习中差异隐私的使用（和滥用）的批判性评论。ACM计算机。Surv。55，1-16（2023）。文章。

Google Scholar

谷歌学者

Tayebi Arasteh, S. et al. Preserving fairness and diagnostic accuracy in private large-scale AI models for medical imaging. Commun. Med. 4, 1–12 (2024).Article

Tayebi-Arasteh，S.等人。在用于医学成像的私有大规模AI模型中保持公平性和诊断准确性。Commun公司。医学杂志4，1-12（2024）。文章

Google Scholar

谷歌学者

Ziller, A. et al. Reconciling privacy and accuracy in AI for medical imaging. Nat. Mach. Intell. 1–11 (2024).Choi, E. et al. Generating multi-label discrete patient records using generative adversarial networks. in Machine learning for healthcare conference 286–305 (2017).Xie, L., Lin, K., Wang, S., Wang, F.

Ziller，A.等人。调和医学成像AI中的隐私和准确性。自然马赫数。因特尔。1-11（2024年）。。在医疗保健机器学习会议286-305（2017）。谢，林，王，王，王，F。

& Zhou, J. Differentially Private Generative Adversarial Network. Preprint at http://arxiv.org/abs/1802.06739 (2018).Baowaly, M. K., Lin, C.-C., Liu, C.-L. & Chen, K.-T. Synthesizing electronic health records using improved generative adversarial networks. J. Am. Med. Inform. Assoc. 26, 228–241 (2019).Article .

&周，J。差异私有生成对抗网络。预印本http://arxiv.org/abs/1802.06739（2018年）。Baowaly，M.K.，Lin，C.-C.，Liu，C.-L.＆Chen，K.-T。使用改进的生成对抗网络合成电子健康记录。J、 上午医疗通知。协会第26228-241号（2019年）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Zhang, Z., Yan, C., Mesa, D. A., Sun, J. & Malin, B. A. Ensuring electronic medical record simulation through better training, modeling, and evaluation. J. Am. Med. Inform. Assoc. 27, 99–108 (2020).Article

Zhang，Z.，Yan，C.，Mesa，D.A.，Sun，J。＆Malin，B.A。通过更好的培训，建模和评估确保电子病历模拟。J、 上午医疗通知。协会第27、99–108号（2020年）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Goodfellow, I. J. et al. Generative adversarial nets. in Advances in neural information processing systems 27: Annual conference on neural information processing systems 2014, december 8-13 2014, montreal, quebec, canada (eds. Ghahramani, Z., Welling, M., Cortes, C., Lawrence, N. D. & Weinberger, K.

Goodfellow，I.J.等人，《生成对抗网》。《神经信息处理系统的进展》27：2014年12月8日至13日在加拿大魁北克蒙特利尔举行的神经信息处理系统年度会议（编辑：Ghahramani，Z.，Welling，M.，Cortes，C.，Lawrence，N.D。和Weinberger，K。

Q.) 2672–2680 (2014).Lin, Y., Wang, Z., Cheng, K.-T. & Chen, H. InsMix: Towards Realistic Generative Data Augmentation for Nuclei Instance Segmentation. in Medical Image Computing and Computer Assisted Intervention – MICCAI 2022 (eds. Wang, L., Dou, Q., Fletcher, P. T., Speidel, S. & Li, S.) vol. 13432 140–149 (Springer Nature Switzerland, Cham, 2022).Salehinejad, H., Valaee, S., Dowdell, T., Colak, E.

Q、 ）2672–2680（2014）。Lin，Y.，Wang，Z.，Cheng，K.-T.＆Chen，H。InsMix：面向核实例分割的现实生成数据增强。在医学图像计算和计算机辅助干预-MICCAI 2022（编辑：Wang，L.，Dou，Q.，Fletcher，P.T.，Speidel，S.＆Li，S。）第13432 140-149卷（瑞士施普林格自然出版社，查姆，2022）。萨利希内贾德（Salehinejad），瓦莱（Valaee），道德尔（Dowdell），科拉克（Colak）。

& Barfett, J. Generalization of Deep Neural Networks for Chest Pathology Classification in X-Rays Using Generative Adversarial Networks. in 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 990–994 (IEEE, Calgary, AB, 2018). https://doi.org/10.1109/ICASSP.2018.8461430.Ratliff, L.

&Barfett，J。使用生成对抗网络对X射线胸部病理学分类的深度神经网络的推广。2018年IEEE声学、语音和信号处理国际会议（ICASSP）990-994（IEEE，卡尔加里，AB，2018）。https://doi.org/10.1109/ICASSP.2018.8461430.Ratliff，L。

J., Burden, S. A. & Sastry, S. S. Characterization and computation of local Nash equilibria in continuous games. in 2013 51st Annual Allerton Conference on Communication, Control, and Computing (Allerton) 917–924 (IEEE, Monticello, IL, 2013). https://doi.org/10.1109/Allerton.2013.6736623.Ioffe, S. & Szegedy, C.

J、 ，Burden，S.A。＆Sastry，S.S。连续博弈中局部纳什均衡的表征和计算。2013年，第51届Allerton通信、控制和计算年度会议（Allerton）917-924（IEEE，伊利诺伊州蒙蒂塞洛，2013）。https://doi.org/10.1109/Allerton.2013.6736623.Ioffe，S.＆Szegedy，C。

Batch normalization: Accelerating deep network training by reducing internal covariate shift. in International conference on machine learning 448–456 (pmlr, 2015).Goodfellow, I. NIPS 2016 Tutorial: Generative Adversarial Networks. Preprint at http://arxiv.org/abs/1701.00160 (2017).Arora, S., Ge, R., Liang, Y., Ma, T.

。在国际机器学习会议448-456（pmlr，2015）中。Goodfellow，I。NIPS 2016教程：生成性对抗网络。预印于http://arxiv.org/abs/1701.00160（2017年）。阿罗拉（Arora，S.），葛（R.），梁（Liang），马（Ma），T。

& Zhang, Y. Generalization and equilibrium in generative adversarial nets (gans). in International conference on machine learni.

&Zhang，Y。生成对抗网（gans）中的推广和平衡。在国际机器学习会议上。

Google Scholar

谷歌学者

Ho, J., Jain, A. & Abbeel, P. Denoising diffusion probabilistic models. in Advances in neural information processing systems 33: Annual conference on neural information processing systems 2020, NeurIPS 2020, december 6-12, 2020, virtual (eds. Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M.-F.

Ho，J.，Jain，A。＆Abbeel，P。去噪扩散概率模型。《神经信息处理系统的进展》33：2020年神经信息处理系统年会，NeurIPS 2020，2020年12月6日至12日，virtual（编辑：Larochelle，H.，Ranzato，M.，Hadsell，R.，Balcan，M.-F。

& Lin, H.-T.) (2020).Dhariwal, P. & Nichol, A. Q. Diffusion models beat GANs on image synthesis. in Advances in neural information processing systems 34: Annual conference on neural information processing systems 2021, NeurIPS 2021, december 6-14, 2021, virtual (eds. Ranzato, M., Beygelzimer, A., Dauphin, Y.

&林，H.-T。）（2020）。Dhariwal，P。＆Nichol，A.Q。扩散模型在图像合成方面击败了GANs。神经信息处理系统进展34：神经信息处理系统2021年年度会议，NeurIPS 20212021年12月6日至14日，虚拟（编辑：Ranzato，M.，Beygelzimer，A.，Dauphin，Y。

N., Liang, P. & Vaughan, J. W.) 8780–8794 (2021).Azizi, S., Kornblith, S., Saharia, C., Norouzi, M. & Fleet, D. J. Synthetic data from diffusion models improves ImageNet classification. Trans. Mach. Learn. Res.Yang, L., Xu, X., Kang, B., Shi, Y. & Zhao, H. Freemask: Synthetic images with dense annotations make stronger segmentation models.

N、 ，Liang，P.＆Vaughan，J.W。）8780-8794（2021）。Azizi，S.，Kornblith，S.，Saharia，C.，Norouzi，M。＆Fleet，D.J。扩散模型的合成数据改进了ImageNet分类。事务处理。马赫。学习。Res.Yang，L.，Xu，X.，Kang，B.，Shi，Y。＆Zhao，H。Freemask：具有密集注释的合成图像可以产生更强的分割模型。

Adv. Neural Inf. Process. Syst. 36, (2024).Tian, Y. et al. Learning vision from models rivals learning vision from data. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 15887–15898 (2024).Hammoud, H. A. A. K., Itani, H., Pizzati, F., Bibi, A. & Ghanem, B. SynthCLIP: Are we ready for a fully synthetic CLIP training? In Synthetic data for computer vision workshop@ CVPR (2024).Pan, S.

高级神经信息处理。系统。36，（2024年）。Tian，Y.等人。从模型中学习愿景与从数据中学习愿景的竞争对手。。Hammoud，H.A.A.K.，Itani，H.，Pizzati，F.，Bibi，A.＆Ghanem，B.SynthCLIP：我们准备好进行全合成剪辑训练了吗？在CVPR计算机视觉研讨会的合成数据中（2024）。潘，S。

et al. 2D medical image synthesis using transformer-based denoising diffusion probabilistic model. Phys. Med. Biol. 68, 105004 (2023).Article .

等人。使用基于变压器的去噪扩散概率模型的2D医学图像合成。物理。医学生物学。68105004（2023）。文章。

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Nguyen, L. X., Sone Aung, P., Le, H. Q., Park, S.-B. & Hong, C. S. A New Chapter for Medical Image Generation: The Stable Diffusion Method. in 2023 International Conference on Information Networking (ICOIN) 483–486 https://doi.org/10.1109/ICOIN56518.2023.10049010 (2023).Khader, F. et al.

。2023年国际信息网络会议（ICOIN）483–486https://doi.org/10.1109/ICOIN56518.2023.10049010（2023年）。Khader，F。等人。

Denoising diffusion probabilistic models for 3D medical image generation. Sci. Rep. 13, 7303 (2023).Article .

用于3D医学图像生成的去噪扩散概率模型。科学。代表137303（2023）。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Harb, R., Pock, T. & Müller, H. Diffusion-based generation of histopathological whole slide images at a gigapixel scale. in Proceedings of the IEEE/CVF winter conference on applications of computer vision 5131–5140 (2024).Peng, W. et al. Generating Realistic Brain MRIs via a Conditional Diffusion Probabilistic Model.

Harb，R.，Pock，T。＆Müller，H。基于扩散的组织病理学全幻灯片图像的生成，以十亿像素为尺度。在IEEE/CVF计算机视觉应用冬季会议论文集5131-5140（2024）中。Peng，W.等人。通过条件扩散概率模型生成真实的大脑MRI。

in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 (eds. Greenspan, H. et al.) 14–24 (Springer Nature Switzerland, Cham, 2023). https://doi.org/10.1007/978-3-031-43993-3_2.Dorjsembe, Z., Odonchimed, S. & Xiao, F. Three-dimensional medical image synthesis with denoising diffusion probabilistic models.

在医学图像计算和计算机辅助干预方面–MICCAI 2023（eds。Greenspan，H。et al。）14-24（Springer Nature Switzerland，Cham，2023）。https://doi.org/10.1007/978-3-031-43993-3_2.Dorjsembe，Z.，Odonchimed，S。＆Xiao，F。具有去噪扩散概率模型的三维医学图像合成。

in Medical imaging with deep learning (2022).Xu, X., Kapse, S., Gupta, R. & Prasanna, P. ViT-DAE: Transformer-Driven Diffusion Autoencoder for Histopathology Image Analysis. in Deep Generative Models (eds. Mukhopadhyay, A., Oksuz, I., Engelhardt, S., Zhu, D. & Yuan, Y.) 66–76 (Springer Nature Switzerland, Cham, 2024).

在医学成像与深度学习（2022）。Xu，X.，Kapse，S.，Gupta，R。＆Prasanna，P。ViT DAE：用于组织病理学图像分析的变压器驱动扩散自动编码器。在深度生成模型中（编辑Mukhopadhyay，A.，Oksuz，I.，Engelhardt，S.，Zhu，D。和Yuan，Y。）66-76（瑞士施普林格自然出版社，查姆，2024）。

https://doi.org/10.1007/978-3-031-53767-7_7.Müller-Franzes, G. et al. A multimodal comparison of latent denoising diffusion probabilistic models and generative adversarial networks for medical image synthesis. Sci. Rep. 13, 12098 (2023).Article .

https://doi.org/10.1007/978-3-031-53767-7_7.Müller-Franzes，G.等人。用于医学图像合成的潜在去噪扩散概率模型和生成对抗网络的多模式比较。科学。第13页，第12098页（2023年）。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Sun, S., Goldgof, G., Butte, A. & Alaa, A. M. Aligning synthetic medical images with clinical knowledge using human feedback. Adv. Neural Inf. Process. Syst. 36, (2024).Takezaki, S. & Uchida, S. An Ordinal Diffusion Model for Generating Medical Images with Different Severity Levels. in 2024 IEEE International Symposium on Biomedical Imaging (ISBI) 1–5 https://doi.org/10.1109/ISBI56570.2024.10635504 (2024).Ye, J., Ni, H., Jin, P., Huang, S.

Sun，S.，Goldgof，G.，Butte，A。＆Alaa，A.M。使用人类反馈将合成医学图像与临床知识对齐。高级神经信息处理。系统。36，（2024年）。Takezaki，S。＆Uchida，S。用于生成具有不同严重程度的医学图像的有序扩散模型。2024年IEEE生物医学成像国际研讨会（ISBI）1-5https://doi.org/10.1109/ISBI56570.2024.10635504（2024年）。叶J.，倪H.，金P.，黄S。

X. & Xue, Y. Synthetic Augmentation with Large-Scale Unconditional Pre-training. in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 (eds. Greenspan, H. et al.) 754–764 (Springer Nature Switzerland, Cham, 2023). https://doi.org/10.1007/978-3-031-43895-0_71.Khosravi, B. et al.

十、 ＆Xue，Y。大规模无条件预训练的综合增强。。https://doi.org/10.1007/978-3-031-43895-0_71.Khosravi，B.等人。

Synthetically enhanced: unveiling synthetic data’s potential in medical imaging research. eBioMedicine 104, 105174 (2024).Article .

综合增强：揭示合成数据在医学成像研究中的潜力。eBioMedicine 104105174（2024）。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Reynaud, H. et al. Feature-Conditioned Cascaded Video Diffusion Models for Precise Echocardiogram Synthesis. in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 (eds. Greenspan, H. et al.) 142–152 (Springer Nature Switzerland, Cham, 2023). https://doi.org/10.1007/978-3-031-43999-5_14.Yoon, J.

Reynaud，H。等人。特征条件级联视频扩散模型，用于精确的超声心动图合成。在医学图像计算和计算机辅助干预方面–MICCAI 2023（eds。Greenspan，H。et al。）142–152（瑞士施普林格自然出版社，Cham，2023）。https://doi.org/10.1007/978-3-031-43999-5_14.Yoon，J。

S., Zhang, C., Suk, H.-I., Guo, J. & Li, X. SADM: Sequence-Aware Diffusion Model for Longitudinal Medical Image Generation. in Information Processing in Medical Imaging (eds. Frangi, A., de Bruijne, M., Wassermann, D. & Navab, N.) 388–400 (Springer Nature Switzerland, Cham, 2023). https://doi.org/10.1007/978-3-031-34048-2_30.Saeed, S.

S、 ，Zhang，C.，Suk，H.-I.，Guo，J。＆Li，X。SADM：用于纵向医学图像生成的序列感知扩散模型。在医学成像中的信息处理（编辑Frangi，A.，de Bruijne，M.，Wassermann，D。＆Navab，N。）388-400（瑞士施普林格自然出版社，查姆，2023）。https://doi.org/10.1007/978-3-031-34048-2_30.Saeed，S。

U. et al. Bi-parametric prostate MR image synthesis using pathology and sequence-conditioned stable diffusion. in Medical imaging with deep learning 814–828 (PMLR, 2024).Weber, T., Ingrisch, M., Bischl, B. & Rügamer, D. Cascaded Latent Diffusion Models for High-Resolution Chest X-ray Synthesis. in Advances in Knowledge Discovery and Data Mining (eds.

U、 等。使用病理学和序列条件稳定扩散的双参数前列腺MR图像合成。深度学习医学成像814-828（PMLR，2024）。Weber，T.，Ingrisch，M.，Bischl，B。＆Rügamer，D。用于高分辨率胸部X射线合成的级联潜在扩散模型。在知识发现和数据挖掘的进展（eds。

Kashima, H., Ide, T. & Peng, W.-C.) 180–191 (Springer Nature Switzerland, Cham, 2023). https://doi.org/10.1007/978-3-031-33380-4_14.Montoya-del-Angel, R., Sam-Millan, K., Vilanova, J. C. & Martí, R. MAM-E: Mammographic Synthetic Image Generation with Diffusion Models. Sensors 24, 2076 (2024).Article .

Kashima，H.，Ide，T.＆Peng，W.-C。）180-191（瑞士施普林格自然，查姆，2023）。https://doi.org/10.1007/978-3-031-33380-4_14.Montoya-del-Angel，R.，Sam Millan，K.，Vilanova，J.C。＆Martí，R.MAM-E：使用扩散模型生成乳腺X线合成图像。。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Xu, Y. et al. MedSyn: Text-guided Anatomy-aware Synthesis of High-Fidelity 3D CT Images. IEEE Trans. Med. Imaging 1–1 (2024) https://doi.org/10.1109/TMI.2024.3415032.Jiang, L., Mao, Y., Wang, X., Chen, X. & Li, C. CoLa-Diff: Conditional Latent Diffusion Model for Multi-modal MRI Synthesis.

Xu，Y。等。MedSyn：高保真3D CT图像的文本引导解剖感知合成。IEEE Trans。医学成像1-1（2024）https://doi.org/10.1109/TMI.2024.3415032.Jiang，L.，Mao，Y.，Wang，X.，Chen，X。＆Li，C。CoLa Diff：用于多模式MRI合成的条件潜在扩散模型。

in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 (eds. Greenspan, H. et al.) 398–408 (Springer Nature Switzerland, Cham, 2023). https://doi.org/10.1007/978-3-031-43999-5_38.Zhu, L. et al. Make-A-Volume: Leveraging Latent Diffusion Models for Cross-Modality 3D Brain MRI Synthesis.

在医学图像计算和计算机辅助干预方面–MICCAI 2023（eds。Greenspan，H。et al。）398–408（Springer Nature Switzerland，Cham，2023）。https://doi.org/10.1007/978-3-031-43999-5_38.Zhu，L。等人。Make-A-Volume：利用潜在扩散模型进行跨模态3D脑MRI合成。

in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 (eds. Greenspan, H. et al.) 592–601 (Springer Nature Switzerland, Cham, 2023). https://doi.org/10.1007/978-3-031-43999-5_56.Sun, S., Goldgof, G. M., Butte, A. & Alaa, A. M. Aligning Synthetic Medical Images with Clinical Knowledge using Human Feedback.Dorjsembe, Z., Pao, H.-K., Odonchimed, S.

在医学图像计算和计算机辅助干预方面–MICCAI 2023（eds。Greenspan，H。et al。）592–601（Springer Nature Switzerland，Cham，2023）。https://doi.org/10.1007/978-3-031-43999-5_56.Sun，S.，Goldgof，G.M.，Butte，A。＆Alaa，A.M。使用人类反馈将合成医学图像与临床知识对齐。Dorjsembe，Z.，Pao，H.-K.，Odonchimed，S。

& Xiao, F. Conditional Diffusion Models for Semantic 3D Brain MRI Synthesis. IEEE J. Biomed. Health Inform. 28, 4084–4093 (2024).Article .

&Xiao，F。语义3D脑MRI合成的条件扩散模型。IEEE J.生物医学。健康信息。284084-4093（2024）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Eschweiler, D. et al. Denoising diffusion probabilistic models for generation of realistic fully-annotated microscopy image datasets. PLOS Comput. Biol. 20, e1011890 (2024).Article

Eschweiler，D.等人。用于生成真实的全注释显微镜图像数据集的去噪扩散概率模型。PLOS计算机。生物学20，e1011890（2024）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Oh, H.-J. & Jeong, W.-K. DiffMix: Diffusion Model-Based Data Synthesis for Nuclei Segmentation and Classification in Imbalanced Pathology Image Datasets. in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 (eds. Greenspan, H. et al.) 337–345 (Springer Nature Switzerland, Cham, 2023).

Oh，H.-J.＆Jeong，W.-K。DiffMix：基于扩散模型的数据合成，用于不平衡病理图像数据集中的细胞核分割和分类。在医学图像计算和计算机辅助干预方面–MICCAI 2023（eds。Greenspan，H。et al。）337–345（Springer Nature Switzerland，Cham，2023）。

https://doi.org/10.1007/978-3-031-43898-1_33.Stojanovski, D., Hermida, U., Lamata, P., Beqiri, A. & Gomez, A. Echo from Noise: Synthetic Ultrasound Image Generation Using Diffusion Models for Real Image Segmentation. in Simplifying Medical Ultrasound (eds. Kainz, B. et al.) 34–43 (Springer Nature Switzerland, Cham, 2023).

https://doi.org/10.1007/978-3-031-43898-1_33.Stojanovski，D.，Hermida，U.，Lamata，P.，Beqiri，A。＆Gomez，A。来自噪声的回波：使用扩散模型进行真实图像分割的合成超声图像生成。简化医学超声（eds。Kainz，B。et al。）34-43（Springer Nature Switzerland，Cham，2023）。

https://doi.org/10.1007/978-3-031-44521-7_4.Zhao, X. & Hou, B. High-fidelity image synthesis from pulmonary nodule lesion maps using semantic diffusion model. in Medical imaging with deep learning, short paper track.Xing, X., Papanastasiou, G., Walsh, S. & Yang, G. Less Is More: Unsupervised Mask-Guided Annotated CT image synthesis with minimum manual segmentations.

https://doi.org/10.1007/978-3-031-44521-7_4.Zhao，X。＆Hou，B。使用语义扩散模型从肺结节病变图进行高保真图像合成。在医学成像与深度学习，短纸轨道。Xing，X.，Papanastasiou，G.，Walsh，S。＆Yang，G。Less Is More：无监督掩模引导的带注释的CT图像合成，具有最小的手动分割。

IEEE Trans. Med. Imaging 42, 2566–2576 (2023).Article .

IEEE Trans。医学成像422566-2576（2023）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Shrivastava, A. & Fletcher, P. T. NASDM: Nuclei-Aware Semantic Histopathology Image Generation Using Diffusion Models. in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 (eds. Greenspan, H. et al.) 786–796 (Springer Nature Switzerland, Cham, 2023). https://doi.org/10.1007/978-3-031-43987-2_76.Zhuang, Y.

Shrivastava，A。＆Fletcher，P.T。NASDM：使用扩散模型生成核感知语义组织病理学图像。在医学图像计算和计算机辅助干预方面–MICCAI 2023（eds。Greenspan，H。et al。）786–796（Springer Nature Switzerland，Cham，2023）。https://doi.org/10.1007/978-3-031-43987-2_76.Zhuang，是的。

et al. Semantic Image Synthesis for Abdominal CT. in Deep Generative Models (eds. Mukhopadhyay, A., Oksuz, I., Engelhardt, S., Zhu, D. & Yuan, Y.) 214–224 (Springer Nature Switzerland, Cham, 2024). https://doi.org/10.1007/978-3-031-53767-7_21.Chen, Q. et al. Towards generalizable tumor synthesis. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 11147–11158 (2024).Huy, P.

等人，《深度生成模型中腹部CT的语义图像合成》（eds.Mukhopadhyay，A.，Oksuz，I.，Engelhardt，S.，Zhu，D.＆Yuan，Y.）214-224（瑞士施普林格自然出版社，查姆，2024）。https://doi.org/10.1007/978-3-031-53767-7_21.Chen，Q。等人，走向可推广的肿瘤合成。在IEEE/CVF计算机视觉和模式识别会议论文集11147-11158（2024）中。休伊，P。

N. & Minh Quan, T. Denoising Diffusion Medical Models. in 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI) 1–5 https://doi.org/10.1109/ISBI53787.2023.10230674 (2023).Aversa, M. et al. Diffinfinite: Large mask-image synthesis via parallel random patch diffusion in histopathology. Adv.

N、 ＆Minh Quan，T。去噪扩散医学模型。2023年IEEE第20届生物医学成像国际研讨会（ISBI）1-5https://doi.org/10.1109/ISBI53787.2023.10230674（2023年）。Aversa，M。等人Diffinfinite：通过组织病理学中的平行随机贴片扩散进行大掩模图像合成。高级。

Neural Inf. Process. Syst. 36, (2024).Go, S., Ji, Y., Park, S. J. & Lee, S. Generation of structurally realistic retinal fundus images with diffusion models. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2335–2344 (2024).Macháček, R. et al. Mask-conditioned latent diffusion for generating gastrointestinal polyp images.

神经信息过程。系统。36，（2024年）。Go，S.，Ji，Y.，Park，S.J。＆Lee，S。使用扩散模型生成结构逼真的视网膜眼底图像。在IEEE/CVF计算机视觉和模式识别会议论文集2335-2344（2024）中。Macháček，R.等人。用于产生胃肠息肉图像的掩模条件潜伏扩散。

in Proceedings of the 4th ACM Workshop on Intelligent Cross-Data Analysis and Retrieval 1–9 (Association for Computing Machinery, New York, NY, USA, 2023). https://doi.org/10.1145/3592571.3592978.Han, K. et al. MedGen3D: A Deep Generative Framework for Paired 3D Image and Mask Generation. in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 (eds.

在第四届ACM智能交叉数据分析和检索研讨会论文集1-9（美国纽约州纽约市计算机械协会，2023年）中。https://doi.org/10.1145/3592571.3592978.Han，K。等人。MedGen3D：用于成对3D图像和掩模生成的深度生成框架。医学图像计算和计算机辅助干预-MICCAI 2023（编辑。

Greenspa.

格林斯帕。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Saragih, D. G., Hibi, A. & Tyrrell, P. N. Using diffusion models to generate synthetic labeled data for medical image segmentation. Int. J. Comput. Assist. Radiol. Surg. 19, 1615–1625 (2024).Article

Saragih，D.G.，Hibi，A。＆Tyrrell，P.N。使用扩散模型生成用于医学图像分割的合成标记数据。国际J.计算机。协助。放射性。外科杂志191615-1625（2024）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Crespi, L., Loiacono, D. & Sartori, P. Are 3D better than 2D Convolutional Neural Networks for Medical Imaging Semantic Segmentation? in 2022 International Joint Conference on Neural Networks (IJCNN) 1–8 (IEEE, Padua, Italy, 2022). https://doi.org/10.1109/IJCNN55064.2022.9892850.Wu, J.

Crespi，L.，Loiacono，D。＆Sartori，P。对于医学成像语义分割，3D优于2D卷积神经网络吗？。https://doi.org/10.1109/IJCNN55064.2022.9892850.Wu，J。

et al. Medsegdiff-v2: Diffusion-based medical image segmentation with transformer. In Proceedings of the AAAI conference on artificial intelligence. vol. 38 6030–6038 (2024).Micikevicius, P. et al. Mixed precision training. in 6th international conference on learning representations, ICLR 2018, Vancouver, BC, Canada, april 30–may 3, 2018, conference track proceedings (OpenReview.net, 2018).Song, J., Meng, C.

等人。Medsegdiff-v2：使用transformer的基于扩散的医学图像分割。在AAAI人工智能会议记录中。。Micikevicius，P。等人。混合精度训练。在2018年4月30日至5月3日于加拿大不列颠哥伦比亚省温哥华举行的第六届国际学习表征会议（ICLR 2018）上，会议记录（OpenReview.net，2018）。宋，J.，孟，C。

& Ermon, S. Denoising diffusion implicit models. in 9th international conference on learning representations, ICLR 2021, virtual event, austria, may 3–7, 2021 (OpenReview.net, 2021).Bao, F., Li, C., Zhu, J. & Zhang, B. Analytic-DPM: an analytic estimate of the optimal reverse variance in diffusion probabilistic models.

&。在2021年5月3日至7日于奥地利举行的第九届国际学习表征会议（ICLR 2021，virtual event）上（OpenReview.net，2021）。Bao，F.，Li，C.，Zhu，J。＆Zhang，B。分析DPM：扩散概率模型中最佳反向方差的分析估计。

in The tenth international conference on learning representations, ICLR 2022, virtual event, april 25-29, 2022 (OpenReview.net, 2022).Rombach, R., Blattmann, A., Lorenz, D., Esser, P. & Ommer, B. High-Resolution Image Synthesis with Latent Diffusion Models. in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 10674–10685 (IEEE, New Orleans, LA, USA, 2022).

在第十届学习表征国际会议上，ICLR 2022，虚拟活动，2022年4月25日至29日（OpenReview.net，2022）。Rombach，R.，Blattmann，A.，Lorenz，D.，Esser，P。＆Ommer，B。具有潜在扩散模型的高分辨率图像合成。2022年IEEE/CVF计算机视觉和模式识别会议（CVPR）10674-10685（IEEE，美国洛杉矶新奥尔良，2022）。

https://doi.org/10.1109/CVPR52688.2022.01042.Arora, A. & Arora, A. Synthetic patient data in health care: a widening legal loophole. Lancet 399, 1601–1602 (2022).Article .

https://doi.org/10.1109/CVPR52688.2022.01042.Arora。柳叶刀3991601-1602（2022）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Appenzeller, A., Leitner, M., Philipp, P., Krempel, E. & Beyerer, J. Privacy and utility of private synthetic data for medical data analyses. Appl. Sci. 12, 12320 (2022).Article

Appenzeller，A.，Leitner，M.，Philipp，P.，Krempel，E。＆Beyerer，J。用于医疗数据分析的私人合成数据的隐私和实用性。应用。科学。11212320（2022年）。文章

Google Scholar

谷歌学者

Giuffrè, M. & Shung, D. L. Harnessing the power of synthetic data in healthcare: innovation, application, and privacy. Npj Digit. Med. 6, 1–8 (2023).Article

Giuffrè，M.和Shung，D.L.利用医疗保健中合成数据的力量：创新，应用和隐私。。医学杂志6，1-8（2023）。文章

Google Scholar

谷歌学者

Teo, C., Abdollahzadeh, M. & Cheung, N.-M. M. On measuring fairness in generative models. Adv. Neural Inf. Process. Syst. 36, (2024).Chen, R. J., Lu, M. Y., Chen, T. Y., Williamson, D. F. K. & Mahmood, F. Synthetic data in machine learning for medicine and healthcare. Nat. Biomed. Eng.

Teo，C.，Abdollahzadeh，M。和Cheung，N.-M.M。关于测量生成模型中的公平性。高级神经信息处理。系统。36，（2024年）。Chen，R.J.，Lu，M.Y.，Chen，T.Y.，Williamson，D.F.K。＆Mahmood，F。医学和医疗保健机器学习中的合成数据。自然生物医学。发动机。

5, 493–497 (2021).Article .

5493-497（2021）。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Roden, A. C. et al. Distribution of mediastinal lesions across multi-institutional, international, radiology databases. J. Thorac. Oncol. 15, 568–579 (2020).Article

Roden，A.C.等人。纵隔病变在多机构，国际放射学数据库中的分布。J、 。Oncol公司。1568-579（2020）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Marx, A. et al. The 2015 World Health Organization Classification of Tumors of the Thymus: Continuity and Changes. J. Thorac. Oncol. 10, 1383–1395 (2015).Article

Marx，A.等人，《2015年世界卫生组织胸腺肿瘤分类：连续性和变化》。J、 。Oncol公司。101383-1395（2015）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Loshchilov, I. & Hutter, F. Decoupled weight decay regularization. in 7th international conference on learning representations, ICLR 2019, new orleans, LA, USA, may 6-9, 2019 (OpenReview.net, 2019).Deng, J. et al. ImageNet: A large-scale hierarchical image database. in 2009 IEEE Conference on Computer Vision and Pattern Recognition 248–255 (IEEE, Miami, FL, 2009).

Loshchilov，I。＆Hutter，F。解耦权重衰减正则化。在2019年5月6日至9日于美国洛杉矶新奥尔良举行的第七届国际学习表征会议（ICLR 2019）上（OpenReview.net，2019）。Deng，J。等人。ImageNet：一个大规模的分层图像数据库。在2009年IEEE计算机视觉和模式识别会议248-255（IEEE，迈阿密，佛罗里达州，2009年）。

https://doi.org/10.1109/CVPR.2009.5206848.Download referencesAcknowledgementsThis study was supported by the National Key R&D Program of China (2018YFA0704000), the National Natural Science Foundation of China (61822111, 62021002), and the Zhejiang Provincial Natural Science Foundation (LDT23F02024F02).

https://doi.org/10.1109/CVPR.2009.5206848.Download参考文献致谢本研究得到了国家重点研发计划（2018YFA0704000），国家自然科学基金（618221162012002）和浙江省自然科学基金（LDT23F02024F02）的支持。

This study was also supported by Institute for Brain and Cognitive Science, Tsinghua University (THUIBCS) and Beijing Laboratory of Brain and Cognitive Intelligence, Beijing Municipal Education Commission (BLBCI).Author informationAuthors and AffiliationsSchool of Software, Tsinghua University, Beijing, ChinaZhanping Zhou, Ruijie Tang & Feng XuBeijing National Research Center for Information Science and Technology (BNRist), Tsinghua University, Beijing, ChinaZhanping Zhou, Yuchen Guo, Ruijie Tang & Feng XuDepartment of Thoracic Oncology and Surgery, China State Key Laboratory of Respiratory Disease & National Clinical Research Center for Respiratory Disease, the First Affiliated Hospital of Guangzhou Medical University, Guangzhou, ChinaHengrui Liang & Jianxing HeAuthorsZhanping ZhouView author publicationsYou can also search for this author in.

这项研究也得到了清华大学脑与认知科学研究所（THUIBCS）和北京市教育委员会北京脑与认知智能实验室（BLBCI）的支持。作者信息作者和附属机构清华大学软件学院，北京，中国周占平，唐瑞杰和冯旭北京国家信息科学与技术研究中心（BNRist），清华大学，北京，中国周占平，郭玉晨，唐瑞杰和冯旭中国呼吸系统疾病国家重点实验室和国家呼吸系统疾病临床研究中心胸腔肿瘤与外科，广州医科大学第一附属医院，广州，中国梁恒瑞和建兴健康作者周占平观点作者出版物您也可以在中搜索这位作者。

PubMed Google ScholarYuchen GuoView author publicationsYou can also search for this author in

PubMed Google ScholarYuchen GuoView作者出版物您也可以在

PubMed Google ScholarRuijie TangView author publicationsYou can also search for this author in

PubMed Google ScholarRuijie TangView作者出版物您也可以在

PubMed Google ScholarHengrui LiangView author publicationsYou can also search for this author in

PubMed谷歌学者梁恒瑞查看作者出版物您也可以在

PubMed Google ScholarJianxing HeView author publicationsYou can also search for this author in

PubMed谷歌学者Jianxing HeView作者出版物您也可以在

PubMed Google ScholarFeng XuView author publicationsYou can also search for this author in

PubMed Google ScholarFeng XuView作者出版物您也可以在

PubMed Google ScholarContributionsZ.Z., Y.G., and F.X. contributed to the conceptualization. H.L. and J.H. supervised the data collection. Z.Z. and R.T. processed the collected data. Z.Z. developed the algorithm, conducted the experiments and performed the analysis. Z.Z. and Y.G.

PubMed谷歌学术贡献。Z、 ，Y.G.和F.X.为概念化做出了贡献。H、 L.和J.H.监督了数据收集。Z、 Z.和R.T.处理了收集的数据。Z、 Z.开发了算法，进行了实验并进行了分析。Z、 Z.和Y.G。

wrote the manuscript. Y.G. and F.X. supervised the study. All authors have read and approved the manuscript.Corresponding authorsCorrespondence to.

写了手稿。Y、 G.和F.X.监督了这项研究。所有作者都阅读并批准了手稿。通讯作者通讯。

Yuchen Guo or Feng Xu.Ethics declarations

郭玉臣或冯旭。道德宣言

Competing interests

相互竞争的利益

The authors declare no competing interests.

。

Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Supplementary informationSupplementary MaterialReporting SummaryRights and permissions

Additional informationPublisher的注释Springer Nature在已发布的地图和机构隶属关系中的管辖权主张方面保持中立。补充信息补充材料报告摘要权利和权限

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material.

。

You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.

根据本许可证，您无权共享源自本文或其部分的改编材料。本文中的图像或其他第三方材料包含在文章的知识共享许可证中，除非该材料的信用额度中另有说明。如果材料未包含在文章的知识共享许可中，并且您的预期用途不受法律法规的许可或超出许可用途，则您需要直接获得版权所有者的许可。

To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/..

要查看此许可证的副本，请访问http://creativecommons.org/licenses/by-nc-nd/4.0/..

Reprints and permissionsAbout this articleCite this articleZhou, Z., Guo, Y., Tang, R. et al. Privacy enhancing and generalizable deep learning with synthetic data for mediastinal neoplasm diagnosis.

转载和许可本文引用本文Zhou，Z.，Guo，Y.，Tang，R。等人。隐私增强和可推广的纵隔肿瘤诊断综合数据深度学习。

npj Digit. Med. 7, 293 (2024). https://doi.org/10.1038/s41746-024-01290-7Download citationReceived: 08 May 2024Accepted: 07 October 2024Published: 20 October 2024DOI: https://doi.org/10.1038/s41746-024-01290-7Share this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard.

npj数字。医学7293（2024）。https://doi.org/10.1038/s41746-024-01290-7Download引文接收日期：2024年5月8日接受日期：2024年10月7日发布日期：2024年10月20日OI：https://doi.org/10.1038/s41746-024-01290-7Share本文与您共享以下链接的任何人都可以阅读此内容：获取可共享链接对不起，本文目前没有可共享的链接。复制到剪贴板。

Provided by the Springer Nature SharedIt content-sharing initiative

由Springer Nature SharedIt内容共享计划提供