WO2024108652A1 - 基于傅里叶增强的半监督支气管分割方法及其系统 - Google Patents

基于傅里叶增强的半监督支气管分割方法及其系统 Download PDF

Info

Publication number
WO2024108652A1
WO2024108652A1 PCT/CN2022/137049 CN2022137049W WO2024108652A1 WO 2024108652 A1 WO2024108652 A1 WO 2024108652A1 CN 2022137049 W CN2022137049 W CN 2022137049W WO 2024108652 A1 WO2024108652 A1 WO 2024108652A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
enhancement
supervised
amplitude
semi
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2022/137049
Other languages
English (en)
French (fr)
Inventor
袁韵怡
王琼
钱银玲
廖祥云
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Publication of WO2024108652A1 publication Critical patent/WO2024108652A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting

Definitions

  • the present invention relates to the technical field of bronchial segmentation, and in particular to a semi-supervised bronchial segmentation method based on Fourier enhancement and a system thereof.
  • bronchial segmentation especially the segmentation of distal bronchioles, is of great significance.
  • Existing bronchial segmentation methods can be divided into traditional methods and deep learning-based methods.
  • the main technical problem solved by the present invention is to provide a semi-supervised bronchus segmentation method based on Fourier enhancement, which utilizes a semi-supervised learning method and a data enhancement method based on Fourier transform to improve the generalization performance of the existing model from the perspective of expanding the data set and improving data enhancement, thereby further improving the segmentation results; a semi-supervised bronchus segmentation system based on Fourier enhancement is also provided.
  • a technical solution adopted by the present invention is: to provide a semi-supervised bronchial segmentation method based on Fourier enhancement, which includes:
  • Step S1 Divide the labeled data into a training set and a validation set, and train the optimal initial model
  • Step S2 Use the initial model to predict the unlabeled data after weak data enhancement, and use the prediction results as pseudo labels to form a new data set;
  • Step S3 The training set and the dataset containing pseudo labels are merged into a new training set.
  • Fourier transform data enhancement is used to train the optimal model.
  • Step S4 Use the optimal model to predict the unknown test data to obtain the final predicted segmentation result.
  • step S3 the input data of the data set containing pseudo labels is clipped into data blocks of the same shape, and after the amplitude information is mixed and modified in the frequency domain using fast Fourier transform, the data restored back to the time domain is regularized for consistency with the unlabeled data that has undergone weak data enhancement processing.
  • the phase and amplitude of the original data are obtained by fast Fourier transform, and then the phase is kept unchanged, the amplitude is mixed with the amplitude of other data, and finally the original phase is combined with the mixed amplitude and inverse fast Fourier transform is performed to obtain the extended data after strong data enhancement.
  • the original data is a data set containing pseudo labels that is cut into data blocks of the same shape.
  • Semi-supervised bronchial segmentation system based on Fourier enhancement including:
  • the training module is used to divide the labeled data into training sets and validation sets, and train the optimal initial model
  • the expansion module is used to predict the unlabeled data after weak data enhancement through the initial model, and use the prediction results as pseudo labels to form a new data set;
  • the enhancement module is used to merge the training set with the dataset containing pseudo labels into a new training set.
  • Fourier transform data enhancement is used to train the optimal model.
  • the segmentation module is used to predict the unknown test data with the optimal model to obtain the final predicted segmentation result.
  • the enhancement module cuts the input data of the data set containing pseudo labels into data blocks of the same shape, and then uses fast Fourier transform to mix and modify the amplitude information in the frequency domain, and then restores the data back to the time domain and performs consistency regularization with the unlabeled data that has undergone weak data enhancement processing.
  • the enhancement module obtains the phase and amplitude of the original data through fast Fourier transform, then keeps the phase unchanged, mixes the amplitude with the amplitude of other data, and finally combines the original phase with the mixed amplitude and performs inverse fast Fourier transform to obtain the extended data after strong data enhancement.
  • the original data is a data set containing pseudo labels that is cut into data blocks of the same shape.
  • the present invention utilizes a semi-supervised learning method and a data enhancement method based on Fourier transform to improve the generalization performance of the existing model from the perspective of expanding the data set and improving data enhancement, thereby further improving the segmentation results.
  • FIG1 is a flowchart of the steps of the semi-supervised bronchial segmentation method of the present invention.
  • FIG2 is a schematic diagram of the bronchial segmentation network structure of the present invention.
  • FIG3 is a schematic diagram of a semi-supervised learning network structure of the present invention.
  • FIG. 4 is a schematic diagram of a data enhancement method based on FFT according to the present invention.
  • the semi-supervised bronchial segmentation method based on Fourier enhancement of the present invention includes:
  • Step S1 Divide the labeled data into a training set and a validation set, and train the optimal initial model
  • Step S2 Use the initial model to predict the unlabeled data after weak data enhancement, and use the prediction results as pseudo labels to form a new data set;
  • Step S3 The training set and the dataset containing pseudo labels are merged into a new training set.
  • Fourier transform data enhancement is used to train the optimal model.
  • Step S4 Use the optimal model to predict the unknown test data to obtain the final predicted segmentation result.
  • the present invention utilizes a semi-supervised learning method and a data enhancement method based on Fourier transform to improve the generalization performance of the existing model from the perspective of expanding the data set and improving data enhancement, thereby further improving the segmentation result.
  • the bronchus segmentation network structure in FIG. 2 i.e. the U-shaped neural network structure shown in FIG. 3
  • a deep supervision mechanism for supervised training i.e. the U-shaped neural network structure shown in FIG. 3
  • the trained learning network is used to predict the unlabeled data, and the prediction result is the pseudo label, which will be directly used as a label in subsequent new training; finally, semi-supervised training is performed as shown in FIG.
  • the input data of the pseudo-label data set is trimmed into data blocks of the same shape, and the amplitude information is mixed and modified in the frequency domain using the fast Fourier transform (FFT), and then the data restored to the time domain is regularized for consistency with the data processed using weak data enhancement.
  • FFT fast Fourier transform
  • step S3 the input data of the dataset containing pseudo labels is clipped into data blocks of the same shape, and after the amplitude information is mixed and modified in the frequency domain using fast Fourier transform, the data restored to the time domain is regularized for consistency with the unlabeled data that has undergone weak data enhancement processing.
  • the consistency regularization method encourages the predictions of unlabeled data before and after perturbation to be the same, so that the decision boundary of learning is located in the low-density area; pseudo-label refers to a false label.
  • the pseudo-label method creates pseudo-labels for unlabeled data and then adds them to model training to obtain a better decision boundary; the steps of using the pseudo-label method are as follows:
  • step 1) The training set in step 1) is combined with the pseudo-labeled data set generated in step 2) as a new training set to train the optimal model;
  • step 3 a special consistency regularization method is designed for the new training set.
  • the model training consistent with step 1) is performed for the original labeled data; for the data with pseudo-labels, FFT-based frequency domain data enhancement is used to obtain new data after data expansion, and the prediction results of the data after model learning are consistent with the pseudo-labels obtained in step 2).
  • the phase and amplitude of the original data are obtained by fast Fourier transform, and then the phase is kept unchanged, the amplitude is mixed with the amplitude of other data, and finally the original phase is combined with the mixed amplitude and an inverse fast Fourier transform is performed, so as to obtain the extended data after strong data enhancement.
  • the frequency domain data enhancement method based on FFT is as follows: the phase information in the frequency domain is usually related to high-frequency details and structural features, while the amplitude information is usually related to low-frequency semantic information.
  • the phase and amplitude of the original data are first obtained by FFT, and then the phase is kept unchanged, the amplitude is mixed with the amplitude of other data with the mixing coefficient ⁇ in Figure 4, and finally the original phase is combined with the mixed amplitude and an inverse fast Fourier transform (IFFT) is performed, so as to obtain the extended data after strong data enhancement;
  • the original data is a data set containing pseudo-labels that is trimmed into data blocks of the same shape.
  • the present invention provides a semi-supervised bronchial segmentation system based on Fourier enhancement, comprising:
  • the training module is used to divide the labeled data into training sets and validation sets, and train the optimal initial model
  • the expansion module is used to predict the unlabeled data after weak data enhancement through the initial model, and use the prediction results as pseudo labels to form a new data set;
  • the enhancement module is used to merge the training set with the dataset containing pseudo labels into a new training set.
  • Fourier transform data enhancement is used to train the optimal model.
  • the segmentation module is used to predict the unknown test data with the optimal model to obtain the final predicted segmentation result.
  • the enhancement module cuts the input data of the dataset containing pseudo-labels into data blocks of the same shape, and then uses fast Fourier transform to mix and modify the amplitude information in the frequency domain, and then restores the data back to the time domain for consistency regularization with the unlabeled data that has undergone weak data enhancement processing.
  • the enhancement module obtains the phase and amplitude of the original data through fast Fourier transform, then keeps the phase unchanged, mixes the amplitude with the amplitude of other data, and finally combines the original phase with the mixed amplitude and performs inverse fast Fourier transform to obtain the extended data after strong data enhancement;
  • the original data is the data set containing pseudo-labels that is trimmed into data blocks of the same shape.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Image Analysis (AREA)

Abstract

本发明涉及支气管分割技术领域，特别涉及基于傅里叶增强的半监督支气管分割方法及其系统；本发明先将有标签的数据分为训练集和验证集，并训练出最优的初始模型；用初始模型对经过弱数据增强处理的无标签数据进行预测，将预测结果作为伪标签构成新的数据集；再将训练集与含伪标签的数据集融合成新的训练集，在新的训练集内，采用傅里叶变换的数据增强方式，进而训练出最优的模型；用最优的模型对未知的测试数据进行预测，得到最终的预测分割结果；本发明利用半监督学习方法及基于傅里叶变换的数据增强方式，从扩充数据集的角度和改进数据增强的角度提高现有模型的泛化性能，从而进一步改善分割结果。

Description

基于傅里叶增强的半监督支气管分割方法及其系统 技术领域
本发明涉及支气管分割技术领域，特别涉及基于傅里叶增强的半监督支气管分割方法及其系统。
 背景技术
在临床诊断中，支气管分割尤其是远端细支气管的分割具有重要意义，现有的支气管分割方法可以分为传统方法和基于深度学习的方法。
在2009年的一次支气管分割挑战赛EXACT中，主办方对参与比赛的15种算法进行了比较，其中大部分算法采用如区域生长、模板匹配、梯度矢量流等传统方法，虽然这些算法可以很好地分割出例如气管、主支气管等厚管状结构，然而由于存在缺乏语义特征，且远端气道壁模糊等问题，限制了他们对远端细支气管的分割。
由于训练深度学习模型使用的数据量较少时，模型性能难以达到理想的状态，而目前可用于支气管分割的含标签数据严重稀缺。
 技术问题
本发明主要解决的技术问题是提供基于傅里叶增强的半监督支气管分割方法，其利用半监督学习方法及基于傅里叶变换的数据增强方式，从扩充数据集的角度和改进数据增强的角度提高现有模型的泛化性能，从而进一步改善分割结果；还提供了基于傅里叶增强的半监督支气管分割系统。
 技术解决方案
为解决上述技术问题，本发明采用的一个技术方案是：提供基于傅里叶增强的半监督支气管分割方法，其中，包括：
步骤S1、将有标签的数据分为训练集和验证集，并训练出最优的初始模型；
步骤S2、用初始模型对经过弱数据增强处理的无标签数据进行预测，将预测结果作为伪标签构成新的数据集；
步骤S3、将训练集与含伪标签的数据集融合成新的训练集，在新的训练集内，采用傅里叶变换的数据增强方式，进而训练出最优的模型；
步骤S4、用最优的模型对未知的测试数据进行预测，得到最终的预测分割结果。
作为本发明的一种改进，在步骤S3内，将含伪标签的数据集的输入数据剪裁成相同形状的数据块，使用快速傅里叶变换在频域上混合修改振幅信息后，恢复回时域的数据与经过弱数据增强处理的无标签数据进行一致性正则化。
作为本发明的进一步改进，通过快速傅里叶变换获得原始数据的相位和振幅，然后保留相位不变，将振幅与其他数据的振幅进行混合，最后将原始的相位与混合后的振幅结合并进行快速傅里叶逆变换，则得到进行强数据增强后的扩展数据。
作为本发明的进一步改进，原始数据为含伪标签的数据集剪裁之后成相同形状的数据块。
基于傅里叶增强的半监督支气管分割系统，其中，包括：
训练模块，用于将有标签的数据分为训练集和验证集，并训练出最优的初始模型；
扩充模块，用于通过初始模型对经过弱数据增强处理的无标签数据进行预测，将预测结果作为伪标签构成新的数据集；
增强模块，用于将训练集与含伪标签的数据集融合成新的训练集，在新的训练集内，采用傅里叶变换的数据增强方式，进而训练出最优的模型；
分割模块，用于将最优的模型对未知的测试数据进行预测，得到最终的预测分割结果。
作为本发明的一种改进，所述增强模块将含伪标签的数据集的输入数据剪裁成相同形状的数据块，再使用快速傅里叶变换在频域上混合修改振幅信息后，恢复回时域的数据与经过弱数据增强处理的无标签数据进行一致性正则化。
作为本发明的进一步改进，所述增强模块通过快速傅里叶变换获得原始数据的相位和振幅，然后保留相位不变，将振幅与其他数据的振幅进行混合，最后将原始的相位与混合后的振幅结合并进行快速傅里叶逆变换，则得到进行强数据增强后的扩展数据。
作为本发明的更进一步改进，原始数据为含伪标签的数据集剪裁之后成相同形状的数据块。
 有益效果
与现有技术相比，本发明利用半监督学习方法及基于傅里叶变换的数据增强方式，从扩充数据集的角度和改进数据增强的角度提高现有模型的泛化性能，从而进一步改善分割结果。
 附图说明
图1为本发明的半监督支气管分割方法的步骤框图；
图2为本发明的支气管分割网络结构示意图；
图3为本发明的半监督学习网络结构示意图；
图4为本发明基于FFT的数据增强方式示意图。
 本发明的实施方式
请参照图1至图4，本发明的基于傅里叶增强的半监督支气管分割方法，包括：
步骤S1、将有标签的数据分为训练集和验证集，并训练出最优的初始模型；
步骤S2、用初始模型对经过弱数据增强处理的无标签数据进行预测，将预测结果作为伪标签构成新的数据集；
步骤S3、将训练集与含伪标签的数据集融合成新的训练集，在新的训练集内，采用傅里叶变换的数据增强方式，进而训练出最优的模型；
步骤S4、用最优的模型对未知的测试数据进行预测，得到最终的预测分割结果。
本发明利用半监督学习方法及基于傅里叶变换的数据增强方式，从扩充数据集的角度和改进数据增强的角度提高现有模型的泛化性能，从而进一步改善分割结果。
在本发明内，首先使用图2中的支气管分割网络结构即图3所示U形神经网络结构，结合深监督机制进行监督训练，然后生成含伪标签的数据集，具体是用训练好的学习网络预测未含标签的数据，预测结果即为伪标签，伪标签将在后续新的训练中直接作为标签使用；最后如图3所示进行半监督训练，保持原本的监督训练不变，同时将伪标签数据集的输入数据剪裁成相同形状的数据块，使用快速傅里叶变换（FFT）在频域上混合修改振幅信息后，恢复回时域的数据与使用弱数据增强处理后的数据进行一致性正则化。
在本发明内，在步骤S3内，将含伪标签的数据集的输入数据剪裁成相同形状的数据块，使用快速傅里叶变换在频域上混合修改振幅信息后，恢复回时域的数据与经过弱数据增强处理的无标签数据进行一致性正则化。
具体地讲，一致性正则化方法通过鼓励无标签数据扰动前后的预测相同使学习的决策边界位于低密度区域；伪标签指的是虚假的标签，伪标签方法通过对无标签数据制作伪标签然后加入模型训练，以得到更好的决策边界；使用伪标签方法的步骤如下：
1）将有标签部分数据分为训练集和验证集，并训练出最优的初始模型；
2）用初始模型对经过弱数据增强处理的无标签数据进行预测，将预测结果作为伪标签，即为图3中的Yw；
3）将步骤1）中的训练集与步骤2）产生的伪标签数据集融合作为新的训练集，训练出最优的模型；
4）再用最优的模型对未知的测试数据进行预测，得到最终的预测结果。
其中，在步骤3）中针对新的训练集设计了特殊的一致性正则化方法，如图3所示，对于原始的有标签数据执行与步骤1）一致的模型训练；对于含伪标签的数据，使用基于FFT的频域数据增强得到数据扩展后新数据，该数据经过模型学习的预测结果与步骤2）得到的伪标签进行一致性正则化。
在本发明内，通过快速傅里叶变换获得原始数据的相位和振幅，然后保留相位不变，将振幅与其他数据的振幅进行混合，最后将原始的相位与混合后的振幅结合并进行快速傅里叶逆变换，则得到进行强数据增强后的扩展数据，具体地讲，基于FFT的频域数据增强方式为：频域的相位信息通常与高频细节及结构特征相关，而振幅信息通常与低频语义信息相关，因此为了获得与原始数据相差较大但支气管结构仍能保持不变的扩展数据，首先使用FFT获得原始数据的相位和振幅，然后保留相位不变，将振幅与其他数据的振幅以图4中的混合系数λ进行混合，最后将原始的相位与混合后的振幅结合并进行快速傅里叶逆变换（IFFT），即可得到进行强数据增强后的扩展数据；在本发明内，原始数据为含伪标签的数据集剪裁之后成相同形状的数据块。
本发明提供了基于傅里叶增强的半监督支气管分割系统，包括：
训练模块，用于将有标签的数据分为训练集和验证集，并训练出最优的初始模型；
扩充模块，用于通过初始模型对经过弱数据增强处理的无标签数据进行预测，将预测结果作为伪标签构成新的数据集；
增强模块，用于将训练集与含伪标签的数据集融合成新的训练集，在新的训练集内，采用傅里叶变换的数据增强方式，进而训练出最优的模型；
分割模块，用于将最优的模型对未知的测试数据进行预测，得到最终的预测分割结果。
其中，增强模块将含伪标签的数据集的输入数据剪裁成相同形状的数据块，再使用快速傅里叶变换在频域上混合修改振幅信息后，恢复回时域的数据与经过弱数据增强处理的无标签数据进行一致性正则化。
在本发明内，增强模块通过快速傅里叶变换获得原始数据的相位和振幅，然后保留相位不变，将振幅与其他数据的振幅进行混合，最后将原始的相位与混合后的振幅结合并进行快速傅里叶逆变换，则得到进行强数据增强后的扩展数据；原始数据为含伪标签的数据集剪裁之后成相同形状的数据块。
以上所述仅为本发明的实施方式，并非因此限制本发明的专利范围，凡是利用本发明说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本发明的专利保护范围内。

Claims (8)

  1. 基于傅里叶增强的半监督支气管分割方法，其特征在于，包括：
    步骤S1、将有标签的数据分为训练集和验证集，并训练出最优的初始模型；
    步骤S2、用初始模型对经过弱数据增强处理的无标签数据进行预测，将预测结果作为伪标签构成新的数据集；
    步骤S3、将训练集与含伪标签的数据集融合成新的训练集，在新的训练集内，采用傅里叶变换的数据增强方式，进而训练出最优的模型；
    步骤S4、用最优的模型对未知的测试数据进行预测，得到最终的预测分割结果。
  2. 根据权利要求1所述的基于傅里叶增强的半监督支气管分割方法，其特征在于，在步骤S3内，将含伪标签的数据集的输入数据剪裁成相同形状的数据块，使用快速傅里叶变换在频域上混合修改振幅信息后，恢复回时域的数据与经过弱数据增强处理的无标签数据进行一致性正则化。
  3. 根据权利要求2所述的基于傅里叶增强的半监督支气管分割方法，其特征在于，通过快速傅里叶变换获得原始数据的相位和振幅，然后保留相位不变，将振幅与其他数据的振幅进行混合，最后将原始的相位与混合后的振幅结合并进行快速傅里叶逆变换，则得到进行强数据增强后的扩展数据。
  4. 根据权利要求3所述的基于傅里叶增强的半监督支气管分割方法，其特征在于，原始数据为含伪标签的数据集剪裁之后成相同形状的数据块。
  5. 基于傅里叶增强的半监督支气管分割系统，其特征在于，包括：
    训练模块，用于将有标签的数据分为训练集和验证集，并训练出最优的初始模型；
    扩充模块，用于通过初始模型对经过弱数据增强处理的无标签数据进行预测，将预测结果作为伪标签构成新的数据集；
    增强模块，用于将训练集与含伪标签的数据集融合成新的训练集，在新的训练集内，采用傅里叶变换的数据增强方式，进而训练出最优的模型；
    分割模块，用于将最优的模型对未知的测试数据进行预测，得到最终的预测分割结果。
  6. 根据权利要求5所述的基于傅里叶增强的半监督支气管分割系统，其特征在于，所述增强模块将含伪标签的数据集的输入数据剪裁成相同形状的数据块，再使用快速傅里叶变换在频域上混合修改振幅信息后，恢复回时域的数据与经过弱数据增强处理的无标签数据进行一致性正则化。
  7. 根据权利要求6所述的基于傅里叶增强的半监督支气管分割系统，其特征在于，所述增强模块通过快速傅里叶变换获得原始数据的相位和振幅，然后保留相位不变，将振幅与其他数据的振幅进行混合，最后将原始的相位与混合后的振幅结合并进行快速傅里叶逆变换，则得到进行强数据增强后的扩展数据。
  8. 根据权利要求7所述的基于傅里叶增强的半监督支气管分割系统，其特征在于，原始数据为含伪标签的数据集剪裁之后成相同形状的数据块。
PCT/CN2022/137049 2022-11-25 2022-12-06 基于傅里叶增强的半监督支气管分割方法及其系统 Ceased WO2024108652A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211489719.7 2022-11-25
CN202211489719.7A CN115861340A (zh) 2022-11-25 2022-11-25 基于傅里叶增强的半监督支气管分割方法及其系统

Publications (1)

Publication Number Publication Date
WO2024108652A1 true WO2024108652A1 (zh) 2024-05-30

Family

ID=85666422

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/137049 Ceased WO2024108652A1 (zh) 2022-11-25 2022-12-06 基于傅里叶增强的半监督支气管分割方法及其系统

Country Status (2)

Country Link
CN (1) CN115861340A (zh)
WO (1) WO2024108652A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118071763B (zh) * 2024-04-16 2024-08-02 浙江大学 一种基于自训练的半监督三维形状分割方法和装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200065969A1 (en) * 2018-08-27 2020-02-27 Siemens Healthcare Gmbh Medical image segmentation from raw data using a deep attention neural network
US20220027741A1 (en) * 2020-07-20 2022-01-27 Korea Advanced Institute Of Science And Technology Unsupervised learning method for general inverse problem and apparatus therefor
CN115100063A (zh) * 2022-06-28 2022-09-23 大连海洋大学 基于自监督的水下图像增强方法、装置及计算机存储介质
CN115147426A (zh) * 2022-09-06 2022-10-04 北京大学 基于半监督学习的模型训练与图像分割方法和系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114022406A (zh) * 2021-09-15 2022-02-08 济南国科医工科技发展有限公司 一种半监督学习的图像分割方法、系统及终端
CN114519677B (zh) * 2022-02-17 2025-12-23 重庆紫光华山智安科技有限公司 数据增强方法、装置、电子设备及存储介质
CN114610967B (zh) * 2022-03-09 2024-10-22 东北大学 一种应用于用户画像领域的数据增广方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200065969A1 (en) * 2018-08-27 2020-02-27 Siemens Healthcare Gmbh Medical image segmentation from raw data using a deep attention neural network
US20220027741A1 (en) * 2020-07-20 2022-01-27 Korea Advanced Institute Of Science And Technology Unsupervised learning method for general inverse problem and apparatus therefor
CN115100063A (zh) * 2022-06-28 2022-09-23 大连海洋大学 基于自监督的水下图像增强方法、装置及计算机存储介质
CN115147426A (zh) * 2022-09-06 2022-10-04 北京大学 基于半监督学习的模型训练与图像分割方法和系统

Also Published As

Publication number Publication date
CN115861340A (zh) 2023-03-28

Similar Documents

Publication Publication Date Title
Guo et al. Accelerating diffusion models via pre-segmentation diffusion sampling for medical image segmentation
Li et al. Hierarchical consistency regularized mean teacher for semi-supervised 3d left atrium segmentation
CN113221902B (zh) 基于数据分布扩充的跨域自适应语义分割方法及系统
CN112861995A (zh) 基于模型无关元学习的无监督少样本图像分类方法、系统及存储介质
CN113222114A (zh) 一种图像数据增广方法及装置
Yang et al. Image synthesis under limited data: A survey and taxonomy
De Palma et al. IBP regularization for verified adversarial robustness via branch-and-bound
Kim et al. Feature statistics mixing regularization for generative adversarial networks
WO2024108652A1 (zh) 基于傅里叶增强的半监督支气管分割方法及其系统
WO2024113395A1 (zh) 一种快速磁共振成像方法及装置
Ye et al. Data augmentation method for insulators based on Cycle GAN
Yang et al. Inversion based on a detached dual-channel domain method for StyleGAN2 embedding
CN116524308B (zh) 基于生成对抗网络的图像处理方法及装置
Hu et al. Image style transfer based on generative adversarial network
CN116883668B (zh) 基于风格自适应和边界增强的域自适应医学语义分割方法
CN104778666A (zh) 一种基于谱聚类和混合模型聚类的图像分割方法
Yu et al. TtfDiffusion: Training-free and text-free image editing in diffusion models with structural and semantic disentanglement
Hu et al. A new framework for predicting tensile stress of natural rubber based on data augmentation and molecular dynamics simulation data
Huang et al. An unsupervised 3D image registration network for brain MRI deformable registration
CN118134950A (zh) 基于双扰动一致性的半监督支气管图像分割方法及装置
Ji et al. A dense-gated U-Net for brain lesion segmentation
CN103559587A (zh) 一种基于行为模式的模型变化区域责任分析系统
CN119919293A (zh) 基于联合频域先验引导扩散的零样本低光图像增强方法
Bouzid et al. 3D semantic segmentation of airway abnormalities on UTE-MRI with reinforcement learning on deep supervision
Wang et al. A feature regularization based meta-learning framework for generalizing prostate mri segmentation

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22966337

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22966337

Country of ref document: EP

Kind code of ref document: A1