CN112862805B - Acoustic neuroma image automatic segmentation method and system - Google Patents
Acoustic neuroma image automatic segmentation method and systemDownload PDFInfo
- Publication number
- CN112862805B CN112862805BCN202110238730.5ACN202110238730ACN112862805BCN 112862805 BCN112862805 BCN 112862805BCN 202110238730 ACN202110238730 ACN 202110238730ACN 112862805 BCN112862805 BCN 112862805B
- Authority
- CN
- China
- Prior art keywords
- image
- segmentation
- network
- magnetic resonance
- neuroma
- 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.)
- Expired - Fee Related
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Medical Informatics (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
Translated fromChinese
Description
Translated fromChinese技术领域technical field
本发明属于图像自动化分割技术领域，特别是基于深度神经网络的听神经瘤图像分割方法。The invention belongs to the technical field of automatic image segmentation, in particular to an acoustic neuroma image segmentation method based on a deep neural network.
背景技术Background technique
听神经瘤是一种并不罕见的颅内肿瘤，患者多为30岁以上的中年人，在肿瘤体积较小时常伴随着耳鸣、恶心等容易被忽视的症状，当发展到疾病中后期时由于肿瘤增大，压迫到了面部神经和小脑，就会出现较为严重的面瘫和四肢不协调，医生对这种良性肿瘤的治疗意见都是尽早切除。Acoustic neuroma is a not uncommon intracranial tumor. Patients are mostly middle-aged people over 30 years old. When the tumor is small in size, it is often accompanied by symptoms that are easily overlooked, such as tinnitus and nausea. As the tumor grows, it compresses the facial nerve and cerebellum, leading to severe facial paralysis and incoordination of limbs. The doctor's treatment advice for this benign tumor is to remove it as soon as possible.
核磁共振成像（Magnetic Resonance Imaging，MRI）在脑部诊疗中使用较为频繁，MR成像具有如下优势：1.MRI可以呈现多样的图像对比度；2.MRI可以实现较为完整的软组织成像；3.MRI分辨率较高。医生能够凭借MRI观察病患大脑、腹部等身体部位的详尽细节，这对疾病的诊疗十分有利。MRI影像一般使用T1、T1ce、T2、FLAIR四个序列，不同的序列可以显示不同的组织特征。在听神经瘤的治疗中，医生常使用T2序列加权的MRI影像作为诊断依据，在这个序列上的肿瘤区域相比周围组织值更大一些，视觉效果上更亮一些。Magnetic Resonance Imaging (MRI) is frequently used in brain diagnosis and treatment. MR imaging has the following advantages: 1. MRI can present a variety of image contrast; 2. MRI can achieve relatively complete soft tissue imaging; 3. MRI can distinguish The rate is higher. Doctors can use MRI to observe the patient's brain, abdomen and other body parts in detail, which is very beneficial to the diagnosis and treatment of diseases. MRI images generally use four sequences: T1, T1ce, T2, and FLAIR, and different sequences can display different tissue characteristics. In the treatment of acoustic neuroma, doctors often use T2 sequence-weighted MRI images as the basis for diagnosis. The tumor area on this sequence is larger than the surrounding tissue, and the visual effect is brighter.
通过医疗影像，医生可以根据自己的知识和经验对肿瘤区域进行人工标注，但在医疗资源分配不均、医疗影像数据连年增长的情况下，人工标注逐渐暴露出了标定周期长、标定精度参差不齐的问题，这也使得自动化标注成为了人们关注的重点。由于医学图像样式单一，界限模糊，传统的语义分割方法在医学图像分割上效果不佳且需要人为辅助，无法实现全自动化的标注。为减轻医生负担、提高分割准确率，近年来以深度学习为基础的计算机辅助诊疗技术逐渐得到了广泛应用，然而单一的分割网络性能提升有限，因此需要提出新的听神经瘤图像自动化分割方法提升分割性能。Through medical imaging, doctors can manually label tumor areas based on their own knowledge and experience. However, under the circumstances of uneven distribution of medical resources and increasing medical image data year after year, manual labeling gradually exposes the long calibration cycle and variable calibration accuracy. Qi problem, which also makes automatic labeling become the focus of people's attention. Due to the single style of medical images and blurred boundaries, traditional semantic segmentation methods are not effective in medical image segmentation and require human assistance, which cannot achieve fully automated annotation. In order to reduce the burden on doctors and improve the accuracy of segmentation, computer-aided diagnosis and treatment technology based on deep learning has been widely used in recent years. However, the performance improvement of a single segmentation network is limited. Therefore, it is necessary to propose a new automatic segmentation method for acoustic neuroma images to improve segmentation. performance.
发明内容Contents of the invention
本发明针对目前单一的听神经瘤图像分割网络性能提升有限的技术问题，提供一种听神经瘤图像自动化分割方法及系统。The present invention provides an automatic acoustic neuroma image segmentation method and system for the technical problem of limited performance improvement of the current single acoustic neuroma image segmentation network.
本发明为实现上述技术目的，采用了以下的技术方案。In order to achieve the above-mentioned technical purpose, the present invention adopts the following technical solutions.
一方面提供一种听神经瘤图像自动化分割方法，包括以下步骤：On the one hand, a method for automatic segmentation of acoustic neuroma images is provided, comprising the following steps:
获取听神经瘤患者的核磁共振图像；Obtaining MRI images of patients with acoustic neuromas;
利用预先训练的听神经瘤图像自动化分割模型对获取的听神经瘤患者的核磁共振图像进行自动化分割获得分割结果，所述听神经瘤图像自动化分割模型采用预处理网络和分割网络的级联结构。The acquired MRI images of acoustic neuroma patients are automatically segmented using a pre-trained automatic segmentation model for acoustic neuroma images to obtain segmentation results. The automatic segmentation model for acoustic neuroma images adopts a cascaded structure of a preprocessing network and a segmentation network.
进一步地，所述预处理网络采用pix2pixGAN网络，所述分割网络采用Unet分割网络，其中：Further, the preprocessing network adopts pix2pixGAN network, and the segmentation network adopts Unet segmentation network, wherein:
pix2pixGAN网络与Unet分割网络均为对称网络结构，均包括下采样部分和上采样部分，所述pix2pixGAN的生成器结构共26层，其下采样部分包含2个dropout层；The pix2pixGAN network and the Unet segmentation network are both symmetrical network structures, including a down-sampling part and an up-sampling part. The generator structure of the pix2pixGAN has a total of 26 layers, and its down-sampling part includes 2 dropout layers;
Unet分割网络共23层，其下采样部分包含10个卷积层，2个最大池化操作，上采样部分包括8个卷积层和4个反卷积层。The Unet segmentation network has a total of 23 layers. The downsampling part includes 10 convolutional layers, 2 maximum pooling operations, and the upsampling part includes 8 convolutional layers and 4 deconvolutional layers.
再进一步地，所述Unet分割网络的训练方法如下：获取听神经瘤患者脑部核磁共振影像样本的训练集；将训练集作为输入数据，该影像样本对应的标注分割结果作为目标数据，对所述Unet分割网络进行训练，多次迭代后选取损失值最小的Unet分割网络作为最终的Unet分割网络。Further, the training method of the Unet segmentation network is as follows: obtain the training set of the brain MRI image samples of patients with acoustic neuroma; use the training set as input data, and use the tagged segmentation results corresponding to the image samples as target data. The Unet segmentation network is trained, and after multiple iterations, the Unet segmentation network with the smallest loss value is selected as the final Unet segmentation network.
再进一步地，所述pix2pixGAN网络的训练方法如下：获取听神经瘤患者脑部核磁共振影像样本的训练集和验证集；基于训练集和Unet分割网络获取训练集中听神经瘤核磁共振影像对应的修正图像；将训练集中的听神经瘤核磁共振影像和对应的修正图像分别作为pix2pixGAN网络的输入数据和目标数据输入pix2pixGAN网络，多次迭代训练，得到生成器网络；使用测试集在所述听神经瘤图像自动化分割模型上进行测试，选取测试性能最优的生成器网络作为听神经瘤图像自动化分割模型的预处理网络。Further, the training method of the pix2pixGAN network is as follows: obtain the training set and verification set of the brain magnetic resonance image samples of patients with acoustic neuroma; obtain the corrected image corresponding to the acoustic neuroma nuclear magnetic resonance image in the training set based on the training set and the Unet segmentation network; The acoustic neuroma MRI images in the training set and the corresponding corrected images are respectively input into the pix2pixGAN network as the input data and target data of the pix2pixGAN network, and the generator network is obtained through multiple iterative training; the automatic segmentation model of the acoustic neuroma image is obtained using the test set The test was carried out on the test, and the generator network with the best test performance was selected as the preprocessing network for the automatic segmentation model of acoustic neuroma images.
再进一步地，获取听神经瘤核磁共振影像对应的修正图像的方法包括：Still further, the method for obtaining the corrected image corresponding to the nuclear magnetic resonance image of the acoustic neuroma includes:
计算训练集中听神经瘤患者脑部核磁共振影像在Unet分割网络中的Dice评估系数；Calculate the Dice evaluation coefficient of the brain MRI images of patients with acoustic neuroma in the training set in the Unet segmentation network;
计算Dice评估系数关于该核磁共振影像的偏微分，将该偏微分值作为当前核磁共振影像的修正值进行叠加；多次迭代得到听神经瘤脑部核磁共振影像最终的修正图像。The partial differential of the Dice evaluation coefficient with respect to the MRI image is calculated, and the partial differential value is superimposed as the correction value of the current MRI image; multiple iterations are performed to obtain the final corrected image of the brain MRI image of the acoustic neuroma.
另一方面，本发明提供了听神经瘤图像自动化分割系统，包括：数据获取模块和听神经瘤图像自动化分割模块，所述数据获取模块，用于获取听神经瘤患者的核磁共振图像；所述听神经瘤图像自动化分割模块用于利用预先训练的听神经瘤图像自动化分割模型对获取的听神经瘤患者的核磁共振图像进行自动化分割获得分割结果，所述听神经瘤图像自动化分割模型采用预处理网络和分割网络的级联结构On the other hand, the present invention provides an automatic acoustic neuroma image segmentation system, comprising: a data acquisition module and an acoustic neuroma image automatic segmentation module, the data acquisition module is used to acquire an MRI image of an acoustic neuroma patient; the acoustic neuroma image The automatic segmentation module is used to use the pre-trained acoustic neuroma image automatic segmentation model to automatically segment the acquired MRI image of the acoustic neuroma patient to obtain the segmentation result, and the acoustic neuroma image automatic segmentation model adopts the cascade of the preprocessing network and the segmentation network structure
本发明还提供了一种计算机可读存储介质，包括：至少一个处理器以及与所述至少一个处理器通信连接的存储器，其中所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使得所述至少一个处理器能够执行如以上技术方案任意一种可能的实时方式所提供的听神经瘤图像自动化分割方法。The present invention also provides a computer-readable storage medium, including: at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, The instructions are executed by the at least one processor, so that the at least one processor can execute the automatic acoustic neuroma image segmentation method provided in any possible real-time manner of the above technical solution.
本发明所取得的有益技术效果：Beneficial technical effect that the present invention obtains:
本发明实施例通过构建以提升模型分割性能为目标的预处理网络提升了模型的分割性能，突破了单一Unet分割模型的局限性，获得了当前最优的听神经瘤图像自动化分割模型，提升了经瘤图像分割方法的性能。The embodiment of the present invention improves the segmentation performance of the model by constructing a preprocessing network aimed at improving the segmentation performance of the model, breaks through the limitations of a single Unet segmentation model, and obtains the current optimal automatic segmentation model for acoustic neuroma images, which improves the performance of the model. Performance of Tumor Image Segmentation Methods.
同时使用后处理模块融合分割掩码与核磁共振图像，使分割结果更加直观，医生可以凭借该结果对肿瘤位置及特征进行较为精确的诊断，有效解决了人工标注标定周期长、标定精度参差不齐的问题。本发明实施例在计算机终端上能以较快的速度实现听神经瘤图像的自动化分割，大幅提升了听神经瘤图像的分割效率，有效简化了诊疗流程。At the same time, the post-processing module is used to fuse the segmentation mask and the MRI image to make the segmentation result more intuitive. Doctors can use the result to make a more accurate diagnosis of the tumor location and characteristics, effectively solving the problem of long manual labeling and calibration cycles and uneven calibration accuracy. The problem. The embodiment of the present invention can realize the automatic segmentation of the acoustic neuroma image at a relatively fast speed on the computer terminal, greatly improves the segmentation efficiency of the acoustic neuroma image, and effectively simplifies the diagnosis and treatment process.
附图说明Description of drawings
图1为本发明实施例的基于梯度下降原理的预处理网络和Unet分割网络级联的听神经瘤图像自动化分割方法流程图；Fig. 1 is the flow chart of the automatic segmentation method for acoustic neuroma image of the cascaded preprocessing network and Unet segmentation network based on the gradient descent principle of the embodiment of the present invention;
图2为本发明实施例的具体分割方案流程图；Fig. 2 is the flow chart of the specific segmentation scheme of the embodiment of the present invention;
图3为听神经瘤患者的二维脑部核磁共振影像；Figure 3 is a two-dimensional brain magnetic resonance image of a patient with acoustic neuroma;
图4为图3对应的真实标注图；Fig. 4 is the real annotation map corresponding to Fig. 3;
图5为本发明实施例Unet分割网络结构图；Fig. 5 is the Unet segmentation network structural diagram of the embodiment of the present invention;
图6为本发明实施例Unet分割网络训练流程图；Fig. 6 is the Unet segmentation network training flowchart of the embodiment of the present invention;
图7为本发明实施例预处理网络结构图；FIG. 7 is a structural diagram of a preprocessing network according to an embodiment of the present invention;
图8为本发明示例预处理网络训练流程图；Fig. 8 is the example preprocessing network training flowchart of the present invention;
图9为本发明实施例自动化分割模型结构图；9 is a structural diagram of an automatic segmentation model according to an embodiment of the present invention;
图10为本发明实施例得到的分割效果示例图。FIG. 10 is an example diagram of the segmentation effect obtained by the embodiment of the present invention.
具体实施方式detailed description
为使本发明的目的、技术方案和优点更加清楚明白，以下结合具体实施例和说明书附图对本发明进一步详细说明。In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with specific embodiments and accompanying drawings.
如图1，本发明实施例提供了一种听神经瘤图像自动化分割方法，包括以下步骤：获取听神经瘤患者的脑部核磁共振成像；As shown in Fig. 1 , an embodiment of the present invention provides a method for automatic segmentation of an acoustic neuroma image, comprising the following steps: acquiring brain magnetic resonance imaging of an acoustic neuroma patient;
使用基于深度学习方法的听神经瘤图像自动化分割模型对获取的核磁共振图像进行自动化分割，所述听神经瘤图像自动化分割模型采用基于梯度下降原理的预处理网络和Unet分割网络的级联结构。The acquired nuclear magnetic resonance images were automatically segmented using an automatic acoustic neuroma image segmentation model based on a deep learning method. The automatic acoustic neuroma image segmentation model used a cascade structure of a preprocessing network and a Unet segmentation network based on the gradient descent principle.
具体地，肿瘤的自动化标注可以被认为是一种图像分割问题，确切地讲，图像分割是对像素进行类别判断的过程，在肿瘤分割中，每个像素只有肿瘤和非肿瘤两种标签，因此肿瘤分割可以看作是对像素进行二分类。由于医学图像样式单一，界限模糊，传统的语义分割方法在医学图像分割上效果不佳且需要人为辅助，无法实现全自动化的标注。为减轻医生负担、提高分割准确率，近年来以深度学习为基础的计算机辅助诊疗技术逐渐得到了广泛应用。目前的医学图像分割算法能够大抵分为典型的影像分割算法和基于神经网络的算法，由于医疗影像架构单一、边缘模糊，传统的语义分割方法在医疗影像分割上效果不佳且需要人为辅助，无法实现全自动标注。近年来，数据驱动的深度学习方法在众多领域得到了广泛应用。在图像处理方面，卷积神经网络(Convolutional Neural Networks,CNNs)可以获取图像的高级特征，已经被应用于以理解自然医学图像为目标的语义分割任务中。目前基于深度学习的语义分割效果提升主要归功于网络结构的改进，通常包括网络深度、宽度、连接方式的改变或新网络层的提出。Unet对称网络包括进行特征提取的下采样操作和将低级特征映射到完整输入图像的上采样操作，与其它编解码网络相比， Unet中加入了跳跃连接（Skip Connection）操作，实现了低分辨率信息与高分辨率信息的结合，在提高模型的分割性能的同时有效消除了编解码结构的梯度消失情况，在生物医学图像分割中取得了较好的效果。Specifically, the automatic labeling of tumors can be considered as an image segmentation problem. Specifically, image segmentation is the process of classifying pixels. In tumor segmentation, each pixel has only two labels: tumor and non-tumor. Therefore, Tumor segmentation can be viewed as binary classification of pixels. Due to the single style of medical images and blurred boundaries, traditional semantic segmentation methods are not effective in medical image segmentation and require human assistance, which cannot achieve fully automated annotation. In order to reduce the burden on doctors and improve the accuracy of segmentation, computer-aided diagnosis and treatment technology based on deep learning has been widely used in recent years. The current medical image segmentation algorithms can be roughly divided into typical image segmentation algorithms and neural network-based algorithms. Due to the single medical image structure and blurred edges, the traditional semantic segmentation methods are not effective in medical image segmentation and require human assistance. Realize fully automatic labeling. In recent years, data-driven deep learning methods have been widely used in many fields. In terms of image processing, Convolutional Neural Networks (CNNs) can obtain advanced features of images and have been applied to semantic segmentation tasks aimed at understanding natural medical images. At present, the improvement of semantic segmentation effect based on deep learning is mainly attributed to the improvement of network structure, which usually includes the change of network depth, width, connection mode or the proposal of new network layer. The Unet symmetric network includes a downsampling operation for feature extraction and an upsampling operation for mapping low-level features to a complete input image. Compared with other codec networks, Unet adds a Skip Connection operation to achieve low resolution. The combination of information and high-resolution information can effectively eliminate the gradient disappearance of the codec structure while improving the segmentation performance of the model, and has achieved good results in biomedical image segmentation.
然而修改Unet网络对模型分割效果的提升是有限的，本发明实施例通过级联一种以提高模型分割性能为目标的预处理网络获得了当前最优的听神经瘤图像自动化分割模型，具体如图2所示。医生可以凭借该结果对肿瘤位置及特征进行较为精确的诊断，大幅提升了听神经瘤疾病的诊疗效率，简化了诊疗流程。However, modifying the Unet network can only improve the segmentation effect of the model to a limited extent. The embodiment of the present invention obtains the current optimal automatic segmentation model for acoustic neuroma images by cascading a preprocessing network aimed at improving the segmentation performance of the model, as shown in the figure 2. Doctors can use the results to make a more accurate diagnosis of tumor location and characteristics, which greatly improves the diagnosis and treatment efficiency of acoustic neuroma and simplifies the diagnosis and treatment process.
进一步地，作为最优实施方式，获取听神经瘤患者的脑部核磁共振成像这一步骤包括：Further, as an optimal implementation mode, the step of obtaining the brain magnetic resonance imaging of the patient with acoustic neuroma includes:
获取听神经瘤患者在T2序列下的脑部核磁共振成像及其标注数据；Obtain brain magnetic resonance imaging and labeling data of patients with acoustic neuroma under T2 sequence;
使用库函数对获取到的核磁共振图像及其标注数据进行格式转换，得到二维可读取听神经瘤数据集；Use the library function to convert the format of the acquired MRI image and its annotation data to obtain a two-dimensional readable acoustic neuroma data set;
使用数据增强方法进行数据扩增，所用数据增强方法包括旋转、平移、翻转。Data augmentation is performed using data augmentation methods, including rotation, translation, and flipping.
具体地，本实施例可获取听神经瘤患者在T1、T1ce、T2、FLAIR四个序列下的脑部核磁共振影像，由于医生在听神经瘤的治疗中常使用T2序列加权的MRI影像作为诊断依据，该序列下的肿瘤区域相比周围组织视觉效果更亮一些，因此使用T2序列下的核磁共振影像作为数据集。Specifically, this embodiment can obtain brain MRI images of patients with acoustic neuroma under the four sequences of T1, T1ce, T2, and FLAIR. Since doctors often use T2 sequence weighted MRI images as a diagnosis basis in the treatment of acoustic neuromas, the The tumor area under the sequence is brighter than the surrounding tissue, so the MRI image under the T2 sequence is used as the data set.
本实施例中进行数据增强的具体方法是使用keras提供的库函数进行图像的旋转、平移和翻转，并以生成器方式供训练函数使用。The specific method for data enhancement in this embodiment is to use the library functions provided by keras to perform image rotation, translation and flipping, and use it as a generator for training functions.
本实施例中听神经瘤图像自动化分割模型中基于梯度下降原理的预处理网络采用pix2pixGAN的生成器结构，分割网络采用Unet架构，网络具体架构如下：In this embodiment, the preprocessing network based on the gradient descent principle in the automatic segmentation model of acoustic neuroma images adopts the generator structure of pix2pixGAN, and the segmentation network adopts the Unet architecture. The specific architecture of the network is as follows:
pix2pixGAN结构的生成器与Unet架构类似，均为对称网络结构，共26层，由下采样部分和上采样部分组成，下采样过程中包含2个dropout层。The generator of the pix2pixGAN structure is similar to the Unet architecture, both of which are symmetrical network structures, with a total of 26 layers, consisting of a downsampling part and an upsampling part. The downsampling process includes 2 dropout layers.
具体地，使用具有Unet结构生成器的pix2pixGAN训练预处理网络，在理想情况下，GAN结构可以使生成器在多次迭代后得到较好的结果，但由于GAN训练过程的特殊性，基于Unet结构的分割网络会使GAN的训练过程不稳定，甚至出现梯度消失问题，为了解决此问题，需要对Unet结构进行修改再将其作为生成器。为了更好地训练GAN，使用如下修改策略：尽量使用LeakyReLU作为激活函数而不是ReLU；用大于1的strides代替最大池化（maxpooling）操作；最后一层的激活函数使用输出值范围在[-1,1]之间的tanh而不是sigmoid；通过增加dropout等方式在模型中引入噪声，增强模型鲁棒性。pix2pixGAN的作者对Unet结构进行了改进，仿照该生成器结构，得到如说明书附图7所示的针对听神经瘤图像分割的生成器结构。在附图7中，各矩形表示多通道特征图，通道数在方框上侧标明，特征图尺寸在方框左侧标明。不同样式的箭头表示不同的计算操作，向右的箭头表示尺寸为4×4的卷积操作，设置strides为2，使用LeakyReLU函数；向左下方的箭头表示进行激活函数为ReLU的反卷积操作，卷积核大小为4×4，这两个操作会分别将数据的长宽修改为原来的二分之一和两倍；向下的箭头表示复制和连接（concatenate）操作，该操作将下采样过程中的浅层数据和上采样后的深层数据在最后一维进行连接；向左的箭头表示一个上采样操作和一个大小为1的卷积操作，经过激活函数tanh，得到与输入图像尺寸相同，各像素值在-1到1范围内的图像，将该图进行标准化后即可得到预测的分割结果。Specifically, the pix2pixGAN with the Unet structure generator is used to train the preprocessing network. Ideally, the GAN structure can enable the generator to obtain better results after multiple iterations. The segmentation network of GAN will make the training process of GAN unstable, and even the problem of gradient disappearance will occur. In order to solve this problem, it is necessary to modify the Unet structure and use it as a generator. In order to better train GAN, use the following modification strategy: try to use LeakyReLU as the activation function instead of ReLU; replace the maximum pooling (maxpooling) operation with strides greater than 1; use the output value range of the last layer of activation function in [-1 ,1] between tanh instead of sigmoid; introduce noise into the model by increasing dropout, etc., to enhance the robustness of the model. The author of pix2pixGAN improved the Unet structure, imitating the generator structure, and obtained the generator structure for acoustic neuroma image segmentation as shown in Figure 7 of the specification. In Figure 7, each rectangle represents a multi-channel feature map, the number of channels is marked on the upper side of the box, and the size of the feature map is marked on the left side of the box. Arrows of different styles indicate different calculation operations. The arrow to the right indicates a convolution operation with a size of 4×4, set the strides to 2, and use the LeakyReLU function; the arrow to the lower left indicates a deconvolution operation whose activation function is ReLU , the size of the convolution kernel is 4×4, and these two operations will modify the length and width of the data to one-half and two times the original; the downward arrow indicates the copy and concatenate operation, which will down The shallow data in the sampling process and the upsampled deep data are connected in the last dimension; the left arrow indicates an upsampling operation and a convolution operation with a size of 1. After the activation function tanh, the size of the input image is obtained. Similarly, for an image whose pixel value is in the range of -1 to 1, the predicted segmentation result can be obtained after normalizing the image.
Unet分割网络共23层，下采样过程包含10个卷积层，2个最大池化操作，上采样过程包括8个卷积层和4个反卷积层。The Unet segmentation network has a total of 23 layers. The downsampling process includes 10 convolutional layers and 2 maximum pooling operations. The upsampling process includes 8 convolutional layers and 4 deconvolutional layers.
具体地，为了增强模型鲁棒性增加dropout操作，得到如说明书附图5所示的具体结构。各矩形表示多通道特征图，通道数在方框右侧标明，特征图尺寸在方框上侧标明，未标明则说明尺寸与其上方方框相同。不同样式的箭头表示不同的计算操作，其中向下的实心实线箭头表示卷积核尺寸为3×3的卷积运算，激活函数为ReLU（Rectified LinearUnit）；向左的箭头表示尺寸为2×2的最大池化，向右的箭头表示进行激活函数为ReLU的反卷积运算，卷积核尺寸为2×2，这两个操作会分别将数据的长宽修改为原来的二分之一和两倍；向下的空心虚线箭头表示复制和连接（concatenate）操作，该操作将下采样过程中的浅层数据和上采样后的深层数据在最后一维进行连接；向下的空心实线箭头表示一个大小为1的卷积操作，激活函数Sigmoid将值映射到[0,1]区间，得到与输入图像分辨率相同，各像素值在0到1区间内的灰度图，该图即为Unet预测的分割结果。Specifically, in order to enhance the robustness of the model, the dropout operation is added to obtain a specific structure as shown in Figure 5 of the specification. Each rectangle represents a multi-channel feature map. The number of channels is marked on the right side of the box, and the size of the feature map is marked on the upper side of the box. If it is not marked, it means that the size is the same as the box above it. Arrows of different styles represent different calculation operations, among which the downward solid arrow represents the convolution operation with a convolution kernel size of 3×3, and the activation function is ReLU (Rectified LinearUnit); the left arrow represents a size of 2× The maximum pooling of 2, the right arrow indicates the deconvolution operation with the activation function ReLU, the convolution kernel size is 2×2, these two operations will respectively modify the length and width of the data to half of the original and twice; the downward hollow dotted arrow indicates the copy and concatenate operation, which connects the shallow data in the downsampling process and the upsampled deep data in the last dimension; the downward hollow solid line The arrow represents a convolution operation with a size of 1. The activation function Sigmoid maps the value to the [0,1] interval to obtain a grayscale image with the same resolution as the input image and each pixel value in the interval from 0 to 1. The image is Segmentation results predicted for Unet.
进一步地，所述使用基于梯度下降原理的预处理网络和Unet分割网络的级联进行听神经瘤图像的自动化分割这一步骤，具体包括：Further, the step of automatic segmentation of acoustic neuroma images using the cascade of the preprocessing network based on the gradient descent principle and the Unet segmentation network specifically includes:
使用划分出的给定样本训练得到听神经瘤自动分割模型。The acoustic neuroma automatic segmentation model is obtained by using the divided given samples for training.
将可读取的听神经瘤患者核磁共振影像输入自动分割模型中，得到听神经瘤的病灶图像分割结果。Input the readable MRI images of acoustic neuroma patients into the automatic segmentation model to obtain the image segmentation results of acoustic neuroma lesions.
进一步地，上述训练得到听神经瘤图像自动化分割模型的具体步骤包括：Further, the specific steps of the above-mentioned training to obtain the automatic segmentation model of acoustic neuroma images include:
将给定样本划分为训练集和测试集。Divide the given sample into training and testing sets.
具体地，原始数据集中包含200余位病患的脑部核磁共振影像数据及其肿瘤标注结果，每位病患的数据由多张二维切片构成。在核磁共振影像数据的多种序列中，T2序列下的肿瘤区域与周遭组织的对比度更高，因此选用T2序列下的900张二维MRI图像及其标注数据作为数据集。为了便于图像的读取与处理，使用pydicom和simpleITK软件库对dicom格式的MRI图形进行处理，并将其转换为jpeg格式，得到尺寸为640*640的脑部MRI图像，如说明书附图3所示。同时将对应的标注数据用掩码图像表示，如说明书附图4所示，掩码图像的尺寸与脑部MRI图像一致。Specifically, the original data set contains brain MRI data and tumor labeling results of more than 200 patients, and each patient's data consists of multiple two-dimensional slices. Among the various sequences of MRI data, the contrast between the tumor area and the surrounding tissue under the T2 sequence is higher, so 900 two-dimensional MRI images and their annotation data under the T2 sequence were selected as the data set. In order to facilitate image reading and processing, use pydicom and simpleITK software library to process MRI graphics in dicom format and convert them into jpeg format to obtain brain MRI images with a size of 640*640, as shown in Figure 3 of the manual Show. At the same time, the corresponding labeled data is represented by a mask image, as shown in Figure 4 of the specification, the size of the mask image is consistent with that of the brain MRI image.
将训练集输入Unet网络进行训练，得到针对听神经瘤图像的Unet自分割模块。The training set is input into the Unet network for training, and the Unet self-segmentation module for acoustic neuroma images is obtained.
将训练集和对应修正图输入pix2pixGAN网络进行训练，得到针对听神经瘤图像进行图像修正的预处理模块。The training set and the corresponding correction map are input into the pix2pixGAN network for training, and a preprocessing module for image correction of acoustic neuroma images is obtained.
进一步地，所述训练Unet网络的具体步骤包括：Further, the concrete steps of described training Unet network include:
将训练集中的听神经瘤患者脑部核磁共振影像作为输入数据，该影像对应的标注分割结果为目标数据。The brain MRI images of acoustic neuroma patients in the training set are used as input data, and the corresponding labeling and segmentation results of the images are the target data.
使用上述样本进行训练，多次迭代后选取损失值最小的模型作为Unet分割模块。Use the above samples for training, and select the model with the smallest loss value as the Unet segmentation module after multiple iterations.
具体地，基于Unet的听神经瘤分割模型训练过程如说明书附图6所示，首先将原始数据集划分为训练数据和测试数据，将训练集进行数据增强并使用新生成的数据进行分割网络的训练，在每个迭代完成后比较该次迭代训练集的损失值与当前最优模型损失值的大小，如果比最优损失值小则更新最优损失值并保存该次迭代的结果为最优模型，若比最优损失值大则直接进行下一次迭代。Specifically, the training process of the Unet-based acoustic neuroma segmentation model is shown in Figure 6 of the specification. First, the original data set is divided into training data and test data, and the training set is subjected to data enhancement and the newly generated data is used to train the segmentation network. , after each iteration is completed, compare the loss value of the iterative training set with the current optimal model loss value, if it is smaller than the optimal loss value, update the optimal loss value and save the result of this iteration as the optimal model , if it is greater than the optimal loss value, proceed to the next iteration directly.
进一步地，所述将训练集和对应修正图输入pix2pixGAN网络进行训练，得到预处理模块的具体步骤包括：Further, the specific steps of inputting the training set and the corresponding correction map into the pix2pixGAN network for training, and obtaining the preprocessing module include:
获取训练集中听神经瘤核磁共振影像对应的修正图像；Obtain the corrected image corresponding to the MRI image of the acoustic neuroma in the training set;
将训练集中的听神经瘤核磁共振影像和对应的修正图像输入pix2pixGAN网络进行训练，得到训练好的生成器和判别器；Input the acoustic neuroma MRI images and corresponding corrected images in the training set into the pix2pixGAN network for training, and obtain the trained generator and discriminator;
提取生成器结构作为听神经瘤图像分割模型的预处理模块。Extracting the generator structure as a preprocessing module for an acoustic neuroma image segmentation model.
进一步地，所述获取训练集听神经瘤核磁共振影像的修正图像这一步骤，具体包括：Further, the step of obtaining the corrected images of the training set acoustic neuroma MRI images specifically includes:
计算训练集中听神经瘤患者脑部核磁共振影像在Unet分割模型中的Dice评估系数；Calculate the Dice evaluation coefficient of the brain MRI images of patients with acoustic neuroma in the training set in the Unet segmentation model;
计算Dice系数关于该核磁共振影像的偏微分，将该偏微分值作为当前核磁共振影像的修正值进'行叠加；Calculate the partial differential of the Dice coefficient with respect to the nuclear magnetic resonance image, and superimpose the partial differential value as the correction value of the current nuclear magnetic resonance image;
多次迭代得到听神经瘤脑部核磁共振影像最终的修正结果。Multiple iterations were performed to obtain the final corrected result of the MRI image of the acoustic neuroma brain.
具体地，将输入图像I的每个像素标注为0和1两个类别，其中0表示该像素属于非肿瘤区域，1表示该像素属于肿瘤区域，由于Unet分割网络最后一层的激活函数是输出范围在[0,1]的sigmoid函数，因此模型的预测输出结果是各像素值在[0,1]范围内的灰度图像，经过归一化后可以得到肿瘤区域是白色，非肿瘤区域为黑色的预测结果。令T表示待分割图像I对应的真实标注数据，分割模型的目标是令P=T，该目标可用dice系数进行衡量，dice系数越大表明P与T越接近，因此本发明实施例预处理算法的目标是找到一个ΔI，使得dice系数越大越好。令iterations为迭代次数，对训练数据中的待分割图像进行iterations次上述计算后可得到更新后的图像I’，在迭代次数达到一定数值后，将得到分割效果优于原图的新图像I’。虽然训练数据可以使用标注图像T对输入数据进行修正，但对测试数据来说标注图像是未知的，无法使用对应的标注数据T对图像进行更新，因此需要一个函数能实现从原始图像I到新图像I’的映射。由于pix2pixGAN结构在图片间的风格转换表现优异，因此使用pix2pixGAN训练I到I’间的映射，该网络的训练数据包括domainA和domainB，其中domainA是分割网络训练集中的待分割图像，domainB是对domainA中数据经过上述预处理算法更新后的图像集合，domainA和domainB中的图像一一对应，训练后的生成器网络可实现从原始图像I到新图像I’的映射。Specifically, each pixel of the input image I is marked as two categories of 0 and 1, where 0 indicates that the pixel belongs to a non-tumor area, and 1 indicates that the pixel belongs to a tumor area. Since the activation function of the last layer of the Unet segmentation network is the output The sigmoid function in the range of [0,1], so the predicted output of the model is a grayscale image with each pixel value in the range of [0,1]. After normalization, it can be obtained that the tumor area is white, and the non-tumor area is Black predicted results. Let T represent the real labeled data corresponding to the image I to be segmented. The goal of the segmentation model is to make P=T, which can be measured by the dice coefficient. The larger the dice coefficient, the closer P is to T. Therefore, the preprocessing algorithm of the embodiment of the present invention The goal of is to find a ΔI such that the larger the dice coefficient, the better. Let iterations be the number of iterations, and perform the above calculations on the image to be segmented in the training data for iterations times to obtain an updated image I'. After the number of iterations reaches a certain value, a new image I' with a better segmentation effect than the original image will be obtained. . Although the training data can use the labeled image T to correct the input data, the labeled image is unknown to the test data, and the corresponding labeled data T cannot be used to update the image. Therefore, a function is needed to realize the conversion from the original image I to the new one. Mapping of image I'. Since the pix2pixGAN structure is excellent in style conversion between pictures, pix2pixGAN is used to train the mapping between I and I'. The training data of the network includes domainA and domainB, where domainA is the image to be segmented in the training set of the segmentation network, and domainB is the image for domainA The image set in which data is updated by the above preprocessing algorithm, the images in domainA and domainB correspond one-to-one, and the trained generator network can realize the mapping from the original image I to the new image I'.
进一步地，所述训练pix2pixGAN网络的具体步骤包括：Further, the concrete steps of described training pix2pixGAN network include:
将训练集中的听神经瘤核磁共振影像及其修正结果分别作为pix2pixGAN网络的输入数据和目标数据，多次迭代训练，得到生成器和判别器；The acoustic neuroma MRI images in the training set and their corrected results are used as the input data and target data of the pix2pixGAN network respectively, and iteratively trained for multiple times to obtain the generator and discriminator;
使用测试数据集在基于梯度下降原理的生成器网络和Unet分割网络级联的听神经瘤图像自动化分割模型上进行测试，选取测试性能最优的生成器网络作为听神经瘤图像自动化分割模型的预处理模块；Use the test data set to test on the automatic segmentation model of acoustic neuroma image based on the cascade of generator network and Unet segmentation network based on the principle of gradient descent, and select the generator network with the best test performance as the preprocessing module of the automatic segmentation model of acoustic neuroma image ;
如图9，将训练得到的预处理模块与Unet分割网络级联得到听神经瘤图像自动化分割模型。As shown in Figure 9, the trained preprocessing module is cascaded with the Unet segmentation network to obtain an automatic acoustic neuroma image segmentation model.
具体地，基于pix2pixGAN的预处理网络训练过程如图8所示，首先构建预处理网络的训练数据，将分割网络训练集中的待分割图像作为domainA，将domainA中数据进行更新后得到的集合作为domainB。然后使用domainA和domainB进行生成对抗网络的训练，在每个迭代完成后比较该次迭代在训练集上的MAE与当前最优模型MAE值的大小，如果比最优MAE值小则更新最优MAE值并保存该次迭代的结果为最优模型，若比最优MAE值大则直接进行下一次迭代。多次迭代后得到更新后的测试数据，使用Unet网络对新图像进行分割得到最终的分割结果。Specifically, the preprocessing network training process based on pix2pixGAN is shown in Figure 8. First, the training data of the preprocessing network is constructed, and the image to be segmented in the segmentation network training set is used as domainA, and the set obtained after updating the data in domainA is used as domainB . Then use domainA and domainB to train the generation confrontation network. After each iteration is completed, compare the MAE of this iteration on the training set with the current optimal model MAE value. If it is smaller than the optimal MAE value, update the optimal MAE Value and save the result of this iteration as the optimal model, if it is greater than the optimal MAE value, proceed to the next iteration directly. After multiple iterations, the updated test data is obtained, and the new image is segmented using the Unet network to obtain the final segmentation result.
采用本具体实施例方法对270个听神经瘤患者在T2序列下的核磁共振图像进行分割，经过评估得到该方法的Dice系数为0.8750，Jaccard系数为0.7988，敏感性为0.9006，特异性为0.9992，可见模型的分割性能较高，模型可靠性较强。The MRI images of 270 acoustic neuroma patients under the T2 sequence were segmented using the method of this specific example. After evaluation, the Dice coefficient of the method was 0.8750, the Jaccard coefficient was 0.7988, the sensitivity was 0.9006, and the specificity was 0.9992. It can be seen that The segmentation performance of the model is high, and the model reliability is strong.
本发明实施例通过构建以提升模型分割性能为目标的预处理网络提升了模型的分割性能，突破了单一Unet分割模型的局限性，获得了当前最优的听神经瘤图像自动化分割模型。同时使用后处理模块融合分割掩码与核磁共振图像，使分割结果更加直观，医生可以凭借该结果对肿瘤位置及特征进行较为精确的诊断，有效解决了人工标注标定周期长、标定精度参差不齐的问题。本发明实施例在计算机终端上能以较快的速度实现听神经瘤图像的自动化分割，大幅提升了听神经瘤疾病的诊疗效率，有效简化了诊疗流程。The embodiment of the present invention improves the segmentation performance of the model by constructing a preprocessing network aimed at improving the segmentation performance of the model, breaks through the limitation of a single Unet segmentation model, and obtains the current optimal automatic segmentation model for acoustic neuroma images. At the same time, the post-processing module is used to fuse the segmentation mask and the MRI image to make the segmentation result more intuitive. Doctors can use the result to make a more accurate diagnosis of the tumor location and characteristics, effectively solving the problem of long manual labeling and calibration cycles and uneven calibration accuracy. The problem. The embodiment of the present invention can realize automatic segmentation of acoustic neuroma images at a relatively fast speed on a computer terminal, greatly improves the diagnosis and treatment efficiency of acoustic neuroma diseases, and effectively simplifies the diagnosis and treatment process.
与以上提供的听神经瘤图像自动化分割方法相对应的，本发明还提供了听神经瘤图像自动化分割系统，包括：数据获取模块和听神经瘤图像自动化分割模块，所述数据获取模块，用于获取听神经瘤患者的核磁共振图像；所述听神经瘤图像自动化分割模块用于利用预先训练的听神经瘤图像自动化分割模型对获取的听神经瘤患者的核磁共振图像进行自动化分割获得分割结果，所述听神经瘤图像自动化分割模型采用预处理网络和分割网络的级联结构。Corresponding to the acoustic neuroma image automatic segmentation method provided above, the present invention also provides an acoustic neuroma image automatic segmentation system, including: a data acquisition module and an acoustic neuroma image automatic segmentation module, the data acquisition module is used to acquire acoustic neuroma images The patient's nuclear magnetic resonance image; the acoustic neuroma image automatic segmentation module is used to use the pre-trained acoustic neuroma image automatic segmentation model to automatically segment the acquired acoustic neuroma patient's nuclear magnetic resonance image to obtain segmentation results, and the acoustic neuroma image automatic segmentation The model adopts a cascaded structure of preprocessing network and segmentation network.
进一步地，所述预处理网络用于对脑部核磁共振影像进行修正，所述分割网络用于对修正后的脑部核磁共振图像进行病灶区域划分，得到肿瘤区域的掩码图像。Further, the preprocessing network is used to correct the brain magnetic resonance image, and the segmentation network is used to divide the corrected brain magnetic resonance image into lesion regions to obtain a mask image of the tumor region.
再进一步地，所述自动化分割系统还包括数据后处理模块，所述数据后处理模块用于融合掩码图像与对应的脑部核磁共振影像，使模型的预测分割结果更为直观。Still further, the automatic segmentation system also includes a data post-processing module, which is used to fuse the mask image and the corresponding brain MRI image, so as to make the predicted segmentation result of the model more intuitive.
需要说明的是，由于所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，本发明中描述的系统，装置/单元的具体工作过程，可以参考前述方法中的对应过程，因此再原申请文件中不再赘述。It should be noted that, as those skilled in the art can clearly understand, for the convenience and brevity of description, the specific working process of the system and device/unit described in the present invention can refer to the corresponding process in the aforementioned method, so again It will not be repeated in the original application documents.
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。The present application is described with reference to flowcharts and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.
尽管已描述了本申请的优选实施例，但本领域内的技术人员一旦得知了基本创造性概念，则可对这些实施例作出另外的变更和修改。所以，所附权利要求意欲解释为包括优选实施例以及落入本申请范围的所有变更和修改。While preferred embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once the basic inventive concept is appreciated. Therefore, it is intended that the appended claims be interpreted to cover the preferred embodiment and all changes and modifications that fall within the scope of the application.
显然，本领域的技术人员可以对本申请进行各种改动和变型而不脱离本申请的精神和范围。这样，倘若本申请的这些修改和变型属于本申请权利要求及其等同技术的范围之内，则本申请也意图包含这些改动和变型在内。Obviously, those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. In this way, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.
Claims (8)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110238730.5ACN112862805B (en) | 2021-03-04 | 2021-03-04 | Acoustic neuroma image automatic segmentation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110238730.5ACN112862805B (en) | 2021-03-04 | 2021-03-04 | Acoustic neuroma image automatic segmentation method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112862805A CN112862805A (en) | 2021-05-28 |
| CN112862805Btrue CN112862805B (en) | 2022-12-06 |
Family
ID=75991519
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110238730.5AExpired - Fee RelatedCN112862805B (en) | 2021-03-04 | 2021-03-04 | Acoustic neuroma image automatic segmentation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112862805B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113409324B (en)* | 2021-07-14 | 2023-03-24 | 清华大学深圳国际研究生院 | Brain segmentation method fusing differential geometric information |
| CN113781403B (en)* | 2021-08-19 | 2023-10-13 | 东北大学 | Chest CT image processing method and device |
| CN113870289B (en)* | 2021-09-22 | 2022-03-15 | 浙江大学 | A method and device for decoupling, dividing and conquering facial nerve segmentation |
| CN114463320B (en)* | 2022-02-17 | 2024-01-26 | 厦门大学 | Magnetic resonance imaging brain glioma IDH gene prediction method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109671018A (en)* | 2018-12-12 | 2019-04-23 | 华东交通大学 | A kind of image conversion method and system based on production confrontation network and ResNets technology |
| CN110009028A (en)* | 2019-03-28 | 2019-07-12 | 北京科技大学 | Microscopic image data enhancement method and device |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110047068A (en)* | 2019-04-19 | 2019-07-23 | 山东大学 | MRI brain tumor dividing method and system based on pyramid scene analysis network |
| CN111354002A (en)* | 2020-02-07 | 2020-06-30 | 天津大学 | Kidney and kidney tumor segmentation method based on deep neural network |
- 2021
- 2021-03-04CNCN202110238730.5Apatent/CN112862805B/ennot_activeExpired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109671018A (en)* | 2018-12-12 | 2019-04-23 | 华东交通大学 | A kind of image conversion method and system based on production confrontation network and ResNets technology |
| CN110009028A (en)* | 2019-03-28 | 2019-07-12 | 北京科技大学 | Microscopic image data enhancement method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112862805A (en) | 2021-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112862805B (en) | Acoustic neuroma image automatic segmentation method and system | |
| CN115205300B (en) | Fundus blood vessel image segmentation method and system based on cavity convolution and semantic fusion | |
| Tang et al. | CT image enhancement using stacked generative adversarial networks and transfer learning for lesion segmentation improvement | |
| CN115170582A (en) | Liver image segmentation method based on multi-scale feature fusion and grid attention mechanism | |
| CN111951288A (en) | A deep learning-based segmentation method for skin cancer lesions | |
| Shan et al. | SCA-Net: A spatial and channel attention network for medical image segmentation | |
| CN111429474A (en) | Mammary gland DCE-MRI image focus segmentation model establishment and segmentation method based on mixed convolution | |
| CN115578384A (en) | UNet brain tumor image segmentation algorithm based on global and local feature fusion | |
| CN112785593A (en) | Brain image segmentation method based on deep learning | |
| Yanling et al. | Segmenting vitiligo on clinical face images using CNN trained on synthetic and internet images | |
| CN118657800B (en) | Joint segmentation method of multiple lesions in retinal OCT images based on hybrid network | |
| CN118314121A (en) | Medical image segmentation method combining curve structure prompt and deep neural network | |
| Martín-Isla et al. | Stacked BCDU-Net with semantic CMR synthesis: Application to myocardial pathology segmentation challenge | |
| CN111091575A (en) | Medical image segmentation method based on reinforcement learning method | |
| CN116563533A (en) | Medical image segmentation method and system based on prior information of target position | |
| CN117633558A (en) | Multi-excitation fusion zero-sample lesion detection method based on visual language model | |
| CN112150564B (en) | Medical image fusion algorithm based on deep convolution neural network | |
| Ruan et al. | An efficient tongue segmentation model based on u-net framework | |
| CN116485853A (en) | A medical image registration method and device based on deep learning neural network | |
| Sattari et al. | Liver margin segmentation in abdominal CT images using U-Net and Detectron2: annotated dataset for deep learning models | |
| CN119180956A (en) | WMH image segmentation system of mixed scale attention | |
| Kaki et al. | Semantic segmentation of dental caries using improved deeplab v3network | |
| CN118823344A (en) | Medical image semantic segmentation method and system based on channel and spatial attention mechanism | |
| Kala et al. | Brain magnetic resonance image inpainting via deep edge region-based generative adversarial network | |
| Shen et al. | Promptable counterfactual diffusion model for unified brain tumor segmentation and generation with mris |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee | Granted publication date:20221206 |