WO2024000161A1 - Ct胰腺肿瘤自动分割方法、系统、终端以及存储介质 - Google Patents

Ct胰腺肿瘤自动分割方法、系统、终端以及存储介质 Download PDF

Info

Publication number
WO2024000161A1
WO2024000161A1 PCT/CN2022/101894 CN2022101894W WO2024000161A1 WO 2024000161 A1 WO2024000161 A1 WO 2024000161A1 CN 2022101894 W CN2022101894 W CN 2022101894W WO 2024000161 A1 WO2024000161 A1 WO 2024000161A1
Authority
WO
WIPO (PCT)
Prior art keywords
layer
dimensional
umrformer
network model
module
Prior art date
Application number
PCT/CN2022/101894
Other languages
English (en)
French (fr)
Inventor
贾富仓
方坤
Original Assignee
中国科学院深圳先进技术研究院
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 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Priority to PCT/CN2022/101894 priority Critical patent/WO2024000161A1/zh
Publication of WO2024000161A1 publication Critical patent/WO2024000161A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • the present application belongs to the field of medical image processing technology, and particularly relates to a CT pancreatic tumor automatic segmentation method, system, terminal and storage medium.
  • Pancreatic cancer is one of the most lethal malignant tumors of the digestive tract, characterized by insidious symptoms, delayed diagnosis, difficult treatment, and high mortality.
  • the five-year survival rate after diagnosis of pancreatic cancer is about 9%, making it one of the malignant tumors with the worst prognosis.
  • the best treatment for pancreatic tumors is tumor resection, which not only requires the use of advanced medical instruments such as computed tomography, magnetic resonance imaging, single photon emission computed tomography, and positron emission tomography to capture and determine the location, size, and location of the tumor. Shape and other information also require professional doctors with rich clinical experience to diagnose. In clinical medicine, pancreatic tumors are mainly marked manually by professional doctors, which requires a lot of time and energy.
  • existing automatic segmentation methods for medical tumors mainly include segmentation methods based on deep learning and segmentation methods based on Transformer networks.
  • segmentation methods based on deep learning are all based on the U-Net structure, such as U-Net++, Attention-UNet, etc.
  • Current segmentation methods based on Transformer networks usually use the following two methods: only using pure Transformer modules to construct the encoder-decoder network, or using Transformer as a single auxiliary module and integrating convolution modules and Transformer to construct segmentation networks.
  • these two methods have some obvious shortcomings, specifically: Pancreatic cancer is closely surrounded by tissues of similar intensity, while existing models use the Transformer module to pay more attention to the global characteristics of the tumor and surrounding tissues, ignoring the relationship with the pancreas. Tumors serve as local characteristic information of small targets.
  • existing methods only consider the combination mode of convolution and Transformer, but ignore the limitation of Transformer itself that lacks attention to spatial local features, which will cause a loss of segmentation accuracy when segmenting pancreatic tumors.
  • This application provides a CT pancreatic tumor automatic segmentation method, system, terminal and storage medium, aiming to solve one of the above technical problems in the prior art at least to a certain extent.
  • An automatic segmentation method for CT pancreatic tumors including:
  • the preprocessed three-dimensional CT image is input into the trained UMRFormer network model.
  • the UMRFormer network model has a U-shaped symmetrical structure.
  • the UMRFormer network model uses 3D CNN to generate feature maps of different scales, and uses a double-layer class
  • the Transformer module encodes the long-range dependency semantic information of the feature map and outputs the tumor segmentation result of the three-dimensional CT image.
  • the technical solution adopted by the embodiment of the present application also includes: the preprocessing of the three-dimensional CT image is specifically:
  • the three-dimensional CT image is enhanced, and the enhanced three-dimensional CT image is cropped to a size consistent with the input of the UMRFormer network model.
  • the UMRFormer network model includes a convolution layer, a multi-head self-attention mechanism layer, a pooling layer, a normalization layer and an upsampling layer, and the UMRFormer network model is U-shaped symmetrical Structure, in the jump connection stage of the U-shaped symmetric network, a double-layer Transformer-like module is embedded;
  • the UMRFormer network model uses a 5-layer 3D CNN to generate feature maps of different scales to capture spatial and depth features, and then uses a double-layer Transformer-like module to provide long-range dependency semantic information on the fourth and fifth layer feature maps in the global space. Encoding is performed, and the upsampling layer and the convolution layer are repeatedly superimposed to obtain the segmentation result of the three-dimensional CT image.
  • the UMRFormer network model includes an encoder and a decoder
  • the encoder includes L MRFormer modules
  • each MRFormer module includes 1 multi-head self-attention module
  • 3 depth modules Convolution layer and 4 post-normalization layers.
  • the depth convolution layer and normalization layer are superimposed as a combination.
  • the residual unit is formed through skip connection. 3 depth convolution layers plus normalization are used.
  • Layer combination replaces the feed-forward network and post-fixes the normalization layer in front of the multi-head self-attention module, so that the normalization layer moves from the beginning of each residual unit to the back end.
  • the training process of the UMRFormer network model includes:
  • LP 1 and LP 2 are linear projection operations, and Represents the position code, and Represents feature embedding;
  • the output formula of the l-th layer (l ⁇ [1,2,..,L]) class Transformer module is:
  • LN (*) is layer normalization
  • z l is the output of the MRFormer module of the l-th layer
  • DConv (*) is the depth convolution.
  • the training process of the UMRFormer network model also includes:
  • a CT pancreatic tumor automatic segmentation system including:
  • Data acquisition module used to acquire three-dimensional CT images to be segmented
  • Data preprocessing module used to preprocess the three-dimensional CT image
  • Data segmentation module used to input the preprocessed three-dimensional CT image into the trained UMRFormer network model.
  • the UMRFormer network model has a U-shaped symmetrical structure.
  • the UMRFormer network model uses 3D CNN to generate feature maps of different scales.
  • a two-layer Transformer-like module is used to encode the long-range dependency semantic information of the feature map, and the tumor segmentation result of the three-dimensional CT image is output.
  • a terminal includes a processor and a memory coupled to the processor, wherein,
  • the memory stores program instructions for implementing the automatic CT pancreatic tumor segmentation method
  • the processor is used to execute the program instructions stored in the memory to control automatic segmentation of CT pancreatic tumors.
  • a storage medium that stores program instructions executable by a processor, and the program instructions are used to execute the automatic CT pancreatic tumor segmentation method.
  • the beneficial effects produced by the embodiments of the present application are that: the CT pancreatic tumor automatic segmentation method, system, terminal and storage medium of the embodiments of the present application propose a new type of deep fusion feature global long-distance correlation and spatial UMRFormer network model of local information. This network model is based on the U-shaped codec structure.
  • a Transformer-like module with the advantages of integrating local and global dependencies of the features of interest is embedded, which enables the network model to learn More feature information of small targets of pancreatic tumors and global long-term dependency feature information between tumors and surrounding tissues to obtain more effective deep feature context information, thereby improving the existing single-class Transformer network framework with certain defects and improving the overall network segmentation accuracy.
  • Figure 1 is a flow chart of the CT pancreatic tumor automatic segmentation method according to the first embodiment of the present application
  • Figure 2 is a flow chart of the CT pancreatic tumor automatic segmentation method according to the second embodiment of the present application.
  • Figure 3 is an architecture diagram of the UMRFormer network model according to the embodiment of this application.
  • Figure 4 is a schematic structural diagram of the MRFormer module according to the embodiment of the present application.
  • Figure 5 is a schematic structural diagram of a CT pancreatic tumor automatic segmentation system according to an embodiment of the present application.
  • Figure 6 is a schematic structural diagram of a terminal according to an embodiment of the present application.
  • Figure 7 is a schematic structural diagram of a storage medium according to an embodiment of the present application.
  • an embodiment means that a particular feature, structure or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application.
  • the appearances of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those skilled in the art understand, both explicitly and implicitly, that the embodiments described herein may be combined with other embodiments.
  • preprocessing the three-dimensional CT image specifically includes: enhancing the three-dimensional CT image, and cropping the enhanced three-dimensional CT image to a size that conforms to the UMRFormer network model input.
  • the UMRFormer network model has a U-shaped symmetrical structure.
  • the UMRFormer network model uses 3D CNN to generate feature maps of different scales, and uses a two-layer Transformer-like module to map the features.
  • the long-range semantic information of the image is encoded, and the tumor segmentation result of the three-dimensional CT image is output.
  • the UMRFormer network model includes a convolution layer, a multi-head self-attention mechanism layer, a pooling layer, a normalization layer and an upsampling layer.
  • the UMRFormer network model has a U-shaped symmetric structure.
  • the UMRFormer network model uses 5-layer 3D CNN to generate feature maps of different scales to capture spatial and depth features, and then uses a double-layer Transformer-like module to map the fourth and fifth layers in the global space.
  • the long-range semantic information of the layer feature map is encoded, and the upsampling layer and the convolution layer are repeatedly superimposed to obtain the segmentation result of the three-dimensional CT image.
  • FIG. 2 is a flow chart of the CT pancreatic tumor automatic segmentation method according to the second embodiment of the present application.
  • the automatic CT pancreatic tumor segmentation method in the second embodiment of the present application includes the following steps:
  • S200 Obtain the CT pancreas data set, and divide the CT pancreas data set into a training set, a verification set and a test set according to the set ratio;
  • the CT pancreas data set used includes 281 CT cases from the MSD pancreas data set.
  • the division ratio of the CT pancreas data set is 6:2:2, that is, the number of training sets, validation sets, and test sets obtained are 187, 47, and 47 respectively.
  • the normalization layer in front of the multi-head self-attention module is post-positioned, that is, the normalization layer is moved from the beginning of each residual unit to the back end, which can produce milder activation values on the network layer, so that Model training is more stable.
  • the UMRFormer network model training process in the embodiment of this application is specifically as follows:
  • LP 1 and LP 2 are linear projection operations, and Represents the position code, and Represents feature embedding.
  • the output formula of the l-th layer (l ⁇ [1,2,..,L]) class Transformer module can be expressed as:
  • the feature sequence output by the double-layer MRFormers module needs to be mapped into a three-dimensional feature map.
  • Two-layer MRFormers module i.e.: and ) are respectively reshaped into and Size.
  • the outputs of Z′ 1 and Z′ 2 are upsampled and convolved respectively, and then fused with encoder features through long-distance jump connections to obtain finer semantics and fine-grained information and a richer space detail.
  • S203 Input the verification set into the CT pancreatic tumor segmentation model, verify the preliminary segmentation results of the CT pancreas/pancreatic tumor of the CT pancreatic tumor segmentation model, and adjust the model hyperparameters;
  • S204 Input the test set into the CT pancreatic tumor segmentation model to test the CT pancreatic/pancreatic tumor segmentation performance of the model.
  • the CT pancreatic tumor automatic segmentation method in the embodiment of this application proposes a new UMRFormer network model that deeply fuses global long-distance correlation of features and spatial local information.
  • the network model is based on a U-shaped codec structure.
  • a Transformer-like module is embedded that has the advantages of integrating local and global dependencies of attention features, which enables the network model to learn more characteristic information of small targets of pancreatic tumors as well as global long-term dependency features between tumors and surrounding tissues. information to obtain more effective deep feature context information, thereby improving the existing single-class Transformer network framework with certain defects and improving the overall segmentation accuracy of the network.
  • FIG. 5 is a schematic structural diagram of a CT pancreatic tumor automatic segmentation system according to an embodiment of the present application.
  • the automatic CT pancreatic tumor segmentation system 40 in the embodiment of the present application includes:
  • Data acquisition module 41 used to acquire three-dimensional CT images to be segmented
  • Data preprocessing module 42 used to preprocess the three-dimensional CT image; wherein, the preprocessing of the three-dimensional CT image specifically includes: enhancing the three-dimensional CT image, and trimming the enhanced three-dimensional CT image to conform to the UMRFormer network model Enter the dimensions.
  • Data segmentation module 43 used to input the preprocessed three-dimensional CT image into the trained UMRFormer network model.
  • the UMRFormer network model has a U-shaped symmetrical structure.
  • the UMRFormer network model uses 3D CNN to generate feature maps of different scales, and uses a double-layer
  • the Transformer-like module encodes the long-range dependent semantic information of the feature map and outputs the tumor segmentation result of the three-dimensional CT image; among them, the UMRFormer network model includes a convolution layer, a multi-head self-attention mechanism layer, a pooling layer, a normalization layer and the above In the sampling layer, the UMRFormer network model has a U-shaped symmetrical structure.
  • a double-layer Transformer-like module is embedded; the UMRFormer network model uses 5 layers of 3D CNN to generate feature maps of different scales to capture space and depth. Features, and then use a two-layer Transformer-like module to encode the long-range dependent semantic information of the fourth and fifth layer feature maps in the global space, and repeatedly superimpose the upsampling layer and the convolution layer to obtain the segmentation of the three-dimensional CT image result.
  • the CT pancreatic tumor automatic segmentation system of the embodiment of the present application proposes a new UMRFormer network model that deeply fuses global long-distance correlation of features and spatial local information.
  • This network model is based on a U-shaped codec structure.
  • a Transformer-like module is embedded that has the advantages of integrating local and global dependencies of attention features, which enables the network model to learn more characteristic information of small targets of pancreatic tumors as well as global long-term dependency features between tumors and surrounding tissues. information to obtain more effective deep feature context information, thereby improving the existing single-class Transformer network framework with certain defects and improving the overall segmentation accuracy of the network.
  • the terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51 .
  • the memory 52 stores program instructions for implementing the above automatic CT pancreatic tumor segmentation method.
  • the processor 51 is used to execute program instructions stored in the memory 52 to control the automatic segmentation of CT pancreatic tumors.
  • the processor 51 can also be called a CPU (Central Processing Unit).
  • the processor 51 may be an integrated circuit chip with signal processing capabilities.
  • the processor 51 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component.
  • DSP digital signal processor
  • ASIC application-specific integrated circuit
  • FPGA off-the-shelf programmable gate array
  • a general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.
  • FIG. 7 is a schematic structural diagram of a storage medium according to an embodiment of the present application.
  • the storage medium of the embodiment of the present application stores a program file 61 that can implement all the above methods.
  • the program file 61 can be stored in the above storage medium in the form of a software product and includes several instructions to make a computer device (can It is a personal computer, server, or network device, etc.) or processor that executes all or part of the steps of the various embodiments of the application.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. , or terminal equipment such as computers, servers, mobile phones, tablets, etc.

Landscapes

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

Abstract

一种CT胰腺肿瘤自动分割方法、系统、终端以及存储介质。方法包括：获取待分割的三维CT图像（S100）；对三维CT图像进行预处理（S101）；将预处理后的三维CT图像输入训练好的UMRFormer网络模型，UMRFormer网络模型生成不同尺度的特征图，并对特征图的长程依赖语义信息进行编码，输出所述三维CT图像的肿瘤分割结果（S102）。这种新型的深度融合特征全局长距离相关性和空间局部信息的UMRFormer网络模型，通过嵌入具有融合关注特征局部和全局依赖性优势的类Transformer模块，使网络模型学习到更多胰腺肿瘤小目标的特征信息以及肿瘤与周边组织间的全局长依赖特征信息，提高网络整体的分割精度。

Description

CT胰腺肿瘤自动分割方法、系统、终端以及存储介质 技术领域
本申请属于医学图像处理技术领域，特别涉及一种CT胰腺肿瘤自动分割方法、系统、终端以及存储介质。
背景技术
胰腺癌是消化道最致命的恶性肿瘤之一，具有症状隐匿、诊断延误、治疗困难以及死亡率高等特点。胰腺癌确诊后的五年生存率约9％，是预后最差的恶性肿瘤之一。目前针对胰腺肿瘤最好的治疗手段是肿瘤切除，不仅要借助计算机断层扫描、磁共振成像、单光子发射计算机断层扫描和正电子发射断层扫描等先进的医学仪器手段去捕捉确定肿瘤的位置，大小和形状等信息，还需要具有丰富临床经验的专业医生进行诊断。在临床医学中，胰腺肿瘤主要通过专业的医生手动标注，需要消耗大量的时间和精力。同时由于人工手动标注往往会带有个人主观性，对同一幅图像的病灶区域多次勾画的结果也可能不同，其一致性和可重复性是难以得到保证的，会造成一定的肿瘤勾画偏差，可能会致使误诊及漏诊的情况。因此，借助计算机辅助分析诊断方法提升胰腺肿瘤的分割精度成为不错的选择。
现有技术中，已有的医学肿瘤自动分割方法主要包括基于深度学习的分割方法和基于Transformer网络的分割方法。其中，基于深度学习的分割方法都是基于U-Net结构，例如U-Net++、Attention-UNet等。当前基于Transformer网络的分割方法通常使用以下两种方法：仅使用纯Transformer模块构造编码器-解码器网络，或单一的将Transformer作为辅助模块,融合卷积模块和Transformer建构分割网络。但是，这两种方法存在一些明显的缺点，具体为： 胰腺癌被相似强度的组织紧密包围，而现有的模型利用Transformer模块更多地关注肿瘤与周边组织的全局性特征，忽略了与胰腺肿瘤作为小目标的局部性特征信息。此外，现有方法只考虑了卷积与Transformer的组合模式，而忽略了Transformer本身缺乏关注空间局部特征的局限性，在分割胰腺肿瘤时，会造成分割精度的损失。
发明内容
本申请提供了一种CT胰腺肿瘤自动分割方法、系统、终端以及存储介质，旨在至少在一定程度上解决现有技术中的上述技术问题之一。
为了解决上述问题，本申请提供了如下技术方案：
一种CT胰腺肿瘤自动分割方法，包括：
获取待分割的三维CT图像；
对所述三维CT图像进行预处理；
将所述预处理后的三维CT图像输入训练好的UMRFormer网络模型，所述UMRFormer网络模型为U形对称结构，所述UMRFormer网络模型利用3D CNN生成不同尺度的特征图，并利用双层的类Transformer模块对所述特征图的长程依赖语义信息进行编码，输出所述三维CT图像的肿瘤分割结果。
本申请实施例采取的技术方案还包括：所述对所述三维CT图像进行预处理具体为：
对所述三维CT图像进行增强处理，并将增强后的三维CT图像剪裁为符合所述UMRFormer网络模型输入的尺寸。
本申请实施例采取的技术方案还包括：所述UMRFormer网络模型包括卷积层、多头自注意力机制层、池化层、归一化层以及上采样层，所述UMRFormer 网络模型为U形对称结构，在所述U型对称网络的跳跃连接阶段，嵌入有双层的类Transformer模块；
所述UMRFormer网络模型利用5层3D CNN生成不同尺度的特征图，捕捉空间和深度特征，然后利用双层的类Transformer模块对全局空间中的第四层和第五层特征图的长程依赖语义信息进行编码，并将上采样层和卷积层反复叠加，得到三维CT图像的分割结果。
本申请实施例采取的技术方案还包括：所述UMRFormer网络模型包括编码器和解码器，所述编码器包括L个MRFormer模块，每个MRFormer模块分别包括1个多头自注意力模块、3个深度卷积层以及4个后置归一化层，将所述深度卷积层和归一化层叠加作为一个组合，通过跳跃连接形成残差单元，采用3个深度卷积层加上归一化层组合替换前馈网络，并将多头自注意力模块前面的归一化层进行后置，使得归一化层从每个残差单元的开始移动到后端。
本申请实施例采取的技术方案还包括：所述UMRFormer网络模型的训练过程包括：
在编码器部分：使用线性投影将前两层的通道大小分别从K 1＝128,K 2＝256增加到d 1＝512,d 2＝1024，将输入的3D特征图的空间维度和深层维度统一压缩为一维，将压缩后的特征图添加到可学习的位置编码中，创建一个特性嵌入的公式:
z 1＝f 1+PE 1＝LP 1×M 1+PE 1
z 2＝f 2+PE 2＝LP 2×M 2+PE 2
其中，LP 1和LP 2为线性投影运算，
Figure PCTCN2022101894-appb-000001
和
Figure PCTCN2022101894-appb-000002
表示位置编码，
Figure PCTCN2022101894-appb-000003
和
Figure PCTCN2022101894-appb-000004
表示特征嵌入；第l层(l∈[1,2,..,L])类Transformer模块的输出公式为:
z′ l＝LN(MHA(z l-1))+z l-1
z l＝LN 3(DConv 3(LN 2(DConv 2(LN 1(DConv 1(z′ l))))))+z′ l
其中，LN (*)为层归一化，z l为第l层的MRFormer模块的输出，DConv (*)为深度卷积。
本申请实施例采取的技术方案还包括：所述UMRFormer网络模型的训练过程还包括：
在解码器部分:将所述双层的MRFormers模块输出的特征序列映射成三维特征映射，对所述三维特征映射进行上采样和卷积，并通过长距离跳转连接与所述编码器特征融合，获取语义和空间细节信息。
本申请实施例采取的另一技术方案为：一种CT胰腺肿瘤自动分割系统，包括：
数据获取模块：用于获取待分割的三维CT图像；
数据预处理模块：用于对所述三维CT图像进行预处理；
数据分割模块：用于将所述预处理后的三维CT图像输入训练好的UMRFormer网络模型，所述UMRFormer网络模型为U形对称结构，所述UMRFormer网络模型利用3D CNN生成不同尺度的特征图，并利用双层的类Transformer模块对所述特征图的长程依赖语义信息进行编码，输出所述三维CT图像的肿瘤分割结果。
本申请实施例采取的又一技术方案为：一种终端，所述终端包括处理器、与所述处理器耦接的存储器，其中，
所述存储器存储有用于实现所述CT胰腺肿瘤自动分割方法的程序指令；
所述处理器用于执行所述存储器存储的所述程序指令以控制CT胰腺肿瘤自动分割。
本申请实施例采取的又一技术方案为：一种存储介质，存储有处理器可运行的程序指令，所述程序指令用于执行所述CT胰腺肿瘤自动分割方法。
相对于现有技术，本申请实施例产生的有益效果在于：本申请实施例的CT胰腺肿瘤自动分割方法、系统、终端以及存储介质提出了一种新型的深度融合特征全局长距离相关性和空间局部信息的UMRFormer网络模型，该网络模型基于u形的编解码器结构，在U型结构的跳跃连接阶段，嵌入具有融合关注特征局部和全局依赖性优势的类Transformer模块，能够使网络模型学习到更多胰腺肿瘤小目标的特征信息以及肿瘤与周边组织间的全局长依赖特征信息，以获取更有效的深层特征上下文信息，从而改善现有的具有一定缺陷的单一类Transformer网络框架，提高网络整体的分割精度。
附图说明
图1是本申请第一实施例的CT胰腺肿瘤自动分割方法的流程图；
图2是本申请第二实施例的CT胰腺肿瘤自动分割方法的流程图；
图3为本申请实施例的UMRFormer网络模型架构图；
图4为本申请实施例的MRFormer模块结构示意图；
图5为本申请实施例的CT胰腺肿瘤自动分割系统结构示意图；
图6为本申请实施例的终端结构示意图；
图7为本申请实施例的存储介质的结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实 施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
本申请中的术语“第一”、“第二”、“第三”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”、“第三”的特征可以明示或者隐含地包括至少一个该特征。本申请的描述中，“多个”的含义是至少两个，例如两个，三个等，除非另有明确具体的限定。本申请实施例中所有方向性指示(诸如上、下、左、右、前、后……)仅用于解释在某一特定姿态(如附图所示)下各部件之间的相对位置关系、运动情况等，如果该特定姿态发生改变时，则该方向性指示也相应地随之改变。此外，术语“包括”和“具有”以及它们任何变形，意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、系统、产品或设备没有限定于已列出的步骤或单元，而是可选地还包括没有列出的步骤或单元，或可选地还包括对于这些过程、方法、产品或设备固有的其它步骤或单元。
在本文中提及“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是，本文所描述的实施例可以与其它实施例相结合。
请参阅图1，是本申请第一实施例的CT胰腺肿瘤自动分割方法的流程图。本申请实施例的CT胰腺肿瘤自动分割方法包括以下步骤：
S100：获取待分割的三维CT图像；
S101：对三维CT图像进行预处理；
本步骤中，对三维CT图像进行预处理具体为：对三维CT图像进行增强处理，并将增强后的三维CT图像剪裁为符合UMRFormer网络模型输入的尺寸。
S102：将预处理后的三维CT图像输入训练好的UMRFormer网络模型，UMRFormer网络模型为U形对称结构，UMRFormer网络模型利用3D CNN生成不同尺度的特征图，并利用双层的类Transformer模块对特征图的长程依赖语义信息进行编码，输出三维CT图像的肿瘤分割结果。
本步骤中，UMRFormer网络模型包括卷积层、多头自注意力机制层、池化层、归一化层以及上采样层，UMRFormer网络模型为U形对称结构，在U型对称网络的跳跃连接阶段，嵌入有双层的类Transformer模块；UMRFormer网络模型利用5层3D CNN生成不同尺度的特征图，捕捉空间和深度特征，然后利用双层的类Transformer模块对全局空间中的第四层和第五层特征图的长程依赖语义信息进行编码，并将上采样层和卷积层反复叠加，得到三维CT图像的分割结果。
请参阅图2，是本申请第二实施例的CT胰腺肿瘤自动分割方法的流程图。本申请第二实施例的CT胰腺肿瘤自动分割方法包括以下步骤：
S200：获取CT胰腺数据集，按照设定比例将CT胰腺数据集划分为训练集、验证集和测试集；
本步骤中，使用的CT胰腺数据集包括来自于MSD胰腺数据集的281个CT案例。CT胰腺数据集的划分比例为6:2:2，即得到的训练集、验证集和测试集数量分别为187、47和47。
S201：对CT胰腺数据集进行预处理，并设置模型训练参数；
本步骤中，数据预处理具体为：对CT胰腺数据进行增强处理，并将增强后的CT胰腺数据剪裁为符合模型输入的尺寸，例如，原始CT胰腺数据的大小为512x512x148，裁剪后的CT胰腺数据的大小为128x128x64，以适应模型训练的输入要求。设置模型训练参数包括但不限于epoch(迭代次数)、batch(批)大小以及学习率等参数，优选地，本申请实施例中设定epoch＝1000，具体可根据应用场景进行设置。
S202：将预处理后的训练集输入UMRFormer网络模型进行迭代训练，并判断迭代次数达到设定次数后，输出CT胰腺肿瘤分割模型；
本步骤中，本申请实施例通过深度融合卷积网络的关注局部信息和Transformer模块的全局特性建构UMRFormer网络模型，如图3所示，为本申请实施例的UMRFormer网络模型架构图。该模型包括卷积层、多头自注意力机 制(MHSA)层、池化层、归一化层以及上采样层，网络模型整体呈U形对称结构，分为编码-解码两个阶段，在U型网络的跳跃连接阶段，嵌入双层的类Transformer模块。给定一个输入的三维CT图像，其空间分辨率为HxW，深度维数为D(切片数)，通道数为C(模态数)，UMRFormer网络模型首先利用5层3D CNN生成不同尺度的特征图，捕捉空间和深度特征，然后利用双层的类Transformer模块对全局空间中的第四层和第五层特征图的长程依赖语义信息进行编码，并将上采样层和卷积层反复叠加，逐渐得到高分辨率的分割结果。
进一步地，本申请实施例中的编码器由L个MRFormer模块组成，具体如图4所示，为本申请实施例的MRFormer模块结构示意图。每个MRFormer模块分别包括一个多头自注意力(MHSA)模块、3个深度卷积层以及4个后置归一化层。相对于现有的Transformer结构，本申请实施例的MRFormer模块将深度卷积层和归一化层叠加作为一个组合，通过跳跃连接形成残差单元，并采用3个深度卷积层加上归一化层组合去替换前馈网络(FFN)，旨在关注特征序列中的局部感受野。另外，将多头自注意力模块前面的归一化层进行后置，即将归一化层从每个残差单元的开始移动到后端，从而可以在网络层上产生更温和的激活值，使得模型训练更加地稳定。
基于上述网络结构，本申请实施例的UMRFormer网络模型训练过程具体为：
在编码器部分：使用线性投影(3x3x3卷积层)将前两层的通道大小分别从K 1＝128,K 2＝256增加到d 1＝512,d 2＝1024。由于双层的MRFormers模块单独需要一个序列作为输入，因此将输入的3D特征图的空间维度和深层维度统一压缩为一维，得到一个
Figure PCTCN2022101894-appb-000005
和
Figure PCTCN2022101894-appb-000006
的特征图f 1和f 2，分别为Q 1d 1维标签和Q 2d 2维标签。将3D特征图压缩为一维后，添加到可学习的位置编码中，创建一个特性嵌入的公式:
z 1＝f 1+PE 1＝LP 1×M 1+PE 1   (1)
z 2＝f 2+PE 2＝LP 2×M 2+PE 2   (2)
其中，LP 1和LP 2为线性投影运算，
Figure PCTCN2022101894-appb-000007
和
Figure PCTCN2022101894-appb-000008
表示位置编码，
Figure PCTCN2022101894-appb-000009
和
Figure PCTCN2022101894-appb-000010
表示特征嵌入。第l层(l∈[1,2,..,L])类Transformer模块的输出公式可表示为:
z′ l＝LN(MHA(z l-1))+z l-1   (3)
z l＝LN 3(DConv 3(LN 2(DConv 2(LN 1(DConv 1(z′ l))))))+z′ l    (4)
其中的LN (*)为层归一化，z l为第l层的MRFormer模块的输出，DConv (*)为深度卷积。
在解码器部分:为了适应解码器的输入尺寸，需要将双层的MRFormers模块输出的特征序列映射成三维特征映射。双层的MRFormers模块(即:
Figure PCTCN2022101894-appb-000011
和
Figure PCTCN2022101894-appb-000012
)分别被重塑到
Figure PCTCN2022101894-appb-000013
和
Figure PCTCN2022101894-appb-000014
尺寸大小。特征映射后，分别对Z′ 1和Z′ 2的输出进行上采样和卷积，然后通过长距离跳转连接与编码器特征融合，以获取更精细的语义和细粒度信息以及更丰富的空间细节。
S203：将验证集输入CT胰腺肿瘤分割模型，对CT胰腺肿瘤分割模型的CT胰腺/胰腺肿瘤的初步分割结果进行验证，并调整模型超参数；
S204：将测试集输入CT胰腺肿瘤分割模型，测试模型的CT胰腺/胰腺肿瘤分割性能。
基于上述，本申请实施例的CT胰腺肿瘤自动分割方法提出了一种新型的深度融合特征全局长距离相关性和空间局部信息的UMRFormer网络模型，该网络模型基于u形的编解码器结构，在U型结构的跳跃连接阶段，嵌入具有融合关注特征局部和全局依赖性优势的类Transformer模块，能够使网络模型学习到更多胰腺肿瘤小目标的特征信息以及肿瘤与周边组织间的全局长依赖特征 信息，以获取更有效的深层特征上下文信息，从而改善现有的具有一定缺陷的单一类Transformer网络框架，提高网络整体的分割精度。
请参阅图5，为本申请实施例的CT胰腺肿瘤自动分割系统结构示意图。本申请实施例的CT胰腺肿瘤自动分割系统40包括：
数据获取模块41：用于获取待分割的三维CT图像；
数据预处理模块42：用于对三维CT图像进行预处理；其中，对三维CT图像进行预处理具体为：对三维CT图像进行增强处理，并将增强后的三维CT图像剪裁为符合UMRFormer网络模型输入的尺寸。
数据分割模块43：用于将预处理后的三维CT图像输入训练好的UMRFormer网络模型，UMRFormer网络模型为U形对称结构，UMRFormer网络模型利用3D CNN生成不同尺度的特征图，并利用双层的类Transformer模块对特征图的长程依赖语义信息进行编码，输出三维CT图像的肿瘤分割结果；其中，UMRFormer网络模型包括卷积层、多头自注意力机制层、池化层、归一化层以及上采样层，UMRFormer网络模型为U形对称结构，在U型对称网络的跳跃连接阶段，嵌入有双层的类Transformer模块；UMRFormer网络模型利用5层3D CNN生成不同尺度的特征图，捕捉空间和深度特征，然后利用双层的类Transformer模块对全局空间中的第四层和第五层特征图的长程依赖语义信息进行编码，并将上采样层和卷积层反复叠加，得到三维CT图像的分割结果。
基于上述，本申请实施例的CT胰腺肿瘤自动分割系统提出了一种新型的深度融合特征全局长距离相关性和空间局部信息的UMRFormer网络模型，该网络模型基于u形的编解码器结构，在U型结构的跳跃连接阶段，嵌入具有融合关注特征局部和全局依赖性优势的类Transformer模块，能够使网络模型学习到更多胰腺肿瘤小目标的特征信息以及肿瘤与周边组织间的全局长依赖特征 信息，以获取更有效的深层特征上下文信息，从而改善现有的具有一定缺陷的单一类Transformer网络框架，提高网络整体的分割精度。
请参阅图6，为本申请实施例的终端结构示意图。该终端50包括处理器51、与处理器51耦接的存储器52。
存储器52存储有用于实现上述CT胰腺肿瘤自动分割方法的程序指令。
处理器51用于执行存储器52存储的程序指令以控制CT胰腺肿瘤自动分割。
其中，处理器51还可以称为CPU(Central Processing Unit，中央处理单元)。处理器51可能是一种集成电路芯片，具有信号的处理能力。处理器51还可以是通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现成可编程门阵列(FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
请参阅图7，为本申请实施例的存储介质的结构示意图。本申请实施例的存储介质存储有能够实现上述所有方法的程序文件61，其中，该程序文件61可以以软件产品的形式存储在上述存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)或处理器(processor)执行本申请各个实施方式方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质，或者是计算机、服务器、手机、平板等终端设备。
以上对发明的具体实施方式进行了详细说明，但其只作为范例，本发明并不限制于以上描述的具体实施方式。对于本领域的技术人员而言，任何对该发明进行的等同修改或替代也都在本发明的范畴之中，因此，在不脱离本发明的精神和原则范围下所作的均等变换和修改、改进等，都应涵盖在本发明的范围内。

Claims (9)

  1. 一种CT胰腺肿瘤自动分割方法，其特征在于，包括：
    获取待分割的三维CT图像；
    对所述三维CT图像进行预处理；
    将所述预处理后的三维CT图像输入训练好的UMRFormer网络模型，所述UMRFormer网络模型为U形对称结构，所述UMRFormer网络模型利用3D CNN生成不同尺度的特征图，并利用双层的类Transformer模块对所述特征图的长程依赖语义信息进行编码，输出所述三维CT图像的肿瘤分割结果。
  2. 根据权利要求1所述的CT胰腺肿瘤自动分割方法，其特征在于，所述对所述三维CT图像进行预处理具体为：
    对所述三维CT图像进行增强处理，并将增强后的三维CT图像剪裁为符合所述UMRFormer网络模型输入的尺寸。
  3. 根据权利要求1或2所述的CT胰腺肿瘤自动分割方法，其特征在于，所述UMRFormer网络模型包括卷积层、多头自注意力机制层、池化层、归一化层以及上采样层，所述UMRFormer网络模型为U形对称结构，在所述U型对称网络的跳跃连接阶段，嵌入有双层的类Transformer模块；
    所述UMRFormer网络模型利用5层3D CNN生成不同尺度的特征图，捕捉空间和深度特征，然后利用双层的类Transformer模块对全局空间中的第四层和第五层特征图的长程依赖语义信息进行编码，并将上采样层和卷积层反复叠加，得到三维CT图像的分割结果。
  4. 根据权利要求3所述的CT胰腺肿瘤自动分割方法，其特征在于，所述UMRFormer网络模型包括编码器和解码器，所述编码器包括L个MRFormer模块， 每个MRFormer模块分别包括1个多头自注意力模块、3个深度卷积层以及4个后置归一化层，将所述深度卷积层和归一化层叠加作为一个组合，通过跳跃连接形成残差单元，采用3个深度卷积层加上归一化层组合替换前馈网络，并将多头自注意力模块前面的归一化层进行后置，使得归一化层从每个残差单元的开始移动到后端。
  5. 根据权利要求4所述的CT胰腺肿瘤自动分割方法，其特征在于，所述UMRFormer网络模型的训练过程包括：
    在编码器部分：使用线性投影将前两层的通道大小分别从K 1＝128,K 2＝256增加到d 1＝512,d 2＝1024，将输入的3D特征图的空间维度和深层维度统一压缩为一维，将压缩后的特征图添加到可学习的位置编码中，创建一个特性嵌入的公式:
    z 1＝f 1+PE 1＝LP 1×M 1+PE 1
    z 2＝f 2+PE 2＝LP 2×M 2+PE 2
    其中，LP 1和LP 2为线性投影运算，
    Figure PCTCN2022101894-appb-100001
    和
    Figure PCTCN2022101894-appb-100002
    表示位置编码，
    Figure PCTCN2022101894-appb-100003
    和
    Figure PCTCN2022101894-appb-100004
    表示特征嵌入；第l层(l∈[1,2,..,L])类Transformer模块的输出公式为:
    z′ l＝LN(MHA(z l-1))+z l-1
    z l＝LN 3(DConv 3(LN 2(DConv 2(LN 1(DConv 1(z′ l))))))+z′ l
    其中，LN (*)为层归一化，z l为第l层的MRFormer模块的输出，DConv (*)为深度卷积。
  6. 根据权利要求5所述的CT胰腺肿瘤自动分割方法，其特征在于，所述UMRFormer网络模型的训练过程还包括：
    在解码器部分:将所述双层的MRFormers模块输出的特征序列映射成三维特征映射，对所述三维特征映射进行上采样和卷积，并通过长距离跳转连接与所述编码器特征融合，获取语义和空间细节信息。
  7. 一种CT胰腺肿瘤自动分割系统，其特征在于，包括：
    数据获取模块：用于获取待分割的三维CT图像；
    数据预处理模块：用于对所述三维CT图像进行预处理；
    数据分割模块：用于将所述预处理后的三维CT图像输入训练好的UMRFormer网络模型，所述UMRFormer网络模型为U形对称结构，所述UMRFormer网络模型利用3D CNN生成不同尺度的特征图，并利用双层的类Transformer模块对所述特征图的长程依赖语义信息进行编码，输出所述三维CT图像的肿瘤分割结果。
  8. 一种终端，其特征在于，所述终端包括处理器、与所述处理器耦接的存储器，其中，
    所述存储器存储有用于实现权利要求1-6任一项所述的CT胰腺肿瘤自动分割方法的程序指令；
    所述处理器用于执行所述存储器存储的所述程序指令以控制CT胰腺肿瘤自动分割。
  9. 一种存储介质，其特征在于，存储有处理器可运行的程序指令，所述程序指令用于执行权利要求1至6任一项所述CT胰腺肿瘤自动分割方法。
PCT/CN2022/101894 2022-06-28 2022-06-28 Ct胰腺肿瘤自动分割方法、系统、终端以及存储介质 WO2024000161A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/101894 WO2024000161A1 (zh) 2022-06-28 2022-06-28 Ct胰腺肿瘤自动分割方法、系统、终端以及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/101894 WO2024000161A1 (zh) 2022-06-28 2022-06-28 Ct胰腺肿瘤自动分割方法、系统、终端以及存储介质

Publications (1)

Publication Number Publication Date
WO2024000161A1 true WO2024000161A1 (zh) 2024-01-04

Family

ID=89383706

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/101894 WO2024000161A1 (zh) 2022-06-28 2022-06-28 Ct胰腺肿瘤自动分割方法、系统、终端以及存储介质

Country Status (1)

Country Link
WO (1) WO2024000161A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117764994A (zh) * 2024-02-22 2024-03-26 浙江首鼎视介科技有限公司 基于人工智能的胆胰成像系统及方法
CN117911705A (zh) * 2024-03-19 2024-04-19 成都理工大学 一种基于GAN-UNet变体网络的脑部MRI肿瘤分割方法
CN118052821A (zh) * 2024-04-15 2024-05-17 苏州凌影云诺医疗科技有限公司 一种针对反流性食管炎的病灶检测和分级方法及装置
CN118229981A (zh) * 2024-05-23 2024-06-21 山东未来网络研究院(紫金山实验室工业互联网创新应用基地) 一种结合卷积网络和Transformer的CT图像肿瘤分割方法、装置和介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113888466A (zh) * 2021-09-03 2022-01-04 武汉科技大学 一种基于ct图像的肺结节图像检测方法及系统
CN114119977A (zh) * 2021-12-01 2022-03-01 昆明理工大学 一种基于图卷积的Transformer胃癌癌变区域图像分割方法
CN114372531A (zh) * 2022-01-11 2022-04-19 北京航空航天大学 一种基于自注意力特征融合的胰腺癌病理图像分类方法
CN114463339A (zh) * 2022-01-10 2022-05-10 武汉大学 一种基于自注意力的医疗影像分割方法
CN114494296A (zh) * 2022-01-27 2022-05-13 复旦大学 一种基于Unet和Transformer相融合的脑部胶质瘤分割方法与系统
CN114519718A (zh) * 2022-02-21 2022-05-20 云南大学 一种腹部多器官ct图像分割方法及系统
US20220180506A1 (en) * 2020-12-03 2022-06-09 Ping An Technology (Shenzhen) Co., Ltd. Method, device, and storage medium for pancreatic mass segmentation, diagnosis, and quantitative patient management

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220180506A1 (en) * 2020-12-03 2022-06-09 Ping An Technology (Shenzhen) Co., Ltd. Method, device, and storage medium for pancreatic mass segmentation, diagnosis, and quantitative patient management
CN113888466A (zh) * 2021-09-03 2022-01-04 武汉科技大学 一种基于ct图像的肺结节图像检测方法及系统
CN114119977A (zh) * 2021-12-01 2022-03-01 昆明理工大学 一种基于图卷积的Transformer胃癌癌变区域图像分割方法
CN114463339A (zh) * 2022-01-10 2022-05-10 武汉大学 一种基于自注意力的医疗影像分割方法
CN114372531A (zh) * 2022-01-11 2022-04-19 北京航空航天大学 一种基于自注意力特征融合的胰腺癌病理图像分类方法
CN114494296A (zh) * 2022-01-27 2022-05-13 复旦大学 一种基于Unet和Transformer相融合的脑部胶质瘤分割方法与系统
CN114519718A (zh) * 2022-02-21 2022-05-20 云南大学 一种腹部多器官ct图像分割方法及系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YOUYANG SHA; YONGHONG ZHANG; XUQUAN JI; LEI HU: "Transformer-Unet: Raw Image Processing with Unet", ARXIV.ORG, 17 September 2021 (2021-09-17), XP091057481 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117764994A (zh) * 2024-02-22 2024-03-26 浙江首鼎视介科技有限公司 基于人工智能的胆胰成像系统及方法
CN117764994B (zh) * 2024-02-22 2024-05-10 浙江首鼎视介科技有限公司 基于人工智能的胆胰成像系统及方法
CN117911705A (zh) * 2024-03-19 2024-04-19 成都理工大学 一种基于GAN-UNet变体网络的脑部MRI肿瘤分割方法
CN117911705B (zh) * 2024-03-19 2024-05-28 成都理工大学 一种基于GAN-UNet变体网络的脑部MRI肿瘤分割方法
CN118052821A (zh) * 2024-04-15 2024-05-17 苏州凌影云诺医疗科技有限公司 一种针对反流性食管炎的病灶检测和分级方法及装置
CN118052821B (zh) * 2024-04-15 2024-06-14 苏州凌影云诺医疗科技有限公司 一种针对反流性食管炎的病灶检测和分级方法及装置
CN118229981A (zh) * 2024-05-23 2024-06-21 山东未来网络研究院(紫金山实验室工业互联网创新应用基地) 一种结合卷积网络和Transformer的CT图像肿瘤分割方法、装置和介质
CN118229981B (zh) * 2024-05-23 2024-07-23 山东未来网络研究院(紫金山实验室工业互联网创新应用基地) 一种结合卷积网络和Transformer的CT图像肿瘤分割方法、装置和介质

Similar Documents

Publication Publication Date Title
WO2024000161A1 (zh) Ct胰腺肿瘤自动分割方法、系统、终端以及存储介质
CN115239637A (zh) 一种ct胰腺肿瘤自动分割方法、系统、终端以及存储介质
WO2021179205A1 (zh) 医学图像分割方法、医学图像分割装置及终端设备
TWI835768B (zh) 用於基於神經網路之快速影像分段及放射性藥品之攝取的測定之系統及方法
WO2022199462A1 (zh) 医学图像报告生成模型的训练方法及图像报告生成方法
WO2022001623A1 (zh) 基于人工智能的图像处理方法、装置、设备及存储介质
Ates et al. Dual cross-attention for medical image segmentation
Xie et al. Semantics lead all: Towards unified image registration and fusion from a semantic perspective
CN114494296A (zh) 一种基于Unet和Transformer相融合的脑部胶质瘤分割方法与系统
CN113888466A (zh) 一种基于ct图像的肺结节图像检测方法及系统
CN113935943A (zh) 颅内动脉瘤识别检测的方法、装置、计算机设备和存储介质
WO2023108526A1 (zh) 一种医学图像分割方法、系统、终端以及存储介质
CN115880317A (zh) 一种基于多分支特征融合精炼的医学图像分割方法
Tang et al. Stroke-based scene text erasing using synthetic data for training
CN116563533A (zh) 基于目标位置先验信息的医学图像分割方法及系统
Zou et al. Graph flow: Cross-layer graph flow distillation for dual efficient medical image segmentation
Sui et al. Cst: A multitask learning framework for colorectal cancer region mining based on transformer
Lin et al. Measurement of Body Surface Area for Psoriasis Using U‐net Models
Hartsock et al. Vision-language models for medical report generation and visual question answering: A review
CN113052857A (zh) 一种基于CovSegNet的肺部病变图像分割方法
Zhang et al. SC-Net: Symmetrical conical network for colorectal pathology image segmentation
Yang et al. Multi‐scale attention network for segmentation of electron dense deposits in glomerular microscopic images
CN116030247A (zh) 一种医学图像样本生成方法、装置、存储介质及电子设备
Teng et al. Blind face restoration via multi-prior collaboration and adaptive feature fusion
Guan et al. MonoPoly: A practical monocular 3D object detector

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: 22948283

Country of ref document: EP

Kind code of ref document: A1