WO2020114332A1 - 基于分割网络的ct肺肿瘤分割方法、装置、设备及介质 - Google Patents

基于分割网络的ct肺肿瘤分割方法、装置、设备及介质 Download PDF

Info

Publication number
WO2020114332A1
WO2020114332A1 PCT/CN2019/122103 CN2019122103W WO2020114332A1 WO 2020114332 A1 WO2020114332 A1 WO 2020114332A1 CN 2019122103 W CN2019122103 W CN 2019122103W WO 2020114332 A1 WO2020114332 A1 WO 2020114332A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
lung
segmentation
preset
sample
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/CN2019/122103
Other languages
English (en)
French (fr)
Inventor
贺宝春
贾富仓
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 WO2020114332A1 publication Critical patent/WO2020114332A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • 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
    • G06T2207/10081—Computed x-ray tomography [CT]
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/30061—Lung
    • 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

Definitions

  • the invention belongs to the technical field of medical image processing, and particularly relates to a CT lung tumor segmentation method, device, equipment and medium based on a segmentation network.
  • Lung cancer is the most common malignant tumor in the world, not only has a high incidence rate, but also ranks first in malignant tumors, and the mortality rate also ranks first.
  • people's living environment and lifestyle have undergone tremendous changes.
  • the smoking population has expanded. People have been under long-term psychological pressure, lack of exercise, and atmospheric pollution has increased, resulting in the continued increase in the incidence of lung cancer.
  • Computer tomography Computer tomography (Computed Tomography, CT for short) technology continues to affect the diagnosis of human diseases, and CT images have become one of the important conventional methods in clinical diagnosis and treatment.
  • lung cancer treatment methods mainly include tumor resection, intervention, and radiation therapy.
  • Tumor resection is the most effective treatment method.
  • CT tumor segmentation is usually used for surgical planning before tumor resection, intraoperative 3D visualization and visualization, surgical resection plan design, and surgical risk assessment. It is used in computer-aided diagnosis to extract features of organs or tissues for qualitative and quantitative analysis.
  • the position, size and shape of lung tumors are variable, and the gray scale is similar to the adjacent pulmonary blood vessels, and small tumors are difficult to identify.
  • Traditional manual manual segmentation requires anatomical knowledge and experience, subjective differences, and it takes a lot of time and effort until the breakthrough progress of deep convolutional neural networks makes automatic segmentation of lung tumors possible.
  • the existing methods of automatic segmentation of lung tumors rely on the segmentation of lung tumors based on the threshold segmentation method. This method is easy to lose the rich texture information of the lungs, so that it is easy to remove the larger lung tumors, resulting in incomplete lung tumor segmentation.
  • the purpose of the present invention is to provide a CT lung tumor segmentation method, device, equipment and storage medium based on a segmentation network, aiming to solve the problem that the prior art cannot provide an effective method for segmenting lung tumors in CT images of lungs , Resulting in low segmentation accuracy and incomplete lung tumors.
  • the present invention provides a CT lung tumor segmentation method based on a segmentation network, the method includes the following steps:
  • a pre-trained second segmentation network is used to obtain a lung tumor segmented image corresponding to the lung CT image.
  • the present invention provides a CT lung tumor segmentation device based on a segmentation network, the device includes:
  • the CT image preprocessing unit is configured to preprocess the lung CT image when receiving a lung tumor segmentation request for the lung CT image to obtain a corresponding lung preprocessed image;
  • a first image segmentation unit configured to perform image segmentation on the lung pre-processed image through a pre-trained first segmentation network to obtain a corresponding lung tumor initial segmented image
  • An image resampling unit for resampling the lung pre-processed image and the lung tumor initial segmented image according to a preset pixel interval to obtain the corresponding lung sampled image and initial segmented sampled image;
  • An image block obtaining unit configured to crop the initial divided sample image according to a preset image block size with the center of mass of the initial divided sample image as the center to obtain a preset number of initial divided image blocks;
  • a second image segmentation unit is used to obtain a lung tumor segmented image corresponding to the lung CT image using a pre-trained second segmentation network based on the lung sampled image and the initial segmented image block.
  • the present invention also provides a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, which is implemented when the processor executes the computer program
  • a computing device including a memory, a processor, and a computer program stored in the memory and executable on the processor, which is implemented when the processor executes the computer program
  • the present invention also provides a computer-readable storage medium that stores a computer program, which when executed by a processor implements the CT lung tumor segmentation method based on the segmentation network as described above The steps described.
  • the present invention uses a first segmentation network to perform image segmentation on the preprocessed lung pre-processed images to obtain an initial lung tumor segmented image, and resamples the lung pre-processed image and the lung tumor initial segmented image according to the pixel pitch to obtain Corresponding lung sampled image and initial segmented sampled image, calculate the centroid of the initial segmented sampled image, crop the initial segmented sampled image to the initial segmented image block of the preset number of image block sizes with the centroid as the center, according to the lung sampled image And the initial segmented image block, the second segmentation network is used to obtain the lung tumor segmentation image corresponding to the lung CT image, so as to locate the lung tumor region in the lung CT image through the first segmentation network, and then the lung tumors through the second segmentation network Segmentation improves the accuracy and integrity of lung tumor segmentation of lung CT images, and then obtains high-precision lung tumor segmentation images, which improves the safety of surgery.
  • FIG. 1 is an implementation flowchart of a segmentation network-based CT lung tumor segmentation method provided in Embodiment 1 of the present invention
  • Embodiment 2 is a flow chart of the implementation of preprocessing of lung CT images provided by Embodiment 2 of the present invention
  • FIG. 3 is a flowchart of implementing a preset segmentation network provided by Embodiment 3 of the present invention.
  • FIG. 4 is a schematic structural diagram of a CT lung tumor segmentation device based on a segmentation network according to Embodiment 4 of the present invention.
  • FIG. 5 is a schematic diagram of a preferred structure of a CT lung tumor segmentation device based on a segmentation network provided by Embodiment 4 of the present invention
  • FIG. 6 is a schematic structural diagram of a CT lung tumor segmentation device based on a segmentation network according to Embodiment 5 of the present invention.
  • FIG. 7 is a schematic structural diagram of a computing device according to Embodiment 6 of the present invention.
  • FIG. 1 shows an implementation process of a CT lung tumor segmentation method based on a segmentation network provided in Embodiment 1 of the present invention.
  • FIG. 1 shows an implementation process of a CT lung tumor segmentation method based on a segmentation network provided in Embodiment 1 of the present invention.
  • the details are as follows:
  • step S101 when receiving a lung tumor segmentation request for a lung CT image, the lung CT image is preprocessed to obtain a corresponding lung preprocessed image.
  • the embodiments of the present invention are applicable to medical image processing platforms, systems or devices, such as personal computers and servers.
  • the CT image is composed of a certain number of pixels with different gray levels from black to white arranged in a matrix, and the different gray levels of the CT image reflect the degree of X-ray absorption by the corresponding organs and tissues
  • the black area indicates a low absorption area, that is, a low-density area, such as a lung containing a large amount of gas
  • the white area indicates a high absorption area, that is, a high-density area, such as bones.
  • the anatomical structure and the characteristics of the diseased tissues with changes in the density of the organs are CT scans of the lungs of the patients to generate CT images of the lungs, so as to find the diseased tissues of the lungs of the patients through the CT images of the lungs, and then treat them.
  • a lung tumor segmentation request is sent, where the user can obtain the patient’s lung CT image from the published medical image database or the surgical image provided by the hospital.
  • the lung CT image input by the user is acquired, and the acquired lung CT image is preprocessed to obtain a corresponding lung preprocessed image.
  • step S102 the pre-trained first segmentation network performs image segmentation on the lung pre-processed image to obtain a corresponding initial segmented image of the lung tumor.
  • image segmentation is a technique and process of dividing an image into several specific areas with unique properties and proposing objects of interest.
  • the lung preprocessed image is input into a pre-trained first segmentation network for image segmentation, and the lung tumor region in the lung preprocessed image is extracted to obtain a corresponding coarse-grained lung tumor initial segmented image.
  • step S103 the lung pre-processed image and the lung tumor initial segmented image are re-sampled according to the preset pixel pitch to obtain the corresponding lung sampled image and initial segmented sampled image.
  • the lung pre-processed image is resampled to a preset pixel interval (ie, pixel space interval) to obtain a lung sampled image
  • the lung tumor initial segmented image is also resampled to the pixel interval To get the corresponding initial segmented sample image.
  • the pixel interval Before re-sampling the lung pre-processed image and the lung tumor initial segmented image according to the preset pixel interval, it is preferable to set the pixel interval to 1mm ⁇ 1mm ⁇ 2mm, so as to increase the lung sampled image and the initial segmented sampling The image information of the image, thereby improving the image quality of the lung sampled image and the initial divided sampled image.
  • step S104 the initial divided sample image is cropped according to the preset image block size with the centroid of the initial divided sample image as the center to obtain a preset number of initial divided image blocks.
  • the centroid of the initial divided sample image is calculated, and a preset number of initial divided image blocks of a preset image block size are extracted from the initial divided sample image centering on the calculated centroid.
  • the image block size is set to 128 ⁇ 128 ⁇ 64, thereby improving the image quality of the initial segmented image block.
  • step S105 based on the lung sampled image and the initial segmented image block, a pre-trained second segmentation network is used to obtain a lung tumor segmented image corresponding to the lung CT image.
  • the lung sampling image and all the initial segmented image blocks are input into the pre-trained second segmentation network, and the lung tumor segmentation image is output through the second segmentation network, that is, the lung CT image The final, fine-grained lung tumor segmentation image obtained by lung tumor region segmentation.
  • the first segmentation network is used to perform image segmentation on the preprocessed lung pre-processed image to obtain an initial lung tumor segmented image, and the lung pre-processed image and the lung tumor initial segmented image are divided according to the pixel distance Perform resampling to obtain the corresponding lung sampled image and initial segmented sampled image, crop the initial segmented sampled image to the initial segmented image block of a preset number of image block sizes centering on the centroid of the initial segmented sampled image, and sample according to the lungs
  • Tumor segmentation improves the accuracy and completeness of lung tumor segmentation of lung CT images, and then obtains high-precision lung tumor segmentation images, improving the safety of surgery.
  • FIG. 2 shows an implementation process of preprocessing lung CT images provided by Embodiment 2 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown. The details are as follows:
  • step S201 the air communication area in the lung CT image is extracted according to a preset seed point.
  • a region growing method is used to extract the air communication region in the lung CT image, which is also the background region of the lung CT image.
  • the first pixel point in the upper left corner of the lung CT image is set as the seed point, thereby improving the accuracy and completeness of the extraction of the air communication area.
  • step S202 the extracted air communication area is removed from the lung CT image to obtain a corresponding body mask area image.
  • the air communication area is cropped from the lung CT image to obtain the corresponding body mask area image, thereby removing the noise in the lung CT image and improving the image quality of the obtained body mask area image .
  • step S203 threshold image segmentation is performed on the body mask region image to obtain the maximum connected region of the left lung and the maximum connected region of the right lung.
  • the body mask region is image segmented by a threshold segmentation method to obtain the left lung maximum connected region and the right lung maximum connected region in the body mask region image.
  • threshold segmentation On the image of the body mask area, it is preferable to perform image segmentation on the body mask area according to the threshold value range [-1024, -400], so as to simplify the image segmentation of the body mask area At the same time, the accuracy and completeness of the segmentation of the left lung maximum connected area and the right lung maximum connected area are improved.
  • step S204 the region of interest of the lung CT image is extracted based on the maximum connected region of the left lung and the maximum connected region of the right lung to obtain a corresponding lung region of interest image.
  • step S205 grayscale transformation is performed on the lung region of interest image according to a preset grayscale transformation formula, and the grayscale transformed lung region of interest image is normalized to obtain a lung CT image corresponding Preprocessed images of the lungs.
  • I(x) is the x-th pixel in the left lung maximum connected area or the right lung maximum connected area
  • the gray value of the point, f(x) is the gray value of the xth pixel after gray conversion, thereby improving the contrast of the lung region of interest image, making the lung region of interest image more clear and delicate , And easy to identify, thereby improving the display effect of the lung region of interest image.
  • the grayscale-transformed lung region of interest image is normalized to [0,1], thereby further improving the lung
  • the contrast of the region of interest image makes the details of the lung region of interest image more clear.
  • the lung CT image is preprocessed by air connected region cropping, region of interest extraction, threshold segmentation, grayscale transformation, grayscale normalization, etc., thereby improving the robustness and execution efficiency of the algorithm .
  • FIG. 3 shows an implementation process for training a preset segmentation network provided by Embodiment 3 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown, and details are as follows:
  • step S301 the pre-trained training samples are re-sampled according to the preset sample sampling size to obtain the first sampling training samples.
  • the pre-processed training samples are resampled to a preset sample sampling size to obtain a first sampling training sample, where the training samples are lung tumor CT image data sets with different shapes and sizes.
  • the sample sampling size is set to 128 ⁇ 128 ⁇ 64, so as to increase the image information of the first sampling training sample, thereby improving the image quality of the first sampling training sample.
  • the training samples are pre-processed, specifically, the training samples are pre-processed through the following steps:
  • the steps 1) to 5) are used to preprocess the training samples, which improves the contrast of the image characteristics of the training samples, makes the training samples clearer, more delicate, and easier to identify, thereby improving the training effect of the subsequent segmentation network.
  • step S302 according to the preset number of training iterations, the preset segmentation network is trained through the first sampling training sample to obtain the first segmentation network.
  • iterative training is performed on the preset segmentation network by a preset number of training iterations through the first sampling training sample to obtain a first segmentation network
  • the preset segmentation network is a fully convolutional network (Fully Convolutional Networks, FCN), SegNet, U-Net or V-Net and other semantic segmentation networks
  • the obtained first segmentation network is obtained by inputting the first sampling training sample into the semantic segmentation network and training the semantic segmentation network, so that the first A segmentation network can roughly locate the lung tumor region in the lung CT image, which improves the integrity of the image segmentation of the lung tumor region in the lung CT image through the first segmentation network.
  • step S303 the training samples are re-sampled according to the pixel pitch to obtain second sampled training samples.
  • the pre-processed training samples are re-sampled to a preset pixel interval (ie, pixel space interval) to obtain a second sampling training sample, where the training sample is the same as the training sample in step S301 .
  • a preset pixel interval ie, pixel space interval
  • step S304 the second sampling training sample is cropped according to the image block size with the preset positioning point as the center to obtain a preset number of training sample image blocks, and the positioning point is one of the tumor regions in the second sampling training sample pixel.
  • a pixel point is randomly selected as the positioning point in the tumor area in the second sampling training sample, and then a predetermined number of preset image block sizes are extracted from the second sampling training sample centering on the positioning point. Training sample image block.
  • the pixel gray value of the positioning point is set to 2, so as to increase the brightness of the pixel at the positioning point, thereby improving the subsequent training effect of the second segmentation network.
  • step S305 according to the number of training iterations, the preset segmentation network is trained by training sample image blocks to obtain a second segmentation network.
  • iterative training is performed on the preset segmentation network by the preset number of training iterations through the training sample image blocks to obtain the second segmentation network, and the obtained second segmentation network inputs all the training sample image blocks to the semantics
  • the semantic segmentation network is trained, so that the second segmentation network can accurately locate the lung tumor in the lung CT image, which improves the lung tumor image segmentation of the lung CT image through the second segmentation network Accuracy.
  • the first segmentation network and the second segmentation network are trained in parallel through lung tumor CT image data sets with different shapes and sizes, thereby improving training efficiency and reducing memory consumption during the training process.
  • FIG. 4 shows the structure of a CT lung tumor segmentation device based on a segmentation network provided in Embodiment 4 of the present invention. For ease of explanation, only parts related to the embodiment of the present invention are shown, including:
  • the CT image pre-processing unit 41 is used to pre-process the lung CT image to obtain the corresponding lung pre-processed image when receiving a lung tumor segmentation request for the lung CT image;
  • the first image segmentation unit 42 is used to perform image segmentation on the lung pre-processed image through a pre-trained first segmentation network to obtain a corresponding lung tumor initial segmented image;
  • the image resampling unit 43 is used to resample the lung pre-processed image and the lung tumor initial segmented image according to the preset pixel interval to obtain the corresponding lung sampled image and initial segmented sampled image;
  • the image block obtaining unit 44 is configured to crop the initial divided sample image according to the preset image block size with the center of mass of the initial divided sample image as the center to obtain a preset number of initial divided image blocks;
  • the second image segmentation unit 45 is used to obtain a lung tumor segmented image corresponding to the lung CT image using a pre-trained second segmentation network based on the lung sampled image and the initial segmented image block.
  • the CT image preprocessing unit 41 includes:
  • the first area extraction unit 411 is used to extract the air-connected area in the lung CT image according to a preset seed point
  • the second area extraction unit 412 is used to remove the extracted air communication area from the lung CT image to obtain the corresponding body mask area image;
  • the image threshold segmentation unit 413 is used to perform threshold segmentation on the body mask region image to obtain the maximum connected region of the left lung and the maximum connected region of the right lung;
  • the third region extraction unit 414 is configured to extract the region of interest from the CT image of the lung based on the maximum connected region of the left lung and the maximum connected region of the right lung to obtain a corresponding lung region of interest image;
  • the normalization processing unit 415 is used for performing gray-scale transformation on the lung region of interest image according to a preset gray-scale transformation formula, and performing normalization processing on the gray region-converted lung region of interest image to obtain the lung Lung pre-processed image corresponding to the CT image.
  • each unit of the CT lung tumor segmentation device based on the segmentation network may be implemented by a corresponding hardware or software unit, and each unit may be an independent software and hardware unit, or may be integrated into one software and hardware unit. This is not to limit the invention. Specifically, for the implementation of each unit, reference may be made to the description of the foregoing method embodiments, and details are not described herein again.
  • FIG. 6 shows the structure of a CT lung tumor segmentation device based on a segmentation network provided in Embodiment 5 of the present invention. For ease of description, only parts related to the embodiment of the present invention are shown, including:
  • the first sample sampling unit 60 is configured to resample the pre-processed training samples according to a preset sample sampling size to obtain first sample training samples;
  • the first network training unit 61 is configured to train the preset segmentation network through the first sampling training sample according to the preset number of training iterations to obtain the first segmentation network;
  • the second sample sampling unit 62 is used to resample the training samples according to the pixel spacing to obtain second sample training samples;
  • the second sample cropping unit 63 is configured to crop the second sampling training sample according to the image block size with the preset positioning point as the center, to obtain a preset number of training sample image blocks, and the positioning point is in the second sampling training sample One pixel in the tumor area;
  • the second network training unit 64 is configured to train the preset segmentation network through training sample image blocks according to the number of training iterations to obtain a second segmentation network;
  • the CT image preprocessing unit 65 is used to preprocess the lung CT image when receiving a lung tumor segmentation request for the lung CT image to obtain the corresponding lung preprocessed image;
  • the first image segmentation unit 66 is used to perform image segmentation on the lung pre-processed image through a pre-trained first segmentation network to obtain a corresponding lung tumor initial segmented image;
  • the image resampling unit 67 is used to resample the lung pre-processed image and the lung tumor initial segmented image according to the preset pixel interval to obtain the corresponding lung sampled image and initial segmented sampled image;
  • the image block obtaining unit 68 is configured to crop the initial divided sample image according to the preset image block size with the center of mass of the initial divided sample image as the center to obtain a preset number of initial divided image blocks;
  • the second image segmentation unit 69 is configured to obtain a lung tumor segmented image corresponding to the lung CT image using a pre-trained second segmentation network based on the lung sampled image and the initial segmented image block.
  • each unit of the CT lung tumor segmentation device based on the segmentation network may be implemented by a corresponding hardware or software unit, and each unit may be an independent software and hardware unit, or may be integrated into one software and hardware unit. This is not to limit the invention. Specifically, for the implementation of each unit, reference may be made to the description of the foregoing method embodiments, and details are not described herein again.
  • FIG. 7 shows the structure of the computing device provided in Embodiment 6 of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown.
  • the computing device 7 of the embodiment of the present invention includes a processor 70, a memory 71, and a computer program 72 stored in the memory 71 and executable on the processor 70.
  • the processor 70 executes the computer program 72, the steps in the above embodiment of the CT lung tumor segmentation method based on the segmentation network are implemented, for example, steps S101 to S105 shown in FIG. 1.
  • the processor 70 executes the computer program 72, the functions of the units in the foregoing device embodiments are realized, for example, the functions of the units 41 to 45 shown in FIG. 4.
  • the first segmentation network is used to perform image segmentation on the preprocessed lung pre-processed image to obtain an initial lung tumor segmented image, and the lung pre-processed image and the lung tumor initial segmented image are divided according to the pixel distance Perform resampling to obtain the corresponding lung sampled image and initial segmented sampled image, crop the initial segmented sampled image to the initial segmented image block of a preset number of image block sizes centering on the centroid of the initial segmented sampled image, and sample according to the lungs
  • Tumor segmentation improves the accuracy and completeness of lung tumor segmentation on lung CT images, and then obtains high-precision lung tumor segmentation images, improving the safety of surgery.
  • the computing device in this embodiment of the present invention may be a personal computer or a server.
  • the processor 70 in the computing device 7 executes the computer program 72 to implement the CT network tumor segmentation method based on the segmentation network, reference may be made to the description of the foregoing method embodiments, which will not be repeated here.
  • a computer-readable storage medium stores a computer program.
  • the computer program is executed by a processor, the above-described CT network tumor segmentation method based on segmentation network embodiment is implemented. Steps, for example, steps S101 to S105 shown in FIG. 1.
  • the functions of the units in the foregoing device embodiments are realized, for example, the functions of the units 41 to 45 shown in FIG. 4.
  • the first segmentation network is used to perform image segmentation on the preprocessed lung pre-processed image to obtain an initial lung tumor segmented image, and the lung pre-processed image and the lung tumor initial segmented image are divided according to the pixel distance Perform resampling to obtain the corresponding lung sampled image and initial segmented sampled image, centering on the centroid of the initial segmented sampled image as the center, and cutting the initial segmented sampled image into an initial segmented image block of a preset number of image block sizes, according to the lung sampling
  • Tumor segmentation improves the accuracy and completeness of lung tumor segmentation on lung CT images, and then obtains high-precision lung tumor segmentation images, improving the safety of surgery.
  • the computer-readable storage medium in the embodiments of the present invention may include any entity or device capable of carrying computer program code, and a recording medium, such as ROM/RAM, magnetic disk, optical disk, flash memory, and other memories.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Image Processing (AREA)

Abstract

一种基于分割网络的CT肺肿瘤分割方法、装置、设备及介质，适用医学图像处理技术领域，该方法包括：通过预先训练好的第一分割网络对肺部预处理图像进行图像分割，得到对应的肺肿瘤初始分割图像（S102），根据预设的像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像（S103），以初始分割采样图像的质心为中心、按照预设图像块大小对初始分割采样图像进行裁剪，得到预设数量个初始分割图像块（S104），根据肺部采样图像和初始分割图像块，采用预先训练好的第二分割网络得到肺部CT图像对应的肺肿瘤分割图像（S105），从而提高了对肺部CT图像进行肺肿瘤分割的精确度和完整性，进而获得高精度的肺肿瘤分割图像，提高手术的安全程度。

Description

基于分割网络的CT肺肿瘤分割方法、装置、设备及介质 技术领域
本发明属于医学图像处理技术领域，尤其涉及一种基于分割网络的CT肺肿瘤分割方法、装置、设备及介质。
背景技术
肺癌是全球最常见的恶性肿瘤，不仅病发率高，居恶性肿瘤榜首，且死亡率也位列第一。近年来，由于经济发展，人们的生活环境和生活方式产生了巨大的改变，吸烟人群扩大，人们长期心理压力过大、缺乏锻炼，大气环境污染加重，导致肺癌发病率持续上升。而计算机断层成像(Computed Tomography，简称CT)技术的高速发展不断影响人体疾病的诊断方式，CT图像已成为临床诊疗中重要的常规手段之一。
目前肺肿瘤治疗手段主要包括肿瘤切除、介入、放射治疗等，肿瘤切除是其中最有效的治疗方式。CT肿瘤分割通常用于肿瘤切除前的手术计划、术中三维可视化直观、手术切除方案设计、手术风险评估等，在计算机辅助诊断中用于提取器官或组织的特征，再进行定性和定量分析，以观察放疗过程中肿瘤的变化情况，然而肺肿瘤的位置、大小及形状多变，灰度与相邻的肺血管近似，小肿瘤难以识别。传统人工手动分割需要具有解剖学知识和经验，具有主观差异性，需要花费大量时间和精力，直到深度卷积神经网络突破性的进展使得肺肿瘤自动分割成为可能。目前已有的肺肿瘤自动分割方法依赖于基于阈值分割法的肺肿瘤分割，该方法容易丢失肺部丰富的纹理信息，从而容易去除掉较大的肺部肿瘤，造成分割的肺肿瘤不完整。
发明内容
本发明的目的在于提供一种基于分割网络的CT肺肿瘤分割方法、装置、设备及存储介质，旨在解决由于现有技术无法提供一种有效的对肺部CT图像中肺肿瘤进行分割的方法，导致分割精度低、分割得到的肺部肿瘤不完整的问题。
一方面，本发明提供了一种基于分割网络的CT肺肿瘤分割方法，所述方法包括下述步骤：
当接收到对肺部CT图像进行肺肿瘤分割请求时，对所述肺部CT图像进行预处理，获得对应的肺部预处理图像；
通过预先训练好的第一分割网络对所述肺部预处理图像进行图像分割，得到对应的肺肿瘤初始分割图像；
根据预设的像元间距将所述肺部预处理图像和所述肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像；
以所述初始分割采样图像的质心为中心、按照预设图像块大小对所述初始分割采样图像进行裁剪，得到预设数量个初始分割图像块；
根据所述肺部采样图像和所述初始分割图像块，采用预先训练好的第二分割网络得到所述肺部CT图像对应的肺肿瘤分割图像。
另一方面，本发明提供了一种基于分割网络的CT肺肿瘤分割装置，所述装置包括：
CT图像预处理单元，用于当接收到对肺部CT图像进行肺肿瘤分割请求时，对所述肺部CT图像进行预处理，获得对应的肺部预处理图像；
第一图像分割单元，用于通过预先训练好的第一分割网络对所述肺部预处理图像进行图像分割，得到对应的肺肿瘤初始分割图像；
图像重采样单元，用于根据预设的像元间距将所述肺部预处理图像和所述肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像；
图像块获得单元，用于以所述初始分割采样图像的质心为中心、按照预设 图像块大小对所述初始分割采样图像进行裁剪，得到预设数量个初始分割图像块；以及
第二图像分割单元，用于根据所述肺部采样图像和所述初始分割图像块，采用预先训练好的第二分割网络得到所述肺部CT图像对应的肺肿瘤分割图像。
另一方面，本发明还提供了一种计算设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如上述基于分割网络的CT肺肿瘤分割方法所述的步骤。
另一方面，本发明还提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现如上述基于分割网络的CT肺肿瘤分割方法所述的步骤。
本发明采用第一分割网络对经过预处理得到的肺部预处理图像进行图像分割，得到肺肿瘤初始分割图像，根据像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像，计算初始分割采样图像的质心，以该质心为中心将初始分割采样图像裁剪为预设数量个图像块大小的初始分割图像块，根据肺部采样图像和初始分割图像块，采用第二分割网络得到肺部CT图像对应的肺肿瘤分割图像，从而通过第一分割网络对肺部CT图像中肺肿瘤区域进行定位，再通过第二分割网络对肺肿瘤进行分割，提高了对肺部CT图像进行肺肿瘤分割的精确度和完整性，进而获得高精度的肺肿瘤分割图像，提高手术的安全程度。
附图说明
图1是本发明实施例一提供的基于分割网络的CT肺肿瘤分割方法的实现流程图；
图2是本发明实施例二提供的对肺部CT图像进行预处理的实现流程图；
图3是本发明实施例三提供的对预设分割网络进行训练的实现流程图；
图4是本发明实施例四提供的基于分割网络的CT肺肿瘤分割装置的结构 示意图；
图5是本发明实施例四提供的基于分割网络的CT肺肿瘤分割装置的优选结构示意图；
图6是本发明实施例五提供的基于分割网络的CT肺肿瘤分割装置的结构示意图；以及
图7是本发明实施例六提供的计算设备的结构示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明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仅用以解释本发明，并不用于限定本发明。
以下结合具体实施例对本发明的具体实现进行详细描述：
实施例一：
图1示出了本发明实施例一提供的基于分割网络的CT肺肿瘤分割方法的实现流程，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S101中，当接收到对肺部CT图像进行肺肿瘤分割请求时，对肺部CT图像进行预处理，获得对应的肺部预处理图像。
本发明实施例适用于医学图像处理平台、系统或设备，例如个人计算机、服务器等。在本发明实施例中，由于CT图像是由一定数目由黑到白不同灰度的像素按矩阵排列所构成，且CT图像的不同灰度反映的是相应器官和组织对X线的吸收程度，黑色区域表示低吸收区，也即低密度区，如含大量气体的肺部，白色区域表示高吸收区，也即高密度区，如骨骼，因此，根据CT图像可以清楚地获取人体某些器官的解剖结构和器官内密度发生变化的病变组织的特性，通过CT设备对患者的肺部进行断层扫描生成肺部CT图像，以通过肺部CT图像发现患者肺部的病变组织，进而进行治疗。当用户需要对患者的肺部CT图像进行肺肿瘤分割时，发送肺肿瘤分割请求，其中，用户可从已公开的医 学图像数据库或医院提供的手术图像中获得患者的肺部CT图像，当接收到该肺肿瘤分割请求时，获取用户输入的肺部CT图像，并对获取到的该肺部CT图像进行预处理，得到对应的肺部预处理图像。
在步骤S102中，通过预先训练好的第一分割网络对肺部预处理图像进行图像分割，得到对应的肺肿瘤初始分割图像。
在本发明实施例中，图像分割就是把图像分成若干个特定的、具有独特性质的区域并提出感兴趣目标的技术和过程。将肺部预处理图像输入到预先训练好的第一分割网络中进行图像分割，并提取肺部预处理图像中的肺肿瘤区域，得到对应的、粗粒度的肺肿瘤初始分割图像。
在步骤S103中，根据预设的像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像。
在本发明实施例中，将肺部预处理图像重采样至预设的像元间距(即像素空间间距)，得到肺部采样图像，并将肺肿瘤初始分割图像也重采样至该像元间距，得到对应的初始分割采样图像。
在根据预设的像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样之前，优选地，将像元间距设置为1mm×1mm×2mm，从而增加肺部采样图像和初始分割采样图像的图像信息，进而提高肺部采样图像和初始分割采样图像的图像质量。
在步骤S104中，以初始分割采样图像的质心为中心、按照预设图像块大小对初始分割采样图像进行裁剪，得到预设数量个初始分割图像块。
在本发明实施例中，计算初始分割采样图像的质心，并以计算得到的该质心为中心从初始分割采样图像中提取预设数量个、预设图像块大小的初始分割图像块。
在对初始分割采样图像进行裁剪之前，优选地，将图像块大小设置为128×128×64，从而提高初始分割图像块的图像质量。
在对初始分割采样图像进行裁剪时，优选地，将初始分割采用图像的质心 的像素灰度值设置为2，从而提高质心处像素的亮度，进而提高后续肺肿瘤分割的精确度。
在步骤S105中，根据肺部采样图像和初始分割图像块，采用预先训练好的第二分割网络得到肺部CT图像对应的肺肿瘤分割图像。
在本发明实施例中，将肺部采样图像和所有的初始分割图像块输入到预先训练好的第二分割网络中，通过第二分割网络输出肺肿瘤分割图像，也即对肺部CT图像中肺肿瘤区域分割得到的最终的、细粒度的肺肿瘤分割图像。
在本发明实施例中，采用第一分割网络对经过预处理得到的肺部预处理图像进行图像分割，得到肺肿瘤初始分割图像，根据像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像，以初始分割采样图像的质心为中心将初始分割采样图像裁剪为预设数量个图像块大小的初始分割图像块，根据肺部采样图像和初始分割图像块，采用第二分割网络得到肺部CT图像对应的肺肿瘤分割图像，从而通过第一分割网络对肺部CT图像中肺肿瘤区域进行定位，再通过第二分割网络对肺肿瘤进行分割，提高了对肺部CT图像进行肺肿瘤分割的精确度和完整性，进而获得高精度的肺肿瘤分割图像，提高手术的安全程度。
实施例二：
图2示出了本发明实施例二提供的对肺部CT图像进行预处理的实现流程，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S201中，根据预设的种子点提取肺部CT图像中的空气连通区域。
在本发明实施例中，以预设的种子点为起点，采用区域生长法提取肺部CT图像中的空气连通区域，空气连通区域也即肺部CT图像的背景区域。
在提取肺部CT图像中的空气连通区域之前，优选地，将肺部CT图像的左上角的第一像素点设置为种子点，从而提高了空气连通区域提取的精确度和完整性。
在步骤S202中，将提取到的空气连通区域从肺部CT图像中进行去除，得 到对应的身体掩膜区域图像。
在本发明实施例中，从肺部CT图像中将空气连通区域裁剪掉，得到对应的身体掩膜区域图像，从而去掉肺部CT图像中的噪声，提高得到的身体掩膜区域图像的图像质量。
在步骤S203中，对身体掩膜区域图像进行阈值分割，得到左肺最大连通区域和右肺最大连通区域。
在本发明实施例中，通过阈值分割法对身体掩膜区域进行图像分割，得到身体掩膜区域图像中的左肺最大连通区域和右肺最大连通区域。
在对身体掩膜区域图像进行阈值分割时，优选地，根据阈值范围[-1024,-400]，通过阈值分割法对身体掩膜区域进行图像分割，从而在简化身体掩膜区域的图像分割的同时，提高左肺最大连通区域和右肺最大连通区域分割的精确度和完整性。
在步骤S204中，根据左肺最大连通区域和右肺最大连通区域，对肺部CT图像进行感兴趣区域提取，得到对应的肺部感兴趣区域图像。
在步骤S205中，根据预设的灰度变换公式对肺部感兴趣区域图像进行灰度变换，并将灰度变换后的肺部感兴趣区域图像进行归一化处理，得到肺部CT图像对应的肺部预处理图像。
在本发明实施例中，优选地，灰度变换公式为f(x)＝0.2*I(x)，其中，I(x)为左肺最大连通区域或右肺最大连通区域中第x个像素点的灰度值，f(x)为灰度变换后第x个像素点的灰度值，从而提高了肺部感兴趣区域图像的对比度，使得肺部感兴趣区域图像变得更加清晰、细腻、且容易识别，进而提高了肺部感兴趣区域图像的显示效果。
在将灰度变换后的肺部感兴趣区域图像进行归一化处理时，优选地，将灰度变换后的肺部感兴趣区域图像归一化至[0,1]，从而进一步提高了肺部感兴趣区域图像的对比度，使得肺部感兴趣区域图像的细节更加清楚。
在本发明实施例中，通过对肺部CT图像进行空气连通区域裁剪、感兴趣 区域提取、阈值分割、灰度变换、灰度归一化等预处理，从而提高算法的鲁棒性和执行效率。
实施例三：
图3示出了本发明实施例三提供的对预设分割网络进行训练的实现流程，为了便于说明，仅示出了与本发明实施例相关的部分，详述如下：
在步骤S301中，根据预设的样本采样大小，将预处理得到的训练样本进行重采样，得到第一采样训练样本。
在本发明实施例中，将预处理得到的训练样本重采样至预设的样本采样大小，得到第一采样训练样本，其中，训练样本为形状大小各异的肺肿瘤CT图像数据集。
在将预处理得到的训练样本进行重采样之前，优选地，将样本采样大小设置为128×128×64，从而增加第一采样训练样本的图像信息，进而提高第一采样训练样本的图像质量。
在将预处理得到的训练样本进行重采样之前，又一优选地，对训练样本进行预处理，具体地，通过下述步骤实现对训练样本进行预处理：
1)根据预设的种子点提取训练样本中的空气连通区域；
2)将提取到的空气连通区域从训练样本中进行去除，得到对应的身体掩膜样本图像；
3)对身体掩膜样本图像进行阈值分割，得到左肺最大连通区域和右肺最大连通区域；
4)根据左肺最大连通区域和右肺最大连通区域，对训练样本进行感兴趣区域提取，得到对应的肺部感兴趣区域样本图像；
5)根据预设的灰度变换公式对肺部感兴趣区域样本图像进行灰度变换，并将灰度变换后的肺部感兴趣区域样本图像进行归一化处理，得到预处理的训练样本。
从而通过步骤1)～5)实现对训练样本的预处理，提高了训练样本的图像 特征对比度，使得训练样本变得更加清晰、细腻、且容易识别，进而提高后续分割网络的训练效果。
在步骤S302中，根据预设的训练迭代次数，通过第一采样训练样本对预设分割网络进行训练，得到第一分割网络。
在本发明实施例中，通过第一采样训练样本对预设分割网络进行预设训练迭代次数的迭代训练，得到第一分割网络，其中，预设分割网络为全卷积网络(Fully Convolutional Networks，FCN)、SegNet、U-Net或者V-Net等语义分割网络，得到的第一分割网络通过将第一采样训练样本输入到语义分割网络中，对语义分割网络进行训练而得到，从而使得该第一分割网络能对肺部CT图像中肺肿瘤区域进行粗略定位，提高了通过第一分割网络对肺部CT图像中的肺肿瘤区域进行图像分割的完整性。
在步骤S303中，根据像元间距，将训练样本进行重采样，得到第二采样训练样本。
在本发明实施例中，将预处理得到的训练样本重采样至预设的像元间距(即像素空间间距)，得到第二采样训练样本，其中，该训练样本与步骤S301中的训练样本相同。
在步骤S304中，以预先设置的定位点为中心、按照图像块大小对第二采样训练样本进行裁剪，得到预设数量个训练样本图像块，定位点为第二采样训练样本中肿瘤区域的一个像素点。
在本发明实施例中，在第二采样训练样本中肿瘤区域随机选取一个像素点作为定位点，再以该定位点为中心从第二采样训练样本中提取预设数量个预设图像块大小的训练样本图像块。
在对第二采样训练样本进行裁剪时，优选地，将定位点的像素灰度值设置为2，从而提高定位点处像素的亮度，进而提高后续第二分割网络的训练效果。
在步骤S305中，根据训练迭代次数，通过训练样本图像块对预设分割网络进行训练，得到第二分割网络。
在本发明实施例中，通过训练样本图像块对预设分割网络进行预设训练迭代次数的迭代训练，得到第二分割网络，得到的第二分割网络通过将所有的训练样本图像块输入到语义分割网络中，对语义分割网络进行训练而得到，从而使得该第二分割网络能对肺部CT图像中肺肿瘤进行精确定位，提高了通过第二分割网络对肺部CT图像进行肺肿瘤图像分割的精确度。
在本发明实施例中，通过形状大小各异的肺肿瘤CT图像数据集对第一分割网络和第二分割网络进行并行训练，从而提高了训练效率，且降低了训练过程中对内存的消耗。
实施例四：
图4示出了本发明实施例四提供的基于分割网络的CT肺肿瘤分割装置的结构，为了便于说明，仅示出了与本发明实施例相关的部分，其中包括：
CT图像预处理单元41，用于当接收到对肺部CT图像进行肺肿瘤分割请求时，对肺部CT图像进行预处理，获得对应的肺部预处理图像；
第一图像分割单元42，用于通过预先训练好的第一分割网络对肺部预处理图像进行图像分割，得到对应的肺肿瘤初始分割图像；
图像重采样单元43，用于根据预设的像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像；
图像块获得单元44，用于以初始分割采样图像的质心为中心、按照预设图像块大小对初始分割采样图像进行裁剪，得到预设数量个初始分割图像块；以及
第二图像分割单元45，用于根据肺部采样图像和初始分割图像块，采用预先训练好的第二分割网络得到肺部CT图像对应的肺肿瘤分割图像。
如图5所示，优选地，CT图像预处理单元41包括：
第一区域提取单元411，用于根据预设的种子点提取肺部CT图像中的空气连通区域；
第二区域提取单元412，用于将提取到的空气连通区域从肺部CT图像中进 行去除，得到对应的身体掩膜区域图像；
图像阈值分割单元413，用于对身体掩膜区域图像进行阈值分割，得到左肺最大连通区域和右肺最大连通区域；
第三区域提取单元414，用于根据左肺最大连通区域和右肺最大连通区域，对肺部CT图像进行感兴趣区域提取，得到对应的肺部感兴趣区域图像；以及
归一化处理单元415，用于根据预设的灰度变换公式对肺部感兴趣区域图像进行灰度变换，并将灰度变换后的肺部感兴趣区域图像进行归一化处理，得到肺部CT图像对应的肺部预处理图像。
在本发明实施例中，基于分割网络的CT肺肿瘤分割装置的各单元可由相应的硬件或软件单元实现，各单元可以为独立的软、硬件单元，也可以集成为一个软、硬件单元，在此不用以限制本发明。具体地，各单元的实施方式可参考前述方法实施例的描述，在此不再赘述。
实施例五：
图6示出了本发明实施例五提供的基于分割网络的CT肺肿瘤分割装置的结构，为了便于说明，仅示出了与本发明实施例相关的部分，其中包括：
第一样本采样单元60，用于根据预设的样本采样大小，将预处理得到的训练样本进行重采样，得到第一采样训练样本；
第一网络训练单元61，用于根据预设的训练迭代次数，通过第一采样训练样本对预设分割网络进行训练，得到第一分割网络；
第二样本采样单元62，用于根据像元间距，将训练样本进行重采样，得到第二采样训练样本；
第二样本裁剪单元63，用于以预先设置的定位点为中心、按照图像块大小对第二采样训练样本进行裁剪，得到预设数量个训练样本图像块，定位点为第二采样训练样本中肿瘤区域的一个像素点；
第二网络训练单元64，用于根据训练迭代次数，通过训练样本图像块对预设分割网络进行训练，得到第二分割网络；
CT图像预处理单元65，用于当接收到对肺部CT图像进行肺肿瘤分割请求时，对肺部CT图像进行预处理，获得对应的肺部预处理图像；
第一图像分割单元66，用于通过预先训练好的第一分割网络对肺部预处理图像进行图像分割，得到对应的肺肿瘤初始分割图像；
图像重采样单元67，用于根据预设的像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像；
图像块获得单元68，用于以初始分割采样图像的质心为中心、按照预设图像块大小对初始分割采样图像进行裁剪，得到预设数量个初始分割图像块；以及
第二图像分割单元69，用于根据肺部采样图像和初始分割图像块，采用预先训练好的第二分割网络得到肺部CT图像对应的肺肿瘤分割图像。
在本发明实施例中，基于分割网络的CT肺肿瘤分割装置的各单元可由相应的硬件或软件单元实现，各单元可以为独立的软、硬件单元，也可以集成为一个软、硬件单元，在此不用以限制本发明。具体地，各单元的实施方式可参考前述方法实施例的描述，在此不再赘述。
实施例六：
图7示出了本发明实施例六提供的计算设备的结构，为了便于说明，仅示出了与本发明实施例相关的部分。
本发明实施例的计算设备7包括处理器70、存储器71以及存储在存储器71中并可在处理器70上运行的计算机程序72。该处理器70执行计算机程序72时实现上述基于分割网络的CT肺肿瘤分割方法实施例中的步骤，例如图1所示的步骤S101至S105。或者，处理器70执行计算机程序72时实现上述各装置实施例中各单元的功能，例如图4所示单元41至45的功能。
在本发明实施例中，采用第一分割网络对经过预处理得到的肺部预处理图像进行图像分割，得到肺肿瘤初始分割图像，根据像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样 图像，以初始分割采样图像的质心为中心将初始分割采样图像裁剪为预设数量个图像块大小的初始分割图像块，根据肺部采样图像和初始分割图像块，采用第二分割网络得到肺部CT图像对应的肺肿瘤分割图像，从而通过第一分割网络对肺部CT图像中肺肿瘤区域进行定位，再通过第二分割网络对肺肿瘤进行分割，提高了对肺部CT图像进行肺肿瘤分割的精确度和完整性，进而获得高精度的肺肿瘤分割图像，提高手术的安全程度。
本发明实施例的计算设备可以为个人计算机、服务器。该计算设备7中处理器70执行计算机程序72时实现基于分割网络的CT肺肿瘤分割方法时实现的步骤可参考前述方法实施例的描述，在此不再赘述。
实施例七：
在本发明实施例中，提供了一种计算机可读存储介质，该计算机可读存储介质存储有计算机程序，该计算机程序被处理器执行时实现上述基于分割网络的CT肺肿瘤分割方法实施例中的步骤，例如，图1所示的步骤S101至S105。或者，该计算机程序被处理器执行时实现上述各装置实施例中各单元的功能，例如图4所示单元41至45的功能。
在本发明实施例中，采用第一分割网络对经过预处理得到的肺部预处理图像进行图像分割，得到肺肿瘤初始分割图像，根据像元间距将肺部预处理图像和肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像，以初始分割采样图像的质心为中心将初始分割采样图像裁剪为预设数量个图像块大小的初始分割图像块，根据肺部采样图像和初始分割图像块，采用第二分割网络得到肺部CT图像对应的肺肿瘤分割图像，从而通过第一分割网络对肺部CT图像中肺肿瘤区域进行定位，再通过第二分割网络对肺肿瘤进行分割，提高了对肺部CT图像进行肺肿瘤分割的精确度和完整性，进而获得高精度的肺肿瘤分割图像，提高手术的安全程度。
本发明实施例的计算机可读存储介质可以包括能够携带计算机程序代码的任何实体或装置、记录介质，例如，ROM/RAM、磁盘、光盘、闪存等存储器。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种基于分割网络的CT肺肿瘤分割方法，其特征在于，所述方法包括下述步骤：
    当接收到对肺部CT图像进行肺肿瘤分割请求时，对所述肺部CT图像进行预处理，获得对应的肺部预处理图像；
    通过预先训练好的第一分割网络对所述肺部预处理图像进行图像分割，得到对应的肺肿瘤初始分割图像；
    根据预设的像元间距将所述肺部预处理图像和所述肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像；
    以所述初始分割采样图像的质心为中心、按照预设图像块大小对所述初始分割采样图像进行裁剪，得到预设数量个初始分割图像块；
    根据所述肺部采样图像和所述初始分割图像块，采用预先训练好的第二分割网络得到所述肺部CT图像对应的肺肿瘤分割图像。
  2. 如权利要求1所述的方法，其特征在于，对获取到的所述肺部CT图像进行预处理的步骤，包括：
    根据预设的种子点提取所述肺部CT图像中的空气连通区域；
    将提取到的所述空气连通区域从所述肺部CT图像中进行去除，得到对应的身体掩膜区域图像；
    对所述身体掩膜区域图像进行阈值分割，得到左肺最大连通区域和右肺最大连通区域；
    根据所述左肺最大连通区域和所述右肺最大连通区域，对所述肺部CT图像进行感兴趣区域提取，得到对应的肺部感兴趣区域图像；
    根据预设的灰度变换公式对所述肺部感兴趣区域图像进行灰度变换，并将灰度变换后的所述肺部感兴趣区域图像进行归一化处理，得到所述肺部CT图像对应的所述肺部预处理图像。
  3. 如权利要求2所述的方法，其特征在于，对所述身体掩膜区域图像进行 阈值分割的步骤，包括
    根据阈值范围[-1024,-400]对所述身体掩膜区域图像进行阈值分割。
  4. 如权利要求2所述的方法，其特征在于，所述灰度变换公式为f(x)＝0.2*I(x)，其中，I(x)为所述左肺最大连通区域或所述右肺最大连通区域中第x个像素点的灰度值，f(x)为灰度变换后第x个像素点的灰度值。
  5. 如权利要求1所述的方法，其特征在于，对所述肺部CT图像进行预处理之前，所述方法还包括：
    根据预设的样本采样大小，将预处理得到的训练样本进行重采样，得到第一采样训练样本；
    根据预设的训练迭代次数，通过所述第一采样训练样本对预设分割网络进行训练，得到所述第一分割网络；
    根据所述像元间距，将所述训练样本进行重采样，得到第二采样训练样本；
    以预先设置的定位点为中心、按照所述图像块大小对所述第二采样训练样本进行裁剪，得到预设数量个训练样本图像块，所述定位点为所述第二采样训练样本中肿瘤区域的一个像素点；
    根据所述训练迭代次数，通过所述训练样本图像块对所述分割网络进行训练，得到所述第二分割网络。
  6. 一种基于分割网络的CT肺肿瘤分割装置，其特征在于，所述装置包括：
    CT图像预处理单元，用于当接收到对肺部CT图像进行肺肿瘤分割请求时，对所述肺部CT图像进行预处理，获得对应的肺部预处理图像；
    第一图像分割单元，用于通过预先训练好的第一分割网络对所述肺部预处理图像进行图像分割，得到对应的肺肿瘤初始分割图像；
    图像重采样单元，用于根据预设的像元间距将所述肺部预处理图像和所述肺肿瘤初始分割图像进行重采样，得到对应的肺部采样图像和初始分割采样图像；
    图像块获得单元，用于以所述初始分割采样图像的质心为中心、按照预设 图像块大小对所述初始分割采样图像进行裁剪，得到预设数量个初始分割图像块；以及
    第二图像分割单元，用于根据所述肺部采样图像和所述初始分割图像块，采用预先训练好的第二分割网络得到所述肺部CT图像对应的肺肿瘤分割图像。
  7. 如权利要求6所述的装置，其特征在于，所述CT图像预处理单元包括：
    第一区域提取单元，用于根据预设的种子点提取所述肺部CT图像中的空气连通区域；
    第二区域提取单元，用于将提取到的所述空气联通区域从所述肺部CT图像中进行去除，得到对应的身体掩膜区域图像；
    图像阈值分割单元，用于对所述身体掩膜区域图像进行阈值分割，得到左肺最大连通区域和右肺最大连通区域；
    第三区域提取单元，用于根据所述左肺最大连通区域和所述右肺最大连通区域，对所述肺部CT图像进行感兴趣区域提取，得到对应的肺部感兴趣区域图像；以及
    归一化处理单元，用于根据预设的灰度变换公式对所述肺部感兴趣区域图像进行灰度变换，并将灰度变换后的所述肺部感兴趣区域图像进行归一化处理，得到所述肺部CT图像对应的所述肺部预处理图像。
  8. 如权利要求6所述的装置，其特征在于，所述装置还包括：
    第一样本采样单元，用于根据预设的样本采样大小，将预处理得到的训练样本进行重采样，得到第一采样训练样本；
    第一网络训练单元，用于根据预设的训练迭代次数，通过所述第一采样训练样本对预设分割网络进行训练，得到所述第一分割网络；
    第二样本采样单元，用于根据所述像元间距，将所述训练样本进行重采样，得到第二采样训练样本；
    第二样本裁剪单元，用于以预先设置的定位点为中心、按照所述图像块大小对所述第二采样训练样本进行裁剪，得到预设数量个训练样本图像块，所述 定位点为所述第二采样训练样本中肿瘤区域的一个像素点；以及
    第二网络训练单元，用于根据所述训练迭代次数，通过所述训练样本图像块对所述分割网络进行训练，得到所述第二分割网络。
  9. 一种计算设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现如权利要求1至5任一项所述方法的步骤。
  10. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至5任一项所述方法的步骤。
PCT/CN2019/122103 2018-12-07 2019-11-29 基于分割网络的ct肺肿瘤分割方法、装置、设备及介质 Ceased WO2020114332A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811493978.0 2018-12-07
CN201811493978.0A CN111292289B (zh) 2018-12-07 2018-12-07 基于分割网络的ct肺肿瘤分割方法、装置、设备及介质

Publications (1)

Publication Number Publication Date
WO2020114332A1 true WO2020114332A1 (zh) 2020-06-11

Family

ID=70975202

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/122103 Ceased WO2020114332A1 (zh) 2018-12-07 2019-11-29 基于分割网络的ct肺肿瘤分割方法、装置、设备及介质

Country Status (2)

Country Link
CN (1) CN111292289B (zh)
WO (1) WO2020114332A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112614143A (zh) * 2020-12-30 2021-04-06 深圳市联影高端医疗装备创新研究院 一种图像分割方法、装置、电子设备及存储介质
CN112767411A (zh) * 2020-12-31 2021-05-07 罗雄彪 一种肺部多器官同步分割方法
CN114445400A (zh) * 2022-03-04 2022-05-06 上海长征医院 肾脏体积测量方法、装置、电子设备及可读存储介质
CN115619804A (zh) * 2022-10-19 2023-01-17 哈尔滨理工大学 一种结合肺肿瘤先验信息的ct肺肿瘤自动分割方法
CN116128887A (zh) * 2021-11-08 2023-05-16 上海微创卜算子医疗科技有限公司 目标器官组织感兴趣区域定位方法、电子设备和存储介质
CN117808830A (zh) * 2023-12-28 2024-04-02 电子科技大学 一种基于医学图像处理的肺栓塞病灶分割方法
US12322111B2 (en) 2020-12-30 2025-06-03 United Imaging Research Institute of Innovative Medical Equipment Image segmentation method, device, equipment and storage medium

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117934480B (zh) * 2024-03-25 2024-05-28 天津市第五中心医院 一种用于肺损伤预测的肺部ct图像处理方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080267499A1 (en) * 2007-04-30 2008-10-30 General Electric Company Method and system for automatic detection of objects in an image
CN103942785A (zh) * 2014-04-09 2014-07-23 苏州大学 一种基于图割的pet和ct图像的肺肿瘤分割方法
CN107392910A (zh) * 2017-07-06 2017-11-24 沈阳东软医疗系统有限公司 一种基于ct图像的肺叶分割方法及装置
CN108537784A (zh) * 2018-03-30 2018-09-14 四川元匠科技有限公司 一种基于深度学习的ct图肺结节检测方法
WO2018189541A1 (en) * 2017-04-11 2018-10-18 Kheiron Medical Technologies Ltd Recist assessment of tumour progression

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8929636B2 (en) * 2012-02-02 2015-01-06 Peter Yim Method and system for image segmentation
CN108492297B (zh) * 2017-12-25 2021-11-19 重庆师范大学 基于深度级联卷积网络的mri脑肿瘤定位与瘤内分割方法
CN108615237B (zh) * 2018-05-08 2021-09-07 上海商汤智能科技有限公司 一种肺部图像处理方法及图像处理设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080267499A1 (en) * 2007-04-30 2008-10-30 General Electric Company Method and system for automatic detection of objects in an image
CN103942785A (zh) * 2014-04-09 2014-07-23 苏州大学 一种基于图割的pet和ct图像的肺肿瘤分割方法
WO2018189541A1 (en) * 2017-04-11 2018-10-18 Kheiron Medical Technologies Ltd Recist assessment of tumour progression
CN107392910A (zh) * 2017-07-06 2017-11-24 沈阳东软医疗系统有限公司 一种基于ct图像的肺叶分割方法及装置
CN108537784A (zh) * 2018-03-30 2018-09-14 四川元匠科技有限公司 一种基于深度学习的ct图肺结节检测方法

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112614143A (zh) * 2020-12-30 2021-04-06 深圳市联影高端医疗装备创新研究院 一种图像分割方法、装置、电子设备及存储介质
US12322111B2 (en) 2020-12-30 2025-06-03 United Imaging Research Institute of Innovative Medical Equipment Image segmentation method, device, equipment and storage medium
CN112767411A (zh) * 2020-12-31 2021-05-07 罗雄彪 一种肺部多器官同步分割方法
CN116128887A (zh) * 2021-11-08 2023-05-16 上海微创卜算子医疗科技有限公司 目标器官组织感兴趣区域定位方法、电子设备和存储介质
CN114445400A (zh) * 2022-03-04 2022-05-06 上海长征医院 肾脏体积测量方法、装置、电子设备及可读存储介质
CN115619804A (zh) * 2022-10-19 2023-01-17 哈尔滨理工大学 一种结合肺肿瘤先验信息的ct肺肿瘤自动分割方法
CN117808830A (zh) * 2023-12-28 2024-04-02 电子科技大学 一种基于医学图像处理的肺栓塞病灶分割方法
CN117808830B (zh) * 2023-12-28 2025-04-25 电子科技大学 一种基于医学图像处理的肺栓塞病灶分割方法

Also Published As

Publication number Publication date
CN111292289A (zh) 2020-06-16
CN111292289B (zh) 2023-09-26

Similar Documents

Publication Publication Date Title
CN111292289B (zh) 基于分割网络的ct肺肿瘤分割方法、装置、设备及介质
CN112508965B (zh) 医学影像中正常器官的轮廓线自动勾画系统
CN109859833B (zh) 消融手术治疗效果的评估方法和装置
CN112288786B (zh) 电子装置及非暂态计算机可读存储介质
WO2021238438A1 (zh) 肿瘤图像的处理方法和装置、电子设备、存储介质
CN110097557B (zh) 基于3D-UNet的医学图像自动分割方法及系统
CN114140378B (zh) 扫描图像处理方法、电子设备和可读介质
WO2020001086A1 (zh) 基于深度学习的心脏ct图像分割方法、装置、设备及介质
CN114037665B (zh) 下颌神经管的分割方法、装置、电子设备及存储介质
CN108288496A (zh) 肿瘤容积智能勾画方法和装置
CN110189324B (zh) 一种医学图像处理方法及处理装置
CN111179373B (zh) 医学图像去骨模型构建方法与骨信息去除方法
CN114283179A (zh) 基于超声图像的骨折远近端空间位姿实时获取与配准系统
CN112561917A (zh) 图像的分段方法、系统、电子设备及可读存储介质
CN115147360B (zh) 一种斑块分割方法、装置、电子设备及可读存储介质
CN113075599B (zh) 磁共振信号采集方法、磁共振系统及介质
CN117911462A (zh) 一种用于自适应放疗中配准引导的自动勾画方法及其应用
CN112686897A (zh) 一种基于弱监督的长短轴辅助肠胃淋巴结像素标注的方法
CN110390701B (zh) 一种伪影的校正方法、伪影校正系数的仿真方法及装置
CN115953363A (zh) 图像处理方法、图像处理装置和存储介质
CN112766333B (zh) 医学影像处理模型训练方法、医学影像处理方法及装置
CN120954614A (zh) 基于ai技术与cbct影像结合的牙齿和骨折线识别治疗方法
TWI838592B (zh) 圈選系統
CN114708415A (zh) 一种人体管道类组织三维模型压缩方法
CN118429367A (zh) 基于点云矩阵的多模态危及器官分割方法及装置

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: 19892018

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 03.11.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19892018

Country of ref document: EP

Kind code of ref document: A1