WO2021184949A1 - 肺炎征象的分割方法、装置、介质及电子设备 - Google Patents

肺炎征象的分割方法、装置、介质及电子设备 Download PDF

Info

Publication number
WO2021184949A1
WO2021184949A1 PCT/CN2021/072515 CN2021072515W WO2021184949A1 WO 2021184949 A1 WO2021184949 A1 WO 2021184949A1 CN 2021072515 W CN2021072515 W CN 2021072515W WO 2021184949 A1 WO2021184949 A1 WO 2021184949A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
pneumonia
images
lung
area
Prior art date
Application number
PCT/CN2021/072515
Other languages
English (en)
French (fr)
Inventor
王瑜
王少康
陈宽
Original Assignee
推想医疗科技股份有限公司
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 推想医疗科技股份有限公司 filed Critical 推想医疗科技股份有限公司
Priority to EP21771562.2A priority Critical patent/EP3971830B1/en
Priority to JP2021571620A priority patent/JP7304437B2/ja
Publication of WO2021184949A1 publication Critical patent/WO2021184949A1/zh
Priority to US17/540,629 priority patent/US20220092788A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration by the use of local operators
    • G06T5/30—Erosion or dilatation, e.g. thinning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/149—Segmentation; Edge detection involving deformable models, e.g. active contour models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20112—Image segmentation details
    • G06T2207/20132—Image cropping
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20172—Image enhancement details
    • G06T2207/20182—Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06T2207/30008—Bone
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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

Definitions

  • This application relates to the field of image processing, in particular to a segmentation method, segmentation device, computer-readable storage medium, and electronic equipment based on pneumonia signs in CT images.
  • Computed Tomography is a three-dimensional radiographic medical image reconstructed after digital geometric processing.
  • This technology mainly irradiates the human body by rotating X-rays on a single axis. Because different tissues have different X-ray absorption capabilities (or rejection rates), the computer’s three-dimensional technology can be used to reconstruct tomographic images. By bit processing, tomographic images of the corresponding tissues can be obtained, and the tomographic images can be stacked layer by layer to form a three-dimensional image.
  • CT images can be used to know whether the subject is a pneumonia patient, especially for the detection of pneumonia patients caused by the new coronavirus.
  • CT image detection is one of the most important and accurate methods.
  • most of the lungs are segmented first, and then the pneumonia lesions or signs are manually segmented, and the pneumonia lesions or signs are used to determine whether it is pneumonia and the degree of pneumonia.
  • This segmentation method is obviously too inefficient, especially It is aimed at the viral pneumonia caused by the current new type of coronavirus. Because it is very contagious, a large number of suspects need to be screened quickly. Therefore, it is necessary to improve efficiency while ensuring accuracy to contain the spread of the virus as soon as possible. Therefore, there is an urgent need for a high-precision and high-efficiency segmentation method for lung lesions or signs.
  • this application proposes a method for segmenting pneumonia signs, a segmentation device, a computer-readable storage medium, and an electronic device.
  • the lung region images in the CT images are input into multiple neural network models. Obtain multiple pneumonia sign images, and then combine the multiple pneumonia sign images to obtain pneumonia syndrome images.
  • the neural network model can efficiently and accurately obtain a large number of pneumonia sign images in the CT images to be detected for subsequent diagnosis Pneumonia provides reliable data basis.
  • a method for segmenting pneumonia signs including: generating multiple pneumonia sign images based on lung region images in CT images; and combining the multiple pneumonia sign images to obtain pneumonia Syndrome image; wherein the method of generating multiple pneumonia sign images respectively includes: inputting the lung region images into multiple neural network models to obtain the multiple pneumonia sign images.
  • the plurality of pneumonia sign images includes any one or a combination of the following sign images: lung consolidation image, ground glass shadow image, mass image, tree bud sign image, nodule image, Hollow image, halo image.
  • the lung area image includes a multi-layer two-dimensional image
  • the inputting the lung area image into multiple neural network models to obtain the multiple pneumonia sign images includes: Part of the multi-layer two-dimensional image is input into the plurality of neural network models to obtain multi-layer two-dimensional sign images corresponding to the plurality of pneumonia sign images; The dimensional sign images are superimposed to obtain the multiple pneumonia sign images.
  • the lung area image includes a multi-layer two-dimensional image
  • the inputting the lung area image into multiple neural network models to obtain the multiple pneumonia sign images includes: The two-dimensional images of the layers are input into the plurality of neural network models respectively to obtain multi-layer two-dimensional symptom images corresponding to the plurality of pneumonia symptom images respectively; and the multi-layer two-dimensional symptom images corresponding to the same pneumonia symptom image are respectively superimposed , To obtain the multiple pneumonia sign images.
  • the method further includes: performing a corrosion expansion operation on the plurality of pneumonia sign images respectively.
  • the method for acquiring the lung region image includes: acquiring a rib region image in the CT image; acquiring a coarse segmented image of the lung region in the CT image; and taking the rib region image as The boundary of the lung region, and the coarse segmented image is used as a seed region, and the lung region is expanded to the boundary of the lung region with a preset step length to obtain the lung region image.
  • the method further includes: performing an erosion operation on the coarse segmented image to obtain a eroded coarse segmented image; Using the rib region image as the boundary of the lung region, and using the coarse segmented image as the seed region, expand to the periphery to the boundary of the lung region at a preset step length to obtain the lung region
  • the image includes: taking the rib area image as the boundary of the lung area, and taking the corroded rough segmented image as the seed area, expanding to the periphery of the lung area with a preset step size, Obtain an image of the lung area.
  • the image of the rib region is used as the boundary of the lung region, and the coarse segmented image is used as the seed region, which is expanded to the periphery with a preset step size.
  • the boundary of the lung area the step of obtaining the image of the lung area.
  • the image of the rib region is used as the boundary of the lung region, and the rough segmented image after corrosion is used as the seed region, and the step size is set to The step of expanding the periphery to the boundary of the lung area to obtain an image of the lung area.
  • the method for acquiring the rib region image includes: acquiring the bone region image in the CT image based on the CT value of the bone; and segmenting the rib region in the bone region image based on the characteristics of the ribs to obtain The image of the rib area.
  • the segmentation method before the obtaining the coarse segmented image of the lung region in the CT image, the segmentation method further includes: preprocessing the CT image, wherein the preprocessing includes the following operations Any one or a combination of multiple items: remove background, remove white noise, crop image, change window width and window level.
  • the segmentation method further includes: smoothing the boundary of the lung region image.
  • a device for segmenting pneumonia signs including: a generation module for generating multiple pneumonia sign images based on lung region images in CT images; and a combination module for combining The multiple pneumonia symptom images are combined to obtain a pneumonia syndrome image; wherein the generating module is further configured to input the lung region images into multiple neural network models to obtain the multiple pneumonia symptom images.
  • a computer-readable storage medium stores a computer program, and the computer program is used to execute the pneumonia sign segmentation method described in any of the above embodiments.
  • an electronic device comprising: a processor; a memory for storing executable instructions of the processor; and the processor for executing any of the above embodiments The segmentation method of pneumonia signs.
  • the present application provides a method for segmenting pneumonia signs, a segmentation device, a computer-readable storage medium, and electronic equipment.
  • a segmentation device By inputting lung region images in CT images into multiple neural network models, multiple signs of pneumonia are obtained. Image, and then combine multiple pneumonia sign images to obtain a pneumonia syndrome image.
  • the neural network model can efficiently and accurately obtain a large number of pneumonia sign images in the CT images to be detected, so as to provide reliable data basis for subsequent diagnosis of pneumonia .
  • Fig. 1 is a schematic flowchart of a method for segmenting pneumonia signs provided by an exemplary embodiment of the present application.
  • Fig. 2 is a schematic flowchart of a method for segmenting pneumonia signs according to another exemplary embodiment of the present application.
  • Fig. 3 is a schematic flowchart of a method for segmenting pneumonia signs according to another exemplary embodiment of the present application.
  • Fig. 4 is a schematic flowchart of a method for segmenting an image of a lung region provided by an exemplary embodiment of the present application.
  • Fig. 5 is a schematic flowchart of a lung region image segmentation method provided by another exemplary embodiment of the present application.
  • Fig. 6 is a schematic flowchart of a method for acquiring an image of a rib region provided by an exemplary embodiment of the present application.
  • Fig. 7 is a schematic flowchart of a lung region image segmentation method provided by another exemplary embodiment of the present application.
  • Fig. 8 is a schematic flowchart of a lung region image segmentation method provided by another exemplary embodiment of the present application.
  • Fig. 9 is a schematic structural diagram of a pneumonia sign segmentation device provided by an exemplary embodiment of the present application.
  • Fig. 10 is a structural diagram of an electronic device provided by an exemplary embodiment of the present application.
  • Pneumonia is an inflammation of the lungs caused by a variety of factors. For patients with pneumonia, severe coughing is often accompanied by the occurrence of the disease. Pneumonia is divided into two types. One is bacterial pneumonia caused by bacteria, that is, pneumonia caused by bacteria invading the lungs. The most common are pneumococcus and streptococcus A hemolyticus. Most of them Bacterial pneumonia is caused by these two kinds of bacteria; the other is viral pneumonia caused by viruses, the most common is influenza caused by first-level macroviruses, such as 2019 new coronavirus, etc., viral pneumonia Compared with bacterial pneumonia, pneumonia is not only more serious, but also more difficult to treat.
  • CT image detection is one of the most important and accurate methods to know whether the subject is a patient with pneumonia, especially for patients with pneumonia caused by the new coronavirus.
  • pneumonia There is pneumonia.
  • the existing work of segmenting pneumonia signs is usually done manually by professional medical staff. Obviously, such efficiency is low, especially for the detection of pneumonia caused by the highly infectious new coronavirus. It is highly infectious and causes a large number of people to have hidden risks of infection. This requires CT image detection on a large number of people, and relatively limited or even scarce medical staff obviously cannot meet the demand for a large number of CT image detection.
  • the larger area of the pneumonia lesion will affect the shape and structure of the lung, which will further increase the difficulty of segmentation of the lung, which will affect the segmentation of pneumonia signs
  • the accuracy of the test results in a certain error, and the transmission and accumulation of the error will eventually affect the doctor’s judgment and cause immeasurable consequences.
  • the present application provides a method for segmenting pneumonia signs, a segmentation device, a computer-readable storage medium, and an electronic device.
  • the lung region images in the CT images are input into multiple neural network models to separate them. Obtain multiple pneumonia sign images, and then combine the multiple pneumonia sign images to obtain pneumonia syndrome images.
  • the neural network model can efficiently and accurately obtain a large number of pneumonia sign images in the CT images to be detected, so as to diagnose pneumonia later Provide reliable data basis.
  • Fig. 1 is a schematic flowchart of a method for segmenting pneumonia signs provided by an exemplary embodiment of the present application. As shown in Figure 1, the method includes the following steps:
  • Step 110 Generate multiple pneumonia sign images based on the lung area image in the CT image; wherein, the method of generating multiple pneumonia sign images respectively includes: input the lung area images into multiple neural network models to obtain multiple pneumonia sign images. Image of pneumonia signs.
  • CT images By taking CT images to obtain the image of the lung area of the subject, and according to the lung area image, it can accurately determine whether the subject has pneumonia.
  • CT image detection which is one of the effective measures to identify patients with pneumonia, is particularly important.
  • the detection of CT images is more complicated, and its efficiency is not high, and because there are multiple signs of pneumonia, it is necessary to know all or multiple signs to comprehensively determine whether you have pneumonia and the extent of the disease. Some of these signs are difficult to obtain, which further increases the difficulty of CT image detection. Therefore, in the embodiments of the present application, by separately inputting lung region images into multiple neural network models, different neural network models are trained for different pneumonia signs, and each pneumonia sign image with higher accuracy can be obtained in a targeted manner, thereby Provide accurate data basis for subsequent pneumonia screening.
  • the pneumonia sign image may include any one or a combination of the following sign images: lung consolidation image, ground glass shadow image, mass image, tree bud sign image, nodule image, cavity image, Halo sign image.
  • Lung consolidation refers to the replacement of the air-containing space on the distal side of the terminal bronchioles by pathological fluids, cells, and tissues. The most important feature is that the lesion area is dense and the blood vessels cannot be visualized.
  • Ground-glass shadow refers to alveolar filling or interstitial thickening caused by various reasons, resulting in a slight increase in lung density, but the shadow of the lesion with vascular texture can still be seen in it.
  • a mass refers to a mass of increased density, and its maximum diameter is ⁇ 3cm.
  • Nodules refer to the shadows of increased density of nodules, the largest diameter of which is less than 3cm.
  • Cavity refers to the transparent area formed by necrosis and liquefaction of the lung lesions, discharged through the drainage bronchus and gas ingress. For infectious diseases, the cavity is clinically prone to purulent infection.
  • Tree bud sign refers to the small nodules formed by the lesions in the terminal bronchioles and the alveolar cavity and the thin line shadows of the branches, which resemble spring branches, and they are mostly 2-4mm-sized nodules and branches at the end of the bronchus at the periphery of the lung. Shaped high-density shadow.
  • Halo sign refers to the ring-like glass-like density shadow surrounding the nodule/cavity, which usually represents exudation, bleeding, or edema.
  • Different neural network models can be set up according to the different characteristics of each sign to segment the lung region image separately to improve the overall segmentation accuracy.
  • the neural network model may be a deep learning neural network model, preferably, the neural network model may be a Unet neural network model.
  • the training method of the neural network model may be: selecting the lung region image that has been segmented by professional medical personnel and marked with pneumonia signs as the training sample of the neural network model to train the neural network model; and the neural network model The segmentation results obtained in the segmentation process can also be verified and modified by a third-party inspection agency, and the modified results can be used as training samples to retrain the neural network model, thereby further improving the segmentation accuracy of the neural network model.
  • the types of the multiple neural network models in the embodiments of the present application may be the same or different, and the embodiments of the present application do not limit the specific types of neural network models for segmenting various signs of pneumonia.
  • Step 120 Combine multiple pneumonia sign images to obtain a pneumonia syndrome image.
  • the determination of pneumonia is based on a comprehensive judgment of multiple pneumonia signs. For example, when there are only nodule images and no other pneumonia signs, it is impossible to determine that the subject is a pneumonia patient. Therefore, multiple pneumonia signs images are obtained separately (Usually the area where each pneumonia sign is marked in the CT image) After combining multiple pneumonia sign images together to obtain a pneumonia syndrome image, it is convenient for medical staff or other testing institutions to accurately judge based on the pneumonia syndrome image Whether the subject is a patient with pneumonia.
  • This application proposes a segmentation method for pneumonia signs.
  • By inputting lung region images in CT images into multiple neural network models multiple pneumonia signs images are obtained respectively, and then the multiple pneumonia signs images are combined.
  • Obtain the pneumonia syndrome image, and the neural network model can efficiently and accurately obtain a large number of pneumonia sign images in the CT images to be detected, so as to provide reliable data basis for the subsequent diagnosis of pneumonia.
  • Fig. 2 is a schematic flowchart of a method for segmenting pneumonia signs according to another exemplary embodiment of the present application.
  • the lung region image includes a multi-layer two-dimensional image.
  • step 110 may specifically include the following sub-steps:
  • Step 111 successively input a part of the multi-layer two-dimensional image into a plurality of neural network models to obtain multi-layer two-dimensional symptom images corresponding to the plurality of pneumonia symptom images.
  • CT images include multi-layer two-dimensional images.
  • the lung area image in CT images also includes multi-layer two-dimensional images. Therefore, in order to improve The efficiency of segmentation can divide the multi-layer two-dimensional image of the lung area image into multiple parts, and input these multiple parts into multiple neural network models by multiple inputs, and the multiple neural network models are divided multiple times. Two-dimensional image to obtain the corresponding multi-layer two-dimensional sign image, thereby improving the efficiency of segmentation.
  • the embodiment of the present application can appropriately select the number of layers of the two-dimensional image input to the neural network model at a time according to the processing capabilities of the neural network or the processing machine. It can be one layer, multiple layers, or all A single input of a two-dimensional image into the neural network model is sufficient, as long as the number of selected layers does not exceed the load that the neural network model or processing machine can carry. The number is not limited.
  • Step 112 superimpose multiple layers of two-dimensional symptom images corresponding to the same pneumonia symptom image to obtain multiple pneumonia symptom images.
  • the multi-layer two-dimensional symptom image of each pneumonia symptom is superimposed to obtain the pneumonia symptom image.
  • the segmentation method of FIG. 1 may further include: performing a corrosion expansion operation on a plurality of pneumonia sign images respectively.
  • Fig. 3 is a schematic flowchart of a method for segmenting pneumonia signs according to another exemplary embodiment of the present application. As shown in FIG. 3, after step 112, the foregoing embodiment may further include:
  • Step 113 Perform corrosion and expansion operations on multiple pneumonia sign images.
  • the erosion operation is a morphological operation, and the specific operation process is: remove pixels along the object boundary in the image and reduce the size of the object, that is, reduce the object boundary to remove the noise of the object in the image.
  • the expansion operation is also a morphological operation, and the specific operation process is exactly the opposite of the erosion operation, that is, adding pixels along the boundary of the object in the image and expanding the size of the object.
  • the noise generated in the segmentation process can be effectively removed.
  • the adjacent two-dimensional sign images are related to each other, by performing the corrosion expansion operation on the superimposed pneumonia sign images, the adjacent ones can be used.
  • the correlation between the two-dimensional sign images can remove the segmentation errors of individual layers. For example, the two-dimensional sign images of the middle layer can be adjusted by the upper and lower two or more layers of the two-dimensional sign images, thereby improving the overall segmentation accuracy of the pneumonia sign images.
  • Fig. 4 is a schematic flowchart of a method for segmenting an image of a lung region provided by an exemplary embodiment of the present application. As shown in Figure 4, the method includes the following steps:
  • Step 410 Obtain an image of the rib area in the CT image.
  • the outer boundary of the lung area is obtained by acquiring the rib area in the CT image, thereby improving the accuracy of segmenting the lung area.
  • Step 420 Obtain a coarse segmented image of the lung region in the CT image.
  • step 420 may be: inputting a CT image into a neural network model to obtain a coarse segmented image of the lung area.
  • the neural network model can be used to identify the lung area in the CT image.
  • the neural network model can be a neural network model such as Unet, and the training method of the neural network model can be: select the lung area that has been identified and marked by professional medical personnel
  • the CT image is used as a training sample of the neural network model to train the neural network model. Since this step is only to obtain a rough segmentation image of the lung area in the CT image, not an accurate segmentation image of the lung area, an appropriate number of training samples can be selected to train the neural network model to improve the segmentation of the entire lung. efficient.
  • the specific implementation of step 420 may also be: according to the CT value of the lung area, a region whose CT value is within the CT value range of the lung area is selected as the coarse segmented image.
  • the CT value is a measurement unit for determining the density of a certain local tissue or organ of the human body, usually called the Hounsfield Unit (HU), where the CT value of air is -1000, and the CT value of dense bone is +1000.
  • the CT value is the corresponding value of each tissue in the CT image and the X-ray attenuation coefficient.
  • the CT value is not an absolutely constant value.
  • CT value of other tissues are below 300 and above -80.
  • CT value of calcification points is 80-300
  • CT value of fat is -20-80. Since the lung area is basically air and its CT value is lower than other tissues, the CT value range can be set, and the connected area with the CT value within the CT value range can be selected as the coarse segmented image of the lung area.
  • Step 430 Taking the rib area image as the boundary of the lung area, and using the coarse segmented image as the seed area, expand to the periphery of the lung area with a preset step length to obtain the lung area image.
  • step 430 may be: taking the rib region image as the boundary, the coarse segmented image as the region of interest, and segmenting the CT image with the preset step size through the active contour model to obtain Image of lung area. Since the lungs are tightly wrapped by ribs, that is, the rib area is the outer boundary of the lung area. Therefore, the coarse segmented image can be used as the seed area or the region of interest, and the rib area image can be used as the boundary.
  • the CT image is segmented with a preset step length, that is, starting from the region of interest, expanding to the surrounding with a preset step length, until it expands to the rib area, so as to obtain an accurately segmented image of the lung area, which is the subsequent lung lobe Segmentation, pneumonia judgment, etc. provide accurate basic image data.
  • the preset step length can be adjusted according to actual needs. In order to obtain higher accuracy, the preset step length can be appropriately reduced.
  • the active contour model may include a Level Set model or a Snake model.
  • the embodiment of the present application can select different active contour models according to the requirements of actual application scenarios, as long as the selected active contour model can use the rib region image as the boundary and the coarse segmentation image as the seed region to obtain an accurate lung region image. That is, the embodiment of the present application does not limit the specific structure of the active contour model.
  • Fig. 5 is a schematic flowchart of a lung region image segmentation method provided by another exemplary embodiment of the present application. As shown in FIG. 5, after step 420, the foregoing embodiment may further include:
  • Step 440 Perform an etching operation on the rough segmented image to obtain a rough segmented image after etching.
  • the coarse segmented image of the lung area is usually not an accurate image of the lung area, for example, the coarse segmented image may contain other areas other than the lung area. This other area is the noise area in the coarse segmented image, and the corrosion operation The noise area in the rough segmented image can be removed to ensure that the corroded rough segmented image contains a part of the lung area and does not include areas other than the lung area.
  • step 430 is adjusted to step 530: take the rib area image as the boundary of the lung area, and use the corroded rough segmented image as the seed area, and expand to the periphery of the lung area with a preset step length to obtain the lung Part area image.
  • the corrosion operation can be used to ensure that the seed area contains only the lung area, thereby avoiding the seed area from being integrated during the expansion process. More non-lung areas. Because the seed area will not remove the area that already exists in the seed area during the expansion process, ensuring that the seed area only contains the lung area is a prerequisite for ensuring the accuracy of segmenting the lung area.
  • Fig. 6 is a schematic flowchart of a method for acquiring an image of a rib region provided by an exemplary embodiment of the present application. As shown in FIG. 6, the obtaining method may include the following steps:
  • Step 411 Obtain an image of the bone area in the CT image based on the CT value of the bone.
  • the image of the bone area with the largest CT value can be obtained.
  • the first CT value threshold may be set based on the CT value of the bone, and the connected area in the CT image whose CT value is greater than or equal to the first CT value threshold is acquired as the bone area image.
  • the first CT value threshold is smaller than the CT value of the bone and larger than the CT value of other tissues.
  • the above method may further include: removing an area in the connected area with an area smaller than a preset area threshold. If the first CT value threshold is set too large, part of the bone area may be missed, and if the first CT value threshold is set too small, due to the large CT value of the calcification point, there may be a certain amount of lung or heart area.
  • the calcified points become the interference noise in the image of the bone area, so it needs to be removed.
  • the area of calcification points is small. Therefore, the interference of calcification points on the image of the bone area can be eliminated by removing the area in the connected area whose area is less than the preset area threshold.
  • the area threshold can be preset according to the actual application.
  • the above method may further include: removing the connected areas located in the coarsely segmented image.
  • Step 412 Based on the characteristics of the ribs, segment the rib area in the bone area image to obtain the rib area image.
  • the implementation of step 412 may be: comparing the standard rib image with the bone region image, and selecting the bone region in the bone region image whose similarity to the standard rib image is greater than the preset similarity as the rib region image .
  • the ribs are relatively regular compared to other bones, the ribs are usually arranged regularly on the outside of the lungs and present a certain arc with a symmetrical structure.
  • the three-dimensional view of the multi-planar reconstruction can distinguish the difference between the ribs and other bones, especially In the sagittal position, the above-mentioned characteristics of the ribs can be clearly understood, that is, the arrangement and shape of the ribs have their particularities.
  • the standard rib image can be compared to select the similarity between the bone area image and the standard rib image.
  • the bone region that meets certain requirements is used to segment the rib region in the bone region image to obtain the rib region image. It should be understood that the embodiment of the present application can also select other methods of obtaining rib area images according to the requirements of actual application scenarios.
  • the rib area image can also be obtained directly through the neural network model, as long as all
  • the selected method for acquiring the image of the rib area only needs to meet the accuracy requirement, and the embodiment of the present application does not limit the specific method for acquiring the image of the rib area.
  • Fig. 7 is a schematic flowchart of a lung region image segmentation method provided by another exemplary embodiment of the present application. As shown in FIG. 7, before step 420, the foregoing embodiment may further include:
  • Step 450 Preprocess the CT image.
  • the preprocessing may include any one or a combination of the following operations: removing background, removing white noise, cropping the image, transforming window width and window level.
  • the specific implementation of removing the background can be: by setting the CT value range, the connected area within the CT value range is obtained, and only the connected area with the largest area in the connected area is retained, and the other areas are set as background areas, thereby eliminating Interference in other areas.
  • a specific implementation manner of removing white noise may be: removing white noise caused in the process of taking a CT image through a Gaussian filter.
  • the specific implementation of cropping the image may be: removing the background and only retaining the effective area, so as to reduce the complexity of subsequent image processing.
  • the specific implementation of changing the window width and window level can be: by setting the value of the window width and window level to highlight the region of interest, so as to avoid the interference of the uninterested region on the subsequent processing.
  • the window level is selected as -500 and the window width is 1500.
  • the setting values of the window level and window width can be adjusted according to the actual situation.
  • step 450 can be set before step 410, and the background and other interference factors in the CT image are eliminated through preprocessing, which can effectively reduce the complexity of subsequent steps and improve the efficiency of lung segmentation.
  • Fig. 8 is a schematic flowchart of a lung region image segmentation method provided by another exemplary embodiment of the present application. As shown in FIG. 8, after step 430, the foregoing embodiment may further include:
  • Step 460 Smoothing the boundary of the lung region image.
  • the boundary of the obtained lung area image may not be smooth. After the lung area image is obtained , The border of the lung area image is smoothed, and a more accurate lung area image can be obtained.
  • FIG. 9 is a schematic structural diagram of a pneumonia sign segmentation device 90 provided by an exemplary embodiment of the present application. As shown in Fig. 9, the device 90 for dividing pneumonia signs includes the following modules:
  • the generating module 91 is used to generate multiple pneumonia sign images based on the lung region image in the CT image; and the combination module 92 is used to combine multiple pneumonia sign images to obtain a pneumonia syndrome image; wherein, the generating module 91 It is further configured to: input lung region images into multiple neural network models to obtain multiple pneumonia sign images.
  • the lung region images in the CT images are input into multiple neural network models through the generating module 91 to obtain multiple pneumonia signs images respectively, and then the multiple pneumonia signs are combined through the combination module 92 The sign images are combined to obtain the pneumonia syndrome image.
  • the neural network model can efficiently and accurately obtain a large number of pneumonia sign images in the CT images to be detected, so as to provide reliable data basis for the subsequent diagnosis of pneumonia.
  • the pneumonia sign image may include any one or a combination of the following sign images: lung consolidation image, ground glass shadow image, mass image, tree bud sign image, nodule image, cavity image, Halo sign image.
  • the neural network model may be a deep learning neural network model, preferably, the neural network model may be a Unet neural network model.
  • the lung region image includes a multi-layer two-dimensional image.
  • the generating module 91 may include the following units: an input unit 911, which is used to input a part of the multi-layer two-dimensional image into multiple neural network models one by one to obtain the multi-layer two images corresponding to the multiple pneumonia sign images.
  • One-dimensional sign image; the superimposing unit 912 is used to superimpose multiple layers of two-dimensional sign images corresponding to the same pneumonia sign image to obtain multiple pneumonia sign images.
  • the lung region image includes a multi-layer two-dimensional image.
  • the generating module 91 may include the following units: an input unit 911, configured to input the multi-layer two-dimensional images into a plurality of neural network models respectively to obtain multi-layer two-dimensional symptom images corresponding to the plurality of pneumonia symptom images;
  • the superimposing unit 912 is configured to superimpose multiple layers of two-dimensional symptom images corresponding to the same pneumonia symptom image to obtain multiple pneumonia symptom images.
  • the generating module 91 may further include: a post-processing unit 913 configured to perform corrosion and expansion operations on multiple pneumonia sign images.
  • the segmentation device 90 may further include: an acquisition module 93 for acquiring an image of the rib area in the CT image; and a coarse segmentation module 94 for acquiring the roughness of the lung area in the CT image. Segmented image; and fine segmentation module 95, used to take the rib region image as the boundary of the lung region, and use the coarse segmented image as the seed region, and expand to the periphery of the lung region with a preset step size to obtain the lung Area image.
  • the coarse segmentation module 94 may be further configured to input the CT image into the neural network model to obtain a coarse segmentation image of the lung region.
  • the coarse segmentation module 94 may be further configured to select an area with a CT value within the CT value range of the lung area as the coarse segmentation image according to the CT value of the lung area.
  • the fine segmentation module 95 may be further configured to: use the rib region image as the boundary, the coarse segmentation image as the region of interest, and segment the CT image with a preset step size through the active contour model to obtain Image of lung area.
  • the active contour model may include a Level Set model or a Snake model.
  • the segmentation device 90 may further include: an erosion module 96 for performing an erosion operation on the rough segmented image to obtain a rough segmented image after corrosion.
  • the fine segmentation module 95 is configured to: take the rib area image as the boundary of the lung area, and use the corroded rough segmented image as the seed area, and expand to the periphery of the lung area at a preset step length to obtain the lung area. Area image.
  • the fine segmentation module 95 may be further configured to: use the rib region image as the boundary, the corroded rough segmented image as the region of interest, and segment the CT image with a preset step size through the active contour model To get an image of the lung area.
  • the active contour model may include a Level Set model or a Snake model.
  • the acquisition module 93 may include: a bone acquisition unit 931, configured to acquire an image of a bone region in a CT image based on the CT value of the bone; , Segment the rib area in the bone area image to get the rib area image.
  • the bone acquiring unit 931 may be further configured to set a first CT value threshold based on the CT value of the bone, and acquire a connected area in the CT image whose CT value is greater than or equal to the first CT value threshold as a bone area image.
  • the acquiring module 93 may be further configured to: after acquiring connected regions with a CT value greater than or equal to the first CT value threshold in the CT image, remove regions with an area smaller than a preset area threshold in the connected region.
  • the acquiring module 93 may be further configured to: after acquiring connected regions with a CT value greater than or equal to the first CT value threshold in the CT image, remove the connected regions located in the coarsely segmented image.
  • the rib acquisition unit 932 may be further configured to: compare the standard rib image with the bone region image, and select the bone region in the bone region image whose similarity to the standard rib image is greater than the preset similarity as the rib region. image.
  • the segmentation device 90 may further include: a preprocessing module 97 for preprocessing the CT image.
  • the preprocessing may include any one or a combination of the following operations: removing background, removing white noise, cropping the image, transforming window width and window level.
  • the segmentation device 90 may further include a smoothing module 98 for smoothing the boundary of the lung region image.
  • the electronic device 10 may be either or both of the first device and the second device, or a stand-alone device independent of them.
  • the stand-alone device may communicate with the first device and the second device to receive collected data from them. To the input signal.
  • FIG. 10 illustrates a block diagram of an electronic device 10 according to an embodiment of the present application.
  • the electronic device 10 includes one or more processors 11 and a memory 12.
  • the processor 11 may be a central processing unit (CPU) or other form of processing unit with data processing capability and/or instruction execution capability, and may control other components in the electronic device 10 to perform desired functions.
  • CPU central processing unit
  • the processor 11 may control other components in the electronic device 10 to perform desired functions.
  • the memory 12 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory.
  • the volatile memory may include random access memory (RAM) and/or cache memory (cache), for example.
  • the non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, and the like.
  • One or more computer program instructions can be stored on the computer-readable storage medium, and the processor 11 can run the program instructions to implement the pneumonia sign segmentation method and/or the various embodiments of the application described above. Or other desired functions.
  • Various contents such as input signals, signal components, noise components, etc. can also be stored in the computer-readable storage medium.
  • the electronic device 10 may further include: an input device 13 and an output device 14, and these components are interconnected by a bus system and/or other forms of connection mechanisms (not shown).
  • the input device 13 may be a CT scanner for acquiring CT images.
  • the input device 13 may be a communication network connector for receiving collected input signals from the first device and the second device.
  • the input device 13 may also include, for example, a keyboard, a mouse, and so on.
  • the output device 14 can output various information to the outside, including the determined image of the lung area, the image of pneumonia signs, and the image of the signs of pneumonia.
  • the output device 14 may include, for example, a display, a speaker, a printer, a communication network and a remote output device connected to it, and so on.
  • the electronic device 10 may also include any other appropriate components.
  • the embodiments of the present application may also be a computer program product, which includes computer program instructions that, when run by a processor, cause the processor to execute the “exemplary method” described above in this specification.
  • the steps in the segmentation method of pneumonia signs according to various embodiments of the present application are described in the section.
  • the computer program product may use any combination of one or more programming languages to write program codes for performing the operations of the embodiments of the present application.
  • the programming languages include object-oriented programming languages, such as Java, C++, etc. , Also includes conventional procedural programming languages, such as "C" language or similar programming languages.
  • the program code can be executed entirely on the user's computing device, partly on the user's device, executed as an independent software package, partly on the user's computing device and partly executed on the remote computing device, or entirely on the remote computing device or server Executed on.
  • the embodiment of the present application may also be a computer-readable storage medium, on which computer program instructions are stored.
  • the processor executes the "exemplary method" part of this specification. The steps in the method for segmenting pneumonia signs according to various embodiments of the present application are described in.
  • the computer-readable storage medium may adopt any combination of one or more readable media.
  • the readable medium may be a readable signal medium or a readable storage medium.
  • the readable storage medium may include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the above, for example. More specific examples (non-exhaustive list) of readable storage media include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable Type programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • each component or each step can be decomposed and/or recombined.
  • decompositions and/or recombinations shall be regarded as equivalent solutions of this application.

Abstract

本发明公开了一种肺炎征象的分割方法、分割装置、计算机可读存储介质以及电子设备，通过将CT图像中的肺部区域图像分别输入多个神经网络模型，以此分别得到多个肺炎征象图像，然后将多个肺炎征象图像进行组合，得到肺炎综合征象图像，利用神经网络模型可以高效且准确的获取大量的待检测CT图像中的肺炎征象图像，从而为后续诊断肺炎提供可靠的数据依据。

Description

肺炎征象的分割方法、装置、介质及电子设备 技术领域
本申请涉及图像处理领域，具体涉及基于CT图像中肺炎征象的分割方法、分割装置、计算机可读存储介质以及电子设备。
发明背景
电脑断层扫描(Computed Tomography，简称CT)是一种利用数位几何处理后重建的三维放射线医学影像。该技术主要通过单一轴面的X射线旋转照射人体，由于不同的组织对X射线的吸收能力(或称阻射率)不同，可以用电脑的三维技术重建出断层面影像，经由窗宽、窗位处理，可以得到相应组织的断层影像，将断层影像层层堆叠，即可形成立体影像。
通过CT图像可以获知被检测者是否为肺炎患者，特别是应对新型冠状病毒所造成的肺炎患者的检测，CT图像检测是最为重要且最为准确的方法之一。目前在获取CT图像后大多是先分割出肺部区域，然后通过人工去分割肺炎病灶或征象，并根据肺炎病灶或征象判断是否为肺炎以及肺炎的程度，这样的分割方式明显效率太低，特别是针对目前新型冠状病毒所造成的病毒性肺炎，因为其传染性很强，需要对大量疑似人员快速的甄别，因此，在保证准确度的同时也要提高效率，以尽早遏制住病毒的传播。因此，目前亟需高精度且高效率的肺部病灶或征象的分割方法。
发明内容
为了解决上述技术问题，本申请提出了一种肺炎征象的分割方法、分割装置、计算机可读存储介质以及电子设备，通过将CT图像中的肺部区域图像分别输入多个神经网络模型，以此分别得到多个肺炎征象图像，然后将多个肺炎征象图像进行组合，得到肺炎综合征象图像，利用神经网络模型可以高效且准确的获取大量的待检测CT图像中的肺炎征象图像，从而为后续诊断肺炎提供可靠的数据依据。
根据本申请的一个方面，提供了一种肺炎征象的分割方法，包括：基于CT图像中的肺部区域图像，分别生成多个肺炎征象图像；以及将所述多个肺炎征象图像组合，得到肺炎综合征象图像；其中，分别生成多个肺炎征象图像的方式包括：将所述肺部区域图像分别输入多个神经网络模型，得到所述多个肺炎征象图像。
在一实施例中，所述多个肺炎征象图像包括如下征象图像中的任一种或多种的组合：肺实变图像、磨玻璃影图像、肿块图像、树芽征图像、结节图像、空洞图像、晕征图像。
在一实施例中，所述肺部区域图像包括多层二维图像；所述将所述肺部区域图像分别输入多个神经网络模型，得到所述多个肺炎征象图像包括：逐次将所述多层二维图像中的一部分分别输入所述多个神经网络模型，得到分别对应所述多个肺炎征象图像的多层二维征象图像；以及分别将对应同一肺炎征象图像的所述多层二维征象图像叠加，得到所述多个肺炎征象图像。
在一实施例中，所述肺部区域图像包括多层二维图像；所述将所述肺部区域图像分别输入多个神经网络模型，得到所述多个肺炎征象图像包括：将所述多层二维图像分别输入所述多个神经网络模型，得到分别对应所述多个肺炎征象图像的多层二维征象图像；以及分别将对应同一肺炎征象图像的所述多层二维征象图像叠加，得到所述多个肺炎征象图像。
在一实施例中，在所述得到所述多个肺炎征象图像之后，所述方法还包括：对所述多个肺炎征象图像分别进行腐蚀膨胀操作。
在一实施例中，所述肺部区域图像的获取方式包括：获取所述CT图像中的肋骨区域图像；获取所述CT图像中肺部区域的粗分割图像；以及以所述肋骨区域图像为所述肺部区域的边界，并且以所述粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像。
在一实施例中，在所述获取所述CT图像中肺部区域的粗分割图像之后，所述方法还包括：对所述粗分割图像进行腐蚀操作，得到腐蚀后的粗分割图像；所述以所述肋骨区域图像为所述肺部区域的边界，并且以所述粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像包括：以所述肋骨区域图像为所述肺部区域的边界，并且以所述腐蚀后的粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像。
在一实施例中，基于活动轮廓模型，执行所述以所述肋骨区域图像为所述肺部区域的边界，并且以所述粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像的步骤。
在一实施例中，基于活动轮廓模型，执行所述以所述肋骨区域图像为所述肺部区域的边界，并且以所述腐蚀后的粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像的步骤。
在一实施例中，所述肋骨区域图像的获取方法包括：基于骨头的CT值，获取所述CT图像中骨头区域图像；基于肋骨的特性，分割出所述骨头区域图 像中的肋骨区域，得到所述肋骨区域图像。
在一实施例中，在所述获取所述CT图像中肺部区域的粗分割图像之前，所述分割方法还包括：对所述CT图像进行预处理，其中，所述预处理包括以下操作中任一项或多项的组合：去除背景、去除白噪声、裁剪图像、变换窗宽和窗位。
在一实施例中，所述分割方法还包括：对所述肺部区域图像的边界进行光滑化处理。
根据本申请的另一个方面，提供了一种肺炎征象的分割装置，包括：生成模块，用于基于CT图像中的肺部区域图像，分别生成多个肺炎征象图像；以及组合模块，用于将所述多个肺炎征象图像组合，得到肺炎综合征象图像；其中，生成模块进一步配置为：将所述肺部区域图像分别输入多个神经网络模型，得到所述多个肺炎征象图像。
根据本申请的另一个方面，提供了一种计算机可读存储介质，所述存储介质存储有计算机程序，所述计算机程序用于执行上述任一实施例所述的肺炎征象的分割方法。
根据本申请的另一个方面，提供了一种电子设备，所述电子设备包括：处理器；用于存储所述处理器可执行指令的存储器；所述处理器，用于执行上述任一实施例所述的肺炎征象的分割方法。
本申请提供的一种肺炎征象的分割方法、分割装置、计算机可读存储介质以及电子设备，通过将CT图像中的肺部区域图像分别输入多个神经网络模型，以此分别得到多个肺炎征象图像，然后将多个肺炎征象图像进行组合，得到肺炎综合征象图像，利用神经网络模型可以高效且准确的获取大量的待检测CT图像中的肺炎征象图像，从而为后续诊断肺炎提供可靠的数据依据。
附图简要说明
通过结合附图对本申请实施例进行更详细的描述，本申请的上述以及其他目的、特征和优势将变得更加明显。附图用来提供对本申请实施例的进一步理解，并且构成说明书的一部分，与本申请实施例一起用于解释本申请，并不构成对本申请的限制。在附图中，相同的参考标号通常代表相同部件或步骤。
图1是本申请一示例性实施例提供的一种肺炎征象的分割方法的流程示意图。
图2是本申请另一示例性实施例提供的一种肺炎征象的分割方法的流程示意图。
图3是本申请另一示例性实施例提供的一种肺炎征象的分割方法的流程示意图。
图4是本申请一示例性实施例提供的一种肺部区域图像分割方法的流程示意图。
图5是本申请另一示例性实施例提供的一种肺部区域图像分割方法的流程示意图。
图6是本申请一示例性实施例提供的一种肋骨区域图像的获取方法的流程示意图。
图7是本申请另一示例性实施例提供的一种肺部区域图像分割方法的流程示意图。
图8是本申请另一示例性实施例提供的一种肺部区域图像分割方法的流程示意图。
图9是本申请一示例性实施例提供的一种肺炎征象的分割装置的结构示意图。
图10是本申请一示例性实施例提供的电子设备的结构图。
实施本发明的方式
下面，将参考附图详细地描述根据本申请的示例实施例。显然，所描述的实施例仅仅是本申请的一部分实施例，而不是本申请的全部实施例，应理解，本申请不受这里描述的示例实施例的限制。
申请概述
肺炎是一种因为多种因素引发的肺部的炎症，对于肺炎患者来说在疾病发生的时候往往会伴随剧烈的咳嗽现象。肺炎分为两种，一种是细菌引起的细菌性的肺炎，即因为细菌侵袭肺部而导致的肺炎，其中最为常见的是肺炎球菌，和甲型溶血性链球菌这两种细菌，大部分的细菌性肺炎都是因为这两种细菌导致的；另一种就是病毒引起的病毒性的肺炎，最常见的就是由一级巨细病毒引发的流感，又例如2019新型冠状病毒等，病毒性的肺炎相对来于细菌性的肺炎来说，不仅更严重一些，治疗的难度相对来说也会更高。
如上所述，CT图像检测是获知被检测者是否为肺炎患者最为重要且最为准确的方法之一，特别是应对新型冠状病毒所造成的肺炎患者。通常得到CT图像后需要对其中的肺部区域进行分割，以得到肺部区域图像，然后对肺部区域内的肺炎征象进行分割，最后由医生根据分割得到的肺炎征象来判断被检测者是否患有肺炎，现有的分割肺炎征象的工作通常是由专业的医护人员手动实现，显然这样的效率是较低的，特别是应对传染性很强的新型冠状病毒所造成的肺炎的检测，因为其具有高传染性而导致大量的人有感染隐患，这就需要对大量的人做CT图像检测，而相对有限甚至紧缺的医务人员显然不能满足大量的CT图像检测的需求。
随着图像处理技术的快速发展，越来越多的医学图像的处理都可以通过计算机进行，例如通过图像分割以得到诊断所需要的感兴趣区域图像或者基础数据等。然而肺炎征象包括多个，其中有的征象在CT图像中与其他组织的密度(在CT图像中表象为CT值)区别较大，可以较为容易的分割出来，而有的征象在CT图像中与其他组织的密度区别较小，很难通过密度的比较进行分割，或者分割的精度不高；同时，由于肺炎患者常常因为咳嗽和吸气不足而导致图像的运动噪声较大，并且肺部内密度升高，从而导致肺部区域分割时存在较大的难度，此外，肺炎病灶的区域较大会影响到肺部的形状和结构，从而导致肺部的分割难度进一步提高，这样会影响到分割肺炎征象的精度，产生一定的误差，并且该误差的传递和累积会最终影响医生的判断，造成不可估量的后果。
为了解决上述问题，本申请提供的一种肺炎征象的分割方法、分割装置、计算机可读存储介质以及电子设备，通过将CT图像中的肺部区域图像分别输入多个神经网络模型，以此分别得到多个肺炎征象图像，然后将多个肺炎征象图像进行组合，得到肺炎综合征象图像，利用神经网络模型可以高效且准确的获取大量的待检测CT图像中的肺炎征象图像，从而为后续诊断肺炎提供可靠的数据依据。
示例性方法
图1是本申请一示例性实施例提供的一种肺炎征象的分割方法的流程示意图。如图1所示，该方法包括如下步骤：
步骤110：基于CT图像中的肺部区域图像，分别生成多个肺炎征象图像；其中，分别生成多个肺炎征象图像的方式包括：将肺部区域图像分别输入多个神经网络模型，得到多个肺炎征象图像。
通过拍摄CT图像来获得被检测者肺部区域图像，并且根据该肺部区域图像可以准确的判断该被检测者是否患有肺炎，对于目前严峻的新型冠状病毒性肺炎，要遏制住病毒的蔓延，就需要将患有肺炎或携带该病毒的患者与其他人群隔离开来，如此就需要对疑似人员进行甄别，而作为甄别肺炎患者的有效措施之一的CT图像检测是特别重要的。然而，CT图像的检测是较为复杂的，其效率不高，并且由于肺炎征象包括多个，需要在获知所有的或者多个征象后才能综合判断是否患有肺炎以及患病的程度，而这多个征象中有些征象的获取难度较大，这也进一步增加了CT图像检测的难度。因此，本申请实施例中通过将肺部区域图像分别输入多个神经网络模型，针对不同的肺炎征象训练得到不同的神经网络模型，可以有针对性的得到精度较高的各个肺炎征象图像，从而为后续肺炎的甄别提供了准确的数据依据。
在一实施例中，肺炎征象图像可以包括如下征象图像中的任一种或多种的组合：肺实变图像、磨玻璃影图像、肿块图像、树芽征图像、结节图像、空洞 图像、晕征图像。肺实变是指终末细支气管远侧的含气腔隙被病理性液体、细胞、组织所代替，其最主要的特点是病变区致密，血管不能显影。磨玻璃影是指各种原因引起的肺泡充填或间质增厚，导致肺密度轻度增高，但其内仍可见血管纹理的病变阴影。肿块是指团块状密度增高影，其最大直径≧3cm。结节是指结节状密度增高影，其最大直径＜3cm。空洞是指肺内病变坏死、液化，经引流支气管排出及气体进入而形成的透亮区，对于感染性疾病，空洞临床上会倾向于化脓性感染。树芽征是指由终末细支气管和肺泡腔内病变形成的小结节影与分支细线影构成的酷似春天的树枝发芽状，多在肺外围支气管末梢呈2-4mm大小结节与树枝状的高密度影。晕征是指结节/空洞周围环绕的类环形玻璃样密度影，通常代表渗出、出血或水肿。可以根据各个征象的不同特征设置不同的神经网络模型来分别对肺部区域图像进行分割，以提高整体的分割精度。
在一实施例中，神经网络模型可以是深度学习神经网络模型，优选地，该神经网络模型可以是Unet神经网络模型。在一实施例中，神经网络模型的训练方式可以是：选取已经由专业医务人员分割并标注出肺炎征象的肺部区域图像作为神经网络模型的训练样本来训练该神经网络模型；并且神经网络模型在分割过程中得到的分割结果也可以由第三方检测机构进行验证和修改，修改后的结果可以作为训练样本再次训练神经网络模型，从而进一步提高神经网络模型的分割精度。应当理解，本申请实施例中的多个神经网络模型的类型可以相同，也可以不同，本申请实施例对于分割各个肺炎征象的神经网络模型的具体类型不做限定。
步骤120：将多个肺炎征象图像组合，得到肺炎综合征象图像。
通常肺炎的确定是基于多个肺炎征象综合判断的，例如当只存在结节图像而不存在其他肺炎征象时，是不能确定被检测者为肺炎患者的，因此，在分别得到多个肺炎征象图像(通常是在CT图像中标注出各个肺炎征象的区域)后，将多个肺炎征象图像组合到一起，得到肺炎综合征象图像，以方便医务人员或者其他的检测机构根据该肺炎综合征象图像准确判断被检测者是否为肺炎患者。
本申请提出了一种肺炎征象的分割方法，通过将CT图像中的肺部区域图像分别输入多个神经网络模型，以此分别得到多个肺炎征象图像，然后将多个肺炎征象图像进行组合，得到肺炎综合征象图像，利用神经网络模型可以高效且准确的获取大量的待检测CT图像中的肺炎征象图像，从而为后续诊断肺炎提供可靠的数据依据。
图2是本申请另一示例性实施例提供的一种肺炎征象的分割方法的流程示意图。肺部区域图像包括多层二维图像，如图2所示，步骤110可以具体包括如下子步骤：
步骤111：逐次将多层二维图像中的一部分输入多个神经网络模型，得到分别对应多个肺炎征象图像的多层二维征象图像。
二维图像的分割要比三维图像的分割难度小、速度快，而CT图像是包括多层二维图像的，CT图像中的肺部区域图像也是包括多层二维图像的，因此，为了提高分割的效率，可以将肺部区域图像的多层二维图像分为多个部分，将这多个部分以多次输入的方式输入多个神经网络模型，由多个神经网络模型分别多次分割二维图像来得到对应的多层二维征象图像，从而提高分割效率。应当理解，本申请实施例可以根据神经网络或处理机器的处理能力适当选取单次输入神经网络模型的二维图像的层数，可以是一层，也可以是多层，还可以是将所有的二维图像单次输入神经网络模型，只要所选取的层数不超过神经网络模型或处理机器所能承载的负荷即可，本申请实施例对于单次输入神经网络模型的二维图像的具体层数不做限定。
步骤112：分别将对应同一肺炎征象图像的多层二维征象图像叠加，得到多个肺炎征象图像。
在得到每个肺炎征象的多层二维征象图像之后，将同一肺炎征象图像的多层二维征象图像叠加起来得到该肺炎征象图像。在一实施例中，相邻次输入神经网络模型的部分二维图像之间存在交叉部分。通过设置交叉部分，可以避免边缘二维图像之间的差异较大，而且可以通过交叉部分的定位更好的实现叠加。
在另一实施例中，图1的分割方法还可以包括：对多个肺炎征象图像分别进行腐蚀膨胀操作。
图3是本申请另一示例性实施例提供的一种肺炎征象的分割方法的流程示意图。如图3所示，在步骤112之后，上述实施例还可以包括：
步骤113：对多个肺炎征象图像分别进行腐蚀膨胀操作。
腐蚀操作为形态学操作，其具体操作过程为：沿着图像中物体边界移除像素并缩小物体的大小，即缩小物体的边界以去除图像中物体的噪声。膨胀操作也是形态学操作，其具体操作过程与腐蚀操作正好相反，即沿着图像中物体边界增加像素并扩大物体的大小。通过腐蚀膨胀操作，可以有效去除分割过程中产生的噪声，同时由于相邻的二维征象图像之间是存在相互关联的，通过对叠加后的肺炎征象图像进行腐蚀膨胀操作，可以利用相邻的二维征象图像之间的相互关联性，去除个别层的分割误差，例如通过上下两层或多层二维征象图像可以调整中间层的二维征象图像，从而提高肺炎征象图像整体的分割精度。
图4是本申请一示例性实施例提供的一种肺部区域图像分割方法的流程示意图。如图4所示，该方法包括如下步骤：
步骤410：获取CT图像中的肋骨区域图像。
由于肋骨是紧紧包裹肺部区域的，并且肺炎患者的肺部形状和部分区域的特性会影响到CT图像的成像，而肺炎患者的肋骨是不会因为患病而发生变化的，因此，可以通过获取CT图像中的肋骨区域来得到肺部区域的外边界，从而提高分割肺部区域的精度。
步骤420：获取CT图像中肺部区域的粗分割图像。
在一实施例中，步骤420的具体实现方式可以为：将CT图像输入神经网络模型，得到肺部区域的粗分割图像。可以利用神经网络模型识别CT图像中的肺部区域，其中神经网络模型可以是Unet等神经网络模型，并且神经网络模型的训练方式可以是：选取已经由专业医务人员识别并标注出肺部区域的CT图像作为神经网络模型的训练样本来训练该神经网络模型。由于该步骤中仅仅是获取CT图像中肺部区域的粗分割图像，并非肺部区域的准确分割图像，因此，可以选取适当数量的训练样本来训练该神经网络模型，从而提高整个肺部分割的效率。
在另一实施例中，步骤420的具体实现方式还可以为：根据肺部区域的CT值，选取CT值在肺部区域的CT值范围内的区域作为粗分割图像。CT值是测定人体某一局部组织或器官密度大小的一种计量单位，通常称亨氏单位(hounsfield unit，HU)，其中，空气的CT值为-1000，致密骨的CT值为+1000。实际上CT值是CT图像中各组织与X线衰减系数相当的对应值，CT值不是绝对不变的数值，它不仅与人体内在因素如呼吸、血流等有关，而且与X线管电压、CT装置、室内温度等外界因素有关。人体组织中除了骨骼以外，其他组织的CT值都在300以下且在-80以上，其中钙化点的CT值为80～300、脂肪的CT值为-20～-80。由于肺部区域内基本为空气，其CT值相对其他组织较低，因此可以设定CT值范围，选取CT值在该CT值范围内的连通区域作为肺部区域的粗分割图像。
步骤430：以肋骨区域图像为肺部区域的边界，并且以粗分割图像为种子区域，以预设的步长向周围扩张至肺部区域的边界，得到肺部区域图像。
在一实施例中，步骤430的具体实现方式可以是：以肋骨区域图像为边界、粗分割图像为感兴趣区域，并通过活动轮廓模型、以预设的步长对CT图像进行分割，以得到肺部区域图像。由于肺部是被肋骨紧紧包裹的，即肋骨区域为肺部区域的外边界，因此，可以将粗分割图像作为种子区域或者感兴趣区域，并且以肋骨区域图像作为边界，通过活动轮廓模型、以预设的步长对CT图像进行分割，即由感兴趣区域开始、以预设的步长向周围扩张，直至扩张到肋骨区域为止，以得到精确分割的肺部区域图像，为后续的肺叶分割、肺炎判断等提供了准确的基础图像数据。其中，预设的步长可以根据实际需求而调整，为 了得到更高的精度，可以适当减小预设的步长。在进一步的实施例中，活动轮廓模型可以包括Level Set模型或Snake模型。应当理解，本申请实施例可以根据实际应用场景的需求而选取不同的活动轮廓模型，只要所选取的活动轮廓模型可以以肋骨区域图像为边界、粗分割图像为种子区域得到精确的肺部区域图像即可，本申请实施例对于活动轮廓模型的具体结构不做限定。
图5是本申请另一示例性实施例提供的一种肺部区域图像分割方法的流程示意图。如图5所示，在步骤420之后，上述实施例还可以包括：
步骤440：对粗分割图像进行腐蚀操作，得到腐蚀后的粗分割图像。
由于肺部区域的粗分割图像通常不是准确的肺部区域图像，例如，粗分割图像中可能会包含肺部区域以外的其他区域，该其他区域即为粗分割图像中的噪声区域，通过腐蚀操作可以将粗分割图像中的噪声区域去除，以保证腐蚀后的粗分割图像包含肺部区域的一部分，而不包含肺部区域以外的区域。
同时，步骤430调整为步骤530：以肋骨区域图像为肺部区域的边界，并且以腐蚀后的粗分割图像为种子区域，以预设的步长向周围扩张至肺部区域的边界，得到肺部区域图像。
由于种子区域只需要是肺部区域的一部分即可，不需要是肺部区域的全部，因此，可以通过腐蚀操作来保证种子区域内只包含肺部区域，从而避免种子区域在扩张的过程中融入更多的非肺部区域。因为种子区域在扩张过程中是不会去除已存在于该种子区域内的区域的，因此，保证种子区域内只包含肺部区域是保证分割肺部区域的精度的前提条件。
图6是本申请一示例性实施例提供的一种肋骨区域图像的获取方法的流程示意图。如图6所示，该获取方法可以包括如下步骤：
步骤411：基于骨头的CT值，获取CT图像中骨头区域图像。
根据CT图像中各个区域的CT值即可获取CT值最大的骨头区域图像。在一实施例中，可以基于骨头的CT值设定第一CT值阈值，获取CT图像中CT值大于或等于第一CT值阈值的连通区域作为骨头区域图像。其中，第一CT值阈值小于骨头的CT值且大于其他组织的CT值。通过设定第一CT值阈值，获取CT图像中CT值大于或等于第一CT值阈值的连通区域，即可得到骨头区域图像。
在一实施例中，在获取CT图像中CT值大于或等于第一CT值阈值的连通区域后，上述方法还可以包括：去除连通区域中面积小于预设的面积阈值的区域。第一CT值阈值设定的过大时可能会遗漏部分骨头区域，而第一CT值阈值设定的过小时，由于钙化点的CT值较大，又可能会出现肺部或者心脏区域存在一定的钙化点成为骨头区域图像中的干扰噪声，因此，需要将其去除。通 常钙化点的面积较小，因此，可以通过去除连通区域中面积小于预设的面积阈值的区域来排出钙化点对骨头区域图像的干扰，其中面积阈值可以根据实际应用而预先设定。
在一实施例中，在获取CT图像中CT值大于或等于第一CT值阈值的连通区域后，上述方法还可以包括：去除位于粗分割图像内的连通区域。通过将粗分割图像内连通区域去除，可以避免肺部或者心脏区域内的钙化点对最终的分割结果造成的影响，从而可以提高后续的分割精度。
步骤412：基于肋骨的特性，分割出骨头区域图像中的肋骨区域，得到肋骨区域图像。
在一实施例中，步骤412的实现方式可以是：将标准肋骨图像与骨头区域图像进行比对，选取骨头区域图像中与标准肋骨图像的相似度大于预设相似度的骨头区域作为肋骨区域图像。由于肋骨相对其他骨头比较规则，肋骨通常是规则的排列在肺部外侧且呈现一定的弧形且为左右对称结构，通过多平面重建的三维视图中可以区分出肋骨与其他骨头的区别，特别是在矢状位上，可以清晰获知肋骨的上述特性，即肋骨的排列和形状有其特殊性，因此，可以通过标准的肋骨图像去比对，选取骨头区域图像中与该标准肋骨图像的相似度满足一定要求(大于预设相似度)的骨头区域，以分割出骨头区域图像中的肋骨区域，得到肋骨区域图像。应当理解，本申请实施例也可以根据实际应用场景的需求而选取其他获取肋骨区域图像的方式，例如可以根据肋骨的弧度是否在预设弧度范围内来判断，也可以根据肋骨之间的间隙是否在预设距离范围内(因为肋骨的排列有一定的规则，相邻的肋骨之间各处的垂直距离都在一定的范围内)，还可以直接通过神经网络模型来获取肋骨区域图像，只要所选取的获取肋骨区域图像的方式满足精度需求即可，本申请实施例对于获取肋骨区域图像的具体方式不做限定。
图7是本申请另一示例性实施例提供的一种肺部区域图像分割方法的流程示意图。如图7所示，在步骤420之前，上述实施例还可以包括：
步骤450：对CT图像进行预处理。
在一实施例中，预处理可以包括以下操作中任一项或多项的组合：去除背景、去除白噪声、裁剪图像、变换窗宽和窗位。其中，去除背景的具体实现方式可以是：通过设置CT值范围，获取该CT值范围内的连通区域，并且只保留连通区域内面积最大的连通区域，其他区域均设置为背景区域，从而排除了其他区域的干扰。去除白噪声的具体实现方式可以是：通过高斯滤波器去除拍摄CT图像过程中引起的白噪声。裁剪图像的具体实现方式可以是：将背景去除，只保留有效区域，以降低后续图像处理的复杂度。变换窗宽和窗位的具体 实现方式可以是：通过设定窗宽和窗位的值来重点突出感兴趣区域，从而避免不感兴趣的区域对后续处理的干扰，在本申请实施例中，可以选取窗位为-500、窗宽为1500，当然应当理解，窗位和窗宽的设定值可以根据实际情况而调整。
应当理解，步骤450可以设置于步骤410之前，通过预处理将CT图像中的背景和其他干扰因素都排出，可以有效降低后续步骤的复杂程度，提高肺部分割的效率。
图8是本申请另一示例性实施例提供的一种肺部区域图像分割方法的流程示意图。如图8所示，在步骤430之后，上述实施例还可以包括：
步骤460：对肺部区域图像的边界进行光滑化处理。
由于根据肋骨区域图像只能确定肺部区域的部分边界，并且是以同一预设的步长通过活动轮廓模型扩张，因而得到的肺部区域图像的边界可能不光滑，在得到肺部区域图像后，对该肺部区域图像的边界进行光滑化处理，可以得到更为精确的肺部区域图像。
示例性装置
图9是本申请一示例性实施例提供的一种肺炎征象的分割装置90的结构示意图。如图9所示，该肺炎征象的分割装置90包括如下模块：
生成模块91，用于基于CT图像中的肺部区域图像，分别生成多个肺炎征象图像；以及组合模块92，用于将多个肺炎征象图像组合，得到肺炎综合征象图像；其中，生成模块91进一步配置为：将肺部区域图像分别输入多个神经网络模型，得到多个肺炎征象图像。
本申请提出的肺炎征象的分割装置，通过生成模块91将CT图像中的肺部区域图像分别输入多个神经网络模型，以此分别得到多个肺炎征象图像，然后通过组合模块92将多个肺炎征象图像进行组合，得到肺炎综合征象图像，利用神经网络模型可以高效且准确的获取大量的待检测CT图像中的肺炎征象图像，从而为后续诊断肺炎提供可靠的数据依据。
在一实施例中，肺炎征象图像可以包括如下征象图像中的任一种或多种的组合：肺实变图像、磨玻璃影图像、肿块图像、树芽征图像、结节图像、空洞图像、晕征图像。在一实施例中，神经网络模型可以是深度学习神经网络模型，优选地，该神经网络模型可以是Unet神经网络模型。
在一实施例中，肺部区域图像包括多层二维图像。如图9所示，生成模块91可以包括如下单元：输入单元911，用于逐次将多层二维图像中的一部分分别输入多个神经网络模型，得到分别对应多个肺炎征象图像的多层二维征象图像；叠加单元912，用于分别将对应同一肺炎征象图像的多层二维征象图像叠 加，得到多个肺炎征象图像。
在一实施例中，肺部区域图像包括多层二维图像。如图9所示，生成模块91可以包括如下单元：输入单元911，用于将多层二维图像分别输入多个神经网络模型，得到分别对应多个肺炎征象图像的多层二维征象图像；叠加单元912，用于分别将对应同一肺炎征象图像的多层二维征象图像叠加，得到多个肺炎征象图像。
在一实施例中，如图9所示，生成模块91还可以包括：后处理单元913，用于对多个肺炎征象图像分别进行腐蚀膨胀操作。
在一实施例中，如图9所示，分割装置90还可以包括：获取模块93，用于获取CT图像中的肋骨区域图像；粗分割模块94，用于获取CT图像中肺部区域的粗分割图像；以及精分割模块95，用于以肋骨区域图像为肺部区域的边界，并且以粗分割图像为种子区域，以预设的步长向周围扩张至肺部区域的边界，得到肺部区域图像。
在一实施例中，粗分割模块94可以进一步配置为：将CT图像输入神经网络模型，得到肺部区域的粗分割图像。
在一实施例中，粗分割模块94可以进一步配置为：根据肺部区域的CT值，选取CT值在肺部区域的CT值范围内的区域作为粗分割图像。
在一实施例中，精分割模块95可以进一步配置为：以肋骨区域图像为边界、粗分割图像为感兴趣区域，并通过活动轮廓模型、以预设的步长对CT图像进行分割，以得到肺部区域图像。其中，活动轮廓模型可以包括Level Set模型或Snake模型。
在一实施例中，如图9所示，该分割装置90还可以包括：腐蚀模块96，用于对粗分割图像进行腐蚀操作，得到腐蚀后的粗分割图像。且精分割模块95配置为：以肋骨区域图像为肺部区域的边界，并且以腐蚀后的粗分割图像为种子区域，以预设的步长向周围扩张至肺部区域的边界，得到肺部区域图像。
在一实施例中，精分割模块95可以进一步配置为：以肋骨区域图像为边界、腐蚀后的粗分割图像为感兴趣区域，并通过活动轮廓模型、以预设的步长对CT图像进行分割，以得到肺部区域图像。其中，活动轮廓模型可以包括Level Set模型或Snake模型。
在一实施例中，如图9所示，获取模块93可以包括：骨头获取单元931，用于基于骨头的CT值，获取CT图像中骨头区域图像；肋骨获取单元932，用于基于肋骨的特性，分割出骨头区域图像中的肋骨区域，得到肋骨区域图像。
在一实施例中，骨头获取单元931可以进一步配置为：基于骨头的CT值设定第一CT值阈值，获取CT图像中CT值大于或等于第一CT值阈值的连通 区域作为骨头区域图像。
在一实施例中，获取模块93可以进一步配置为：在获取CT图像中CT值大于或等于第一CT值阈值的连通区域后，去除连通区域中面积小于预设的面积阈值的区域。
在一实施例中，获取模块93可以进一步配置为：在获取CT图像中CT值大于或等于第一CT值阈值的连通区域后，去除位于粗分割图像内的连通区域。
在一实施例中，肋骨获取单元932可以进一步配置为：将标准肋骨图像与骨头区域图像进行比对，选取骨头区域图像中与标准肋骨图像的相似度大于预设相似度的骨头区域作为肋骨区域图像。
在一实施例中，如图9所示，该分割装置90还可以包括：预处理模块97，用于对CT图像进行预处理。在一实施例中，预处理可以包括以下操作中任一项或多项的组合：去除背景、去除白噪声、裁剪图像、变换窗宽和窗位。
在一实施例中，如图9所示，该分割装置90还可以包括：光滑化模块98，用于对肺部区域图像的边界进行光滑化处理。
示例性电子设备
下面，参考图10来描述根据本申请实施例的电子设备10。该电子设备10可以是第一设备和第二设备中的任一个或两者、或与它们独立的单机设备，该单机设备可以与第一设备和第二设备进行通信，以从它们接收所采集到的输入信号。
图10图示了根据本申请实施例的电子设备10的框图。
如图10所示，电子设备10包括一个或多个处理器11和存储器12。
处理器11可以是中央处理单元(CPU)或者具有数据处理能力和/或指令执行能力的其他形式的处理单元，并且可以控制电子设备10中的其他组件以执行期望的功能。
存储器12可以包括一个或多个计算机程序产品，所述计算机程序产品可以包括各种形式的计算机可读存储介质，例如易失性存储器和/或非易失性存储器。所述易失性存储器例如可以包括随机存取存储器(RAM)和/或高速缓冲存储器(cache)等。所述非易失性存储器例如可以包括只读存储器(ROM)、硬盘、闪存等。在所述计算机可读存储介质上可以存储一个或多个计算机程序指令，处理器11可以运行所述程序指令，以实现上文所述的本申请的各个实施例的肺炎征象的分割方法以及/或者其他期望的功能。在所述计算机可读存储介质中还可以存储诸如输入信号、信号分量、噪声分量等各种内容。
在一个示例中，电子设备10还可以包括：输入装置13和输出装置14，这些组件通过总线系统和/或其他形式的连接机构(未示出)互连。
例如，在该电子设备10是第一设备或第二设备时，该输入装置13可以是CT扫描仪，用于获取CT图像。在该电子设备10是单机设备时，该输入装置13可以是通信网络连接器，用于从第一设备和第二设备接收所采集的输入信号。
此外，该输入装置13还可以包括例如键盘、鼠标等等。
该输出装置14可以向外部输出各种信息，包括确定出的肺部区域图像、肺炎征象图像、肺炎综合征象图像等。该输出装置14可以包括例如显示器、扬声器、打印机、以及通信网络及其所连接的远程输出设备等等。
当然，为了简化，图10中仅示出了该电子设备10中与本申请有关的组件中的一些，省略了诸如总线、输入/输出接口等等的组件。除此之外，根据具体应用情况，电子设备10还可以包括任何其他适当的组件。
示例性计算机程序产品和计算机可读存储介质
除了上述方法和设备以外，本申请的实施例还可以是计算机程序产品，其包括计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行本说明书上述“示例性方法”部分中描述的根据本申请各种实施例的肺炎征象的分割方法中的步骤。
所述计算机程序产品可以以一种或多种程序设计语言的任意组合来编写用于执行本申请实施例操作的程序代码，所述程序设计语言包括面向对象的程序设计语言，诸如Java、C++等，还包括常规的过程式程序设计语言，诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算设备上执行、部分地在用户设备上执行、作为一个独立的软件包执行、部分在用户计算设备上部分在远程计算设备上执行、或者完全在远程计算设备或服务器上执行。
此外，本申请的实施例还可以是计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行本说明书上述“示例性方法”部分中描述的根据本申请各种实施例的肺炎征象的分割方法中的步骤。
所述计算机可读存储介质可以采用一个或多个可读介质的任意组合。可读介质可以是可读信号介质或者可读存储介质。可读存储介质例如可以包括但不限于电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。
以上结合具体实施例描述了本申请的基本原理，但是，需要指出的是，在本申请中提及的优点、优势、效果等仅是示例而非限制，不能认为这些优点、优势、 效果等是本申请的各个实施例必须具备的。另外，上述公开的具体细节仅是为了示例的作用和便于理解的作用，而非限制，上述细节并不限制本申请为必须采用上述具体的细节来实现。
本申请中涉及的器件、装置、设备、系统的方框图仅作为例示性的例子并且不意图要求或暗示必须按照方框图示出的方式进行连接、布置、配置。如本领域技术人员将认识到的，可以按任意方式连接、布置、配置这些器件、装置、设备、系统。诸如“包括”、“包含”、“具有”等等的词语是开放性词汇，指“包括但不限于”，且可与其互换使用。这里所使用的词汇“或”和“和”指词汇“和/或”，且可与其互换使用，除非上下文明确指示不是如此。这里所使用的词汇“诸如”指词组“诸如但不限于”，且可与其互换使用。
还需要指出的是，在本申请的装置、设备和方法中，各部件或各步骤是可以分解和/或重新组合的。这些分解和/或重新组合应视为本申请的等效方案。
提供所公开的方面的以上描述以使本领域的任何技术人员能够做出或者使用本申请。对这些方面的各种修改对于本领域技术人员而言是非常显而易见的，并且在此定义的一般原理可以应用于其他方面而不脱离本申请的范围。因此，本申请不意图被限制到在此示出的方面，而是按照与在此公开的原理和新颖的特征一致的最宽范围。
为了例示和描述的目的已经给出了以上描述。此外，此描述不意图将本申请的实施例限制到在此公开的形式。尽管以上已经讨论了多个示例方面和实施例，但是本领域技术人员将认识到其某些变型、修改、改变、添加和子组合。

Claims (15)

  1. 一种肺炎征象的分割方法，其特征在于，包括：
    基于CT图像中的肺部区域图像，分别生成多个肺炎征象图像；以及
    将所述多个肺炎征象图像组合，得到肺炎综合征象图像；
    其中，分别生成所述多个肺炎征象图像的方式包括：
    将所述肺部区域图像分别输入多个神经网络模型，得到所述多个肺炎征象图像。
  2. 根据权利要求1所述的分割方法，其特征在于，所述多个肺炎征象图像包括如下征象图像中的任一种或多种的组合：
    肺实变图像、磨玻璃影图像、肿块图像、树芽征图像、结节图像、空洞图像、晕征图像。
  3. 根据权利要求1或2所述的分割方法，其特征在于，所述肺部区域图像包括多层二维图像；所述将所述肺部区域图像分别输入多个神经网络模型，得到所述多个肺炎征象图像包括：
    逐次将所述多层二维图像中的一部分分别输入所述多个神经网络模型，得到分别对应所述多个肺炎征象图像的多层二维征象图像；以及
    分别将对应同一肺炎征象图像的所述多层二维征象图像叠加，得到所述多个肺炎征象图像。
  4. 根据权利要求1或2所述的分割方法，其特征在于，所述肺部区域图像包括多层二维图像；所述将所述肺部区域图像分别输入多个神经网络模型，得到所述多个肺炎征象图像包括：
    将所述多层二维图像分别输入所述多个神经网络模型，得到分别对应所述多个肺炎征象图像的多层二维征象图像；以及
    分别将对应同一肺炎征象图像的所述多层二维征象图像叠加，得到所述多个肺炎征象图像。
  5. 根据权利要求1至4中任一项所述的分割方法，其特征在于，在所述得到所述多个肺炎征象图像之后，还包括：
    对所述多个肺炎征象图像分别进行腐蚀膨胀操作。
  6. 根据权利要求1至5中任一项所述的分割方法，其特征在于，所述肺部区域图像的获取方式包括：
    获取所述CT图像中的肋骨区域图像；
    获取所述CT图像中肺部区域的粗分割图像；以及
    以所述肋骨区域图像为所述肺部区域的边界，并且以所述粗分割图像为种 子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像。
  7. 根据权利要求6所述的分割方法，其特征在于，在所述获取所述CT图像中肺部区域的粗分割图像之后，还包括：
    对所述粗分割图像进行腐蚀操作，得到腐蚀后的粗分割图像；
    所述以所述肋骨区域图像为所述肺部区域的边界，并且以所述粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像包括：
    以所述肋骨区域图像为所述肺部区域的边界，并且以所述腐蚀后的粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像。
  8. 根据权利要求6所述的分割方法，其特征在于，基于活动轮廓模型，执行所述以所述肋骨区域图像为所述肺部区域的边界，并且以所述粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像的步骤。
  9. 根据权利要求7所述的分割方法，其特征在于，基于活动轮廓模型，执行所述以所述肋骨区域图像为所述肺部区域的边界，并且以所述腐蚀后的粗分割图像为种子区域，以预设的步长向周围扩张至所述肺部区域的边界，得到所述肺部区域图像的步骤。
  10. 根据权利要求6至9中任一项所述的分割方法，其特征在于，所述肋骨区域图像的获取方法包括：
    基于骨头的CT值，获取所述CT图像中骨头区域图像；
    基于肋骨的特性，分割出所述骨头区域图像中的肋骨区域，得到所述肋骨区域图像。
  11. 根据权利要求6至10中任一项所述的分割方法，其特征在于，在所述获取所述CT图像中肺部区域的粗分割图像之前，所述分割方法还包括：
    对所述CT图像进行预处理，其中，所述预处理包括以下操作中任一项或多项的组合：去除背景、去除白噪声、裁剪图像、变换窗宽和窗位。
  12. 根据权利要求6至11中任一项所述的分割方法，其特征在于，还包括：
    对所述肺部区域图像的边界进行光滑化处理。
  13. 一种肺炎征象的分割装置，其特征在于，包括：
    生成模块，用于基于CT图像中的肺部区域图像，分别生成多个肺炎征象图像；以及
    组合模块，用于将所述多个肺炎征象图像组合，得到肺炎综合征象图像；
    其中，所述生成模块进一步配置为：
    将所述肺部区域图像分别输入多个神经网络模型，得到所述多个肺炎征象图像。
  14. 一种计算机可读存储介质，所述存储介质存储有计算机程序，所述计算机程序用于执行上述权利要求1至12中任一项所述的肺炎征象的分割方法。
  15. 一种电子设备，所述电子设备包括：
    处理器；
    用于存储所述处理器可执行指令的存储器；
    所述处理器，用于执行上述权利要求1至12中任一项所述的肺炎征象的分割方法。
PCT/CN2021/072515 2020-03-17 2021-01-18 肺炎征象的分割方法、装置、介质及电子设备 WO2021184949A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP21771562.2A EP3971830B1 (en) 2020-03-17 2021-01-18 Pneumonia sign segmentation method and apparatus, medium and electronic device
JP2021571620A JP7304437B2 (ja) 2020-03-17 2021-01-18 肺炎徴候の分割方法、装置、媒体及び電子デバイス
US17/540,629 US20220092788A1 (en) 2020-03-17 2021-12-02 Segmentation method for pneumonia sign, medium, and electronic device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010189089.6A CN111415356B (zh) 2020-03-17 2020-03-17 肺炎征象的分割方法、装置、介质及电子设备
CN202010189089.6 2020-03-17

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/540,629 Continuation US20220092788A1 (en) 2020-03-17 2021-12-02 Segmentation method for pneumonia sign, medium, and electronic device

Publications (1)

Publication Number Publication Date
WO2021184949A1 true WO2021184949A1 (zh) 2021-09-23

Family

ID=71494475

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/072515 WO2021184949A1 (zh) 2020-03-17 2021-01-18 肺炎征象的分割方法、装置、介质及电子设备

Country Status (5)

Country Link
US (1) US20220092788A1 (zh)
EP (1) EP3971830B1 (zh)
JP (1) JP7304437B2 (zh)
CN (1) CN111415356B (zh)
WO (1) WO2021184949A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111415356B (zh) * 2020-03-17 2020-12-29 推想医疗科技股份有限公司 肺炎征象的分割方法、装置、介质及电子设备
WO2022225794A1 (en) * 2021-04-23 2022-10-27 The Johns Hopkins University Systems and methods for detecting and characterizing covid-19
CN115496703A (zh) * 2021-06-18 2022-12-20 富联精密电子(天津)有限公司 肺炎区域检测方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140074493A1 (en) * 2012-01-03 2014-03-13 David Lyle Schneider Medical protection lockout system for procedures and devices
CN110503640A (zh) * 2018-08-21 2019-11-26 深圳科亚医疗科技有限公司 对医学图像进行分析的装置、系统及计算机可读介质
CN110706200A (zh) * 2019-09-02 2020-01-17 杭州深睿博联科技有限公司 数据预测的方法及装置
CN111415356A (zh) * 2020-03-17 2020-07-14 北京推想科技有限公司 肺炎征象的分割方法、装置、介质及电子设备
CN111415743A (zh) * 2020-03-17 2020-07-14 北京推想科技有限公司 肺炎分类方法、装置、介质及电子设备

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002087444A1 (fr) * 2001-04-26 2002-11-07 Teijin Limited Procede de mesure d'une structure articulaire tridimensionnelle
WO2006093523A2 (en) * 2004-07-15 2006-09-08 Kenji Suzuki Computerized scheme for distinction between benign and malignant nodules in thoracic low-dose ct
WO2010009735A2 (en) * 2008-07-23 2010-01-28 Dako Denmark A/S Combinatorial analysis and repair
EA035265B1 (ru) * 2013-10-02 2020-05-21 Бёрингер Ингельхайм Энимал Хелс Ю-Эс-Эй Инк. Вариант белка opc2 pcv2 и содержащие его вирусоподобные частицы
JP6906347B2 (ja) * 2017-04-10 2021-07-21 富士フイルム株式会社 医用画像分類装置、方法およびプログラム
CN107220373B (zh) * 2017-06-19 2019-09-17 太原理工大学 一种基于医学征象和卷积神经网络的肺结节ct图像哈希检索方法
CN107392910B (zh) * 2017-07-06 2020-01-07 东软医疗系统股份有限公司 一种基于ct图像的肺叶分割方法及装置
CN107563383A (zh) * 2017-08-24 2018-01-09 杭州健培科技有限公司 一种医学影像辅助诊断及半监督样本生成系统
CN110838124B (zh) * 2017-09-12 2021-06-18 深圳科亚医疗科技有限公司 用于分割具有稀疏分布的对象的图像的方法、系统和介质
CN107845091B (zh) * 2017-10-31 2019-02-15 北京青燕祥云科技有限公司 基于预筛查的快速标注方法和系统
JP2019169049A (ja) 2018-03-26 2019-10-03 富士フイルム株式会社 医用画像特定装置、方法およびプログラム
US10304193B1 (en) * 2018-08-17 2019-05-28 12 Sigma Technologies Image segmentation and object detection using fully convolutional neural network
CN109727253A (zh) * 2018-11-14 2019-05-07 西安大数据与人工智能研究院 基于深度卷积神经网络自动分割肺结节的辅助检测方法
CN109636802A (zh) * 2019-01-18 2019-04-16 天津工业大学 基于深度卷积神经网络的肺实质经ct图像分割方法
CN110517225B (zh) * 2019-07-19 2023-07-11 平安科技（深圳）有限公司 Ai影像识别方法、装置、设备及存储介质
CN110599448B (zh) * 2019-07-31 2022-03-15 浙江工业大学 基于MaskScoring R-CNN网络的迁移学习肺部病变组织检测系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140074493A1 (en) * 2012-01-03 2014-03-13 David Lyle Schneider Medical protection lockout system for procedures and devices
CN110503640A (zh) * 2018-08-21 2019-11-26 深圳科亚医疗科技有限公司 对医学图像进行分析的装置、系统及计算机可读介质
CN110706200A (zh) * 2019-09-02 2020-01-17 杭州深睿博联科技有限公司 数据预测的方法及装置
CN111415356A (zh) * 2020-03-17 2020-07-14 北京推想科技有限公司 肺炎征象的分割方法、装置、介质及电子设备
CN111415743A (zh) * 2020-03-17 2020-07-14 北京推想科技有限公司 肺炎分类方法、装置、介质及电子设备

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
DENG LING-BO , ZHOU WEN , ZENG QIAO-LING , YUAN ZHI-DONG , CHENG GUAN-XUN: "Clinical and Chest CT Features of Suspected and Confirmed NovelCoronary Pneumonia", MEDICAL INFORMATION, vol. 33, no. 4, 15 February 2020 (2020-02-15), CN, pages 4 - 7, XP055852127, ISSN: 1006-1959, DOI: 10.3969/j.issn.1006-1959.2020.04.002 *

Also Published As

Publication number Publication date
EP3971830B1 (en) 2023-09-20
CN111415356B (zh) 2020-12-29
JP7304437B2 (ja) 2023-07-06
EP3971830C0 (en) 2023-09-20
US20220092788A1 (en) 2022-03-24
EP3971830A4 (en) 2022-09-14
CN111415356A (zh) 2020-07-14
EP3971830A1 (en) 2022-03-23
JP2022535811A (ja) 2022-08-10

Similar Documents

Publication Publication Date Title
WO2021184949A1 (zh) 肺炎征象的分割方法、装置、介质及电子设备
CN110969622B (zh) 用于辅助肺炎诊断的图像处理方法和系统
CN111415743B (zh) 肺炎分类方法、装置、介质及电子设备
Pu et al. CT based computerized identification and analysis of human airways: a review
Pu et al. A computational geometry approach to automated pulmonary fissure segmentation in CT examinations
JP5701138B2 (ja) 医用画像処理装置および方法、並びにプログラム
US20110243403A1 (en) Medical image processing apparatus and method, and computer readable recording medium on which is recorded program for the same
JP2009011828A (ja) 血管系の中の障害を検出するための方法及びシステム
JP2014161734A (ja) 医療画像を整合する方法及びその装置
JP2008503303A (ja) 気管気管支樹を表示する方法及び装置
JP5943353B2 (ja) 画像処理装置及び画像処理方法
JP2005199057A (ja) 肺画像データの位置合わせのための方法及び装置
US20050063579A1 (en) Method of automatically detecting pulmonary nodules from multi-slice computed tomographic images and recording medium in which the method is recorded
CN111260669A (zh) 基于ct图像的肺叶分割方法和装置
JP2009028161A (ja) 医用画像表示装置及び医用画像表示方法
WO2021209887A1 (en) Rapid, accurate and machine-agnostic segmentation and quantification method and device for coronavirus ct-based diagnosis
JP2007526033A (ja) 構造化オブジェクトの画像を登録するための装置及び方法
US20100266173A1 (en) Computer-aided detection (cad) of a disease
Rebouças Filho et al. 3D segmentation and visualization of lung and its structures using CT images of the thorax
CN111311612A (zh) 肺部分割方法、装置、介质及电子设备
Chiplunkar et al. Segmentation and quantitation of the primary human airway tree
WO2022164374A1 (en) Automated measurement of morphometric and geometric parameters of large vessels in computed tomography pulmonary angiography
US20050002548A1 (en) Automatic detection of growing nodules
JP2006102508A (ja) Ｃａｄシステムにおける解剖学的形状の検出のための方法及びシステム
CN111415341A (zh) 肺炎阶段的评估方法、装置、介质及电子设备

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: 21771562

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021571620

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2021771562

Country of ref document: EP

Effective date: 20211217

NENP Non-entry into the national phase

Ref country code: DE