WO2021139447A1 - 一种宫颈异常细胞检测装置及方法 - Google Patents

一种宫颈异常细胞检测装置及方法 Download PDF

Info

Publication number
WO2021139447A1
WO2021139447A1 PCT/CN2020/132474 CN2020132474W WO2021139447A1 WO 2021139447 A1 WO2021139447 A1 WO 2021139447A1 CN 2020132474 W CN2020132474 W CN 2020132474W WO 2021139447 A1 WO2021139447 A1 WO 2021139447A1
Authority
WO
WIPO (PCT)
Prior art keywords
cervical
slide image
cytopathological
cells
slide
Prior art date
Application number
PCT/CN2020/132474
Other languages
English (en)
French (fr)
Inventor
初晓
郭冰雪
李亘杰
Original Assignee
平安科技(深圳)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 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2021139447A1 publication Critical patent/WO2021139447A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/11Region-based segmentation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/50Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/90Dynamic range modification of images or parts thereof
    • G06T5/92Dynamic range modification of images or parts thereof based on global image properties
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • G06T7/0012Biomedical image inspection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30004Biomedical image processing
    • G06T2207/30024Cell structures in vitro; Tissue sections in vitro
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30004Biomedical image processing
    • G06T2207/30096Tumor; Lesion

Definitions

  • This application relates to the field of medical technology, and in particular to a device and method for detecting abnormal cervical cells.
  • Cervical cancer is one of the malignant tumors that cause serious harm to women's life and health, and its incidence ranks second among female malignant tumors.
  • cervical cancer is currently the only cancer that can be detected and cured early, so early screening is critical to the treatment of cervical cancer.
  • Cervical liquid-based cell examination is currently the most commonly used cervical cancer screening method.
  • the screening rate for cervical cancer is very low, so various intelligent auxiliary screening equipment systems are gradually appearing.
  • the inventor realized that most of the current intelligent assisted screening systems for cervical cancer have low detection accuracy for abnormal cells such as atypical squamous cells of undetermined signature (ASC-US).
  • ASC-US atypical squamous cells of undetermined signature
  • Some of the abnormal cervical cells have a low degree of deformation (multiple increase of nucleus to cytoplasm and degree of nuclear anomaly, etc.), and some of them are in the form of single small cells rather than clusters of cells. Therefore, in the current cervical cancer intelligent auxiliary screen It is difficult to detect abnormal cervical cells during the examination, resulting in low accuracy in detecting abnormal cervical cells.
  • the present application provides a device and method for detecting abnormal cervical cells, which is beneficial to improve the accuracy of detecting abnormal cervical cells.
  • the first aspect of the present application provides a device for detecting abnormal cervical cells, including:
  • Obtaining module used to obtain cervical cytopathological slide images
  • a segmentation module configured to segment the cervical cytopathological slide image into a plurality of cervical cytopathological slide image blocks
  • the preprocessing module is used to preprocess each cervical cytopathological slide image block to obtain multiple target cervical cytopathological slide image blocks;
  • the processing module is used to process each target cervical cytopathological slide image block to obtain the cervical cell characteristics in each target cervical cytopathological slide image block, wherein the cervical cell characteristics are determined by the height of the cervical cells.
  • the determining module is configured to determine the abnormal probability of the cervical cells in the cervical cytopathological slide image according to the characteristics of the cervical cells in each target cervical cytopathological slide image block.
  • the second aspect of the present application provides a method for detecting abnormal cervical cells, including:
  • Each target cervical cytopathological slide image block is processed to obtain the cervical cell characteristics in each target cervical cytopathological slide image block, wherein the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • the third aspect of the present application provides an electronic device that includes a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and are The configuration is performed by the processor to implement the following methods:
  • Each target cervical cytopathological slide image block is processed to obtain the cervical cell characteristics in each target cervical cytopathological slide image block, wherein the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • a fourth aspect of the present application provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the following method:
  • Each target cervical cytopathological slide image block is processed to obtain the cervical cell characteristics in each target cervical cytopathological slide image block, where the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • this application When acquiring cervical cell characteristics, this application performs two-way feature weighted fusion of high-level features to low-level features and low-level features to high-level features of cervical cells, which can fuse features of different levels and enrich the expression ability of features. In this way, the detailed information of the cells can be better extracted, and the detection accuracy of ASC-US similar to normal cervical cells can be improved, that is, the detection accuracy of abnormal cervical cells can be effectively improved.
  • Figure 1 is a schematic diagram of a feature pyramid network
  • FIG. 2 is a schematic flowchart of a method for detecting abnormal cervical cells according to an embodiment of the application
  • FIG. 3 is a schematic diagram of bidirectional feature fusion in the target pyramid network provided by an embodiment of this application.
  • FIG. 4 is a schematic flowchart of another method for detecting abnormal cervical cells provided by an embodiment of the application.
  • FIG. 5 is a schematic diagram of a device for detecting abnormal cervical cells according to an embodiment of the application.
  • FIG. 6 is a schematic diagram of the structure of an electronic device in a hardware operating environment involved in an embodiment of the application.
  • the device and method for detecting abnormal cervical cells provided in the embodiments of the present application are beneficial to improve the accuracy of detecting abnormal cervical cells.
  • At least one refers to one or more, and “multiple” refers to two or more.
  • And/or describes the association relationship of the associated objects, indicating that there can be three relationships, for example, A and/or B, which can mean: A alone exists, A and B exist at the same time, and B exists alone, where A, B can be singular or plural.
  • the character “/” generally indicates that the associated objects before and after are in an “or” relationship.
  • "The following at least one item (a)” or similar expressions refers to any combination of these items, including any combination of a single item (a) or a plurality of items (a).
  • first, second, third, and fourth in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than describing a specific order, Timing, priority, or importance.
  • first information and the second information are only for distinguishing different information, but do not indicate the difference in content, priority, sending order, or importance of the two types of information.
  • the terms “including” and “having” and any variations of them are intended to cover non-exclusive inclusions.
  • a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes unlisted steps or units, or optionally also includes Other steps or units inherent to these processes, methods, products or equipment.
  • the technical solution of this application can be applied to the fields of artificial intelligence, smart city, digital medical and/or blockchain technology, for example, it can specifically involve neural network technology to realize abnormality detection.
  • the data involved in this application such as images and/or abnormal probability, can be stored in a database, or can be stored in a blockchain, such as distributed storage through a blockchain, which is not limited in this application.
  • Cervical cancer is one of the malignant tumors that cause serious harm to women's life and health, and its incidence ranks second among female malignant tumors.
  • Common cervical cell lesions include: atypical squamous cells-atypical squamous cells of undetermined significance (ASC-US), low-grade squamous intraepithelial lesions (LSIL), atypical squamous cells of undetermined significance
  • ASC-US atypical squamous cells-atypical squamous cells of undetermined significance
  • LSIL low-grade squamous intraepithelial lesions
  • HSIL high-grade squamous intraepithelial lesions
  • AGC atypical glandular epithelial cells
  • Cervical liquid-based cell examination is currently the most commonly used cervical cancer screening method, and most of the current cervical cancer intelligent assisted screening systems have low accuracy in detecting abnormal cells such as ASC-US.
  • Some of the abnormal cervical cells have a low degree of deformation (multiple increase of nucleus to cytoplasm and degree of nuclear anomaly, etc.), and some of them are in the form of single small cells rather than clusters of cells. Therefore, in the current cervical cancer intelligent auxiliary screen It is difficult to detect abnormal cervical cells during the examination, resulting in low accuracy in detecting abnormal cervical cells.
  • FPN Feature pyramid networks
  • FIG. 1 is a schematic diagram of a feature pyramid network.
  • FPN will use the information of each layer in the CNN network to generate the final feature combination, so that the final output feature can better represent the information of each level of the input image.
  • the basic process of FPN includes: bottom-up feature generation at different levels, top-down feature fusion, and association expression between CNN network layer features and the final output features at each level. That is to say, the features generated after top-down processing also have an association relationship.
  • the features of the upper layer will affect the feature expression of the lower layer, and finally all the features are used together as the input for the next task of target detection or category analysis.
  • FIG. 2 is a schematic flowchart of a method for detecting abnormal cervical cells according to an embodiment of the application.
  • a method for detecting abnormal cervical cells provided by an embodiment of the present application may include:
  • the initial cervical cytopathological slide image obtained by scanning the cervical cytopathological slide by a scanner is obtained, and the region of interest in the initial cervical cytopathological slide image is extracted to obtain the cervical cytopathological slide.
  • Slice image is obtained.
  • the cervical cytopathology slide is first obtained, and then the cervical cytopathology slide is processed to obtain the digital information of the slide, that is, the initial cervical cytopathology slide image.
  • a scanner may be used to scan the cervical cytopathology slide to obtain an initial cervical cytopathology slide image, where the initial cervical cytopathology slide image includes multiple formats, for example, svs format , Kfb format, ndpi format, tif format, sdpc format, etc.
  • the Hough transform is used to find the foreground area where the cervical cells are located in the initial cervical cell pathology slide image, so as to extract the region of interest (ROI) where the cervical cells are located, and get Cervical cell pathology slide image.
  • ROI region of interest
  • the cervical cytopathological slide After acquiring the cervical cytopathological slide image, the cervical cytopathological slide is divided into multiple cervical cytopathological slide image blocks.
  • the areas of the multiple cervical cytopathological slide image blocks may be the same or different.
  • the cervical cytopathology slide image is divided into multiple cervical cytopathology slide image blocks, and each image block can be processed separately in the subsequent calculation process, thereby increasing the processing speed.
  • the cervical cytopathology slide image block Before detecting the cervical cytopathology slide image block, it is necessary to preprocess the cervical cytopathology slide image block, including Gamma transformation and contrast enhancement processing, to correct the image that is too bright or too dark, and at the same time, Improve the contrast of the image, improve the visual effect of the image, and reduce the difference of the digitized slide image scanned by different scanners.
  • each target cervical cell pathology slide image block Process each target cervical cell pathology slide image block to obtain cervical cell characteristics in each target cervical cell pathology slide image block, wherein the cervical cell characteristics range from the high-level features of the cervical cells to the cervical cell characteristics. Low-level features and two-way feature weighted fusion from low-level features to high-level features are obtained.
  • each target cervical cell pathological slide image block into the convolutional neural network to obtain the multi-level cell characteristics of the cervical cells; perform high-level features to low-level features of the multi-level cell features of the cervical cells Feature weighted fusion of features, and then feature weighted fusion from low-level features to high-level features, to obtain cervical cell features.
  • the features of different levels reflect different image features.
  • the lower-level features reflect the shallower-level image features, such as edges, and the higher-level features reflect the deeper-level image features, such as object contours.
  • the multi-level cell features of the cervical cells are weighted and fused from the high-level features to the low-level features through the target pyramid network, and then the low-level features to the high-level features are combined. Feature weighted fusion, and finally the cervical cell features are obtained.
  • the target pyramid network provided by the embodiment of this application is an improvement over the existing pyramid network structure. See FIG. 3, which is a schematic diagram of the two-way feature fusion in the target pyramid network provided by the embodiment of the application. As shown in FIG. 3, When performing feature weighted fusion from high-level features to low-level features, there is an association relationship between the features generated after top-down processing.
  • the upper-level features will affect the lower-level feature expression, and the features from low-level features to high-level features are performed.
  • the features generated after bottom-up processing also have an association relationship, and the features of the lower layer will affect the feature expression of the upper layer.
  • the high-level features to low-level features and low-level features to high-level features of cervical cells are subjected to two-way feature weighted fusion, which can fuse different levels of features together, enrich the expression ability of features, and better Extracting the detailed information of cells can improve the detection accuracy for ASC-US cells that are similar to normal cervical cells.
  • the weights corresponding to the cell features of multiple levels of cervical cells in the feature-weighted fusion are obtained.
  • the features of multiple levels are often simply averaged fusion.
  • the weights corresponding to the features of different levels may be different. That is to say, weighted fusion is performed in the feature fusion process.
  • the abnormal probability of the cervical cells is determined according to the characteristics of the cervical cells; when the abnormal probability of the cervical cells is not less than a preset abnormal probability threshold, it is determined that the cervical cells are abnormal cervical cells.
  • each target cervical cell pathology slide image block the characteristics of the cervical cells therein are acquired, so as to determine the abnormal probability of the cervical cells in each image block.
  • the abnormal probability is not less than the preset abnormal probability threshold, the cervical cells can be considered abnormal, so that abnormal cervical cells in each image block can be detected.
  • the abnormal cervical cells in the cervical cytopathological slide image can be detected, and the automatic detection of abnormal cervical cells can be realized.
  • the cervical cells are abnormal cervical cells
  • the abnormal probability of cervical cells is acquired.
  • the cervical cell can be considered abnormal
  • the position information of the cervical cell in the corresponding target cervical cell pathological slide image block is obtained, and the position information and abnormal probability of the cervical cell are retained
  • the location information and abnormal probability of abnormal cervical cells in each image block can be detected.
  • the position of abnormal cervical cells in the cervical cytopathology slide image can be determined.
  • it can be displayed according to the position of the abnormal cervical cells in the cervical cytopathological slide image, and the corresponding abnormal probability can also be displayed.
  • the high-level features to low-level features and low-level features to high-level features of cervical cells are subjected to bidirectional feature weighted fusion, which can be The features of different levels are merged together to enrich the expression ability of the features, so as to better extract the detailed information of the cells.
  • the detection accuracy of ASC-US similar to normal cervical cells can be improved, thereby effectively improving the abnormal cervical cells The detection accuracy.
  • the solution of this application can also be applied to the field of smart medical care.
  • a cervical cell pathological slide image input by a doctor is received, and the abnormal probability of cervical cells in the cervical cytopathological slide image is determined by the method for detecting abnormal cervical cells of the present application. Since the cervical abnormal cell detection method of the present application can determine the abnormal probability of cervical cells more accurately, this can provide a more accurate judgment basis for the doctor's diagnosis, thereby improving the doctor's diagnosis efficiency and accuracy.
  • FIG. 4 is a schematic flowchart of another method for detecting abnormal cervical cells provided by an embodiment of the application.
  • another method for detecting abnormal cervical cells provided in an embodiment of the present application may include:
  • the initial cervical cytopathological slide image obtained by scanning the cervical cytopathological slide by a scanner is obtained, and the region of interest in the initial cervical cytopathological slide image is extracted to obtain the cervical cytopathological slide.
  • Slice image is obtained.
  • the cervical cytopathology slide is first obtained, and then the cervical cytopathology slide is processed to obtain the digital information of the slide, that is, the initial cervical cytopathology slide image.
  • a scanner may be used to scan the cervical cytopathology slide to obtain an initial cervical cytopathology slide image, where the initial cervical cytopathology slide image includes multiple formats, for example, svs format , Kfb format, ndpi format, tif format, sdpc format, etc.
  • the initial cervical cytopathological slide image is obtained, the initial cervical cytopathological slide image is enlarged, where the magnification may be preset. After the magnification process, the black and white image of the magnified initial cervical cell pathological slide image is successively subjected to dilation operation and corrosion operation.
  • the Hough transform is used to find the foreground area where the cervical cells are located in the initial cervical cell pathology slide image processed by the expansion and corrosion operation, so as to extract the region of interest (ROI) where the cervical cells are located, and obtain the cervical cell pathology glass Slice image.
  • ROI region of interest
  • the cervical cytopathological slide After acquiring the cervical cytopathological slide image, the cervical cytopathological slide is divided into multiple cervical cytopathological slide image blocks.
  • the areas of the multiple cervical cytopathological slide image blocks may be the same or different.
  • the cervical cytopathology slide image is divided into multiple cervical cytopathology slide image blocks, and each image block can be processed separately in the subsequent calculation process, thereby increasing the processing speed.
  • the areas of the multiple cervical cytopathological slide image blocks are the same size, and during the segmentation process, a certain size window is used to segment the cervical cytopathology slide image, so as to obtain the same size Multiple image blocks of cervical cell pathology slides.
  • the size of the window used for segmentation may be 5120*5120.
  • the resulting cervical cell pathological slide image may appear too bright or too dark. Therefore, before detecting the cervical cytopathology slide image block, it is necessary to preprocess the cervical cytopathology slide image block, including performing Gamma transformation, correcting the image that is too bright or too dark, so that the image changes from the exposure intensity. The linear response becomes closer to the response felt by the human eye, thereby improving the visual effect of the image and reducing the difference in the digitized slide image scanned by different scanners.
  • V out V in ⁇
  • V represents the R, G, and B channels of the cervical cell pathology slide image block.
  • the resulting cervical cell pathological slide image may appear blurry. Therefore, before the cervical cytopathology slide image block is tested, it is necessary to preprocess the cervical cytopathology slide image block, including contrast enhancement processing, to increase the contrast of the image, thereby improving the visual effect of the image, and reducing the difference The difference of the digitized slide image scanned by the scanner.
  • V represents the R, G, and B channels of the image block.
  • R, G, and B channel data of the cervical cytopathology slide image block subtract the minimum value of all the data in the channel respectively, and then divide by the maximum value of all the data in the channel minus the minimum value , And finally multiply by 255 to restore it to the value range of [0, 255].
  • This can make the R, G, B three-channel data of the cervical cytopathology slide image block more evenly distributed from 0 to 255, improve the contrast of the image, and achieve the purpose of improving the subjective visual effect of the image and enhancing the details of the image.
  • each target cervical cell pathological slide image block into the convolutional neural network to obtain multiple levels of cellular characteristics of cervical cells.
  • the features of different levels reflect different image features.
  • the lower-level features reflect the shallower-level image features, such as edges, and the higher-level features reflect the deeper-level image features, such as object contours.
  • the multi-level cell features of the cervical cells are weighted and fused from high-level features to low-level features through the target pyramid network, and then low-level features to high-level features are combined.
  • the feature weighted fusion of the first-order features, and finally the cervical cell features are obtained.
  • the target pyramid network provided by the embodiments of the present application is an improvement on the existing pyramid network structure.
  • the feature weighted fusion of high-order features to low-order features there is an association relationship between the features generated after processing from top to bottom.
  • the features of the upper layer will affect the feature expression of the lower layer.
  • the features generated after the bottom-up processing also have an association relationship, and the features of the lower layer will affect the feature expression of the upper layer.
  • the high-level features to low-level features and low-level features to high-level features of cervical cells are subjected to two-way feature weighted fusion, which can fuse different levels of features together, enrich the expression ability of features, and better Extracting the detailed information of cells can improve the detection accuracy for ASC-US cells that are similar to normal cervical cells.
  • the weights corresponding to the cell features of multiple levels of cervical cells in the feature weighted fusion are obtained respectively.
  • the features of multiple levels are often simply averaged fusion.
  • the weights corresponding to the features of different levels may be different. That is to say, weighted fusion is performed in the feature fusion process.
  • the characteristics of the cervical cells therein are acquired, so as to determine the abnormal probability of the cervical cells in each image block.
  • the cervical cell characteristics are compared with those of normal cervical cells to determine the abnormal probability of cervical cells.
  • the cervical cells when the abnormality probability is not less than the preset abnormality probability threshold, the cervical cells can be considered to be abnormal, so that abnormal cervical cells in each image block can be detected. Combining the detection conditions in each target cervical cytopathological slide image block, the abnormal cervical cells in the cervical cytopathological slide image can be detected, and the automatic detection of abnormal cervical cells can be realized.
  • the cervical cells after determining that the cervical cells are abnormal cervical cells, obtain the position information of the cervical cells in the corresponding target cervical cytopathological slide image block to obtain the position of the abnormal cervical cells in the cervical cytopathological slide image .
  • the abnormal probability of cervical cells is acquired.
  • the cervical cell can be considered abnormal
  • the position information of the cervical cell in the corresponding target cervical cell pathological slide image block is obtained, and the position information and abnormal probability of the cervical cell are retained
  • the location information and abnormal probability of abnormal cervical cells in each image block can be detected.
  • the position of abnormal cervical cells in the cervical cytopathology slide image can be determined.
  • it can be displayed according to the position of the abnormal cervical cells in the cervical cytopathological slide image, and the corresponding abnormal probability can also be displayed.
  • the image is corrected by Gamma transformation and contrast enhancement, and the contrast of the image is increased at the same time, thereby improving the visual effect of the image, and Reduce the difference on the digitized slide image obtained by scanning with different scanners.
  • the high-level features to low-level features and low-level features to high-level features of cervical cells are subjected to two-way feature weighted fusion, which can fuse different levels of features together, enrich the expression ability of features, and thereby improve
  • the detailed information of cells can be extracted well, and the detection accuracy of ASC-US, which is similar to normal cervical cells, can be improved, thereby effectively improving the detection accuracy of abnormal cervical cells.
  • FIG. 5 is a schematic diagram of a device for detecting abnormal cervical cells according to an embodiment of the application.
  • a device for detecting abnormal cervical cells provided in an embodiment of the present application may include:
  • the obtaining module 501 is used to obtain cervical cell pathological slide images
  • the segmentation module 502 is configured to segment the cervical cytopathology slide image into multiple cervical cytopathology slide image blocks;
  • the preprocessing module 503 is used to preprocess each cervical cytopathological slide image block to obtain multiple target cervical cytopathological slide image blocks;
  • the processing module 504 is configured to process each target cervical cytopathological slide image block to obtain the cervical cell characteristics in each target cervical cytopathological slide image block, wherein the cervical cell characteristics are derived from cervical cells Two-way feature weighted fusion from high-level features to low-level features and low-level features to high-level features;
  • the determining module 505 is configured to determine the abnormal probability of cervical cells in the cervical cytopathological slide image according to the characteristics of the cervical cells in each target cervical cytopathological slide image block.
  • the acquiring module 501 is specifically configured to: acquire an initial cervical cytopathological slide image obtained by scanning the cervical cytopathological slide by a scanner; and extract the initial cervical cytopathological slide image The region of interest where the cervical cells are located to obtain the pathological slide image of the cervical cells.
  • the preprocessing module 503 is specifically configured to: perform gamma Gamma transformation on each cervical cytopathological slide image block to obtain multiple Gamma-transformed cervical cytopathological slide images Block; Contrast enhancement processing is performed on each Gamma-transformed cervical cytopathology slide image block to obtain the multiple target cervical cytopathology slide image blocks.
  • the processing module 504 is specifically configured to: input each target cervical cell pathological slide image block into a convolutional neural network to obtain multiple levels of cervical cell characteristics;
  • the multi-level cell features of the cervical cells are subjected to feature-weighted fusion from high-level features to low-level features, and then feature-weighted fusion from low-level features to high-level features is performed to obtain the cervical cell features.
  • the acquiring module 501 is further configured to obtain the weights corresponding to the cell features of the cervical cells at multiple levels in the feature-weighted fusion through the convolutional neural network.
  • the determining module 505 is further configured to: when the abnormal probability of the cervical cells is not less than a preset abnormal probability threshold, determine that the cervical cells are abnormal cervical cells.
  • the acquiring module 501 is further configured to: acquire the position information of the abnormal cervical cells in the corresponding target cervical cytopathological slide image block, so as to obtain the abnormal cervical cells in the The location of the cervical cell pathology slide image.
  • FIG. 6 is a schematic structural diagram of an electronic device in a hardware operating environment involved in an embodiment of the application.
  • the electronic device of the hardware operating environment involved in the embodiment of the present application may include:
  • the processor 601 is, for example, a CPU.
  • the memory 602 optionally, the memory may be a high-speed RAM memory, or a stable memory, such as a disk memory.
  • the communication interface 603 is used to implement connection and communication between the processor 601 and the memory 602.
  • FIG. 6 does not constitute a limitation on the electronic device, and may include more or fewer components than shown in the figure, or a combination of certain components, or different component arrangements. .
  • the memory 602 may include an operating system, a network communication module, and a cervical abnormal cell detection program.
  • the operating system is a program that manages and controls the hardware and software resources of electronic devices, and supports the operation of the cervical abnormal cell detection program and other software or programs.
  • the network communication module is used to implement communication between various components in the memory 602 and communication with other hardware and software in the electronic device.
  • the processor 601 is configured to execute the cervical abnormal cell detection program stored in the memory 602, and implement the following steps:
  • Each target cervical cytopathological slide image block is processed to obtain the cervical cell characteristics in each target cervical cytopathological slide image block, wherein the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • Another embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the following steps:
  • Each target cervical cytopathological slide image block is processed to obtain the cervical cell characteristics in each target cervical cytopathological slide image block, wherein the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • the cervical cell characteristics range from high-level features to low-level features of cervical cells.
  • the storage medium involved in this application such as a computer-readable storage medium, may be non-volatile or volatile.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Investigating Or Analysing Biological Materials (AREA)
  • Image Analysis (AREA)

Abstract

一种宫颈异常细胞检测装置及方法。该装置包括:获取模块(501),用于获取宫颈细胞病理玻片图像;分割模块(502),用于将宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;预处理模块(503),用于对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;处理模块(504),用于对每个目标宫颈细胞病理玻片图像块进行处理,得到宫颈细胞特征,其中,宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;确定模块(505),用于根据宫颈细胞特征,确定宫颈细胞病理玻片图像中的宫颈细胞的异常概率。该装置及方法有利于提高宫颈异常细胞的检测精度。

Description

一种宫颈异常细胞检测装置及方法
本申请要求于2020年9月30日提交中国专利局、申请号为202011062981.4,发明名称为“一种宫颈异常细胞检测装置及方法”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及医疗科技领域,尤其涉及一种宫颈异常细胞检测装置及方法。
背景技术
宫颈癌作为对女性生命健康造成严重危害的恶性肿瘤之一,发病率在女性恶性肿瘤中位居第二。并且,宫颈癌是目前唯一可以早发现并治愈的癌症,因此早期筛查对于宫颈癌的治疗相当关键。
宫颈液基细胞检查方法是目前最常用的宫颈癌筛查方法,但由于缺乏病理医生和细胞学检测设备,对宫颈癌的普查率很低,因此各种智能辅助筛查的设备系统渐渐出现。但发明人意识到,目前的宫颈癌智能辅助筛查系统大多对非典型鳞状上皮细胞(atypical squamous cells of undetermined significance,ASC-US)等异常细胞的检测精度较低。宫颈异常细胞中部分异常细胞的形变程度(核质比增大倍数以及细胞核异形程度等)较低,且有的是以单个小细胞而不是成团细胞的形式存在,因此在目前的宫颈癌智能辅助筛查过程中很难检测到宫颈异常细胞,导致宫颈异常细胞的检测精度较低。
发明内容
本申请提供了一种宫颈异常细胞检测装置及方法,有利于提高宫颈异常细胞的检测精度。
本申请第一方面提供了一种宫颈异常细胞检测装置,包括:
获取模块,用于获取宫颈细胞病理玻片图像;
分割模块,用于将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
预处理模块,用于对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
处理模块,用于对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
确定模块,用于根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
本申请第二方面提供了一种宫颈异常细胞检测方法,包括:
获取宫颈细胞病理玻片图像;
将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
本申请第三方面提供了一种电子设备,所述电子设备包括处理器、存储器、通信接口以及一个或多个程序,其中,所述一个或多个程序被存储在所述存储器中,并且被配置由所述处理器执行,以实现以下方法:
获取宫颈细胞病理玻片图像;
将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
本申请第四方面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以实现以下方法:
获取宫颈细胞病理玻片图像;
将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
本申请在获取宫颈细胞特征时,将宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征进行双向特征加权融合,能够将不同层次的特征融合在一起,丰富特征的表达能力,从而更好地提取细胞的细节信息,对于与宫颈正常细胞相似的ASC-US这一类细胞能够提高检测精度,即可以有效提高宫颈异常细胞的检测精度。
附图说明
为了更清楚地说明本申请实施例中的技术方案,下面将对实施例中所需使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1为特征金字塔网络的示意图;
图2为本申请实施例提供的一种宫颈异常细胞检测方法的流程示意图;
图3为本申请实施例提供的目标金字塔网络中双向特征融合的示意图;
图4为本申请实施例提供的另一种宫颈异常细胞检测方法的流程示意图;
图5为本申请实施例提供的一种宫颈异常细胞检测装置的示意图;
图6为本申请实施例涉及的硬件运行环境的电子设备结构示意图。
具体实施方式
本申请实施例提供的宫颈异常细胞检测装置及方法,有利于提高宫颈异常细胞的检测精度。
为了使本技术领域的人员更好地理解本申请方案,下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本申请一部分的实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都应当属于本申请保护的范围。
本申请实施例中涉及的“至少一个”是指一个或者多个,“多个”是指两个或两个以上。“和/或”,描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B的情况,其中A,B可以是单数或者复数。字符“/”一般表示前后关联对象是一种“或”的关系。“以下至少一项(个)”或其类似表达,是指的这些项中的任意组合,包括单项(个)或复数项(个)的任意组合。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等是用于区别不同对象,而不是用于描述特定顺序、时序、优先级或者重要程度。例如, 第一信息和第二信息,只是为了区分不同的信息,而并不是表示这两种信息的内容、优先级、发送顺序或者重要程度的不同。此外,术语“包括”和“具有”以及它们任何变形,意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、系统、产品或设备没有限定于已列出的步骤或单元,而是可选地还包括没有列出的步骤或单元,或可选地还包括对于这些过程、方法、产品或设备固有的其它步骤或单元。
本申请的技术方案可应用于人工智能、智慧城市、数字医疗和/或区块链技术领域,如可具体涉及神经网络技术,以实现异常检测。可选的,本申请涉及的数据如图像和/或异常概率等可存储于数据库中,或者可以存储于区块链中,比如通过区块链分布式存储,本申请不做限定。
为了便于理解本申请,首先对本申请涉及的概念进行解释。
宫颈癌作为对女性生命健康造成严重危害的恶性肿瘤之一,发病率在女性恶性肿瘤中位居第二。常见的宫颈细胞病变包括:非典型鳞状上皮细胞-意义不明确(atypical squamous cells of undetermined significance,ASC-US)、低度鳞状上皮病变(low-grade squamous intraepithelial lesion,LSIL)、非典型鳞状上皮细胞-不排除高级鳞状上皮病变(atypical squamous cells cannot exclude high-gradesquamous intraepithelial lesion,ASC-H)、高度鳞状上皮病变(high-grade squamous intraepithelial lesion,HSIL)、非典型腺上皮细胞(atypical glandular cells,AGC)等等。
宫颈液基细胞检查方法是目前最常用的宫颈癌筛查方法,目前的宫颈癌智能辅助筛查系统大多对ASC-US等异常细胞的检测精度较低。宫颈异常细胞中部分异常细胞的形变程度(核质比增大倍数以及细胞核异形程度等)较低,且有的是以单个小细胞而不是成团细胞的形式存在,因此在目前的宫颈癌智能辅助筛查过程中很难检测到宫颈异常细胞,导致宫颈异常细胞的检测精度较低。
特征金字塔网络(feature pyramid networks,FPN):FPN是一种利用卷积神经网络(convolutional neural networks,CNN)来高效提取图片中各层次特征的方法。FPN通过利用常规CNN模型内部从底至上各个层次对同一图像不同层次的特征表达结构,提出了一种可有效在单一图像视图下生成对其的多层次特征表达的方法。
参见图1,图1为特征金字塔网络的示意图。其中,如图1所示,FPN会使用CNN网络中每一层的信息来生成最后的特征组合,使得最终输出的特征可以更好的表示出输入图像各个层次的信息。具体的,FPN的基本过程包括:自下而上的不同层次特征生成、自上而下的特征融合、CNN网络层特征与最终输出的各层次特征之间的关联表达。也就是说,自上至下处理后生成的特征之间也有关联关系,上层的特征会影响下层的特征表达,最终所有的特征一起用来作为下一步的目标检测或者类别分析等任务的输入。
如上介绍了本申请的背景技术,下面介绍本申请实施例的技术特征。
参见图2,图2为本申请实施例提供的一种宫颈异常细胞检测方法的流程示意图。其中,如图2所示,本申请实施例提供的一种宫颈异常细胞检测方法可以包括:
201、获取宫颈细胞病理玻片图像。
可选的,获取通过扫描仪对宫颈细胞病理玻片进行扫描得到的初始宫颈细胞病理玻片图像,提取该初始宫颈细胞病理玻片图像中宫颈细胞所在的感兴趣区域,以得到宫颈细胞病理玻片图像。
具体的,在进行宫颈异常细胞检测时,首先获取宫颈细胞病理玻片,然后对该宫颈细胞病理玻片进行处理,得到玻片数字化信息,也即初始宫颈细胞病理玻片图像。在一种可能的实施方式中,可以采用扫描仪对宫颈细胞病理玻片进行扫描,得到初始宫颈细胞病理玻片图像,其中,该初始宫颈细胞病理玻片图像包括多种格式,例如,svs格式、kfb格式、ndpi格式、tif格式、sdpc格式等。
获取初始宫颈细胞病理玻片图像后,采用霍夫变换在该初始宫颈细胞病理玻片图像中寻找宫颈细胞所在的前景区域,从而提取宫颈细胞所在的感兴趣区域(region of interest,ROI),得到宫颈细胞病理玻片图像。通过提取宫颈细胞所在的感兴趣区域,可以去除外围的不含有效信息的背景区域,从而在之后的计算过程中节省计算时间。
202、将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块。
获取宫颈细胞病理玻片图像后,将该宫颈细胞病理玻片分割为多个宫颈细胞病理玻片图像块。可选的,该多个宫颈细胞病理玻片图像块的面积可以是一样大的,也可以是不一样大的。将该宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块,在之后的计算过程中可以分别对每个图像块进行处理,从而提高处理速度。
203、对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块。
可选的,对每个宫颈细胞病理玻片图像块进行伽马Gamma变换,得到多个Gamma变换后的宫颈细胞病理玻片图像块;对每个Gamma变换后的宫颈细胞病理玻片图像块进行对比度增强处理,得到多个目标宫颈细胞病理玻片图像块。
具体的,在对宫颈细胞病理玻片进行扫描时,采用不同扫描仪扫描得到的数字化玻片图像上会有一定差异,不利于进行统一自动检测。除此之外,在扫描时,受外界因素例如光线等影响,得到的宫颈细胞病理玻片图像可能会出现过亮或者过暗的情况。因此,在对宫颈细胞病理玻片图像块进行检测之前,要对宫颈细胞病理玻片图像块进行预处理,包括进行Gamma变换和对比度增强处理,对过亮或过暗的图像进行校正,同时,提高图像的对比度,改善图像的视觉效果,并降低通过不同扫描仪扫描得到的数字化玻片图像的差异。
204、对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到。
可选的,将每个目标宫颈细胞病理玻片图像块输入卷积神经网络,得到宫颈细胞的多个层次的细胞特征;对该宫颈细胞的多个层次的细胞特征进行高阶特征到低阶特征的特征加权融合,再进行低阶特征到高阶特征的特征加权融合,得到宫颈细胞特征。
具体的,在进行宫颈异常细胞检测之前,首先通过目标金字塔网络提取宫颈细胞特征,根据宫颈细胞特征,检测是否异常。在特征提取过程中,首先将目标宫颈细胞病理玻片图像块输入目标金字塔网络结构中的卷积神经网络,得到宫颈细胞的多个层次的细胞特征,其中,不同层次的特征反映的图像特征不同,较低层次的特征反映了较浅层次的图像特征,例如边缘等,较高层次的特征反映了较深层次的图像特征例如物体轮廓等。
获取宫颈细胞的多个层次的细胞特征后,通过目标金字塔网络对该宫颈细胞的多个层次的细胞特征进行高阶特征到低阶特征的特征加权融合,再进行低阶特征到高阶特征的特征加权融合,最终得到宫颈细胞特征。本申请实施例提供的目标金字塔网络在现有金字塔网络结构上进行了改进,参见图3,图3为本申请实施例提供的目标金字塔网络中双向特征融合的示意图,如图3所示,在进行高阶特征到低阶特征的特征加权融合时,自上至下处理后生成的特征之间有关联关系,上层的特征会影响下层的特征表达,在进行低阶特征到高阶特征的特征加权融合时,自下而上处理后生成的特征之间也有关联关系,下层的特征会影响上层的特征表达。
通过目标金字塔网络,将宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征进行双向特征加权融合,能够将不同层次的特征融合在一起,丰富特征的表达能力,从而更好地提取细胞的细节信息,对于与宫颈正常细胞相似的ASC-US这一类细胞能够提高检测精度。
可选的,通过卷积神经网络中的权重层,得到宫颈细胞的多个层次的细胞特征在特征 加权融合中分别对应的权重。具体的,现有金字塔网络中,多个层次的特征往往是简单的平均融合,在本申请实施例改进后的目标金字塔网络中,进行特征融合时,不同层次的特征对应的权重可能不同,也就是说在特征融合过程中进行加权融合。
205、根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
可选的,根据宫颈细胞特征,确定宫颈细胞的异常概率;当该宫颈细胞的异常概率不小于预设异常概率阈值时,确定该宫颈细胞为宫颈异常细胞。
具体的,对于每个目标宫颈细胞病理玻片图像块,获取其中的宫颈细胞特征,从而确定每个图像块中宫颈细胞的异常概率。当异常概率不小于预设异常概率阈值时,可以认为该宫颈细胞异常,这样可以检测出每个图像块中异常的宫颈细胞。结合每个目标宫颈细胞病理玻片图像块中的检测情况,可以检测出宫颈细胞病理玻片图像中的宫颈异常细胞,实现宫颈异常细胞的自动检测。
可选的,在确定该宫颈细胞为宫颈异常细胞之后,获取该宫颈异常细胞在对应的目标宫颈细胞病理玻片图像块中的位置信息,以得到宫颈异常细胞在宫颈细胞病理玻片图像中的位置。
具体的,对于每个目标宫颈细胞病理玻片图像块,获取宫颈细胞的异常概率。当异常概率不小于预设异常概率阈值时,可以认为该宫颈细胞异常,获取该宫颈细胞在对应的目标宫颈细胞病理玻片图像块中的位置信息,并且保留该宫颈细胞的位置信息和异常概率,这样可以检测出每个图像块中宫颈异常细胞的位置信息和异常概率。结合每个目标宫颈细胞病理玻片图像块中的检测情况,可以确定宫颈异常细胞在宫颈细胞病理玻片图像中的位置。在显示时,可以根据宫颈异常细胞在宫颈细胞病理玻片图像中的位置进行显示,并且还可以显示对应的异常概率。
可以看出,通过本申请实施例提供的宫颈异常细胞检测方法,在获取宫颈细胞特征时,将宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征进行双向特征加权融合,能够将不同层次的特征融合在一起,丰富特征的表达能力,从而更好地提取细胞的细节信息,对于与宫颈正常细胞相似的ASC-US这一类细胞能够提高检测精度,从而有效提高宫颈异常细胞的检测精度。
在本申请的一个实施方式中,本申请的方案还可以应用到智慧医疗领域。比如,接收医生输入的宫颈细胞病理玻片图像,通过本申请的宫颈异常细胞检测方法,确定宫颈细胞病理玻片图像中的宫颈细胞的异常概率。由于通过本申请的宫颈异常细胞检测方法,可以较为精准的确定宫颈细胞的异常概率,这样可以为医生的诊断提供较为精准的判断依据,从而提高医生的诊断效率和精准度。
参见图4,图4为本申请实施例提供的另一种宫颈异常细胞检测方法的流程示意图。其中,如图4所示,本申请实施例提供的另一种宫颈异常细胞检测方法可以包括:
401、获取宫颈细胞病理玻片图像。
可选的,获取通过扫描仪对宫颈细胞病理玻片进行扫描得到的初始宫颈细胞病理玻片图像,提取该初始宫颈细胞病理玻片图像中宫颈细胞所在的感兴趣区域,以得到宫颈细胞病理玻片图像。
具体的,在进行宫颈异常细胞检测时,首先获取宫颈细胞病理玻片,然后对该宫颈细胞病理玻片进行处理,得到玻片数字化信息,也即初始宫颈细胞病理玻片图像。在一种可能的实施方式中,可以采用扫描仪对宫颈细胞病理玻片进行扫描,得到初始宫颈细胞病理玻片图像,其中,该初始宫颈细胞病理玻片图像包括多种格式,例如,svs格式、kfb格式、ndpi格式、tif格式、sdpc格式等。
获取初始宫颈细胞病理玻片图像后,对该初始宫颈细胞病理玻片图像进行放大,其中, 放大倍数可以是预先设定。经过放大处理后,对放大后的初始宫颈细胞病理玻片图像的黑白图像先后进行膨胀操作和腐蚀操作。
采用霍夫变换在经过膨胀、腐蚀操作处理的初始宫颈细胞病理玻片图像中寻找宫颈细胞所在的前景区域,从而提取宫颈细胞所在的感兴趣区域(region of interest,ROI),得到宫颈细胞病理玻片图像。通过提取宫颈细胞所在的感兴趣区域,可以去除外围的不含有效信息的背景区域,从而在之后的计算过程中节省计算时间。
402、将宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块。
获取宫颈细胞病理玻片图像后,将该宫颈细胞病理玻片分割为多个宫颈细胞病理玻片图像块。可选的,该多个宫颈细胞病理玻片图像块的面积可以是一样大的,也可以是不一样大的。将该宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块,在之后的计算过程中可以分别对每个图像块进行处理,从而提高处理速度。
在一种可能的实施方式中,该多个宫颈细胞病理玻片图像块的面积等大,在分割过程中,采用一定大小的划窗对宫颈细胞病理玻片图像进行分割,从而得到等大的多个宫颈细胞病理玻片图像块。例如,用于分割的划窗的大小可以是5120*5120。
403、对每个宫颈细胞病理玻片图像块进行伽马Gamma变换,得到多个Gamma变换后的宫颈细胞病理玻片图像块。
具体的,在对宫颈细胞病理玻片进行扫描时,采用不同扫描仪扫描得到的数字化玻片图像上会有一定差异,不利于进行统一自动检测。除此之外,在扫描时,受外界因素例如光线等影响,得到的宫颈细胞病理玻片图像可能会出现过亮或者过暗的情况。因此,在对宫颈细胞病理玻片图像块进行检测之前,要对宫颈细胞病理玻片图像块进行预处理,包括进行Gamma变换,对过亮或过暗的图像进行校正,使得图像从曝光强度的线性响应变得更接近人眼感受的响应,从而改善图像的视觉效果,并降低通过不同扫描仪扫描得到的数字化玻片图像的差异。
对宫颈细胞病理玻片图像块进行Gamma变换的公式如下所示:
V out=V in γ
其中,V分别代表宫颈细胞病理玻片图像块的R、G、B三通道。
404、对每个Gamma变换后的宫颈细胞病理玻片图像块进行对比度增强处理,得到多个目标宫颈细胞病理玻片图像块。
具体的,在对宫颈细胞病理玻片进行扫描时,采用不同扫描仪扫描得到的数字化玻片图像上会有一定差异,不利于进行统一自动检测。除此之外,在扫描时,受外界因素例如光线等影响,得到的宫颈细胞病理玻片图像可能会出现较模糊的情况。因此,在对宫颈细胞病理玻片图像块进行检测之前,要对宫颈细胞病理玻片图像块进行预处理,包括进行对比度增强处理,提高图像的对比度,从而改善图像的视觉效果,并降低通过不同扫描仪扫描得到的数字化玻片图像的差异。
对每个Gamma变换后的宫颈细胞病理玻片图像块进行对比度增强处理的公式如下所示:
Figure PCTCN2020132474-appb-000001
其中,V分别代表图像块的R、G、B三通道。也就是说,对于宫颈细胞病理玻片图像块的R、G、B通道数据,分别对其进行减去该通道所有数据中的最小值,然后除以该通道所有数据中的最大值减最小值,最后再乘以255将其恢复到[0,255]的取值范围的操作。这样可以使得宫颈细胞病理玻片图像块的R、G、B三通道数据在0-255上分布更均匀,提高 了图像的对比度,达到改善图像主观视觉效果以及增强图像细节的目的。
405、将每个目标宫颈细胞病理玻片图像块输入卷积神经网络,得到宫颈细胞的多个层次的细胞特征。
具体的,在进行宫颈异常细胞检测之前,首先通过目标金字塔网络提取宫颈细胞特征,根据宫颈细胞特征,检测是否异常。在特征提取过程中,首先将目标宫颈细胞病理玻片图像块输入目标金字塔网络结构中的卷积神经网络,得到宫颈细胞的多个层次的细胞特征,其中,不同层次的特征反映的图像特征不同,较低层次的特征反映了较浅层次的图像特征,例如边缘等,较高层次的特征反映了较深层次的图像特征例如物体轮廓等。
406、对该宫颈细胞的多个层次的细胞特征进行高阶特征到低阶特征的特征加权融合,再进行低阶特征到高阶特征的特征加权融合,得到宫颈细胞特征。
具体的,获取宫颈细胞的多个层次的细胞特征后,通过目标金字塔网络对该宫颈细胞的多个层次的细胞特征进行高阶特征到低阶特征的特征加权融合,再进行低阶特征到高阶特征的特征加权融合,最终得到宫颈细胞特征。
本申请实施例提供的目标金字塔网络在现有金字塔网络结构上进行了改进,在进行高阶特征到低阶特征的特征加权融合时,自上至下处理后生成的特征之间有关联关系,上层的特征会影响下层的特征表达,在进行低阶特征到高阶特征的特征加权融合时,自下而上处理后生成的特征之间也有关联关系,下层的特征会影响上层的特征表达。
通过目标金字塔网络,将宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征进行双向特征加权融合,能够将不同层次的特征融合在一起,丰富特征的表达能力,从而更好地提取细胞的细节信息,对于与宫颈正常细胞相似的ASC-US这一类细胞能够提高检测精度。
可选的,通过卷积神经网络中的权重层,得到宫颈细胞的多个层次的细胞特征在特征加权融合中分别对应的权重。具体的,现有金字塔网络中,多个层次的特征往往是简单的平均融合,在本申请实施例改进后的目标金字塔网络中,进行特征融合时,不同层次的特征对应的权重可能不同,也就是说在特征融合过程中进行加权融合。
407、根据宫颈细胞特征,确定宫颈细胞的异常概率。
具体的,对于每个目标宫颈细胞病理玻片图像块,获取其中的宫颈细胞特征,从而确定每个图像块中宫颈细胞的异常概率。
在一种可能的实施方式中,将宫颈细胞特征与宫颈正常细胞的细胞特征进行对比,从而确定宫颈细胞的异常概率。
408、当该宫颈细胞的异常概率不小于预设异常概率阈值时,确定该宫颈细胞为宫颈异常细胞。
具体的,当异常概率不小于预设异常概率阈值时,可认为该宫颈细胞异常,这样可以检测出每个图像块中异常的宫颈细胞。结合每个目标宫颈细胞病理玻片图像块中的检测情况,可以检测出宫颈细胞病理玻片图像中的宫颈异常细胞,实现宫颈异常细胞的自动检测。
可选的,在确定该宫颈细胞为宫颈异常细胞之后,获取该宫颈细胞在对应的目标宫颈细胞病理玻片图像块中的位置信息,以得到宫颈异常细胞在宫颈细胞病理玻片图像中的位置。
具体的,对于每个目标宫颈细胞病理玻片图像块,获取宫颈细胞的异常概率。当异常概率不小于预设异常概率阈值时,可以认为该宫颈细胞异常,获取该宫颈细胞在对应的目标宫颈细胞病理玻片图像块中的位置信息,并且保留该宫颈细胞的位置信息和异常概率,这样可以检测出每个图像块中宫颈异常细胞的位置信息和异常概率。结合每个目标宫颈细胞病理玻片图像块中的检测情况,可以确定宫颈异常细胞在宫颈细胞病理玻片图像中的位置。在显示时,可以根据宫颈异常细胞在宫颈细胞病理玻片图像中的位置进行显示,并且 还可以显示对应的异常概率。
可以看出,通过本申请实施例提供的宫颈异常细胞检测方法,在对图像进行特征提取之前,先通过Gamma变换以及对比度增强对图像进行矫正,同时提高图像的对比度,从而改善图像视觉效果,并降低通过不同扫描仪进行扫描得到的数字化玻片图像上的差异。在获取宫颈细胞特征时,将宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征进行双向特征加权融合,能够将不同层次的特征融合在一起,丰富特征的表达能力,从而更好地提取细胞的细节信息,对于与宫颈正常细胞相似的ASC-US这一类细胞能够提高检测精度,从而有效提高宫颈异常细胞的检测精度。
参见图5,图5为本申请实施例提供的一种宫颈异常细胞检测装置的示意图。其中,如图5所示,本申请实施例提供的一种宫颈异常细胞检测装置可以包括:
获取模块501,用于获取宫颈细胞病理玻片图像;
分割模块502,用于将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
预处理模块503,用于对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
处理模块504,用于对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
确定模块505,用于根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
在一种可能的实施方式中,所述获取模块501具体用于:获取通过扫描仪对宫颈细胞病理玻片进行扫描得到的初始宫颈细胞病理玻片图像;提取所述初始宫颈细胞病理玻片图像中宫颈细胞所在的感兴趣区域,以得到所述宫颈细胞病理玻片图像。
在一种可能的实施方式中,所述预处理模块503具体用于:对所述每个宫颈细胞病理玻片图像块进行伽马Gamma变换,得到多个Gamma变换后的宫颈细胞病理玻片图像块;对每个Gamma变换后的宫颈细胞病理玻片图像块进行对比度增强处理,得到所述多个目标宫颈细胞病理玻片图像块。
在一种可能的实施方式中,所述处理模块504具体用于:将所述每个目标宫颈细胞病理玻片图像块输入卷积神经网络,得到宫颈细胞的多个层次的细胞特征;对所述宫颈细胞的多个层次的细胞特征进行高阶特征到低阶特征的特征加权融合,再进行低阶特征到高阶特征的特征加权融合,得到所述宫颈细胞特征。
在一种可能的实施方式中,所述获取模块501还用于:通过所述卷积神经网络,得到所述宫颈细胞的多个层次的细胞特征在特征加权融合中分别对应的权重。
在一种可能的实施方式中,所述确定模块505还用于:当所述宫颈细胞的异常概率不小于预设异常概率阈值时,确定所述宫颈细胞为宫颈异常细胞。
在一种可能的实施方式中,所述获取模块501还用于:获取所述宫颈异常细胞在对应的目标宫颈细胞病理玻片图像块中的位置信息,以得到所述宫颈异常细胞在所述宫颈细胞病理玻片图像中的位置。
本申请实施例中宫颈异常细胞检测装置的具体实施可参见上述宫颈异常细胞检测方法的各实施例,在此不做赘述。
参见图6,图6为本申请的实施例涉及的硬件运行环境的电子设备结构示意图。其中,如图6所示,本申请的实施例涉及的硬件运行环境的电子设备可以包括:
处理器601,例如CPU。
存储器602,可选的,存储器可以为高速RAM存储器,也可以是稳定的存储器,例如 磁盘存储器。
通信接口603,用于实现处理器601和存储器602之间的连接通信。
本领域技术人员可以理解,图6中示出的电子设备的结构并不构成对电子设备的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
如图6所示,存储器602中可以包括操作系统、网络通信模块以及宫颈异常细胞检测程序。操作系统是管理和控制电子设备硬件和软件资源的程序,支持宫颈异常细胞检测程序以及其他软件或程序的运行。网络通信模块用于实现存储器602内部各组件之间的通信,以及与电子设备中其他硬件和软件之间通信。
在图6所示的电子设备中,处理器601用于执行存储器602中存储的宫颈异常细胞检测程序,实现以下步骤:
获取宫颈细胞病理玻片图像;
将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
本申请实施例中电子设备的具体实施可参见上述宫颈异常细胞检测方法的各实施例,在此不做赘述。
本申请的另一个实施例提供了一种计算机可读存储介质,计算机可读存储介质存储有计算机程序,计算机程序被处理器执行以实现以下步骤:
获取宫颈细胞病理玻片图像;
将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
本申请实施例中计算机可读存储介质的具体实施可参见上述宫颈异常细胞检测方法的各实施例,在此不做赘述。
可选的,本申请涉及的存储介质如计算机可读存储介质可以是非易失性的,也可以是易失性的。
还需要说明的是,对于前述的各方法实施例,为了简单描述,故将其都表述为一系列的动作组合,但是本领域技术人员应该知悉,本申请并不受所描述的动作顺序的限制,因为依据本申请,某些步骤可以采用其他顺序或者同时进行。其次,本领域技术人员也应该知悉,说明书中所描述的实施例均属于优选实施例,所涉及的动作和模块并不一定是本申请所必须的。在上述实施例中,对各个实施例的描述都各有侧重,某个实施例中没有详述的部分,可以参见其他实施例的相关描述。
以上所述,以上实施例仅用以说明本申请的技术方案,而非对其限制;尽管参照前述实施例对本申请进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本申请各实施例技术方案的范围。

Claims (20)

  1. 一种宫颈异常细胞检测装置,包括:
    获取模块,用于获取宫颈细胞病理玻片图像;
    分割模块,用于将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
    预处理模块,用于对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
    处理模块,用于对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
    确定模块,用于根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
  2. 根据权利要求1所述的装置,其中,所述获取模块具体用于:
    获取通过扫描仪对宫颈细胞病理玻片进行扫描得到的初始宫颈细胞病理玻片图像;
    提取所述初始宫颈细胞病理玻片图像中宫颈细胞所在的感兴趣区域,以得到所述宫颈细胞病理玻片图像。
  3. 根据权利要求1或2所述的装置,其中,所述预处理模块具体用于:
    对所述每个宫颈细胞病理玻片图像块进行伽马Gamma变换,得到多个Gamma变换后的宫颈细胞病理玻片图像块;
    对每个Gamma变换后的宫颈细胞病理玻片图像块进行对比度增强处理,得到所述多个目标宫颈细胞病理玻片图像块。
  4. 根据权利要求1所述的装置,其中,所述处理模块具体用于:
    将所述每个目标宫颈细胞病理玻片图像块输入卷积神经网络,得到宫颈细胞的多个层次的细胞特征;
    对所述宫颈细胞的多个层次的细胞特征进行高阶特征到低阶特征的特征加权融合,再进行低阶特征到高阶特征的特征加权融合,得到所述宫颈细胞特征。
  5. 根据权利要求4所述的装置,其中,所述获取模块还用于:
    通过所述卷积神经网络,得到所述宫颈细胞的多个层次的细胞特征在特征加权融合中分别对应的权重。
  6. 根据权利要求1所述的装置,其中,所述确定模块还用于:
    当所述宫颈细胞的异常概率不小于预设异常概率阈值时,确定所述宫颈细胞为宫颈异常细胞。
  7. 根据权利要求6所述的装置,其中,所述获取模块还用于:
    获取所述宫颈异常细胞在对应的目标宫颈细胞病理玻片图像块中的位置信息,以得到所述宫颈异常细胞在所述宫颈细胞病理玻片图像中的位置。
  8. 一种宫颈异常细胞检测方法,包括:
    获取宫颈细胞病理玻片图像;
    将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
    对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
    对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
    根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
  9. 一种电子设备,所述电子设备包括处理器、存储器、通信接口以及一个或多个程序, 其中,所述一个或多个程序被存储在所述存储器中,并且被配置由所述处理器执行,以实现以下方法:
    获取宫颈细胞病理玻片图像;
    将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
    对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
    对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
    根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
  10. 根据权利要求9所述的电子设备,其中,所述获取宫颈细胞病理玻片图像时,具体实现:
    获取通过扫描仪对宫颈细胞病理玻片进行扫描得到的初始宫颈细胞病理玻片图像;
    提取所述初始宫颈细胞病理玻片图像中宫颈细胞所在的感兴趣区域,以得到所述宫颈细胞病理玻片图像。
  11. 根据权利要求9或10所述的电子设备,其中,所述对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块时,具体实现:
    对所述每个宫颈细胞病理玻片图像块进行伽马Gamma变换,得到多个Gamma变换后的宫颈细胞病理玻片图像块;
    对每个Gamma变换后的宫颈细胞病理玻片图像块进行对比度增强处理,得到所述多个目标宫颈细胞病理玻片图像块。
  12. 根据权利要求9所述的电子设备,其中,所述对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征时,具体实现:
    将所述每个目标宫颈细胞病理玻片图像块输入卷积神经网络,得到宫颈细胞的多个层次的细胞特征;
    对所述宫颈细胞的多个层次的细胞特征进行高阶特征到低阶特征的特征加权融合,再进行低阶特征到高阶特征的特征加权融合,得到所述宫颈细胞特征。
  13. 根据权利要求12所述的电子设备,其中,所述处理器还用于执行:
    通过所述卷积神经网络,得到所述宫颈细胞的多个层次的细胞特征在特征加权融合中分别对应的权重。
  14. 根据权利要求9所述的电子设备,其中,所述处理器还用于执行:
    当所述宫颈细胞的异常概率不小于预设异常概率阈值时,确定所述宫颈细胞为宫颈异常细胞。
  15. 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以实现以下方法:
    获取宫颈细胞病理玻片图像;
    将所述宫颈细胞病理玻片图像分割为多个宫颈细胞病理玻片图像块;
    对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块;
    对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,其中,所述宫颈细胞特征由宫颈细胞的高阶特征到低阶特征以及低阶特征到高阶特征的双向特征加权融合得到;
    根据所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征,确定所述宫颈细胞病理玻片图像中的宫颈细胞的异常概率。
  16. 根据权利要求15所述的计算机可读存储介质,其中,所述获取宫颈细胞病理玻片 图像时,具体实现:
    获取通过扫描仪对宫颈细胞病理玻片进行扫描得到的初始宫颈细胞病理玻片图像;
    提取所述初始宫颈细胞病理玻片图像中宫颈细胞所在的感兴趣区域,以得到所述宫颈细胞病理玻片图像。
  17. 根据权利要求15或16所述的计算机可读存储介质,其中,所述对每个宫颈细胞病理玻片图像块进行预处理,得到多个目标宫颈细胞病理玻片图像块时,具体实现:
    对所述每个宫颈细胞病理玻片图像块进行伽马Gamma变换,得到多个Gamma变换后的宫颈细胞病理玻片图像块;
    对每个Gamma变换后的宫颈细胞病理玻片图像块进行对比度增强处理,得到所述多个目标宫颈细胞病理玻片图像块。
  18. 根据权利要求15所述的计算机可读存储介质,其中,所述对每个目标宫颈细胞病理玻片图像块进行处理,得到所述每个目标宫颈细胞病理玻片图像块中的宫颈细胞特征时,具体实现:
    将所述每个目标宫颈细胞病理玻片图像块输入卷积神经网络,得到宫颈细胞的多个层次的细胞特征;
    对所述宫颈细胞的多个层次的细胞特征进行高阶特征到低阶特征的特征加权融合,再进行低阶特征到高阶特征的特征加权融合,得到所述宫颈细胞特征。
  19. 根据权利要求18所述的计算机可读存储介质,其中,所述计算机程序被处理器执行时还用于实现:
    通过所述卷积神经网络,得到所述宫颈细胞的多个层次的细胞特征在特征加权融合中分别对应的权重。
  20. 根据权利要求15所述的计算机可读存储介质,其中,所述计算机程序被处理器执行时还用于实现:
    当所述宫颈细胞的异常概率不小于预设异常概率阈值时,确定所述宫颈细胞为宫颈异常细胞。
PCT/CN2020/132474 2020-09-30 2020-11-27 一种宫颈异常细胞检测装置及方法 WO2021139447A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011062981.4 2020-09-30
CN202011062981.4A CN112184733A (zh) 2020-09-30 2020-09-30 一种宫颈异常细胞检测装置及方法

Publications (1)

Publication Number Publication Date
WO2021139447A1 true WO2021139447A1 (zh) 2021-07-15

Family

ID=73948267

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/132474 WO2021139447A1 (zh) 2020-09-30 2020-11-27 一种宫颈异常细胞检测装置及方法

Country Status (2)

Country Link
CN (1) CN112184733A (zh)
WO (1) WO2021139447A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114399494A (zh) * 2022-01-14 2022-04-26 平安科技(深圳)有限公司 异常细胞检测分割方法、装置、设备及存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116630341B (zh) * 2023-04-20 2023-12-01 广州方信医疗技术有限公司 病理数字化玻片的处理方法及系统

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108510482A (zh) * 2018-03-22 2018-09-07 姚书忠 基于阴道镜图像的宫颈癌检测方法、装置、设备及介质
CN109118485A (zh) * 2018-08-13 2019-01-01 复旦大学 基于多任务神经网络的消化道内镜图像分类及早癌检测系统
CN109272492A (zh) * 2018-08-24 2019-01-25 深思考人工智能机器人科技(北京)有限公司 一种细胞病理涂片的处理方法及系统
CN110705583A (zh) * 2019-08-15 2020-01-17 平安科技(深圳)有限公司 细胞检测模型训练方法、装置、计算机设备及存储介质
CN110992303A (zh) * 2019-10-29 2020-04-10 平安科技(深圳)有限公司 一种异常细胞筛选方法、装置、电子设备和存储介质
WO2020078252A1 (en) * 2018-10-16 2020-04-23 The Chinese University Of Hong Kong Method, apparatus and system for automatic diagnosis
CN111340059A (zh) * 2018-12-19 2020-06-26 北京嘀嘀无限科技发展有限公司 图像特征提取方法、装置、电子设备及存储介质
CN111340044A (zh) * 2018-12-19 2020-06-26 北京嘀嘀无限科技发展有限公司 图像处理方法、装置、电子设备及存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110415212A (zh) * 2019-06-18 2019-11-05 平安科技(深圳)有限公司 异常细胞检测方法、装置及计算机可读存储介质
CN110598561A (zh) * 2019-08-15 2019-12-20 平安科技(深圳)有限公司 基于机器学习的细胞玻片分析方法、装置及存储介质
CN110648322B (zh) * 2019-09-25 2023-08-15 杭州智团信息技术有限公司 一种子宫颈异常细胞检测方法及系统

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108510482A (zh) * 2018-03-22 2018-09-07 姚书忠 基于阴道镜图像的宫颈癌检测方法、装置、设备及介质
CN109118485A (zh) * 2018-08-13 2019-01-01 复旦大学 基于多任务神经网络的消化道内镜图像分类及早癌检测系统
CN109272492A (zh) * 2018-08-24 2019-01-25 深思考人工智能机器人科技(北京)有限公司 一种细胞病理涂片的处理方法及系统
WO2020078252A1 (en) * 2018-10-16 2020-04-23 The Chinese University Of Hong Kong Method, apparatus and system for automatic diagnosis
CN111340059A (zh) * 2018-12-19 2020-06-26 北京嘀嘀无限科技发展有限公司 图像特征提取方法、装置、电子设备及存储介质
CN111340044A (zh) * 2018-12-19 2020-06-26 北京嘀嘀无限科技发展有限公司 图像处理方法、装置、电子设备及存储介质
CN110705583A (zh) * 2019-08-15 2020-01-17 平安科技(深圳)有限公司 细胞检测模型训练方法、装置、计算机设备及存储介质
CN110992303A (zh) * 2019-10-29 2020-04-10 平安科技(深圳)有限公司 一种异常细胞筛选方法、装置、电子设备和存储介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114399494A (zh) * 2022-01-14 2022-04-26 平安科技(深圳)有限公司 异常细胞检测分割方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN112184733A (zh) 2021-01-05

Similar Documents

Publication Publication Date Title
WO2021139258A1 (zh) 基于图像识别的细胞识别计数方法、装置和计算机设备
CN111986150B (zh) 一种数字病理图像的交互式标注精细化方法
CN103164692B (zh) 一种基于计算机视觉的特种车辆仪表自动识别系统及方法
CN111862044B (zh) 超声图像处理方法、装置、计算机设备和存储介质
TW202014984A (zh) 一種圖像處理方法、電子設備及存儲介質
WO2021189771A1 (zh) 玻片数字化信息质量检测方法、装置、设备及介质
US11501431B2 (en) Image processing method and apparatus and neural network model training method
CN101901342A (zh) 影像目标区域的提取方法和装置
CN110648330B (zh) 摄像头玻璃的缺陷检测方法
US20230051951A1 (en) Method for training image processing model
CN111127400A (zh) 一种乳腺病变检测方法和装置
CN112330613A (zh) 一种细胞病理数字图像质量的评价方法及系统
US20220319208A1 (en) Method and apparatus for obtaining feature of duct tissue based on computer vision, and intelligent microscope
CN113256634A (zh) 基于深度学习的宫颈癌tct切片排阴方法和系统
JP6819445B2 (ja) 情報処理装置、制御方法、及びプログラム
CN113724235B (zh) 镜下环境改变条件时半自动化的Ki67/ER/PR阴性、阳性细胞计数系统及方法
WO2021139447A1 (zh) 一种宫颈异常细胞检测装置及方法
CN113470041B (zh) 免疫组化细胞图像细胞核分割与计数方法和系统
Athinarayanan et al. COMPUTER AIDED DIAGNOSIS FOR DETECTION AND STAGE IDENTIFICATION OF CERVICAL CANCER BY USING PAP SMEAR SCREENING TEST IMAGES.
CN110348353B (zh) 一种图像处理方法及装置
CN116228767A (zh) 基于计算机视觉的x光肺部肿块图像处理方法
CN113763315B (zh) 玻片图像的信息获取方法、装置、设备及介质
WO2022126923A1 (zh) Asc-us诊断结果分流方法、装置、计算机设备及存储介质
Abdullah et al. An accurate thresholding-based segmentation technique for natural images
CN113920100A (zh) 基于知识蒸馏的弱监督骨扫描图像热点分割方法及系统

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: 20912294

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20912294

Country of ref document: EP

Kind code of ref document: A1