Cell segmentation method and device and electronic equipment

Abstract

The invention discloses a cell segmentation method and device and electronic equipment, and relates to the technical field of medical image processing, and the method comprises the steps: carrying out the preprocessing of medical image data of papillary thyroid carcinoma cells, and obtaining the preprocessed image data; performing spatial domain, frequency domain and time domain feature extraction and fusion on the preprocessed image data to generate a cross-domain spatial-temporal feature map; performing cell segmentation on the cross-domain spatial-temporal feature map through a pre-constructed convolutional neural network to obtain a cell segmentation result; and removing artifacts and noise in the cell segmentation result to obtain a target cell segmentation result. According to the cell segmentation scheme provided by the invention, the cell segmentation precision and robustness can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a cell segmentation method and device, and electronic equipment. Background Art

[0002] The segmentation of thyroid papillary cancer cells is of great significance in medical image analysis, but its accurate segmentation faces many challenges. The segmentation accuracy is affected by many factors, including: Factor 1. Complexity of image data: The morphology of thyroid papillary cancer cells is complex and the boundaries are blurred, making it difficult to accurately extract their boundaries. At the same time, the image data often contains a lot of noise and artifacts. Factor 2. Difficulty in extracting frequency domain features: Although frequency domain features can effectively capture periodic structures and edge features in images, the noise and high-frequency components after frequency domain conversion interfere with feature extraction. Factor 3. Integration problem of self-attention mechanism: The self-attention mechanism is good at capturing multi-scale information, but combining it with frequency domain features while retaining the original information still needs to be studied. Although the multi-head design can process complex information, it increases the computational complexity and difficulty of feature fusion. Factor 4. Limitations of convolutional neural networks (CNN): Convolutional neural networks have limitations in processing frequency domain information and restoring high-resolution feature maps. Traditional convolutional layers and pooling layers cannot fully utilize frequency domain features, affecting segmentation accuracy. Factor 5: Post-segmentation processing requirements: In the segmentation results, artifacts or incomplete cell boundaries may appear due to the presence of noise and artifacts.

[0003] Existing thyroid papillary cancer cell segmentation methods have the following defects: Defect 1: Strong dependence on local features: Traditional segmentation methods, such as those based on threshold segmentation, region growing, and edge detection, usually rely on local features of the image and are easily disturbed by noise and artifacts, resulting in unstable segmentation results. They have poor robustness when facing complex medical imaging data and have difficulty processing cancer cells with fuzzy boundaries and complex morphology.

[0004] Defect 2: Limitations of artificial feature design: It relies on artificially designed features, such as morphological features, texture features, etc. However, artificially designed features are difficult to cover all important information in the image, especially for cancer cell images with diversity and complexity. The limitations of artificial feature design lead to insufficient segmentation accuracy and difficulty in adapting to different types of medical images. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a cell segmentation method and device, and an electronic device, which can solve the problems of low segmentation accuracy and poor robustness in existing thyroid papillary cancer cell segmentation methods.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: An embodiment of the present invention provides a cell segmentation method, wherein the method comprises: Preprocessing the medical image data of thyroid papillary cancer cells to obtain preprocessed image data, wherein the preprocessing includes at least one of the following: image denoising, image standardization, and image contrast enhancement; Extract and fuse the spatial, frequency and time domain features of the preprocessed image data to generate a cross-domain spatiotemporal feature map; Performing cell segmentation on the cross-domain spatiotemporal feature map through a pre-built convolutional neural network to obtain a cell segmentation result; Artifacts and noise in the cell segmentation result are removed to obtain a target cell segmentation result.

[0007] Optionally, the step of extracting and fusing the preprocessed image data in the spatial domain, frequency domain and time domain to generate a cross-domain spatiotemporal feature map includes: Through adaptive frequency domain feature extraction, the pre-processed image data is converted into the frequency domain, adaptively adjusted according to the local frequency domain information, and key features are extracted; Through the frequency domain-time domain joint enhancement mechanism, the dynamic characteristics of the pre-processed image data are captured and enhanced; Generate a cross-domain spatiotemporal feature map based on the key characteristics and the enhanced dynamic features; Combined with the multi-head self-attention mechanism, multi-scale feature enhancement is performed on the cross-domain spatiotemporal feature map, and the original feature information is retained.

[0008] Optionally, the step of performing cell segmentation on the cross-domain spatiotemporal feature map by using a pre-built convolutional neural network to obtain a cell segmentation result includes: Extracting the spatial information of the medical image data through a convolutional layer in a pre-constructed convolutional neural network; wherein the convolutional neural network includes: a convolutional layer, a pooling layer, and an upsampling layer; Reducing the size of the cross-domain spatiotemporal feature map through the pooling layer and retaining key information; The size of the intermediate feature map after the size reduction is restored to the original size through the upsampling layer to obtain the thyroid papillary cancer cell segmentation result.

[0009] Optionally, performing image denoising processing on medical imaging data of thyroid papillary cancer cells includes: Using a Gaussian kernel function to perform image smoothing on the medical imaging data of the thyroid papillary cancer cells; Use bilateral filtering to remove image noise; Use median filtering to replace the pixel value in an image with the median value of its neighborhood pixels.

[0010] Optionally, performing image standardization processing on medical imaging data of thyroid papillary cancer cells includes: Gray value normalization was performed on medical imaging data of thyroid papillary cancer cells.

[0011] Optionally, performing image enhancement contrast processing on medical imaging data of thyroid papillary cancer cells includes: The medical image data of the thyroid papillary cancer cells is subjected to contrast enhancement processing by using histogram equalization or adaptive histogram equalization.

[0012] Optionally, the step of removing artifacts and noise in the cell segmentation result to obtain a target cell segmentation result includes: The artifacts and noise in the cell segmentation result are removed by using a preset morphological operation, wherein the preset morphological operation includes: a dilation operation, an erosion operation, and an opening operation and a closing operation.

[0013] Optionally, the medical imaging data includes at least one of the following: an ultrasound image, a CT image, and a magnetic resonance imaging image.

[0014] An embodiment of the present invention provides a cell segmentation device, wherein the device comprises: A preprocessing module, used for preprocessing the medical imaging data of the thyroid papillary cancer cells to obtain preprocessed imaging data, wherein the preprocessing includes at least one of the following: image denoising, image standardization and image contrast enhancement; A generation module is used to extract and fuse the spatial, frequency and time domain features of the pre-processed image data to generate a cross-domain spatiotemporal feature map; A segmentation module, used to perform cell segmentation on the cross-domain spatiotemporal feature map through a pre-built convolutional neural network to obtain a cell segmentation result; The post-processing module is used to remove artifacts and noise in the cell segmentation result to obtain the target cell segmentation result.

[0015] Optionally, the generating module includes: The first submodule is used to convert the pre-processed image data into the frequency domain through adaptive frequency domain feature extraction, perform adaptive adjustment according to local frequency domain information, and extract key features; The second submodule is used to capture the dynamic characteristics of the pre-processed image data and perform enhancement processing through a frequency domain-time domain joint enhancement mechanism; The third submodule is used to generate a cross-domain spatiotemporal feature map based on the key characteristics and the enhanced dynamic features; The fourth submodule is used to combine the multi-head self-attention mechanism to perform multi-scale feature enhancement on the cross-domain spatiotemporal feature map and retain the original feature information.

[0016] Optionally, the segmentation module includes: A fifth submodule, configured to extract spatial information of the medical image data through a convolutional layer in a pre-constructed convolutional neural network; wherein the convolutional neural network includes: a convolutional layer, a pooling layer, and an upsampling layer; A sixth submodule, configured to reduce the size of the cross-domain spatiotemporal feature map through the pooling layer and retain key information; The seventh submodule is used to restore the size of the intermediate feature map after the size reduction to the original size through the upsampling layer to obtain the thyroid papillary cancer cell segmentation result.

[0017] Optionally, when the preprocessing module performs image denoising on the medical imaging data of thyroid papillary cancer cells, it is specifically used to: Using a Gaussian kernel function to perform image smoothing on the medical imaging data of the thyroid papillary cancer cells; Use bilateral filtering to remove image noise; Use median filtering to replace the pixel value in an image with the median value of its neighborhood pixels.

[0018] Optionally, when the preprocessing module performs image standardization processing on the medical imaging data of thyroid papillary cancer cells, it is specifically used to: Gray value normalization was performed on medical imaging data of thyroid papillary cancer cells.

[0019] Optionally, when the preprocessing module performs image enhancement contrast processing on the medical imaging data of thyroid papillary cancer cells, it is specifically used to: The medical image data of the thyroid papillary cancer cells is subjected to contrast enhancement processing by using histogram equalization or adaptive histogram equalization.

[0020] Optionally, the post-processing module is specifically used for: The artifacts and noise in the cell segmentation result are removed by using a preset morphological operation, wherein the preset morphological operation includes: a dilation operation, an erosion operation, and an opening operation and a closing operation.

[0021] Optionally, the medical imaging data includes at least one of the following: an ultrasound image, a CT image, and a magnetic resonance imaging image.

[0022] An embodiment of the present invention also provides an electronic device, wherein the electronic device includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the cell segmentation method described in any of the above embodiments when executing the program stored in the memory.

[0023] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the cell segmentation method as described in any of the above embodiments is implemented.

[0024] The cell segmentation scheme provided in the embodiment of the present invention pre-processes the medical imaging data of thyroid papillary cancer cells to obtain pre-processed imaging data; extracts and fuses the spatial, frequency and time domain features of the pre-processed imaging data to generate a cross-domain spatiotemporal feature map; performs cell segmentation on the cross-domain spatiotemporal feature map through a pre-built convolutional neural network to obtain a cell segmentation result; removes artifacts and noise in the cell segmentation result to obtain a target cell segmentation result. The cell segmentation scheme provided in the embodiment of the present invention can improve the accuracy of cell segmentation and has strong universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flowchart showing the steps of a cell segmentation method according to an embodiment of the present application; Figure 2 is a structural block diagram of a cell segmentation device according to an embodiment of the present application; Figure 3 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0027] The cell segmentation method provided by the present invention aims to improve the accuracy and robustness of cell segmentation. The method is a thyroid papillary cancer cell segmentation method based on adaptive frequency domain features and self-attention mechanism, comprising the following steps: first, obtaining and preprocessing medical imaging data of thyroid papillary carcinoma; then, extracting key features of spatial domain, frequency domain and time domain simultaneously through a dynamic cross-domain spatiotemporal feature fusion module, and generating a comprehensive feature map using a frequency domain-time domain joint enhancement mechanism. Next, the feature map is multi-scale enhanced using a self-attention mechanism to optimize feature expression. Finally, a convolutional neural network is used to perform cell segmentation on the enhanced feature map, and morphological operations are performed for post-processing to further improve the accuracy of the segmentation results. By integrating multi-domain features and advanced enhancement mechanisms, this method shows excellent results in cell segmentation tasks under complex backgrounds, and is suitable for early diagnosis of thyroid papillary carcinoma and formulation of treatment plans.

[0028] The cell segmentation scheme provided in the embodiment of the present application is described in detail below through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0029] As attached Figure 1 As shown, the cell segmentation method of the embodiment of the present application includes the following steps: Step 101: pre-processing the medical image data of the thyroid papillary cancer cells to obtain pre-processed image data.

[0030] The medical image data includes at least one of the following: ultrasound images, CT images, and magnetic resonance imaging images. The preprocessing includes at least one of the following: image denoising, image standardization, and image contrast enhancement.

[0031] In an optional embodiment, a method of performing image denoising processing on medical imaging data of thyroid papillary cancer cells includes: The image in the medical imaging data of the thyroid papillary cancer cells is smoothed using a Gaussian kernel function; image noise is removed using a bilateral filter; and pixel values ​​in the image are replaced with the median values ​​of their neighborhood pixels using a median filter.

[0032] Image denoising includes denoising using Gaussian filtering, bilateral filtering or median filtering. Gaussian filtering uses a Gaussian kernel function to convolve the image to smooth the image; bilateral filtering removes noise while maintaining edge details; median filtering reduces edge noise by replacing pixel values ​​with the median value in its neighborhood.

[0033] In an optional embodiment, the method of performing image normalization processing on the medical imaging data of the papillary thyroid cancer cells includes: performing grayscale value normalization processing on the medical imaging data of the papillary thyroid cancer cells.

[0034] In the actual implementation process, when the grayscale value of the image data is normalized, its grayscale value range is between 0 and 1. The specific method includes subtracting the minimum grayscale value of the image from the grayscale value of each pixel, and then dividing it by the grayscale value range of the image, so as to linearly map the grayscale value to the range between 0 and 1.

[0035] In an optional embodiment, a method of performing image enhancement contrast processing on medical imaging data of thyroid papillary cancer cells includes: Histogram equalization or adaptive histogram equalization is used to perform contrast enhancement on medical imaging data of thyroid papillary cancer cells.

[0036] Histogram equalization enhances the global contrast of an image by stretching the grayscale value distribution of the image; adaptive histogram equalization enhances the contrast in local areas to improve local details of the image.

[0037] Step 102: Extract and fuse the spatial, frequency and time domain features of the pre-processed image data to generate a cross-domain spatiotemporal feature map.

[0038] In this step, by constructing a dynamic cross-domain spatiotemporal feature fusion module, the pre-processed image data is simultaneously subjected to spatial, frequency and temporal feature extraction and fusion. The dynamic cross-domain spatiotemporal feature fusion module may further include a frequency domain filter to suppress high-frequency noise in the frequency domain. Specifically, the frequency domain filter retains only the information of the low-frequency part in the frequency domain by designing a low-pass filter, thereby reducing the influence of high-frequency noise on image feature extraction.

[0039] In an optional embodiment, the method of performing spatial domain, frequency domain and time domain feature extraction and fusion on the pre-processed image data to generate a cross-domain spatiotemporal feature map may include the following sub-steps: Sub-step 1: Through adaptive frequency domain feature extraction, the pre-processed image data is converted to the frequency domain, adaptively adjusted according to the local frequency domain information, and key features are extracted.

[0040] Converting the preprocessed image data to the frequency domain and performing adaptive adjustments based on the local frequency domain information can suppress high-frequency noise.

[0041] Sub-step 2: Capture the dynamic features of the pre-processed image data and perform enhancement processing through a frequency domain-time domain joint enhancement mechanism.

[0042] Sub-step 3: Generate a cross-domain spatiotemporal feature map based on the key features and enhanced dynamic features.

[0043] Sub-step 4: Combine the multi-head self-attention mechanism to perform multi-scale feature enhancement on the cross-domain spatiotemporal feature map and retain the original feature information.

[0044] The multi-head self-attention mechanism uses multiple sets of attention heads in parallel to independently calculate features of different scales and concatenate the results. Each attention head calculates the attention weights of the input features separately, and then applies these weights to the feature map. Finally, the outputs of multiple heads are concatenated together to generate an enhanced feature map.

[0045] Step 103: Perform cell segmentation on the cross-domain spatiotemporal feature map through a pre-built convolutional neural network to obtain a cell segmentation result.

[0046] An optional method of performing cell segmentation on the cross-domain spatiotemporal feature map through a pre-built convolutional neural network to obtain the cell segmentation result can be as follows: First, the spatial information of the medical image data is extracted through the convolutional layer in the pre-built convolutional neural network; wherein the convolutional neural network includes: a convolutional layer, a pooling layer, and an upsampling layer; The convolution layer can use 3×3 or 5×5 convolution kernels for feature extraction to balance the accuracy and computational complexity of feature extraction. The choice of convolution kernel is adjusted according to the requirements of feature extraction. The 3×3 convolution kernel is suitable for extracting detailed features, and the 5×5 convolution kernel is suitable for extracting feature information in a larger range.

[0047] Secondly, the pooling layer is used to reduce the size of the cross-domain spatiotemporal feature map and retain key information; The pooling layer can use maximum pooling or average pooling to reduce the size of the feature map. Maximum pooling selects the maximum value in the pooling window to represent the window and retains the main features; average pooling calculates the average value in the pooling window to represent the window and reduce the impact of noise.

[0048] Finally, the size of the intermediate feature map after the size reduction is restored to the original size through the upsampling layer to obtain the thyroid papillary cancer cell segmentation result.

[0049] The upsampling layer can use deconvolution or bilinear interpolation methods to restore the spatial resolution of the feature map. Deconvolution restores the high-resolution information of the feature map by learning weights; bilinear interpolation generates a high-resolution feature map by interpolating the low-resolution feature map.

[0050] Step 104: remove artifacts and noise in the cell segmentation result to obtain the target cell segmentation result.

[0051] An optional way to remove artifacts and noise in the cell segmentation result and obtain the target cell segmentation result may be: using a preset morphological operation to remove artifacts and noise in the cell segmentation result.

[0052] The preset morphological operations include: dilation operation, erosion operation, opening operation and closing operation. The dilation operation fills small holes by expanding the foreground area in the image; the erosion operation removes small noise points by shrinking the foreground area; the opening operation and closing operation are erosion followed by dilation and dilation followed by erosion, which are used to remove small objects and fill small holes respectively.

[0053] The segmentation results obtained in this step are used to assist doctors in diagnosing and formulating treatment plans for papillary thyroid cancer. The segmentation results can accurately locate the location and range of cancer cells, help doctors make accurate diagnoses, and formulate personalized treatment plans to improve treatment outcomes.

[0054] The cell segmentation method provided in the embodiment of the present invention pre-processes the medical imaging data of thyroid papillary cancer cells to obtain pre-processed imaging data; extracts and fuses the spatial, frequency and time domain features of the pre-processed imaging data to generate a cross-domain spatiotemporal feature map; performs cell segmentation on the cross-domain spatiotemporal feature map through a pre-constructed convolutional neural network to obtain a cell segmentation result; removes artifacts and noise in the cell segmentation result to obtain a target cell segmentation result. The cell segmentation method provided in the embodiment of the present invention can improve the accuracy of cell segmentation and has strong universality.

[0055] The cell segmentation method provided by the embodiment of the present application is described below with a specific example. This embodiment specifically provides a thyroid papillary cancer cell segmentation method based on adaptive frequency domain features and self-attention mechanism, comprising the following steps: Step 1: Obtain medical imaging data of thyroid papillary cancer cells.

[0056] Medical imaging data of thyroid papillary cancer cells may include ultrasound images, CT images, and MRI (magnetic resonance imaging) images. These imaging data can be obtained through the hospital's medical imaging system or a public medical imaging database. The quality and resolution of the images must be ensured during data acquisition for subsequent processing.

[0057] Step 2: Preprocess the acquired medical image data to improve image quality and standardize the input data.

[0058] Preprocessing of medical image data includes image denoising and image standardization.

[0059] (1) The image denoising process is as follows: i. Gaussian filtering: Use Gaussian filter to smooth the image data and reduce high-frequency noise.

[0060] ii. Bilateral filtering: Bilateral filtering can smooth the image while retaining edge information, and is an effective method in the denoising process.

[0061] iii. Median filtering: Median filtering of image data can effectively remove salt and pepper noise.

[0062] (2) The image standardization process is as follows: i. Grayscale normalization: Normalize the grayscale value range of the image data to between 0 and 1 to eliminate the grayscale differences between different images. This can be achieved using the following formula: ; in, I is the original image, and are the minimum and maximum grayscale values ​​of the image, respectively.

[0063] Step 3: Construct a dynamic cross-domain spatiotemporal feature fusion module, which simultaneously processes and fuses the features in the spatial domain, frequency domain, and time domain.

[0064] The specific process is as follows: (1) Spatial domain feature extraction: Extract the spatial features of the image, such as shape and texture. The spatial domain features can be expressed as: ; in, represents the extracted spatial domain features, It is a convolutional neural network for spatial feature extraction.

[0065] (2) Frequency domain feature extraction: The image is converted from the spatial domain to the frequency domain through Fourier transform, which is expressed as: ; Where F represents 2D Discrete Fourier Transform. In the frequency domain, the adaptive adjustment formula is: ; in, It is an adaptive weight matrix calculated based on the local frequency domain information of the image and is used to optimize the feature extraction process.

[0066] (3) Temporal feature extraction: Extract temporal features from dynamic image sequences (such as videos or time series images), expressed as: ; Among them, RNN represents a recurrent neural network for time domain feature extraction. It is time series image data.

[0067] (4) Multi-domain feature fusion: The spatial, frequency and time domain features are fused through the frequency-time domain joint enhancement mechanism. The formula is: ; in, , and is an adaptive weight used to adjust the contribution of features in different domains and generate a cross-domain spatiotemporal feature map .

[0068] Step 4: Construct a self-attention mechanism module. The fused feature map enters the self-attention mechanism module and features are enhanced through the multi-head self-attention mechanism.

[0069] The calculation formula of the multi-head self-attention mechanism is: ; Where Q, K and V are query, key and value vectors respectively. is the dimension of the key vector. The final output is an enhanced feature map obtained by concatenating the results of multiple attention heads: ; Here, h represents the number of attention heads.

[0070] Step 5: Perform cell segmentation.

[0071] The enhanced feature map is input into the convolutional neural network (CNN) and cell segmentation is performed through multi-layer convolution, pooling and upsampling. The final result of cell segmentation is expressed as: ; in, is a convolutional neural network for cell segmentation, Represents the segmentation result.

[0072] Step 6: Post-processing of segmentation results.

[0073] Morphological operations such as dilation, erosion, opening and closing are performed on the segmentation results to further improve the segmentation accuracy. These operations can be implemented through morphological filters.

[0074] The thyroid papillary cancer cell segmentation method based on adaptive frequency domain features and self-attention mechanism provided in the embodiment of the present application has been significantly improved in the following two aspects: (1) The significant improvement in segmentation accuracy is reflected in the following two aspects: i. Ability to capture details: Traditional segmentation methods often rely on local features and may not be able to effectively capture subtle changes in cell boundaries. Through the adaptive frequency domain feature extraction module, this method can better identify and adjust subtle features in the image in the frequency domain, ensuring that details in low-contrast areas and edge areas are fully captured. In addition, the multi-head self-attention mechanism can process features of different scales in parallel during the feature enhancement process, so that the boundaries and morphology of cells can be segmented more accurately, avoiding the problem of missing or misclassification caused by insufficient features in traditional methods.

[0075] ii. Background noise suppression: By adaptively adjusting the frequency domain features and applying frequency domain filters, this method can effectively suppress high-frequency noise and interference in the image, making the features of the cell area more prominent. This noise suppression capability improves the accuracy of cell segmentation and reduces the impact of artifacts and background noise on the segmentation results.

[0076] (2) The enhancement of model robustness is reflected in the following two points: i. Multi-scale feature integration: The multi-head self-attention mechanism processes features of different scales in parallel and integrates these features together, so that the model can better adapt to various changes and inconsistencies in the image. This mechanism not only improves the model's adaptability to different types of image data, but also enhances the model's robustness in processing complex and deformable cells.

[0077] ii. Feature preservation and enhancement: Residual connections are added during feature enhancement to effectively preserve the original feature information. This design enables the model to maintain sensitivity to the original image information while enhancing features, reducing segmentation errors caused by feature loss or insufficient information, thereby enhancing the stability of the model when facing different image qualities and complex backgrounds.

[0078] Figure 2 A structural block diagram of a cell segmentation device for implementing an embodiment of the present application.

[0079] The cell segmentation device provided in the embodiment of the present application includes the following functional modules: A preprocessing module 201 is used to preprocess the medical image data of the thyroid papillary cancer cells to obtain preprocessed image data, wherein the preprocessing includes at least one of the following: image denoising, image standardization and image contrast enhancement; A generation module 202 is used to extract and fuse the spatial domain, frequency domain and time domain features of the pre-processed image data to generate a cross-domain spatiotemporal feature map; A segmentation module 203 is used to perform cell segmentation on the cross-domain spatiotemporal feature map through a pre-built convolutional neural network to obtain a cell segmentation result; The post-processing module 204 is used to remove artifacts and noise in the cell segmentation result to obtain a target cell segmentation result.

[0080] Optionally, the generating module includes: The first submodule is used to convert the pre-processed image data into the frequency domain through adaptive frequency domain feature extraction, perform adaptive adjustment according to local frequency domain information, and extract key features; The second submodule is used to capture the dynamic characteristics of the pre-processed image data and perform enhancement processing through a frequency domain-time domain joint enhancement mechanism; The third submodule is used to generate a cross-domain spatiotemporal feature map based on the key characteristics and the enhanced dynamic features; The fourth submodule is used to combine the multi-head self-attention mechanism to perform multi-scale feature enhancement on the cross-domain spatiotemporal feature map and retain the original feature information.

[0081] Optionally, the segmentation module includes: A fifth submodule, configured to extract spatial information of the medical image data through a convolutional layer in a pre-constructed convolutional neural network; wherein the convolutional neural network comprises: a convolutional layer, a pooling layer, and an upsampling layer; A sixth submodule, configured to reduce the size of the cross-domain spatiotemporal feature map through the pooling layer and retain key information; The seventh submodule is used to restore the size of the intermediate feature map after the size reduction to the original size through the upsampling layer to obtain the thyroid papillary cancer cell segmentation result.

[0082] Optionally, when the preprocessing module performs image denoising on the medical imaging data of thyroid papillary cancer cells, it is specifically used to: Using a Gaussian kernel function to perform image smoothing on the medical imaging data of the thyroid papillary cancer cells; Use bilateral filtering to remove image noise; Use median filtering to replace the pixel value in an image with the median value of its neighborhood pixels.

[0083] Optionally, when the preprocessing module performs image standardization processing on the medical imaging data of thyroid papillary cancer cells, it is specifically used to: Gray value normalization was performed on medical imaging data of thyroid papillary cancer cells.

[0084] Optionally, when the preprocessing module performs image enhancement contrast processing on the medical imaging data of thyroid papillary cancer cells, it is specifically used to: The medical image data of the thyroid papillary cancer cells is subjected to contrast enhancement processing by using histogram equalization or adaptive histogram equalization.

[0085] Optionally, the post-processing module is specifically used for: The artifacts and noise in the cell segmentation result are removed by using a preset morphological operation, wherein the preset morphological operation includes: a dilation operation, an erosion operation, and an opening operation and a closing operation.

[0086] Optionally, the medical imaging data includes at least one of the following: an ultrasound image, a CT image, and a magnetic resonance imaging image.

[0087] The embodiments of the present application provide Figure 2 The cell segmentation device shown can achieve Figure 1 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0088] The cell segmentation device provided in the embodiment of the present invention pre-processes the medical imaging data of thyroid papillary cancer cells to obtain pre-processed imaging data; extracts and fuses the spatial, frequency and time domain features of the pre-processed imaging data to generate a cross-domain spatiotemporal feature map; performs cell segmentation on the cross-domain spatiotemporal feature map through a pre-constructed convolutional neural network to obtain a cell segmentation result; removes artifacts and noise in the cell segmentation result to obtain a target cell segmentation result. The cell segmentation device provided in the embodiment of the present invention can improve the accuracy of cell segmentation and has strong universality.

[0089] The embodiment of the present invention further provides an electronic device, such as Figure 3 As shown, it includes a processor 301 , a communication interface 302 , a memory 303 and a communication bus 304 , wherein the processor 301 , the communication interface 302 , and the memory 303 communicate with each other via the communication bus 304 .

[0090] Memory 303, used for storing computer programs; The processor 301 is used to implement each cell segmentation method step in the above method embodiment when executing the program stored in the memory 303.

[0091] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0092] The communication interface is used for communication between the above terminal and other devices.

[0093] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0094] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0095] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0096] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0097] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A cell segmentation method, characterized in that: The method comprises: Preprocessing the medical image data of thyroid papillary cancer cells to obtain preprocessed image data, wherein the preprocessing includes at least one of the following: image denoising, image standardization, and image contrast enhancement; Extract and fuse the spatial, frequency and time domain features of the preprocessed image data to generate a cross-domain spatiotemporal feature map; Performing cell segmentation on the cross-domain spatiotemporal feature map through a pre-built convolutional neural network to obtain a cell segmentation result; Artifacts and noise in the cell segmentation result are removed to obtain a target cell segmentation result.

2. The method according to claim 1, characterized in that The steps of extracting and fusing the spatial, frequency and time domain features of the preprocessed image data to generate a cross-domain spatiotemporal feature map include: Through adaptive frequency domain feature extraction, the pre-processed image data is converted into the frequency domain, adaptively adjusted according to the local frequency domain information, and key features are extracted; Through the frequency domain-time domain joint enhancement mechanism, the dynamic characteristics of the pre-processed image data are captured and enhanced; Generate a cross-domain spatiotemporal feature map based on the key features and the dynamic features of the enhanced image data; Combined with the multi-head self-attention mechanism, multi-scale feature enhancement is performed on the cross-domain spatiotemporal feature map, and the original feature information is retained.

3. The method according to claim 1, characterized in that The step of performing cell segmentation on the cross-domain spatiotemporal feature map by using a pre-built convolutional neural network to obtain a cell segmentation result comprises: Extracting the spatial information of the medical image data through a convolutional layer in a pre-constructed convolutional neural network; wherein the convolutional neural network includes: a convolutional layer, a pooling layer, and an upsampling layer; Reducing the size of the cross-domain spatiotemporal feature map through the pooling layer and retaining key information; The size of the cross-domain spatiotemporal feature map after the size reduction is restored to the original size through the upsampling layer to obtain the thyroid papillary cancer cell segmentation result.

4. The method according to claim 1, characterized in that Image denoising of medical imaging data of thyroid papillary cancer cells includes: Using a Gaussian kernel function to perform image smoothing on the medical imaging data of the thyroid papillary cancer cells; Use bilateral filtering to remove image noise; Use median filtering to replace the pixel value in an image with the median value of its neighborhood pixels.

5. The method according to claim 1, characterized in that Image standardization processing of medical imaging data of thyroid papillary cancer cells includes: Gray value normalization was performed on medical imaging data of thyroid papillary cancer cells.

6. The method according to claim 1, characterized in that Image enhancement contrast processing of medical imaging data of thyroid papillary cancer cells includes: The medical image data of the thyroid papillary cancer cells is subjected to contrast enhancement processing by using histogram equalization or adaptive histogram equalization.

7. The method according to claim 1, characterized in that The step of removing artifacts and noise in the cell segmentation result to obtain the target cell segmentation result comprises: The artifacts and noise in the cell segmentation result are removed by using a preset morphological operation, wherein the preset morphological operation includes: a dilation operation, an erosion operation, and an opening operation and a closing operation.

8. The method according to claim 1, characterized in that The medical imaging data includes at least one of the following: an ultrasound image, a CT image, and a magnetic resonance imaging image.

9. A cell segmentation device, characterized in that: The device comprises: A preprocessing module, used for preprocessing the medical imaging data of the thyroid papillary cancer cells to obtain preprocessed imaging data, wherein the preprocessing includes at least one of the following: image denoising, image standardization and image contrast enhancement; A generation module is used to extract and fuse the spatial, frequency and time domain features of the pre-processed image data to generate a cross-domain spatiotemporal feature map; A segmentation module, used to perform cell segmentation on the cross-domain spatiotemporal feature map through a pre-built convolutional neural network to obtain a cell segmentation result; The post-processing module is used to remove artifacts and noise in the cell segmentation result to obtain the target cell segmentation result.

10. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the cell segmentation method according to any one of claims 1 to 8 when executing the program stored in the memory.

Images