Ear-nose-throat department CT examination image segmentation method and system

Abstract

The embodiment of the invention relates to the technical field of image segmentation, and particularly discloses a CT examination image segmentation method and system for the otolaryngology department. According to the embodiment of the invention, on the basis of a big data technology, public CT images of the otolaryngology department are collected, and privacy, extension and labeling processing is carried out; selecting a basic segmentation framework, and constructing an image segmentation model; performing image preprocessing and initial segmentation processing on the target CT image to generate a plurality of initial segmentation images; and importing the images into an image segmentation model, deriving a plurality of model segmentation images, and performing fine adjustment to generate a plurality of fine segmentation images. According to the method, the image segmentation model can be constructed, the target CT image is subjected to image preprocessing and initial segmentation processing, then image segmentation is performed through the image segmentation model, fine adjustment is performed, a plurality of fine segmentation images are generated, automatic and accurate CT examination image segmentation of the otolaryngology department can be realized, the cost is reduced, the consumed time is short, and the efficiency is high. And real-time requirements in a clinical environment can be met.

Description

Technical Field

[0001] The present invention belongs to the technical field of image segmentation, and in particular relates to an ENT CT examination image segmentation method and system. Background Art

[0002] ENT CT image segmentation is an important field in medical image processing. Its goal is to segment different anatomical structures (such as nasal cavity, larynx, cochlea, etc.) in ENT CT images through algorithms or artificial intelligence technology to provide support for clinical diagnosis and treatment.

[0003] In the existing technology, there is no accurate ENT CT examination image segmentation technology, and it is often only possible to perform rough CT examination image segmentation, and then manually optimize and refine the image segmentation, which greatly increases the labor cost of ENT CT examination image segmentation and is time-consuming, and cannot meet the real-time needs in the clinical environment. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide an ENT CT examination image segmentation method and system, aiming to solve the problems raised in the background technology.

[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions: A method for segmenting CT images of ENT department, the method specifically comprising the following steps: Based on big data technology, multiple public CT images of otolaryngology are collected, and multiple public CT images are privatized, expanded, and annotated to generate multiple annotated CT images; Selecting a basic segmentation architecture, training and validating the basic segmentation architecture based on the plurality of annotated CT images, and constructing an image segmentation model; Acquire a target CT image of the otolaryngology department, perform image preprocessing and initial segmentation processing on the target CT image, and generate a plurality of initial segmentation images; The multiple initial segmented images are imported into the image segmentation model, multiple model segmented images are exported, and fine adjustments are performed to generate multiple fine segmented images.

[0006] An ENT CT examination image segmentation system, the system comprises a public image processing unit, a model training and verification unit, an image preprocessing segmentation unit and a segmentation fine adjustment unit, wherein: A public image processing unit, used to collect a plurality of public CT images of otolaryngology based on big data technology, perform privacy protection, expansion and annotation processing on the plurality of public CT images, and generate a plurality of annotated CT images; A model training and verification unit, used for selecting a basic segmentation architecture, training and verifying the basic segmentation architecture based on the plurality of annotated CT images, and constructing an image segmentation model; An image preprocessing and segmentation unit is used to obtain a target CT image of the otolaryngology department, perform image preprocessing and initial segmentation processing on the target CT image, and generate a plurality of initial segmented images; The segmentation fine adjustment unit is used to import the multiple initial segmentation images into the image segmentation model, export multiple model segmentation images, and perform fine adjustments to generate multiple fine segmentation images.

[0007] Compared with the prior art, the present invention has the following beneficial effects: The embodiment of the present invention collects public CT images of the otolaryngology department based on big data technology, performs privacy protection, expansion and annotation processing; selects a basic segmentation architecture to build an image segmentation model; performs image preprocessing and initial segmentation processing on the target CT image to generate multiple initial segmented images; imports them into the image segmentation model, exports multiple model segmentation images, and performs fine adjustments to generate multiple fine segmented images. It is possible to build an image segmentation model, perform image preprocessing and initial segmentation processing on the target CT image, and then perform image segmentation through the image segmentation model, and perform fine adjustments to generate multiple fine segmented images, which can realize automatic and accurate image segmentation of otolaryngology CT examinations, reduce costs, and save time, and can meet real-time needs in clinical environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0009] Figure 1 A flow chart of a method provided by an embodiment of the present invention is shown.

[0010] Figure 2 A flow chart of generating multiple annotated CT images in the method provided in an embodiment of the present invention is shown.

[0011] Figure 3 A flow chart of constructing an image segmentation model in the method provided in an embodiment of the present invention is shown.

[0012] Figure 4 The flowchart of image preprocessing and initial segmentation processing in the method provided by the embodiment of the present invention is shown.

[0013] Figure 5 A flow chart of initial segmentation processing in the method provided by an embodiment of the present invention is shown.

[0014] Figure 6 A flow chart of generating multiple fine segmentation images in the method provided by an embodiment of the present invention is shown.

[0015] Figure 7 The application architecture diagram of the system provided by the embodiment of the present invention is shown.

[0016] Figure 8 A structural block diagram of a public image processing unit in a system provided by an embodiment of the present invention is shown.

[0017] Fig. 9 The structure block diagram of the image preprocessing and segmentation unit in the system provided by the embodiment of the present invention is shown.

[0018] Fig.10 The structure block diagram of the segmentation fine adjustment unit in the system provided by the embodiment of the present invention is shown. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] It is understandable that in the prior art, there is no accurate ENT CT examination image segmentation technology, and often only rough CT examination image segmentation can be performed, and then the image segmentation is manually optimized and refined, which greatly increases the labor cost of ENT CT examination image segmentation and is time-consuming, and cannot meet the real-time needs in the clinical environment.

[0021] To solve the above problems, the embodiment of the present invention collects multiple public CT images of otolaryngology based on big data technology, performs privacy, expansion and annotation processing on multiple public CT images, and generates multiple annotated CT images; selects a basic segmentation architecture, trains and verifies the basic segmentation architecture based on multiple annotated CT images, and builds an image segmentation model; obtains the target CT image of otolaryngology, performs image preprocessing and initial segmentation processing on the target CT image, and generates multiple initial segmentation images; imports multiple initial segmentation images into the image segmentation model, exports multiple model segmentation images, and performs fine adjustment to generate multiple fine segmentation images. It is possible to build an image segmentation model, perform image preprocessing and initial segmentation processing on the target CT image, and then perform image segmentation through the image segmentation model, and perform fine adjustment to generate multiple fine segmentation images, which can realize automatic and accurate otolaryngology CT examination image segmentation, reduce costs, and save time, and can meet real-time needs in clinical environments.

[0022] Figure 1 A flow chart of a method provided by an embodiment of the present invention is shown.

[0023] Specifically, in one embodiment of the present invention, a method for segmenting an ENT CT examination image comprises the following steps: Step S101, based on big data technology, collect multiple public CT images of otolaryngology, perform privacy protection, expansion and annotation processing on the multiple public CT images, and generate multiple annotated CT images.

[0024] In an embodiment of the present invention, based on big data technology, multiple publishing data sources such as multiple public medical databases and / or hospital clinical databases are determined, and then multiple public CT images of otolaryngology are collected from the multiple publishing data sources. The multiple public CT images are privacy-processed to remove private data such as names and hospitals in the multiple public CT images to generate multiple private CT images. Then, the multiple private CT images are extended by performing geometric enhancement and intensity enhancement processing through rotation, translation, scaling, flipping, brightness adjustment, Gaussian noise adjustment, etc. to generate multiple extended CT images. Then, segmentation annotations for the multiple extended CT images are received to generate multiple annotated CT images.

[0025] Specifically, Figure 2 A flow chart of generating multiple annotated CT images in the method provided in an embodiment of the present invention is shown.

[0026] Among them, in the preferred embodiment provided by the present invention, the method of collecting a plurality of public CT images of otolaryngology based on big data technology, performing privacy protection, expansion and annotation processing on the plurality of public CT images, and generating a plurality of annotated CT images specifically comprises the following steps: Step S1011, determining multiple publishing data sources based on big data technology; Step S1012, collecting a plurality of public CT images of otolaryngology from a plurality of the publishing data sources; Step S1013, performing privacy processing on the plurality of public CT images to generate a plurality of private CT images; Step S1014, performing geometric enhancement and intensity enhancement expansion processing on the plurality of privacy CT images to generate a plurality of expanded CT images; Step S1015 , receiving segmentation annotations for the plurality of extended CT images, and generating a plurality of annotated CT images.

[0027] Among them, in the preferred embodiment provided by the present invention, the expansion processing of geometric enhancement and intensity enhancement of the plurality of privacy CT images to generate a plurality of extended CT images specifically comprises the following steps: According to the position coordinates of each pixel point of the privacy CT image, a two-dimensional rotation transformation is performed on the privacy CT image to obtain a transformed privacy CT image; The transformed private CT image is subjected to denoising by using Gaussian smoothing to obtain a denoised private CT image; Performing two-dimensional fast Fourier transform on the denoised privacy CT image to achieve frequency domain enhancement and obtain a geometrically enhanced privacy CT image; The grayscale value of each pixel of the privacy CT image is nonlinearly transformed using the gamma correction method, and then the nonlinear transformation result is adjusted using exponential decay to obtain the privacy CT image with enhanced intensity. The privacy CT image after intensity enhancement is decomposed into high-frequency components and low-frequency components by wavelet decomposition; Gaussian noise is added to the sensitive area corresponding to the high-frequency component for privacy protection, and the blurred high-frequency component is obtained; Perform inverse wavelet transform on the low-frequency component and the blurred high-frequency component to obtain the privacy CT image with enhanced intensity; The geometrically enhanced privacy CT image and the intensity enhanced privacy CT image are weighted fused, and the fusion result is nonlinearly adjusted to obtain the extended CT image.

[0028] In an embodiment of the present invention, a combination of geometric enhancement and intensity enhancement is used to optimize not only the geometric characteristics of the image (such as rotation, translation, scaling, etc.), but also the pixel intensity characteristics (such as grayscale value, contrast, etc.) are enhanced. This two-way combination can comprehensively improve the quality of the image and make the model robust to both spatial changes and intensity changes. In addition, frequency domain processing (such as Fourier transform) is introduced in the geometric enhancement process, which can effectively extract high-frequency features (such as edges, textures) and low-frequency features (such as overall shape) hidden in the image. Compared with processing only in the spatial domain, this method can refine the image features at a deeper level and improve the model's ability to capture important features.

[0029] Frequency domain processing can more efficiently enhance specific parts (such as edges), while the spatial domain is responsible for maintaining the overall visualization quality.

[0030] In addition, in order to ensure the effect of intensity enhancement and privacy protection, wavelet transform is used for multi-scale feature decomposition, which brings more detail levels to the image. Gaussian blur is added to the corresponding sensitive areas in the high-frequency component for privacy protection. This multi-scale feature extraction method is very suitable for medical image processing, because anatomical structures usually require multi-level representation and patient privacy needs to be guaranteed.

[0031] In a preferred embodiment of the present invention, the step of receiving segmentation annotations for the plurality of extended CT images and generating the plurality of annotated CT images specifically comprises the following steps: Selecting a CT image from a plurality of extended CT images as a current image; Use the pre-trained U-Net to segment the current image and obtain a segmentation probability map; Introducing manual annotation, based on the results of multiple annotators, by calculating the confidence intervals of different annotation schemes to generate optimized manual annotations; The optimized manual annotation is converted into a discrete region division image, the confidence is dynamically calculated according to whether the discrete region division image is a boundary region, and the confidence is converted into a weight to obtain an artificial confidence weight; According to the probability value of the segmentation probability map, the corresponding confidence is assigned to obtain the intelligent segmentation confidence, and the intelligent segmentation confidence is converted into a weight to obtain the model confidence weight; The segmentation probability map is weighted with the model confidence weight to obtain a first weighted result, the manual annotation is weighted with the manual confidence weight to obtain a second weighted result, and the current image is fused with the corresponding first weighted result and second weighted result to obtain a fused labeled image; The fused annotated image is optimized at multiple scales to obtain the final annotated CT image.

[0032] In an embodiment of the present invention, the segmentation model provides the ability to automatically generate preliminary results, greatly reducing the workload of manual annotation. Manual annotation experts only need to correct specific areas without having to annotate from scratch, which significantly improves the annotation efficiency. Manual annotation may have certain subjective differences in medical images (for example, different annotators have inconsistent judgments on boundaries). By introducing confidence analysis and probability fusion, the system can smooth the differences between annotators and generate more consistent annotation results. At the same time, the results of the segmentation model can also be used as a reference to reduce omissions in manual annotation. Finally, through probability fusion, the system can dynamically adjust the weights according to the confidence of intelligent segmentation and manual annotation. Compared with the simple method of "manual annotation covers intelligent segmentation" or "intelligent segmentation replaces manual annotation", probability fusion ensures the flexibility and robustness of the processing results. The dynamic allocation of confidence weights enables the system to make optimal decisions based on the characteristics of different regions (high confidence regions or low confidence regions).

[0033] Furthermore, the ENT CT examination image segmentation method further includes the following steps: Step S102: selecting a basic segmentation architecture, training and verifying the basic segmentation architecture based on the plurality of annotated CT images, and constructing an image segmentation model.

[0034] In an embodiment of the present invention, a basic segmentation architecture is selected from a plurality of preset network architectures such as U-Net, 3D U-Net, Hybrid CNN-Transformer and Swin UNETR, and a data division ratio is obtained. Then, according to the data division ratio, a plurality of annotated CT images are divided into a model training set and a model verification set. Then, based on the model training set and the model verification set, the basic segmentation architecture is trained and verified to construct an image segmentation model.

[0035] Specifically, Figure 3 A flow chart of constructing an image segmentation model in the method provided in an embodiment of the present invention is shown.

[0036] Among them, in the preferred embodiment provided by the present invention, the selecting of the basic segmentation architecture, training and verifying the basic segmentation architecture based on the plurality of annotated CT images, and constructing the image segmentation model specifically include the following steps: Step S1021, selecting a basic segmentation architecture from a plurality of preset network architectures; Step S1022, obtaining a data division ratio; Step S1023, dividing the plurality of annotated CT images into a model training set and a model verification set according to the data division ratio; Step S1024: Based on the model training set and the model verification set, the basic segmentation architecture is trained and verified to construct an image segmentation model.

[0037] Among them, in the preferred embodiment provided by the present invention, the basic segmentation architecture is trained and verified based on the model training set and the model verification set, and the image segmentation model is constructed, which specifically includes the following steps: Divide the model training set into multiple small batches, each batch contains several training samples; Perform data augmentation operations on training samples to obtain enhanced images; The enhanced image is input into the basic segmentation architecture for forward propagation to perform segmentation prediction and auxiliary task prediction, thereby obtaining the classification result and the target boundary contour; The main task uses Dice loss to evaluate the segmentation result, and the auxiliary task uses cross entropy loss to evaluate the boundary detection result. The Dice loss and cross entropy loss are weighted summed, and regularization loss is introduced to obtain a mixed loss function. Input the hybrid loss function into the Adam optimizer for optimization, set the weight decay strategy and learning parameters of the Adam optimizer; train the basic segmentation architecture by updating the weights and learning parameters to minimize the loss, and obtain the trained basic segmentation architecture; After the training is completed, the validation set is used to verify the performance of the trained basic segmentation architecture, and the image segmentation model is obtained after the verification is completed.

[0038] Among them, in the preferred embodiment provided by the present invention, the step of inputting the enhanced image into the basic segmentation architecture for forward propagation to perform segmentation prediction and auxiliary task prediction to obtain the classification result and the target boundary contour specifically includes the following steps: Step S102401, the basic segmentation architecture is composed of an encoder and a decoder, and the encoder and the decoder are each provided with a number of layers; Step S102402, inputting the enhanced image into the first layer encoder, and extracting the local feature map of the enhanced image through the convolutional neural network; Step S102403, sending the multi-scale local detail features into the Transformer module for global modeling to capture long-range dependencies and obtain a global feature map; Step S102404, performing global average pooling on the local feature map and the global feature map respectively to obtain the average value of each channel; Step S102405, inputting the average value of each channel into a shared fully connected network to generate coefficients for dynamically adjusting the weights of local features and global features; Step S102406, compressing the coefficients of dynamically adjusting the weights of local features and global features to a range of [0, 1] through a Sigmoid activation function to obtain local feature weights and global feature weights; Step S102407, each channel of the local feature map is multiplied by the corresponding weight to achieve dynamic adjustment of the local features to obtain a weighted local feature map; Step S102408, each channel of the global feature map is multiplied by the corresponding weight to achieve dynamic adjustment of the global feature to obtain a weighted global feature map; Step S102409, concatenating the local feature map and the global feature map in the channel dimension, and performing channel compression through a 1×1 convolution operation to obtain a fused feature map; Step S102410, using the fused feature map as the input of the next layer decoder, repeating steps S102401 to S102410 several times in an iterative manner to obtain the final fused feature; Step S102411, sending the final fused features to the decoder, performing a 3×3 convolution operation on the final fused features to obtain a convolution feature map; Step S102412, performing global average pooling on the convolution feature map to obtain a global pooling result; Step S102413, passing the global pooling result through a fully connected layer and a Sigmoid activation function to generate channel weights; Step S102414, performing a weighted operation on the convolution feature map and the channel weights channel by channel to obtain the first layer decoder output; Step S102415, using dense skip connection to splice the output of the previous layer decoder and the output of the corresponding layer encoder in the channel dimension as the input of the next layer decoder, and iteratively repeating steps S102411 to S102414 for several times to obtain the final output of the decoder; Step S102416: The decoder finally outputs a segmentation map with the same resolution as the input image after an upsampling operation, and obtains a high-resolution feature map of multi-scale feature aggregation. The high-resolution feature map of multi-scale feature aggregation is sent to the classification head to generate a category probability distribution and obtain a classification result. Step S102417, take the output of the middle layer decoder to perform boundary detection operation, generate a boundary probability image, and obtain the target boundary contour.

[0039] In the embodiment of the present invention, the advantages of convolutional neural network (CNN) and lightweight Transformer are combined, and a feature fusion mechanism is introduced. Among them, the CNN part is good at extracting local information of the image, such as edges, textures, small lesions, etc.

[0040] The Transformer part captures long-range dependencies through the self-attention mechanism and understands the global context of the image (such as the overall morphology of the anatomical structure, left-right symmetry, etc.).

[0041] The fused hybrid features can take into account both the details of small targets and the global structural information in the pixel-level segmentation task, thereby improving the overall segmentation accuracy of the model. In medical imaging, everything from the outline of the lesion to the shape and position of the entire anatomical area is very important. Through this hybrid extraction method, we can find small lesions while maintaining awareness of the global anatomical structure.

[0042] And in the fusion process, dynamic weight adjustment is introduced to adjust the relative importance between CNN and Transformer features according to task requirements. When it comes to segmentation tasks with fuzzy boundaries, it may rely more on global Transformer features; when it comes to segmentation tasks with rich details, it may rely more on CNN features. This dynamic mechanism makes the model more adaptable and can achieve better results in different tasks and data distributions. In CT images, the boundaries of lesions may be blurred or mixed with surrounding tissues, and the contextual information of anatomical structures can help the model determine the location of the boundaries. This dynamic adjustment ensures that the model can both accurately segment the boundaries and maintain global consistency of the context.

[0043] Furthermore, the ENT CT examination image segmentation method further includes the following steps: Step S103, obtaining a target CT image of the ENT department, performing image preprocessing and initial segmentation processing on the target CT image, and generating a plurality of initial segmented images.

[0044] In an embodiment of the present invention, a target CT image of an otolaryngology department that needs to be segmented is obtained, and noise removal, image enhancement, resolution adjustment, and standardized image preprocessing are performed on the target CT image to generate a standard CT image. Then, segmentation candidate regions of the standard CT image are identified, and the segmentation candidate regions in the standard CT image are separately cut out to obtain a candidate CT image, thereby eliminating irrelevant image regions, and then a preset image input size is determined, and the candidate CT image is size-segmented according to the image input size to obtain a plurality of initial segmented images.

[0045] Specifically, Figure 4 The flowchart of image preprocessing and initial segmentation processing in the method provided by the embodiment of the present invention is shown.

[0046] Among them, in the preferred embodiment provided by the present invention, the acquisition of the target CT image of the otolaryngology department, the image preprocessing and initial segmentation processing of the target CT image, and the generation of multiple initial segmented images specifically include the following steps: Step S1031, obtaining a target CT image of the otolaryngology department; Step S1032, performing image preprocessing of noise removal, image enhancement, resolution adjustment and standardization on the target CT image to generate a standard CT image; Step S1033, performing initial segmentation processing on the standard CT image to generate a plurality of initial segmented images.

[0047] Specifically, Figure 5 A flow chart of initial segmentation processing in the method provided by an embodiment of the present invention is shown.

[0048] Among them, in the preferred embodiment provided by the present invention, the initial segmentation processing of the standard CT image to generate a plurality of initial segmented images specifically includes the following steps: Step S10331, identifying the segmentation candidate region of the standard CT image; Step S10332, performing interception processing on the standard CT image according to the segmented candidate region to obtain a candidate CT image; Step S10333, determining a preset image input size, and segmenting the candidate CT image into different sizes to obtain a plurality of initial segmented images.

[0049] Furthermore, the ENT CT examination image segmentation method further includes the following steps: Step S104, importing the plurality of initial segmented images into the image segmentation model, exporting a plurality of model segmented images, and performing fine adjustments to generate a plurality of fine segmented images.

[0050] In an embodiment of the present invention, multiple initial segmented images are imported into an image segmentation model, and automatic image segmentation processing is performed through the image segmentation model to export multiple model segmented images. Then, the multiple model segmented images are eroded, expanded, opened, and closed to repair holes and / or isolated areas to generate multiple repaired segmented images. Then, the multiple repaired segmented images are refined to smooth their edges to generate multiple refined segmented images.

[0051] Specifically, Figure 6 A flow chart of generating multiple fine segmentation images in the method provided by an embodiment of the present invention is shown.

[0052] Among them, in the preferred embodiment provided by the present invention, the step of importing the multiple initial segmented images into the image segmentation model, deriving multiple model segmented images, and performing fine adjustments to generate multiple fine segmented images specifically includes the following steps: Step S1041, importing a plurality of the initial segmented images into the image segmentation model; Step S1042, exporting multiple model segmentation images; Step S1043, performing erosion, expansion, opening and closing operations on the plurality of model segmented images, repairing holes and / or isolated areas, and generating a plurality of repaired segmented images; Step S1044, performing a smooth edge refinement process on the plurality of repaired segmented images to generate a plurality of refined segmented images.

[0053] Furthermore, Figure 7 The application architecture diagram of the system provided by the embodiment of the present invention is shown.

[0054] Among them, in another preferred embodiment provided by the present invention, an ENT CT examination image segmentation system includes: The public image processing unit 101 is used to collect a plurality of public CT images of otolaryngology based on big data technology, perform privacy protection, expansion and annotation processing on the plurality of public CT images, and generate a plurality of annotated CT images.

[0055] In the embodiment of the present invention, the public image processing unit 101 determines multiple publishing data sources such as multiple public medical databases and / or hospital clinical databases based on big data technology, and then collects multiple public CT images of otolaryngology from the multiple publishing data sources, performs privacy processing on the multiple public CT images, removes private data such as names and hospitals in the multiple public CT images, and generates multiple private CT images. Then, the multiple private CT images are extended by performing geometric enhancement and intensity enhancement processing through rotation, translation, scaling, flipping, brightness adjustment, Gaussian noise adjustment, etc., to generate multiple extended CT images, and then receives segmentation annotations for the multiple extended CT images to generate multiple annotated CT images.

[0056] Specifically, Figure 8 It shows a structural block diagram of the public image processing unit 101 in the system provided by the embodiment of the present invention.

[0057] Among them, in the preferred embodiment provided by the present invention, the public image processing unit 101 specifically includes: The data source determination module 1011 is used to determine multiple publishing data sources based on big data technology; An image collection module 1012, for collecting a plurality of public CT images of otolaryngology from a plurality of said publishing data sources; A privacy processing module 1013, configured to perform privacy processing on the plurality of public CT images to generate a plurality of private CT images; An expansion processing module 1014 is used to perform expansion processing of geometric enhancement and intensity enhancement on the plurality of privacy CT images to generate a plurality of expanded CT images; The segmentation and annotation module 1015 is used to receive segmentation and annotation of the plurality of extended CT images and generate a plurality of annotated CT images.

[0058] Furthermore, the ENT CT examination image segmentation system also includes: The model training and verification unit 102 is used to select a basic segmentation architecture, train and verify the basic segmentation architecture based on the multiple annotated CT images, and construct an image segmentation model.

[0059] In an embodiment of the present invention, the model training and verification unit 102 selects a basic segmentation architecture from a plurality of preset network architectures such as U-Net, 3D U-Net, HybridCNN-Transformer and Swin UNETR, and obtains a data division ratio, and then divides a plurality of annotated CT images into a model training set and a model verification set according to the data division ratio, and then trains and verifies the basic segmentation architecture based on the model training set and the model verification set to construct an image segmentation model.

[0060] The image preprocessing and segmentation unit 103 is used to obtain a target CT image of the ENT department, perform image preprocessing and initial segmentation processing on the target CT image, and generate a plurality of initial segmented images.

[0061] In the embodiment of the present invention, the image preprocessing and segmentation unit 103 obtains a target CT image of the otolaryngology department that needs to be segmented, performs noise removal, image enhancement, resolution adjustment and standardized image preprocessing on the target CT image to generate a standard CT image, then identifies the segmentation candidate area of ​​the standard CT image, and separately extracts the segmentation candidate area in the standard CT image to obtain a candidate CT image, thereby eliminating irrelevant image areas, and then determines a preset image input size, and size-segmenting the candidate CT image according to the image input size to obtain multiple initial segmented images.

[0062] Specifically, Fig. 9 The structure block diagram of the image preprocessing and segmentation unit 103 in the system provided by the embodiment of the present invention is shown.

[0063] Among them, in the preferred embodiment provided by the present invention, the image preprocessing and segmentation unit 103 specifically includes: An image acquisition module 1031 is used to acquire a target CT image of the ENT department; An image preprocessing module 1032 is used to perform image preprocessing such as noise removal, image enhancement, resolution adjustment and standardization on the target CT image to generate a standard CT image; The initial segmentation module 1033 is used to perform initial segmentation processing on the standard CT image to generate a plurality of initial segmented images.

[0064] Furthermore, the ENT CT examination image segmentation system also includes: The segmentation fine adjustment unit 104 is used to import the multiple initial segmentation images into the image segmentation model, export multiple model segmentation images, and perform fine adjustments to generate multiple fine segmentation images.

[0065] In an embodiment of the present invention, the segmentation fine adjustment unit 104 imports multiple initial segmented images into the image segmentation model, performs automatic image segmentation processing through the image segmentation model, exports multiple model segmentation images, and then performs erosion, expansion, opening and closing operations on the multiple model segmentation images to repair holes and / or isolated areas to generate multiple repaired segmented images, and then performs smooth edge refinement processing on the multiple repaired segmented images to generate multiple fine segmented images.

[0066] Specifically, Fig.10 It shows a structural block diagram of the segmentation fine adjustment unit 104 in the system provided by an embodiment of the present invention.

[0067] In a preferred embodiment of the present invention, the segmentation fine adjustment unit 104 specifically includes: An image import module 1041, used for importing a plurality of the initial segmented images into the image segmentation model; An image export module 1042, used to export multiple model segmentation images; An image restoration module 1043 is used to perform corrosion, expansion, opening and closing operations on the plurality of model segmented images to restore holes and / or isolated areas, and generate a plurality of restored segmented images; The refinement processing module 1044 is used to perform refinement processing on the multiple repaired segmented images to obtain smooth edges, thereby generating multiple refined segmented images.

[0068] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0069] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0070] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for segmenting CT images of ENT department, characterized in that: The method specifically comprises the following steps: Based on big data technology, multiple public CT images of otolaryngology are collected, and multiple public CT images are privatized, expanded, and annotated to generate multiple annotated CT images; Selecting a basic segmentation architecture, training and validating the basic segmentation architecture based on the plurality of annotated CT images, and constructing an image segmentation model; Acquire a target CT image of the otolaryngology department, perform image preprocessing and initial segmentation processing on the target CT image, and generate a plurality of initial segmentation images; Importing the plurality of initial segmented images into the image segmentation model, exporting a plurality of model segmented images, and performing fine adjustments to generate a plurality of fine segmented images; The method of collecting a plurality of public CT images of the otolaryngology department based on big data technology, performing privacy protection, expansion and annotation processing on the plurality of public CT images, and generating a plurality of annotated CT images specifically comprises the following steps: Based on big data technology, determine multiple publishing data sources; Collecting a plurality of public CT images of otolaryngology from a plurality of said published data sources; Performing privacy processing on the plurality of public CT images to generate a plurality of private CT images; Performing geometric enhancement and intensity enhancement expansion processing on the plurality of privacy CT images to generate a plurality of expanded CT images; Segmentation annotations for the plurality of extended CT images are received, and a plurality of annotated CT images are generated.

2. The method for segmenting CT images of ENT department according to claim 1, characterized in that: The step of performing geometric enhancement and intensity enhancement on the plurality of privacy CT images to generate a plurality of extended CT images specifically comprises the following steps: According to the position coordinates of each pixel point of the privacy CT image, a two-dimensional rotation transformation is performed on the privacy CT image to obtain a transformed privacy CT image; The transformed private CT image is subjected to denoising by using Gaussian smoothing to obtain a denoised private CT image; Performing two-dimensional fast Fourier transform on the denoised privacy CT image to achieve frequency domain enhancement and obtain a geometrically enhanced privacy CT image; The grayscale value of each pixel of the privacy CT image is nonlinearly transformed using the gamma correction method, and then the nonlinear transformation result is adjusted using exponential decay to obtain the privacy CT image with enhanced intensity. The privacy CT image after intensity enhancement is decomposed into high-frequency components and low-frequency components by wavelet decomposition; Gaussian noise is added to the sensitive area corresponding to the high-frequency component for privacy protection, and the blurred high-frequency component is obtained; Perform inverse wavelet transform on the low-frequency component and the blurred high-frequency component to obtain the privacy CT image with enhanced intensity; The geometrically enhanced privacy CT image and the intensity enhanced privacy CT image are weighted fused, and the fusion result is nonlinearly adjusted to obtain the extended CT image.

3. The method for segmenting CT images of ENT department according to claim 2, characterized in that: The selecting of the basic segmentation architecture, the receiving of the segmentation annotations for the plurality of extended CT images, and the generating of the plurality of annotated CT images specifically include the following steps: Selecting a CT image from a plurality of extended CT images as a current image; Use the pre-trained U-Net to segment the current image and obtain a segmentation probability map; Introducing manual annotation, based on the results of multiple annotators, by calculating the confidence intervals of different annotation schemes to generate optimized manual annotations; The optimized manual annotation is converted into a discrete region division image, the confidence is dynamically calculated according to whether the discrete region division image is a boundary region, and the confidence is converted into a weight to obtain an artificial confidence weight; According to the probability value of the segmentation probability map, the corresponding confidence is assigned to obtain the intelligent segmentation confidence, and the intelligent segmentation confidence is converted into a weight to obtain the model confidence weight; The segmentation probability map is weighted with the model confidence weight to obtain a first weighted result, the manual annotation is weighted with the manual confidence weight to obtain a second weighted result, and the current image is fused with the corresponding first weighted result and second weighted result to obtain a fused labeled image; The fused annotated image is optimized at multiple scales to obtain the final annotated CT image.

4. The method for segmenting CT images of ENT department according to claim 3, characterized in that: The selecting of the basic segmentation architecture, training and verifying the basic segmentation architecture based on the plurality of annotated CT images, and constructing the image segmentation model specifically include the following steps: Select the basic segmentation architecture from multiple preset network architectures; Get the data partition ratio; According to the data division ratio, the plurality of annotated CT images are divided into a model training set and a model verification set; Based on the model training set and the model verification set, the basic segmentation architecture is trained and verified to construct an image segmentation model.

5. The method for segmenting CT images of ENT department according to claim 4, characterized in that: The training and verification of the basic segmentation architecture based on the model training set and the model verification set, and the construction of the image segmentation model specifically include the following steps: Divide the model training set into multiple small batches, each batch contains several training samples; Perform data augmentation operations on training samples to obtain enhanced images; The enhanced image is input into the basic segmentation architecture for forward propagation to perform segmentation prediction and auxiliary task prediction, thereby obtaining the classification result and the target boundary contour; The main task uses Dice loss to evaluate the segmentation result, and the auxiliary task uses cross entropy loss to evaluate the boundary detection result. The Dice loss and cross entropy loss are weighted summed, and regularization loss is introduced to obtain a mixed loss function. Input the hybrid loss function into the Adam optimizer for optimization, set the weight decay strategy and learning parameters of the Adam optimizer; train the basic segmentation architecture by updating the weights and learning parameters to minimize the loss, and obtain the trained basic segmentation architecture; After the training is completed, the validation set is used to verify the performance of the trained basic segmentation architecture, and the image segmentation model is obtained after the verification is completed.

6. The method for segmenting CT images of ENT department according to claim 5, characterized in that: The step of inputting the enhanced image into the basic segmentation architecture for forward propagation to perform segmentation prediction and auxiliary task prediction to obtain the classification result and the target boundary contour specifically includes the following steps: Step S102401, the basic segmentation architecture is composed of an encoder and a decoder, and the encoder and the decoder are each provided with a number of layers; Step S102402, inputting the enhanced image into the first layer encoder, and extracting the local feature map of the enhanced image through the convolutional neural network; Step S102403, sending the multi-scale local detail features into the Transformer module for global modeling to capture long-range dependencies and obtain a global feature map; Step S102404, performing global average pooling on the local feature map and the global feature map respectively to obtain the average value of each channel; Step S102405, inputting the average value of each channel into a shared fully connected network to generate coefficients for dynamically adjusting the weights of local features and global features; Step S102406, compressing the coefficients of dynamically adjusting the weights of local features and global features to a range of [0, 1] through a Sigmoid activation function to obtain local feature weights and global feature weights; Step S102407, each channel of the local feature map is multiplied by the corresponding weight to achieve dynamic adjustment of the local features to obtain a weighted local feature map; Step S102408, each channel of the global feature map is multiplied by the corresponding weight to achieve dynamic adjustment of the global feature to obtain a weighted global feature map; Step S102409, concatenating the local feature map and the global feature map in the channel dimension, and performing channel compression through a 1×1 convolution operation to obtain a fused feature map; Step S102410, using the fused feature map as the input of the next layer decoder, repeating steps S102401 to S102410 several times in an iterative manner to obtain the final fused feature; Step S102411, sending the final fused features to the decoder, performing a 3×3 convolution operation on the final fused features to obtain a convolution feature map; Step S102412, performing global average pooling on the convolution feature map to obtain a global pooling result; Step S102413, passing the global pooling result through a fully connected layer and a Sigmoid activation function to generate channel weights; Step S102414, performing a weighted operation on the convolution feature map and the channel weights channel by channel to obtain the first layer decoder output; Step S102415, using dense skip connection to splice the output of the previous layer decoder and the output of the corresponding layer encoder in the channel dimension as the input of the next layer decoder, and iteratively repeating steps S102411 to S102414 for several times to obtain the final output of the decoder; Step S102416: The decoder finally outputs a segmentation map with the same resolution as the input image after an upsampling operation, and obtains a high-resolution feature map of multi-scale feature aggregation. The high-resolution feature map of multi-scale feature aggregation is sent to the classification head to generate a category probability distribution and obtain a classification result. Step S102417, take the output of the middle layer decoder to perform boundary detection operation, generate a boundary probability image, and obtain the target boundary contour.

7. The method for segmenting CT images of ENT department according to claim 6, characterized in that: The step of obtaining a target CT image of the ENT department, performing image preprocessing and initial segmentation processing on the target CT image, and generating a plurality of initial segmented images specifically comprises the following steps: Acquire target CT images of ENT; Performing image preprocessing including noise removal, image enhancement, resolution adjustment and standardization on the target CT image to generate a standard CT image; Performing initial segmentation processing on the standard CT image to generate a plurality of initial segmentation images.

8. The method for segmenting CT images of ENT department according to claim 7, characterized in that: The initial segmentation process of the standard CT image to generate a plurality of initial segmented images specifically comprises the following steps: Identifying segmentation candidate regions of the standard CT image; According to the segmentation candidate region, the standard CT image is intercepted to obtain a candidate CT image; A preset image input size is determined, and the candidate CT image is segmented into different sizes to obtain a plurality of initial segmented images.

9. The method for segmenting CT images of ENT department according to claim 8, characterized in that: The step of importing the plurality of initial segmented images into the image segmentation model, deriving a plurality of model segmented images, and performing fine adjustment to generate a plurality of fine segmented images specifically comprises the following steps: Importing a plurality of the initial segmented images into the image segmentation model; Export multiple model segmentation images; Performing erosion, expansion, opening and closing operations on the plurality of model segmented images to repair holes and / or isolated areas, and generating a plurality of repaired segmented images; The multiple repaired segmented images are subjected to a refinement process for smooth edges to generate multiple refined segmented images.

10. An ENT CT examination image segmentation system, the system being applied to the ENT CT examination image segmentation method according to any one of claims 1 to 9, characterized in that: The system comprises a public image processing unit, a model training and verification unit, an image preprocessing and segmentation unit, and a segmentation fine adjustment unit, wherein: A public image processing unit, used to collect a plurality of public CT images of otolaryngology based on big data technology, perform privacy protection, expansion and annotation processing on the plurality of public CT images, and generate a plurality of annotated CT images; A model training and verification unit, used for selecting a basic segmentation architecture, training and verifying the basic segmentation architecture based on the plurality of annotated CT images, and constructing an image segmentation model; An image preprocessing and segmentation unit is used to obtain a target CT image of the otolaryngology department, perform image preprocessing and initial segmentation processing on the target CT image, and generate a plurality of initial segmented images; The segmentation fine adjustment unit is used to import the multiple initial segmentation images into the image segmentation model, export multiple model segmentation images, and perform fine adjustments to generate multiple fine segmentation images.

Images