Gynecological tumor image processing system and method

Abstract

The present application discloses a gynecological tumor image processing system and method thereof, which uses imaging technology to collect gynecological tumor images, and introduces image processing and analysis algorithms at the back end to automatically analyze gynecological tumor images to determine the type of gynecological tumor. In this way, automatic diagnosis and suggestion output of gynecological tumor images can be achieved, providing auxiliary decision-making for doctors.

Description

Technical Field

[0001] The present application relates to the field of image processing, and more specifically, to a gynecological tumor image processing system and method thereof. Background Art

[0002] Gynecological tumors refer to various benign and malignant tumors that occur in the female reproductive system. Common gynecological tumors include cervical cancer, ovarian cancer, endometrial cancer, endometriosis, etc. Early diagnosis of these tumors is crucial for treatment and prognosis.

[0003] However, traditional gynecological tumor diagnosis usually requires imaging examinations, which use imaging technology to obtain images of gynecological tumors and rely on the doctor's subjective diagnosis to assess the presence and status of gynecological tumors. However, different doctors may have different diagnostic results, resulting in inconsistency and reduced reliability of diagnostic results.

[0004] In recent years, image processing and machine learning technologies have been widely used in the field of medical imaging, providing new possibilities for the early diagnosis and treatment of gynecological tumors. However, traditional gynecological tumor image processing systems can only provide surface information of the tumor and lack detailed analysis of the internal structure and tissue of the tumor, which limits the accurate assessment of the nature and extent of the tumor. In addition, when faced with complex image situations, traditional gynecological tumor image processing systems often have a high misdiagnosis rate, which may lead to delayed treatment or unnecessary surgery.

[0005] Therefore, an optimized gynecological tumor image processing system is desired. Summary of the invention

[0006] In order to solve the above technical problems, this application is proposed. The embodiment of this application provides a gynecological tumor image processing system and method thereof, which uses imaging technology to collect gynecological tumor images, and introduces image processing and analysis algorithms at the back end to automatically analyze gynecological tumor images to determine the type of gynecological tumors. In this way, automatic diagnosis and suggestion output of gynecological tumor images can be achieved, providing auxiliary decision-making for doctors.

[0007] According to one aspect of the present application, a gynecological tumor image processing system is provided, comprising:

[0008] An image acquisition module, used to acquire gynecological tumor images;

[0009] An image preprocessing module, used for performing image preprocessing on the gynecological tumor image to obtain a preprocessed gynecological tumor image;

[0010] An image semantic segmentation module, used for performing image segmentation on the pre-processed gynecological tumor image based on edge detection to obtain a region of interest image and a background image;

[0011] An image semantic feature analysis module, used for performing image feature analysis on the image of the region of interest to obtain semantic features of the tumor image;

[0012] The gynecological tumor type detection module is used to determine the type label of the gynecological tumor based on the semantic features of the tumor image.

[0013] According to another aspect of the present application, a gynecological tumor image processing method is provided, which includes:

[0014] Acquire gynecological tumor images;

[0015] Performing image preprocessing on the gynecological tumor image to obtain a preprocessed gynecological tumor image;

[0016] Performing image segmentation on the preprocessed gynecological tumor image based on edge detection to obtain a region of interest image and a background image;

[0017] Performing image feature analysis on the region of interest image to obtain semantic features of the tumor image;

[0018] Based on the semantic features of the tumor image, a type label of the gynecological tumor is determined.

[0019] Compared with the prior art, the present application provides a gynecological tumor image processing system and method, which uses imaging technology to collect gynecological tumor images, and introduces image processing and analysis algorithms at the back end to automatically analyze gynecological tumor images to determine the type of gynecological tumor. In this way, automatic diagnosis and suggestion output of gynecological tumor images can be achieved, providing auxiliary decision-making for doctors. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0021] Figure 1 is a block diagram of a gynecological tumor image processing system according to an embodiment of the present application;

[0022] Figure 2 is a system architecture diagram of a gynecological tumor image processing system according to an embodiment of the present application;

[0023] Figure 3is a block diagram of the training phase of the gynecological tumor image processing system according to an embodiment of the present application;

[0024] Figure 4 is a block diagram of an image semantic feature analysis module in a gynecological tumor image processing system according to an embodiment of the present application;

[0025] Figure 5 Flow chart of a gynecological tumor image processing method according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0027] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0028] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0029] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0030] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0031] Traditional gynecological tumor diagnosis usually requires imaging examinations, which use imaging technology to obtain images of gynecological tumors and rely on the doctor's subjective diagnosis to evaluate whether there is a gynecological tumor and the status of the gynecological tumor. However, different doctors may have different diagnostic results, resulting in inconsistency and reduced reliability of the diagnostic results. In recent years, image processing and machine learning technologies have been widely used in the field of medical imaging, providing new possibilities for the early diagnosis and treatment of gynecological tumors. However, traditional gynecological tumor image processing systems can often only provide surface information of the tumor and lack detailed analysis of the internal structure and tissue of the tumor, which limits the accurate assessment of the nature and extent of the tumor. In addition, traditional gynecological tumor image processing systems often have a high misdiagnosis rate when faced with complex image situations, which may lead to delayed treatment or unnecessary surgery. Therefore, an optimized gynecological tumor image processing system is desired.

[0032] In the technical solution of the present application, a gynecological tumor image processing system is proposed. Figure 1 4 is a block diagram of a gynecological tumor image processing system according to an embodiment of the present application. Figure 2 FIG. 1 is a system architecture diagram of a gynecological tumor image processing system according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to an embodiment of the present application, the gynecological tumor image processing system 300 includes: an image acquisition module 310, used to acquire a gynecological tumor image; an image preprocessing module 320, used to perform image preprocessing on the gynecological tumor image to obtain a preprocessed gynecological tumor image; an image semantic segmentation module 330, used to perform image segmentation on the preprocessed gynecological tumor image based on edge detection to obtain a region of interest image and a background image; an image semantic feature analysis module 340, used to perform image feature analysis on the region of interest image to obtain a semantic feature of the tumor image; and a gynecological tumor type detection module 350, used to determine a type label of the gynecological tumor based on the semantic feature of the tumor image.

[0033] In particular, the image acquisition module 310 is used to obtain gynecological tumor images. In the technical solution of the present application, gynecological tumor images are usually obtained through medical imaging technology (such as ultrasound, computed tomography (CT), magnetic resonance imaging (MRI), etc.) for detecting, diagnosing and monitoring gynecological tumors. These images can provide detailed information about gynecological tumors and provide doctors with the visual information needed to evaluate and diagnose gynecological tumors. These images are very important for formulating individualized treatment plans, monitoring disease progression and evaluating treatment effects.

[0034] In particular, the image preprocessing module 320 is used to perform image preprocessing on the gynecological tumor image to obtain a preprocessed gynecological tumor image. Considering that the gynecological tumor image may be interfered by noise from different sources during the acquisition process, such as electronic noise, artifacts, etc. Therefore, it is necessary to perform denoising and enhancement processing on the image before feature extraction, that is, perform image preprocessing on the gynecological tumor image to obtain a preprocessed gynecological tumor image to improve the quality and analyzability of the image. In addition, during the image preprocessing process, the difference between the tumor and the surrounding tissue can be made more obvious by adjusting the brightness and contrast of the image.

[0035] It is worth noting that image preprocessing refers to a series of preprocessing steps on the original image before image analysis and processing to improve image quality, reduce noise, enhance features, etc. The following are some common image preprocessing steps: Image denoising: Use filters (such as Gaussian filters, median filters, etc.) to remove noise from the image to reduce interference in subsequent processing; Image enhancement: Enhance the visual effect of the image and the visibility of features by adjusting parameters such as contrast, brightness, color balance, etc. Common enhancement methods include histogram equalization, contrast stretching, etc.; image resizing: adjust the image size according to specific needs to adapt to the requirements of subsequent processing algorithms or display devices; image cropping: crop part of the image, remove irrelevant areas or adjust the composition of the image as needed; image rotation and flipping: rotate, flip, and other operations on the image as needed to adjust the direction or viewing angle of the image; image smoothing: use smoothing filters (such as mean filters, Gaussian filters, etc.) to reduce noise and details in the image to make the image smoother; image edge detection: use edge detection algorithms (such as Sobel operators, Canny operators, etc.) to detect edges in the image to extract edge information; image grayscale or binarization: convert color images into grayscale images or binary images to facilitate subsequent image analysis and processing.

[0036] In particular, the image semantic segmentation module 330 is used to perform image segmentation on the pre-processed gynecological tumor image based on edge detection to obtain a region of interest image and a background image. Considering that there will be a large amount of tissue interference in the tumor image, in order to further improve the accuracy of gynecological tumor feature recognition, in the technical solution of the present application, the pre-processed gynecological tumor image is further segmented based on edge detection to obtain a region of interest image and a background image. In this way, the region of interest (i.e., the tumor area) in the image can be distinguished from the background area, that is, the gynecological tumor can be extracted from the surrounding tissues and structures to facilitate subsequent feature extraction and classification.

[0037] It is worth noting that edge detection is a commonly used technique in image processing to detect boundaries or contours between objects in an image. Edges usually indicate where features such as brightness, color, or texture in an image change suddenly. The purpose of edge detection is to find the location of these changes. Common edge detection algorithms include: Roberts operator, Sobel operator, Prewitt operator, and Canny edge detection. These algorithms have different performance and applicable scenarios in edge detection. The selection of a suitable edge detection algorithm depends on the characteristics of the image, the noise situation, and the application requirements. Edge detection can be used for tasks such as image segmentation, target detection, and edge tracking, providing a basis for further image analysis and processing.

[0038] Accordingly, in a possible implementation, the pre-processed gynecological tumor image can be segmented based on edge detection to obtain a region of interest image and a background image through the following steps, for example: pre-processing the gynecological tumor image, including denoising, enhancement, resizing and other steps to improve the image quality and edge detection effect; using an appropriate edge detection algorithm (such as the Canny operator) to perform edge detection on the pre-processed image to find the boundary in the image; performing edge enhancement processing on the edge image, and morphological operations (such as dilation and corrosion) can be used to connect broken edges, fill holes or remove small noise; performing edge connection operations on the enhanced edge image to connect adjacent edges into continuous boundaries to obtain a more complete region of interest boundary; filling the region of interest according to the obtained boundary to obtain a region of interest image; obtaining a background image by subtracting the region of interest image from the original image. This can be achieved by setting the pixels in the region of interest to the background color or by pixel-level differences; post-processing the segmentation results, including operations such as noise removal and boundary smoothing to obtain more accurate and reliable segmentation results.

[0039] In particular, the image semantic feature analysis module 340 is used to perform image feature analysis on the region of interest image to obtain the semantic features of the tumor image. In particular, in a specific example of the present application, Figure 4As shown, the image semantic feature analysis module 340 includes: a tumor image multi-scale feature extraction unit 341, which is used to perform image multi-scale feature extraction on the region of interest image through a tumor image multi-scale feature extractor based on a deep neural network model to obtain a tumor image shallow feature map, a tumor image middle-layer feature map and a tumor image deep feature map; a tumor image middle-layer semantic feature propagation and fusion unit 342, which is used to perform semantic feature propagation and fusion on the tumor image middle-layer feature map and the tumor image deep feature map to obtain a semantically fused tumor image middle-layer feature map; a tumor image shallow-layer semantic feature propagation and fusion unit 343, which is used to perform semantic feature propagation and fusion on the semantically fused tumor image middle-layer feature map and the tumor image shallow-layer feature map to obtain a semantic multi-scale fused tumor image shallow-layer feature map as the tumor image semantic feature.

[0040] Specifically, the tumor image multi-scale feature extraction unit 341 is used to perform image multi-scale feature extraction on the region of interest image through a tumor image multi-scale feature extractor based on a deep neural network model to obtain a shallow feature map of the tumor image, a middle feature map of the tumor image, and a deep feature map of the tumor image. That is, a convolutional neural network model with excellent performance in implicit feature extraction of images is used to perform feature mining on the region of interest image. In particular, considering that when extracting features of gynecological tumors for type detection, it is necessary not only to focus on the deep implicit feature information about the gynecological tumor in the region of interest image, but also to focus on the feature information such as color, texture and shape of the gynecological tumor. The pyramid network mainly solves the multi-scale problem in target detection. It can simultaneously utilize the semantic information of low-level features and high-level features, and achieve a good expression effect by fusing the features of these different layers. Based on this, in the technical solution of the present application, the region of interest image is passed through a tumor image multi-scale feature extractor based on a pyramid network to obtain a shallow feature map of the tumor image, a middle feature map of the tumor image, and a deep feature map of the tumor image. By extracting feature maps of different levels in the region of interest image, the image information can be expressed and analyzed at multiple levels, so that the region of interest image of the gynecological tumor area can be better understood, providing more accurate and comprehensive information for subsequent gynecological tumor type detection tasks.

[0041] It is worth noting that the Pyramid Network is a deep learning network structure designed to process image data with information of different scales. Its concept and structure are derived from the concept in pyramid image processing, where images are represented by pyramids at multiple scales to capture details at different levels. The main goal of the Pyramid Network is to address the limitations of traditional convolutional neural networks (CNNs) in processing multi-scale information. Traditional CNNs usually only focus on features of fixed scales when performing convolution and pooling operations, while ignoring information at other scales. This may lead to ineffective detection and representation of objects or features of different scales. The Pyramid Network solves this problem by introducing multiple branches or convolutional layers of multiple scales. Each branch or convolutional layer of a scale can specialize in processing features of the corresponding scale. Typically, low-level branches process details of smaller scales, while high-level branches process global features of larger scales. In this way, the Pyramid Network can capture features at different scales simultaneously, thereby improving the detection and representation capabilities of multi-scale objects.

[0042] Specifically, the tumor image mid-level semantic feature propagation and fusion unit 342 is used to perform semantic feature propagation and fusion on the tumor image mid-level feature map and the tumor image deep-level feature map to obtain a semantically fused tumor image mid-level feature map. In particular, in a specific example of the present application, a first semantic propagation and fusion module is used to fuse the tumor image mid-level feature map and the tumor image deep-level feature map to obtain the semantically fused tumor image mid-level feature map. Among them, the first semantic propagation and fusion module can learn and propagate the deep semantic feature information of the region of interest image of the gynecological tumor, and fuse it into the tumor image mid-level feature map, so that the fused semantically fused tumor image mid-level feature map has richer semantic information about gynecological tumor cells, so as to improve the subsequent detection accuracy of the type of gynecological tumors.

[0043] It is worth mentioning that in other specific examples of the present application, semantic feature propagation and fusion can be performed on the mid-layer feature map of the tumor image and the deep feature map of the tumor image in other ways to obtain a semantically fused mid-layer feature map of the tumor image, for example: using a pre-trained convolutional neural network (such as ResNet, VGG, etc.) to extract features of the tumor image to obtain a layer feature map and a deep feature map. The layer feature map is usually the output of a shallow convolutional layer, and the deep feature map is the output of a deep convolutional layer; the deep feature map is propagated to the layer feature map to fuse semantic information. A commonly used method is to use an upsampling operation (such as deconvolution or interpolation) to adjust the size of the deep feature map to the same as the layer feature map, and then perform element-by-element addition or cascade operation with the layer feature map to achieve the propagation of semantic features; the propagated deep feature map is fused with the layer feature map to obtain a semantically fused mid-layer feature map of the tumor image. The fusion can be performed by simple element-by-element addition, cascading, or using attention mechanisms to perform weighted fusion according to the importance of the features; necessary post-processing operations such as normalization and activation functions are performed on the fused feature maps to enhance the expressiveness and robustness of the features.

[0044] Specifically, the tumor image shallow semantic feature propagation fusion unit 343 is used to perform semantic feature propagation fusion on the semantically fused tumor image middle-layer feature map and the tumor image shallow-layer feature map to obtain a semantic multi-scale fused tumor image shallow-layer feature map as the tumor image semantic feature. In particular, in a specific example of the present application, a second semantic propagation fusion module is used to fuse the semantically fused tumor image middle-layer feature map and the tumor image shallow-layer feature map to obtain the semantic multi-scale fused tumor image shallow-layer feature map. It should be understood that since the tumor image shallow-layer feature map usually contains low-level image features, such as edges, textures, etc., these features are very important for describing the details and local information of the tumor image, which is beneficial for the detection of the type of gynecological tumors. Therefore, in the technical solution of the present application, a second semantic propagation fusion module is used to fuse the semantically fused tumor image middle-layer feature map and the tumor image shallow-layer feature map to obtain a semantic multi-scale fused tumor image shallow-layer feature map. It is worth mentioning that here, the second semantic propagation fusion module can learn and propagate the feature information in the middle-level feature map of the semantic fusion tumor image, and fuse it into the shallow feature map of the tumor image, so that the fused semantic multi-scale fusion tumor image shallow feature map has more comprehensive and accurate feature expression ability and discrimination about gynecological tumor cells, so as to better capture the important features of tumor images at different scales, which helps to improve the accuracy and performance of the gynecological tumor image processing system.

[0045] It is worth mentioning that in other specific examples of the present application, the image feature analysis of the region of interest image can also be performed in other ways to obtain the semantic features of the tumor image, for example: necessary preprocessing of the region of interest image, such as denoising, enhancement, resizing, etc., to improve the effect of subsequent feature analysis; extracting image features related to tumor features from the region of interest image. Common features include: shape features: such as the size, perimeter, area, roundness, etc. of the tumor; texture features: such as the texture roughness, contrast, energy, entropy, etc. of the tumor; color features: such as the color histogram, color mean, color variance, etc. of the tumor; edge features: such as the shape and curvature of the edge of the tumor. Feature extraction can use various computer vision and image processing techniques, such as gray-level co-occurrence matrix, wavelet transform, histogram, etc.; according to the specific task requirements and the importance of the features, the extracted features are selected and screened to reduce redundancy and noise and improve the accuracy of classification or recognition; the selected features are properly represented, such as standardizing, normalizing or reducing the dimension of the feature vector so that subsequent classification, clustering or machine learning algorithms can process it; the represented features can be analyzed using statistical methods, machine learning algorithms or deep learning models to understand the semantic characteristics of tumor images. This may include analyzing the relationship between different features, the importance of features, and the accuracy of classification or recognition.

[0046] In particular, the gynecological tumor type detection module 350 is used to determine the type label of the gynecological tumor based on the semantic features of the tumor image. In particular, in a specific example of the present application, the semantic multi-scale fusion tumor image shallow feature map is passed through a classifier to obtain a classification result, and the classification result is used to represent the type label of the gynecological tumor. In the technical solution of the present application, the classification label of the classifier is the type label of the gynecological tumor. Therefore, after obtaining the classification result, the automatic detection and judgment of the type of gynecological tumor can be realized based on the classification result, providing auxiliary decision-making basis for doctors. Specifically, the semantic multi-scale fusion tumor image shallow feature map is expanded into a semantic multi-scale fusion tumor image shallow feature vector based on a row vector or a column vector; the semantic multi-scale fusion tumor image shallow feature vector is fully connected encoded using multiple fully connected layers of the classifier to obtain an encoded classification feature vector; and the encoded classification feature vector is passed through the Softmax classification function of the classifier to obtain the classification result.

[0047] A classifier is a machine learning model or algorithm that is used to classify input data into different categories or labels. A classifier is part of supervised learning and performs classification tasks by learning the mapping relationship from input data to output categories.

[0048] A fully connected layer is a common type of layer in a neural network. In a fully connected layer, each neuron is connected to all neurons in the previous layer, and each connection has a weight. This means that each neuron in a fully connected layer receives inputs from all neurons in the previous layer, and performs a weighted sum of these inputs through weights, and then passes the result to the next layer.

[0049] The Softmax classification function is a commonly used activation function for multi-classification problems. It converts each element of the input vector into a probability value between 0 and 1, and the sum of these probability values ​​is equal to 1. The Softmax function is often used in the output layer of a neural network, especially for multi-classification problems, because it can map the network output to the probability distribution of each category. During the training process, the output of the Softmax function can be used to calculate the loss function and update the network parameters through the back propagation algorithm. It is worth noting that the output of the Softmax function does not change the relative size relationship between the elements, but only normalizes them. Therefore, the Softmax function does not change the characteristics of the input vector, but only converts it into a probability distribution form.

[0050] It is worth mentioning that in other specific examples of the present application, the type label of gynecological tumors can also be determined based on the semantic features of the tumor image in other ways, for example: collect a data set containing gynecological tumor images and corresponding labels. Ensure that each image in the data set has the correct gynecological tumor type label; use a pre-trained convolutional neural network (such as ResNet, VGG, etc.) to extract features from gynecological tumor images to obtain a semantic feature representation of the image. The previously mentioned layer feature map, deep feature map, or multi-scale fused feature map can be used as input; use a machine learning algorithm (such as a support vector machine, random forest, neural network, etc.) to train and learn the extracted semantic features to establish a classification model; divide the data set into a training set and a test set. Usually, most of the data is used for training, and a small part of the data is used for testing to evaluate the performance of the model; use the training set to train the classification model. According to the selected machine learning algorithm, adjust the parameters and hyperparameters of the model to improve the classification performance; use the test set to evaluate the performance of the trained classification model. Common evaluation indicators include accuracy, precision, recall, F1 score, etc. Use the trained classification model to predict new tumor images and obtain the type label of gynecological tumors. Input the semantic features of the image into the classification model, and the model will output the predicted type label.

[0051] It should be understood that before using the above-mentioned neural network model for inference, it is necessary to train the multi-scale feature extractor of the tumor image based on the pyramid network, the first semantic propagation fusion module, the second semantic propagation fusion module and the classifier. That is, according to the gynecological tumor image processing system 300 of the present application, it also includes a training stage 400 for training the multi-scale feature extractor of the tumor image based on the pyramid network, the first semantic propagation fusion module, the second semantic propagation fusion module and the classifier.

[0052] Figure 3 FIG. 4 is a block diagram of the training phase of the gynecological tumor image processing system according to an embodiment of the present application. Figure 3 As shown, according to the embodiment of the present application, the gynecological tumor image processing system 300 includes: a training stage 400, including: a training data acquisition module 410, used to obtain training data, the training data including training gynecological tumor images, and the true value of the type label of the gynecological tumor; a training image preprocessing module 420, used to perform image preprocessing on the training gynecological tumor image to obtain a training preprocessed gynecological tumor image; a training image semantic segmentation module 430, used to perform image segmentation on the training preprocessed gynecological tumor image based on edge detection to obtain a training region of interest image and a training background image; a training image multi-scale semantic feature extraction module 440, used to pass the training region of interest image through the pyramid network-based tumor image multi-scale feature extractor to obtain a training tumor image shallow feature map, a training tumor image middle feature map and a training tumor image deep feature map; a training image middle-layer semantic feature propagation module 450, used to fuse the training tumor image with the first semantic propagation fusion module The invention relates to a method for extracting a multi-scale feature map of a tumor image from a pyramid network, and a method for extracting a multi-scale feature map of a tumor image from a pyramid network. The method comprises the following steps: integrating the middle-layer feature map of the image and the deep-layer feature map of the training tumor image to obtain the middle-layer feature map of the training semantic fusion tumor image; a training image shallow-layer semantic feature propagation module 460, which is used to use the second semantic propagation fusion module to fuse the middle-layer feature map of the training semantic fusion tumor image and the shallow-layer feature map of the training tumor image to obtain the shallow-layer feature map of the training semantic multi-scale fusion tumor image; and a classification loss module 470, which is used to pass the shallow-layer feature map of the training semantic multi-scale fusion tumor image through a classifier to obtain a classification loss function value; and a model training module 480, which is used to train the multi-scale feature extractor of the tumor image based on the pyramid network, the first semantic propagation fusion module, the second semantic propagation fusion module and the classifier based on the classification loss function value and through back propagation of gradient descent, wherein in each round of iteration of the training process, the shallow-layer feature vector of the training semantic multi-scale fusion tumor image obtained by expanding the shallow-layer feature map of the training semantic multi-scale fusion tumor image is trained and optimized.

[0053] In particular, in the technical solution of the present application, when the training region of interest image passes through a pyramid network-based tumor image multi-scale feature extractor, the pyramid network-based tumor image multi-scale feature extractor will extract the local image semantic association features at the corresponding scale of the convolution kernel based on the convolution kernels of each layer, thereby, each feature value of the training tumor image shallow feature map, the training tumor image middle feature map and the training tumor image deep feature map has a feature expression super-resolution corresponding to the local image semantic association features at the scale of each convolution kernel, thereby, when the semantic propagation fusion module is used to fuse the training tumor image shallow feature map, the training tumor image middle feature map and the training tumor image deep feature map, the training semantic multi-scale fused tumor image shallow feature map will have the super-resolution expression characteristics of the multi-dimensional context under the local spatial context association of image semantic features of different scales and different depths, which makes the overall feature distribution of the training semantic multi-scale fused tumor image shallow feature map have more significant inconsistency and instability, thereby affecting the stability of the training semantic multi-scale fused tumor image shallow feature map through the classifier for classification training.

[0054] Based on this, when the shallow feature vector of the training semantic multi-scale fusion tumor image obtained after expanding the shallow feature map of the training semantic multi-scale fusion tumor image is classified through a classifier, the shallow feature vector of the training semantic multi-scale fusion tumor image is trained and optimized in each iteration.

[0055] In a specific example of the present application, in each iteration of the training process, the shallow feature vector of the training semantic multi-scale fusion tumor image obtained by expanding the shallow feature map of the training semantic multi-scale fusion tumor image is trained and optimized using the following optimization formula to obtain an optimized shallow feature vector of the training semantic multi-scale fusion tumor image;

[0056] Wherein, the optimization formula is:

[0057]

[0058] Where V is the shallow feature vector of the training semantic multi-scale fusion tumor image, v i is the eigenvalue of the ith position of the shallow feature vector V of the training semantic multi-scale fusion tumor image, ||·||1 and ||·||2 are the 1-norm and 2-norm of the shallow feature vector V of the training semantic multi-scale fusion tumor image, respectively, L is the length of the shallow feature vector V of the training semantic multi-scale fusion tumor image, and α is i The associated weight hyperparameter, v' i is the eigenvalue of the ith position of the shallow feature vector of the optimized training semantic multi-scale fusion tumor image.

[0059] Here, the overall feature distribution of the shallow feature vector V of the training semantic multi-scale fusion tumor image is represented by the structural consistency and stability under the rigid structure of absolute distance and the non-rigid structure of spatial distance, so that the global feature distribution of the shallow feature vector V of the training semantic multi-scale fusion tumor image has a certain repeatability for local pattern changes, so that when the shallow feature vector V of the training semantic multi-scale fusion tumor image is classified by the classifier, the global feature distribution is robust to the scale and rotation changes of the weight matrix of the classifier, thereby improving the stability of classification training. In this way, automatic diagnosis and suggestion output of gynecological tumor images can be achieved to determine the type of gynecological tumors and provide doctors with auxiliary decision-making basis.

[0060] As described above, the gynecological tumor image processing system 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a gynecological tumor image processing algorithm. In a possible implementation, the gynecological tumor image processing system 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the gynecological tumor image processing system 300 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the gynecological tumor image processing system 300 can also be one of the many hardware modules of the wireless terminal.

[0061] Alternatively, in another example, the gynecological tumor image processing system 300 and the wireless terminal may also be separate devices, and the gynecological tumor image processing system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0062] Furthermore, a gynecological tumor image processing method is also provided.

[0063] Figure 5 FIG. 1 is a flow chart of a gynecological tumor image processing method according to an embodiment of the present application. Figure 5 As shown, according to the embodiment of the present application, the gynecological tumor image processing method includes the following steps: S1, acquiring a gynecological tumor image; S2, performing image preprocessing on the gynecological tumor image to obtain a preprocessed gynecological tumor image; S3, performing image segmentation on the preprocessed gynecological tumor image based on edge detection to obtain a region of interest image and a background image; S4, performing image feature analysis on the region of interest image to obtain semantic features of the tumor image; S5, determining the type label of the gynecological tumor based on the semantic features of the tumor image.

[0064] In summary, the gynecological tumor image processing method according to the embodiment of the present application is explained, which uses imaging technology to collect gynecological tumor images, and introduces image processing and analysis algorithms at the back end to automatically analyze gynecological tumor images to determine the type of gynecological tumor. In this way, automatic diagnosis and suggestion output of gynecological tumor images can be achieved, providing auxiliary decision-making for doctors.

[0065] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A gynecological tumor image processing system, characterized in that: include: An image acquisition module, used to acquire gynecological tumor images; An image preprocessing module, used for performing image preprocessing on the gynecological tumor image to obtain a preprocessed gynecological tumor image; An image semantic segmentation module, used for performing image segmentation on the pre-processed gynecological tumor image based on edge detection to obtain a region of interest image and a background image; An image semantic feature analysis module, used for performing image feature analysis on the image of the region of interest to obtain semantic features of the tumor image; A gynecological tumor type detection module, used to determine the type label of the gynecological tumor based on the semantic features of the tumor image; Wherein, the image semantic feature analysis module includes: A tumor image multi-scale feature extraction unit, used to perform image multi-scale feature extraction on the region of interest image through a tumor image multi-scale feature extractor based on a deep neural network model to obtain a tumor image shallow feature map, a tumor image middle feature map and a tumor image deep feature map; A tumor image mid-layer semantic feature propagation and fusion unit, used for performing semantic feature propagation and fusion on the tumor image mid-layer feature map and the tumor image deep-layer feature map to obtain a semantically fused tumor image mid-layer feature map; A tumor image shallow semantic feature propagation and fusion unit, used for performing semantic feature propagation and fusion on the semantic fused tumor image middle-layer feature map and the tumor image shallow-layer feature map to obtain a semantic multi-scale fused tumor image shallow-layer feature map as the tumor image semantic feature; Wherein, the deep neural network model is a pyramid network; The tumor image mid-layer semantic feature propagation and fusion unit is used to: use a first semantic propagation and fusion module to fuse the tumor image mid-layer feature map and the tumor image deep-layer feature map to obtain the semantically fused tumor image mid-layer feature map; The tumor image shallow semantic feature propagation fusion unit is used to: use a second semantic propagation fusion module to fuse the semantic fused tumor image middle-layer feature map and the tumor image shallow feature map to obtain the semantic multi-scale fused tumor image shallow feature map; The gynecological tumor type detection module is used to: pass the shallow feature map of the semantic multi-scale fusion tumor image through a classifier to obtain a classification result, and the classification result is used to represent the type label of the gynecological tumor; Among them, it also includes a training module for training the pyramid network-based tumor image multi-scale feature extractor, the first semantic propagation fusion module, the second semantic propagation fusion module and the classifier; Wherein, the training module includes: A training data acquisition module, used to acquire training data, wherein the training data includes training gynecological tumor images and true values ​​of type labels of the gynecological tumors; A training image preprocessing module, used for performing image preprocessing on the training gynecological tumor image to obtain a training preprocessed gynecological tumor image; A training image semantic segmentation module, used for performing image segmentation on the training pre-processed gynecological tumor image based on edge detection to obtain a training region of interest image and a training background image; A training image multi-scale semantic feature extraction module, used for passing the training region of interest image through the pyramid network-based tumor image multi-scale feature extractor to obtain a training tumor image shallow feature map, a training tumor image middle feature map and a training tumor image deep feature map; A training image mid-level semantic feature propagation module, configured to use the first semantic propagation fusion module to fuse the training tumor image mid-level feature map and the training tumor image deep-level feature map to obtain a training semantic fused tumor image mid-level feature map; a training image shallow semantic feature propagation module, configured to use the second semantic propagation fusion module to fuse the training semantic fused tumor image middle-layer feature map and the training tumor image shallow feature map to obtain a training semantic multi-scale fused tumor image shallow feature map; and A classification loss module, used for passing the shallow feature map of the training semantic multi-scale fusion tumor image through a classifier to obtain a classification loss function value; A model training module, used to train the pyramid network-based tumor image multi-scale feature extractor, the first semantic propagation fusion module, the second semantic propagation fusion module and the classifier based on the classification loss function value and through gradient descent back propagation, wherein in each round of iteration of the training process, the shallow feature vector of the training semantic multi-scale fusion tumor image obtained by expanding the shallow feature map of the training semantic multi-scale fusion tumor image is trained and optimized; In each iteration of the training process, the shallow feature vector of the training semantic multi-scale fusion tumor image obtained by expanding the shallow feature map of the training semantic multi-scale fusion tumor image is trained and optimized by the following optimization formula to obtain the optimized shallow feature vector of the training semantic multi-scale fusion tumor image; Wherein, the optimization formula is: in, is the shallow feature vector of the training semantic multi-scale fusion tumor image, is the shallow feature vector of the training semantic multi-scale fusion tumor image No. The eigenvalues ​​at the positions, and They are respectively the shallow feature vectors of the training semantic multi-scale fusion tumor image The 1-norm and 2-norm of is the shallow feature vector of the training semantic multi-scale fusion tumor image The length of is with The relevant weight hyperparameters, This is the first step of optimizing the shallow feature vector of the semantic multi-scale fusion tumor image. The feature value at each position.

2. A gynecological tumor image processing method, using the gynecological tumor image processing system according to claim 1, characterized in that: include: Acquire gynecological tumor images; Performing image preprocessing on the gynecological tumor image to obtain a preprocessed gynecological tumor image; Performing image segmentation on the preprocessed gynecological tumor image based on edge detection to obtain a region of interest image and a background image; Performing image feature analysis on the region of interest image to obtain semantic features of the tumor image; Based on the semantic features of the tumor image, a type label of the gynecological tumor is determined.

Images