Craniocerebral CT image tumor identification method and system based on computer vision

Abstract

The invention provides a computer vision-based craniocerebral CT image tumor identification method and system, and the method comprises the steps: obtaining a craniocerebral CT image set of a to-be-detected patient, and recognizing a brain tissue region image in the CT image set according to a connected domain matching method; performing image saliency enhancement on the brain tissue region image to obtain an enhanced image; taking a neighborhood window by taking any point on the enhanced image as a center, and calculating a gray average value of all pixel points in the neighborhood window; taking the gray average value of the corresponding pixel points as the output of a central pixel point to obtain a mean value image; obtaining an optimal segmentation threshold value according to the relevancy between the mean value image and the enhanced image; and segmenting the enhanced image by using the optimal segmentation threshold to obtain a segmented image, and inputting the segmented image into a preset tumor recognition model to obtain a recognition result. According to the invention, through saliency enhancement processing, the visual quality of the image is improved, the tumor features are easier to identify, and thus the performance of a subsequent classification model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision, and in particular to a method and system for identifying tumors in cranial CT images based on computer vision. Background Art

[0002] With the rapid development of medical imaging technology, computed tomography (CT) has become one of the indispensable tools in clinical diagnosis. Cranial CT can clearly display the brain tissue structure and adjacent anatomical relationships in different cross sections. Cranial CT is of great significance for the diagnosis of most skull, brain tissue, and scalp diseases. It is widely used in the preliminary examination of central nervous system tumors. Cranial CT can clearly identify the location, number, size, contour structure, tumor density, hemorrhage, calcification, and invasion range of the tumor. However, due to the complex morphology, different sizes, and low contrast with surrounding tissues of the tumor, the traditional manual reading method has certain limitations, which can easily lead to missed diagnosis or misdiagnosis. Therefore, how to use advanced computer vision technology to improve the accuracy and efficiency of tumor recognition in cranial CT images has become a hot topic in current research.

[0003] Computer vision is a science that studies how to make computers "understand" images, involving multiple fields such as image processing, pattern recognition, and machine learning. In recent years, breakthroughs in deep learning technology have provided new opportunities for the development of computer vision. In particular, the emergence of convolutional neural networks (CNNs) has enabled computers to make significant progress in tasks such as image classification and target detection. Image recognition methods based on deep learning can automatically extract multi-level features of images and have strong robustness and generalization capabilities.

[0004] In the field of tumor CT image recognition, researchers have developed a variety of automated recognition methods based on computer vision. These methods usually include the following steps: first, preprocessing the CT images to improve image quality and contrast; second, using deep learning models to extract and classify images; and finally, optimizing the recognition results through post-processing techniques. However, existing methods still face some challenges, such as the need for a large amount of labeled data for model training, and the limited recognition accuracy due to the complexity of the model and the consumption of computing resources.

[0005] The invention patent with application number 202410030706.6 discloses a method for CT image recognition of multi-phase pancreatic tumors based on deep learning. The method includes S1, constructing a CT image data set of pancreatic tumor samples and normal samples and marking image information; S2, constructing a registration model of pancreatic CT image data based on the CT image data set; S3, constructing an image classification model based on a cross transformer model; S4, training the registration model and the image classification model; S5, applying the registration model and the image classification model for CT image recognition. This scheme can diagnose tumor images more accurately. However, the registration model may face the problem of insufficient registration accuracy when processing images of different patients or different scanning conditions. In particular, when the tumor morphology is complex or the organ position changes, the registration result may not be accurate enough, affecting the input quality of the subsequent classification model, and the technology does not use a suitable denoising algorithm, which will lead to the loss of image details or the blurring of important features. Common denoising methods in the prior art, such as mean filtering, median filtering or Gaussian filtering, may not be able to effectively handle complex noise, especially when the tumor boundary is blurred. Summary of the invention

[0006] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a method and system for tumor recognition in cranial CT images based on computer vision, which improves the overall accuracy and reliability of cranial CT image recognition, thereby providing more effective technical support for clinical diagnosis.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for identifying tumors in cranial CT images based on computer vision, comprising:

[0009] Acquire a cranial CT image set of a patient to be tested, and identify a brain tissue region image in the cranial CT image set according to a connected domain matching method;

[0010] Performing image significance enhancement on the brain tissue region image to obtain an enhanced image;

[0011] Taking any point on the enhanced image as the center, take a neighborhood window and calculate the grayscale average of all pixels in the neighborhood window;

[0012] The grayscale average of the corresponding pixels is used as the output of the central pixel to obtain the mean image;

[0013] Obtaining an optimal segmentation threshold according to the correlation between the mean image and the enhanced image;

[0014] Segmenting the enhanced image using the optimal segmentation threshold to obtain a segmented image;

[0015] The segmented image is input into a preset tumor recognition model to obtain a recognition result.

[0016] Preferably, identifying images in a cranial CT image set according to a connected domain labeling method includes:

[0017] Performing threshold segmentation on each image in the CT image set to obtain a binary image;

[0018] Performing standardization processing on each of the binary images to obtain a standardized image;

[0019] Using a connected domain labeling method to perform region recognition on the standardized image to obtain various connected regions;

[0020] Using a depth-first search method to mark each connected region, so that each of the connected regions is assigned a unique identifier;

[0021] Extracting possible brain tissue regions based on image features of connected regions; the image features include: size, shape and contour;

[0022] The possible regions are screened according to a preset brain tissue region feature threshold to obtain the brain tissue region image; the brain tissue region feature threshold includes an area value and a perimeter value.

[0023] Preferably, performing image significance enhancement on the brain tissue region image to obtain an enhanced image includes:

[0024] Based on fuzzy theory, fuzzy enhancement functions of different pixels are constructed; the formula of the fuzzy enhancement function is: Among them, μ dark represents the first fuzzy enhancement function, μ birg represents the second fuzzy enhancement function, μ gray represents the third fuzzy enhancement function, x ij Represents the gray value at the position (i, j);

[0025] Construct a saliency enhancement model based on the fuzzy enhancement functions of different pixels;

[0026] The brain tissue region image is processed according to the significance enhancement model to obtain the enhanced image.

[0027] Preferably, the formula of the significance enhancement model is:

[0028]

[0029] Among them, x i ' j Indicates that when the input pixel is x ijWhen the pixel value after saliency enhancement is performed, ν d represents the first weight, ν b represents the second weight, ν g Represents the third weight.

[0030] Preferably, obtaining an optimal segmentation threshold according to the correlation between the mean image and the enhanced image includes:

[0031] Extracting the grayscale values ​​at the same position on the mean image and the enhanced image to form a grayscale array;

[0032] constructing a segmentation function using the grayscale array;

[0033] Get the preset segmentation array, and continuously adjust the preset segmentation array until the value of the segmentation function is maximized;

[0034] The segmentation array corresponding to the maximum value of the segmentation function is taken as the optimal segmentation threshold.

[0035] Preferably, constructing a segmentation function using the grayscale array includes:

[0036] Get the array probability according to the number of times the grayscale array appears;

[0037] The grayscale distribution probability of the mean image and the enhanced image is obtained according to the array probability; wherein the grayscale distribution probability is:

[0038]

[0039] Among them, μ Tm represents the grayscale distribution probability of the enhanced image, μ Tn Represents the grayscale distribution probability of the mean image, p mn represents the probability of the array, M represents the length of the enhanced image, N represents the width of the enhanced image, and F mn Represents the number of occurrences of the grayscale array (m,n), where m,n=0,1,2,…L;

[0040] A segmentation function is constructed according to the grayscale distribution probability.

[0041] Preferably, the segmentation function is:

[0042]

[0043] Among them, (s, t) represents the initial segmentation array, represents the target grayscale distribution probability, Represents the background grayscale distribution probability.

[0044] Preferably, the method for constructing the tumor recognition model comprises:

[0045] Obtaining a sample segmentation image data set, and dividing the sample segmentation image data set into a training set, a test set, and a validation set;

[0046] Constructing an initial network model based on a CNN framework; the initial network model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a flattening layer, a first fully connected layer, and a second fully connected layer connected in sequence;

[0047] The initial network model is trained, tested and verified by using the training set, the test set and the verification set to obtain a trained tumor recognition model.

[0048] A computer vision-based brain CT image tumor recognition system, comprising:

[0049] A brain tissue image acquisition module is used to acquire a set of cranial CT images of a patient to be tested, and identify a brain tissue region image in the CT image set according to a connected domain matching method;

[0050] An image enhancement module is used to enhance the image significance of the brain tissue area image to obtain an enhanced image;

[0051] A gray value calculation module is used to take a neighborhood window with any point on the enhanced image as the center and calculate the gray value average of all pixels in the neighborhood window;

[0052] The mean image acquisition module is used to take the grayscale average value of the corresponding pixel points as the output of the central pixel point to obtain the mean image;

[0053] A threshold determination module, used to obtain an optimal segmentation threshold according to the correlation between the mean image and the enhanced image;

[0054] A segmentation module, used to segment the enhanced image using the optimal segmentation threshold to obtain a segmented image;

[0055] The recognition module is used to input the segmented image into a preset tumor recognition model to obtain a recognition result.

[0056] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0057] The present invention provides a method and system for identifying tumors in cranial CT images based on computer vision, the method comprising: obtaining a cranial CT image set of a patient to be tested, and identifying a brain tissue region image in the cranial CT image set according to a connected domain matching method; performing image significance enhancement on the brain tissue region image to obtain an enhanced image; taking a neighborhood window with any point on the enhanced image as the center, and calculating the grayscale average of all pixels in the neighborhood window; taking the grayscale average of the corresponding pixels as the output of the central pixel to obtain a mean image; obtaining an optimal segmentation threshold according to the correlation between the mean image and the enhanced image; segmenting the enhanced image using the optimal segmentation threshold to obtain a segmented image; inputting the segmented image into a preset tumor recognition model to obtain a recognition result. The present invention improves the visual quality of the image through significance enhancement processing, makes tumor features easier to identify, thereby improving the performance of the subsequent classification model, and ensures accurate segmentation of the tumor area by dynamically determining the segmentation threshold and the neighborhood grayscale average, reducing misidentification caused by fuzzy boundaries. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0059] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] The purpose of the present invention is to provide a method and system for tumor recognition in cranial CT images based on computer vision. Through significance enhancement processing, the visual quality of the image is improved, making tumor features easier to identify, thereby improving the performance of the subsequent classification model. By dynamically determining the segmentation threshold and the neighborhood grayscale average, the accurate segmentation of the tumor area is ensured, reducing the misidentification caused by fuzzy boundaries.

[0063] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0064] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a method for identifying tumors in cranial CT images based on computer vision, comprising:

[0065] Step 100: obtaining a cranial CT image set of a patient to be tested, and identifying a brain tissue region image in the cranial CT image set according to a connected domain matching method;

[0066] Step 200: performing image significance enhancement on the brain tissue region image to obtain an enhanced image;

[0067] Step 300: taking a neighborhood window with any point on the enhanced image as the center, and calculating the grayscale average of all pixels in the neighborhood window;

[0068] Step 400: taking the grayscale average of the corresponding pixels as the output of the central pixel to obtain a mean image;

[0069] Step 500: Obtaining an optimal segmentation threshold according to the correlation between the mean image and the enhanced image;

[0070] Step 600: Segmenting the enhanced image using the optimal segmentation threshold to obtain a segmented image;

[0071] Step 700: Input the segmented image into a preset tumor recognition model to obtain a recognition result.

[0072] Preferably, identifying the brain tissue region image in the cranial CT image set according to the connected domain labeling method includes:

[0073] Perform threshold segmentation on each image in the cranial CT image set to obtain a binary image;

[0074] Performing standardization processing on each of the binary images to obtain a standardized image;

[0075] Using a connected domain labeling method to perform region recognition on the standardized image to obtain various connected regions;

[0076] Using a depth-first search method to mark each connected region, so that each of the connected regions is assigned a unique identifier;

[0077] Extracting brain tissue regions according to image features of connected regions; the image features include: size, shape and contour;

[0078] The possible organ regions are screened according to a preset organ region feature threshold to obtain the brain tissue region image; the brain tissue region feature threshold includes an area value and a perimeter value.

[0079] Specifically, this embodiment uses the Otsu method to set the part with a pixel value higher than a threshold value to 1 (foreground) and the part with a pixel value lower than the threshold value to 0 (background). And this embodiment uses a connected domain labeling algorithm (such as an 8-neighborhood or a 4-neighborhood) to perform region recognition on the standardized image. The algorithm traverses each pixel in the image and marks the connected foreground pixels as the same connected region. This embodiment applies a depth-first search (DFS) algorithm to each connected region, traversing the pixels of each connected region to ensure that each region is assigned a unique identifier. DFS can be implemented by recursion or a stack.

[0080] Furthermore, this embodiment extracts image features from each connected region, including:

[0081] The number of pixels in the connected area is calculated by the number of boundary pixels or by using an edge detection algorithm (such as Canny edge detection), and the shape features are described by calculating the moment of the shape (such as Hu moment).

[0082] Optionally, this embodiment further screens the extracted features according to preset brain tissue region feature thresholds (such as area value and perimeter value), sets a reasonable threshold range, and screens out organ regions that meet the conditions.

[0083] Preferably, performing image significance enhancement on the brain tissue region image to obtain an enhanced image comprises:

[0084] Based on fuzzy theory, fuzzy enhancement functions of different pixels are constructed; the formula of the fuzzy enhancement function is: Among them, μ dark represents the first fuzzy enhancement function, μ birg represents the second fuzzy enhancement function, μ gray represents the third fuzzy enhancement function, x ij Represents the gray value at the position (i, j);

[0085] Construct a saliency enhancement model based on the fuzzy enhancement functions of different pixels;

[0086] The brain tissue region image is processed according to the significance enhancement model to obtain the enhanced image.

[0087] Preferably, the formula of the significance enhancement model is:

[0088]

[0089] Among them, x i' j It means that when the input pixel is x ij When the pixel value after saliency enhancement is performed, ν d represents the first weight, ν b represents the second weight, ν g Represents the third weight.

[0090] Specifically, this embodiment is aimed at medical CT images. Motion artifacts, noise interference, equipment performance and other reasons can cause the image to be blurred. Based on this, the fuzzy theory is used to enhance the image. Based on the fuzzy theory, the contrast between the organ area and the background area is increased, making the organ area brighter and the background area darker. Therefore, the following fuzzy rules are specified:

[0091] (1) If the current pixel is dark, make this pixel darker.

[0092] (2) If the current pixel is gray, leave it unchanged.

[0093] (3) If the current pixel is bright, make this pixel brighter.

[0094] Preferably, obtaining an optimal segmentation threshold according to the correlation between the mean image and the enhanced image includes:

[0095] Extracting the grayscale values ​​at the same position on the mean image and the enhanced image to form a grayscale array;

[0096] constructing a segmentation function using the grayscale array;

[0097] Get the preset segmentation array, and continuously adjust the preset segmentation array until the value of the segmentation function is maximized;

[0098] The segmentation array corresponding to the maximum value of the segmentation function is taken as the optimal segmentation threshold.

[0099] In practical applications, the pixel grayscale values ​​of the enhanced image and the mean image at position (i, j) are assumed to be m and n respectively, so the grayscale values ​​of the enhanced image and the mean image at the same position constitute a grayscale array (m, n).

[0100] Preferably, constructing a segmentation function using the grayscale array includes:

[0101] Get the array probability according to the number of times the grayscale array appears;

[0102] The grayscale distribution probability of the mean image and the enhanced image is obtained according to the array probability; wherein the grayscale distribution probability is:

[0103]

[0104] Among them, μ Tm represents the grayscale distribution probability of the enhanced image, μ Tn Represents the grayscale distribution probability of the mean image, p mn represents the probability of the array, M represents the length of the enhanced image, N represents the width of the enhanced image, and F mn Represents the number of occurrences of the grayscale array (m,n), where m,n=0,1,2,…L;

[0105] A segmentation function is constructed according to the grayscale distribution probability.

[0106] Preferably, the segmentation function is:

[0107]

[0108] Among them, (s, t) represents the initial segmentation array, represents the target grayscale distribution probability, Represents the background grayscale distribution probability.

[0109] Specifically, this embodiment segments the image based on the principle of histogram, and can obtain the optimal segmentation threshold as a whole according to the probability of the gray value distribution of the image. Using this segmentation threshold to segment the image can separate the background area and the target area of ​​the image, which is convenient for doctors to identify and extract the tumor lesion area of ​​the image.

[0110] Preferably, the method for constructing the tumor recognition model comprises:

[0111] Obtaining a sample segmentation image data set, and dividing the sample segmentation image data set into a training set, a test set, and a validation set;

[0112] Constructing an initial network model based on a CNN framework; the initial network model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a flattening layer, a first fully connected layer, and a second fully connected layer connected in sequence;

[0113] The initial network model is trained, tested and verified by using the training set, the test set and the verification set to obtain a trained tumor recognition model.

[0114] Specifically, the network structure of this embodiment includes:

[0115] 1. Input layer, which receives the segmented CT image; the input parameter is the size of the segmented image, such as (H, W, C), where H is the height, W is the width, and C is the number of channels (usually 1 or 3).

[0116] 2. The first convolutional layer (Conv2D) extracts low-level features (such as edges and textures). The convolutional layer parameters include the number of filters (such as 32), the filter size (such as 3x3), and the activation function (such as ReLU). The output parameter is the feature map (Feature Map), with a size of (H-2, W-2, 32).

[0117] 3. The first pooling layer (MaxPooling2D) is used to downsample, reduce the size of the feature map, reduce the amount of calculation and the risk of overfitting.

[0118] 4. The second convolutional layer (Conv2D) is used to extract more complex features. The convolutional layer parameters are the number of filters (such as 64), the filter size (such as 3x3), and the activation function (such as ReLU).

[0119] 5. The second pooling layer (MaxPooling2D) is used for further downsampling.

[0120] 6. The third convolutional layer (Conv2D) is used to extract higher-level features. The convolutional layer parameters are the number of filters (such as 128), the filter size (such as 3x3), and the activation function (such as ReLU).

[0121] 7. The third pooling layer (MaxPooling2D) is used for further downsampling.

[0122] 8. Flatten layer, which is used to flatten the multi-dimensional feature map into a one-dimensional array for input into the fully connected layer.

[0123] 9. The first fully connected layer (Dense) is used to learn global features. Its parameters include the number of neurons (such as 256) and the activation function (such as ReLU).

[0124] 10. The second fully connected layer (Dense) is used as the output layer to provide classification information. Parameters include the number of neurons (such as 2, for normal and tumor classification), activation function (such as Softmax or Sigmoid, depending on the task).

[0125] like Figure 2 As shown, this embodiment also provides a tumor CT image recognition system based on computer vision, including:

[0126] An organ image acquisition module is used to acquire a CT image set of a patient to be tested, and identify a brain tissue region image in the CT image set according to a connected domain matching method;

[0127] An image enhancement module, used for performing image significance enhancement on the brain tissue region image to obtain an enhanced image;

[0128] A gray value calculation module is used to take a neighborhood window with any point on the enhanced image as the center and calculate the gray value average of all pixels in the neighborhood window;

[0129] The mean image acquisition module is used to take the grayscale average value of the corresponding pixel points as the output of the central pixel point to obtain the mean image;

[0130] A threshold determination module, used to obtain an optimal segmentation threshold according to the correlation between the mean image and the enhanced image;

[0131] A segmentation module, used to segment the enhanced image using the optimal segmentation threshold to obtain a segmented image;

[0132] The recognition module is used to input the segmented image into a preset tumor recognition model to obtain a recognition result.

[0133] The beneficial effects of the present invention are as follows:

[0134] (1) The present invention identifies organ regions through connected domain matching, significantly improving the registration accuracy of images of different patients and under different scanning conditions, and providing a reliable basis for tumor identification.

[0135] (2) The significance enhancement processing of the present invention improves the visual quality of the image, making tumor features easier to identify, thereby improving the performance of subsequent classification models.

[0136] (3) The present invention ensures accurate segmentation of the tumor area by dynamically determining the segmentation threshold and the neighborhood grayscale average value, thereby reducing misidentification caused by fuzzy boundaries.

[0137] (4) Due to the use of adaptive denoising and segmentation methods, the robustness of the model of the present invention on images of different situations and patients is improved, reducing the need for manual adjustment of appropriate parameters, thereby enhancing the convenience of clinical application.

[0138] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0139] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for identifying tumors in cranial CT images based on computer vision, characterized in that: include: Acquire a cranial CT image set of a patient to be tested, and identify a brain tissue region image in the cranial CT image set according to a connected domain matching method; Performing image significance enhancement on the brain tissue region image to obtain an enhanced image; Taking any point on the enhanced image as the center, take a neighborhood window and calculate the grayscale average of all pixels in the neighborhood window; The grayscale average of the corresponding pixels is used as the output of the central pixel to obtain the mean image; Obtaining an optimal segmentation threshold according to the correlation between the mean image and the enhanced image; Segmenting the enhanced image using the optimal segmentation threshold to obtain a segmented image; The segmented image is input into a preset tumor recognition model to obtain a recognition result.

2. The method for identifying tumors in cranial CT images based on computer vision according to claim 1, characterized in that: Identifying the brain tissue region image in the cranial CT image set according to the connected domain labeling method includes: Performing threshold segmentation on each image in the CT image set to obtain a binary image; Performing standardization processing on each of the binary images to obtain a standardized image; Using a connected domain labeling method to perform region recognition on the standardized image to obtain various connected regions; Using a depth-first search method to mark each connected region, so that each of the connected regions is assigned a unique identifier; Extracting possible organ regions based on image features of connected regions; the image features include: size, shape and contour; The possible organ regions are screened according to a preset organ region feature threshold to obtain the brain tissue region image; the organ region feature threshold includes an area value and a perimeter value.

3. The method for identifying tumors in cranial CT images based on computer vision according to claim 1, characterized in that: Performing image significance enhancement on the brain tissue region image to obtain an enhanced image includes: Based on fuzzy theory, fuzzy enhancement functions of different pixels are constructed; the formula of the fuzzy enhancement function is: Among them, μ dark represents the first fuzzy enhancement function, μ birg represents the second fuzzy enhancement function, μ gray represents the third fuzzy enhancement function, x ij Represents the gray value at the position (i, j); Construct a saliency enhancement model based on the fuzzy enhancement functions of different pixels; The brain tissue region image is processed according to the significance enhancement model to obtain the enhanced image.

4. The method for tumor CT image recognition based on computer vision according to claim 3, characterized in that: The formula of the significance enhancement model is: Among them, x i ' j Indicates that when the input pixel is x ij When the pixel value after saliency enhancement is performed, ν d represents the first weight, ν b represents the second weight, ν g Represents the third weight.

5. The method for identifying tumors in cranial CT images based on computer vision according to claim 1, characterized in that: Obtaining an optimal segmentation threshold according to the correlation between the mean image and the enhanced image, including: Extracting the grayscale values ​​at the same position on the mean image and the enhanced image to form a grayscale array; Constructing a segmentation function using the grayscale array; Get the preset segmentation array, and continuously adjust the preset segmentation array until the value of the segmentation function is maximized; The segmentation array corresponding to the maximum value of the segmentation function is taken as the optimal segmentation threshold.

6. The method for identifying tumors in cranial CT images based on computer vision according to claim 5, characterized in that: The segmentation function is constructed using the grayscale array, including: Get the array probability according to the number of times the grayscale array appears; The grayscale distribution probability of the mean image and the enhanced image is obtained according to the array probability; wherein the grayscale distribution probability is: Among them, μ Tm represents the grayscale distribution probability of the enhanced image, μ Tn Represents the grayscale distribution probability of the mean image, p mn represents the probability of the array, M represents the length of the enhanced image, N represents the width of the enhanced image, and F mn Represents the number of occurrences of the grayscale array (m,n), where m,n=0,1,2,…L; A segmentation function is constructed according to the grayscale distribution probability.

7. The method for identifying tumors in cranial CT images based on computer vision according to claim 6, characterized in that: The segmentation function is: Among them, (s, t) represents the initial segmentation array, represents the target grayscale distribution probability, Represents the background grayscale distribution probability.

8. The method for identifying tumors in cranial CT images based on computer vision according to claim 6, characterized in that: The method for constructing the tumor recognition model comprises: Obtaining a sample segmentation image data set, and dividing the sample segmentation image data set into a training set, a test set, and a validation set; Constructing an initial network model based on a CNN framework; the initial network model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a flattening layer, a first fully connected layer, and a second fully connected layer connected in sequence; The initial network model is trained, tested and verified by using the training set, the test set and the verification set to obtain a trained tumor recognition model.

9. A computer vision-based brain CT image tumor recognition system, characterized in that: include: An organ image acquisition module is used to acquire a set of cranial CT images of a patient to be tested, and identify a brain tissue region image in the cranial CT image set according to a connected domain matching method; An image enhancement module, used for performing image significance enhancement on the brain tissue region image to obtain an enhanced image; A gray value calculation module is used to take a neighborhood window with any point on the enhanced image as the center and calculate the gray value average of all pixels in the neighborhood window; The mean image acquisition module is used to take the grayscale average value of the corresponding pixel points as the output of the central pixel point to obtain the mean image; A threshold determination module, used to obtain an optimal segmentation threshold according to the correlation between the mean image and the enhanced image; A segmentation module, used to segment the enhanced image using the optimal segmentation threshold to obtain a segmented image; The recognition module is used to input the segmented image into a preset tumor recognition model to obtain a recognition result.

Images