Breast lump image processing and classifying method and system

Abstract

The invention relates to the technical field of image processing, and discloses a breast lump image processing and classification method and system, and the method comprises the steps: collecting breast lump image data, and carrying out the spatial registration and fusion of breast lump images; converting the two-dimensional slice image stack into a three-dimensional model, and detecting and marking a suspicious lump area according to the three-dimensional model; and separating the lump area from the background, and displaying the segmented lumps. According to the method, imaging data of two-dimensional ultrasound, CT or MRI and the like are converted into an intuitive three-dimensional model through three-dimensional reconstruction, a doctor can observe a lump and the relation between the lump and surrounding tissues from any angle, and more comprehensive space information is provided. The three-dimensional reconstruction can more accurately measure lump data, is beneficial to improving the benign and malignant judgment precision, and carries out refined reconstruction on the lump edge, so that a complex boundary which is difficult to identify in a two-dimensional image becomes clear, and is beneficial to evaluating the wettability of the lump and the invasion condition of surrounding blood vessels and lymph nodes.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for processing and classifying breast mass images. Background Art

[0002] The research and development background of breast mass image processing and classification system stems from the development of medical imaging technology and the growing demand for early detection and diagnosis of breast diseases. With the improvement of resolution and technical level of medical imaging equipment (such as ultrasound, X-ray photography, magnetic resonance imaging MRI, etc.), the visualization of internal structure of breast tissue has become clearer and more accurate, but it also brings a large amount of data analysis challenges. On the one hand, early detection of breast masses, especially malignant tumors, is crucial to treatment effect and patient survival rate, while traditional clinical examination methods have certain limitations in identifying tiny lesions or judging the nature of masses. Therefore, it is an urgent need to automate mass detection, localization, feature extraction and classification of breast images through computer-aided diagnosis system. On the other hand, the rapid development of deep learning and machine learning technology provides a powerful toolbox for breast mass image processing. These technologies can learn and extract key features from a large amount of breast image data, and perform accurate mass segmentation, classification and prediction of benign and malignant. At the same time, combined with three-dimensional reconstruction, registration algorithm and other means, the three-dimensional morphology and internal structure of the mass can be more comprehensively displayed, thereby assisting doctors in formulating diagnosis and treatment plans. Summary of the invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by the present invention is that the projection effect in the prior art may lead to inaccurate size estimation and difficulty in boundary identification of irregularly shaped or non-axial masses.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: a method for processing and classifying breast mass images, comprising: collecting breast mass image data, performing spatial registration and fusion of breast mass images; converting a two-dimensional slice image stack into a three-dimensional model, detecting and marking suspicious mass areas according to the three-dimensional model; separating the mass area from the background, and displaying the segmented mass.

[0006] As a preferred solution of the breast mass image processing and classification method described in the present invention, the acquisition of breast mass image data includes querying breast images of a specific patient from a medical imaging device to obtain breast mass image data that meets the requirements.

[0007] As a preferred solution of the breast mass image processing and classification method of the present invention, after collecting the breast mass image data, the method further includes preprocessing the breast mass image data and extracting significant feature points.

[0008] As a preferred embodiment of the method for processing and classifying breast mass images of the present invention, the breast mass image data includes but is not limited to imaging data such as two-dimensional ultrasound, CT or MRI; the spatial registration and fusion of the breast mass images includes setting a rough transformation matrix T, selecting a set of point sets S from the source image, mapping them to the target image coordinate system through the current transformation matrix T, obtaining a preliminary corresponding point set S', finding the nearest neighbor point N of each mapped point S' in the target image to form a target point set N, and calculating an objective function that minimizes the distance between point pairs; optimizing the transformation matrix T according to the above objective function, obtaining the optimal rotation and translation parameters during rigid body registration, updating the transformation matrix T according to the new transformation parameters obtained by minimizing the error function, and continuously iterating and optimizing the transformation matrix T until the termination condition is met. After the registration is completed, the breast images of different modal angles are superimposed and fused according to the registration results.

[0009] As a preferred solution of the breast mass image processing and classification method described in the present invention, the step of converting the two-dimensional slice image stack into a three-dimensional model includes: sorting the superimposed and fused continuous two-dimensional slice images according to the depth Z axis to form a stack, performing grayscale value interpolation, generating continuous spatial sampling, and calculating the grayscale value of each target position based on linear interpolation; mapping each pixel to a voxel in the three-dimensional space, constructing a three-dimensional array to represent the entire volume; determining the internal structure information of each voxel through projections on different slices; applying maximum intensity projection, selecting the maximum grayscale value encountered along the line of sight as the display color, and using minimum intensity projection and average intensity projection to select the minimum and average grayscale values; based on the ray projection model, calculating the cumulative effect of each beam of light after passing through the volume; extracting the boundary of the internal structure of the entity to form a closed surface mesh model, rendering the extracted surface model, and completing the construction of the three-dimensional model.

[0010] As a preferred embodiment of the breast lump image processing and classification method described in the present invention, the detection and marking of suspicious lump areas includes: using threshold segmentation on a three-dimensional model to obtain candidate areas of existing lumps; further analyzing the candidate areas, calculating shape index, texture features, and size, combining a machine learning classifier to determine whether it is a lump, and highlighting the identified lump areas; the shape index includes but is not limited to circularity, aspect ratio, and boundary roughness.

[0011] As a preferred embodiment of the breast lump image processing and classification method described in the present invention, the separation of the lump area from the background includes: after determining the suspicious lump area, using threshold segmentation for preliminary segmentation, after the preliminary segmentation, using dilation and erosion to close the holes and remove noise, and optimizing the contour through an edge detection algorithm, and then performing iterative optimization to segment the lump area; based on the segmented lump area, calculating the shape index, and extracting texture features reflecting the changes in the internal tissue structure of the lump.

[0012] A breast mass image processing and classification system using any of the methods described in the present invention, wherein: a data acquisition module collects breast mass image data from medical imaging equipment in real time and in batches; an image registration and fusion module performs spatial registration on breast images of the same patient at different angles and modalities to achieve multi-source information fusion; a three-dimensional reconstruction module converts a two-dimensional slice image stack into a three-dimensional stereo model; a mass recognition module detects and marks suspicious mass areas in the breast according to the model reconstructed by the three-dimensional reconstruction module; a mass segmentation module separates the mass area from the background; and a mass display module performs scaling, rotation and translation display operations on the segmented mass.

[0013] A computer device comprises: a memory and a processor; the memory stores a computer program, comprising: the steps of implementing any one of the methods of the present invention when the processor executes the computer program.

[0014] A computer-readable storage medium stores a computer program, comprising: when the computer program is executed by a processor, the steps of implementing any one of the methods of the present invention are implemented.

[0015] Beneficial effects of the present invention: The method of the present invention converts two-dimensional ultrasound, CT or MRI imaging data into an intuitive three-dimensional model through three-dimensional reconstruction. Doctors can observe the mass and its relationship with surrounding tissues from any angle, providing more comprehensive spatial information. Compared with traditional two-dimensional images, three-dimensional reconstruction can more accurately measure the volume of the mass, shape parameters such as long diameter, short diameter, and aspect ratio, which helps to improve the accuracy of benign and malignant judgments. At the same time, the edge of the mass is finely reconstructed, making the complex boundaries that are difficult to identify in two-dimensional images clear, which is conducive to evaluating the invasiveness of the mass and the invasion of surrounding blood vessels and lymph nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. 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 creative work. Among them:

[0017] Figure 1 An overall flow chart of a method for processing and classifying breast mass images provided by one embodiment of the present invention;

[0018] Figure 2 A system schematic diagram of a breast mass image processing and classification method provided in a third embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0020] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a method for processing and classifying breast mass images, comprising:

[0021] S1: Collect breast mass image data and perform spatial registration and fusion of breast mass images.

[0022] Furthermore, the client software and API supporting the DICOM protocol are accessed through the medical equipment access module, a connection request is initiated to the medical imaging device in the network, and the service provider of the target device is found and identified through the service user function of DICOM. Then, C-ECHO verification is performed, C-ECHO operation is performed, the communication link with the device is verified and tested, and C-FIND is used to query the breast imaging study of a specific patient. Based on the results of C-FIND, a C-MOVE request is sent to obtain a breast mass image sequence that meets the requirements.

[0023] Furthermore, the breast image data is preprocessed, and then the significant feature points of the contour points and boundary points are extracted. The gradient amplitude and direction are calculated by the first-order and second-order derivative Sobel operators, and non-maximum suppression and double threshold processing are applied to extract continuous and significant edges. The formula for calculating the gradient amplitude is:

[0024]

[0025] Among them, G x and G y are the gradients in the horizontal and vertical directions respectively; G M Represents the gradient magnitude.

[0026] Then, on the binary image after edge detection, use the findContours function in the OpenCV library to find and extract all contour lines. The points contained in these contour lines are used as contour point candidates. By calculating the covariance matrix of the local area and its determinant and trace, find the point with the largest response value as the feature point. The Harris response function R is given by the formula:

[0027] R = det(M) - k (trace(M)) 2

[0028] Wherein, M represents the covariance matrix calculated based on the grayscale difference within the local window of the image; k represents the empirical constant; det(M) represents the determinant of the matrix M; trace(M) represents the trace of the matrix M.

[0029] For the extracted contour points, their positions on the contour lines and their relative relationships with other points, endpoints, and inflection points are analyzed to determine them as boundary feature points.

[0030] Among them, breast mass image data can be imaging data such as two-dimensional ultrasound, CT or MRI. The preprocessing operations include graying, denoising, and enhancement. The preprocessing operations and steps are determined according to the different breast mass image data obtained. For example, two-dimensional ultrasound, CT or MRI data are grayscale images themselves and do not need to be grayed; while color Doppler ultrasound, spectral imaging, etc. need to be grayed. Secondly, the preprocessing focus of different imaging data is different. For example, two-dimensional ultrasound has more noise, and the focus is on denoising and enhancing edges; CT has high resolution, and the focus is on enhancing edges and extracting contours; MRI has high contrast but blurred images, and the focus is on noise reduction and texture-based feature extraction.

[0031] Specifically, taking the processing of two-dimensional ultrasound data as an example, the two-dimensional ultrasound data is obtained by connecting the medical equipment access module to the ultrasound equipment. The two-dimensional ultrasound image itself is a grayscale image and does not need to be grayscaled. The preprocessing operations are denoising and enhancement.

[0032] The noise in two-dimensional ultrasound images is relatively obvious. The common type of noise is speckle noise. Median filtering or adaptive filtering is used to remove the noise, and then histogram equalization or CLAHE (contrast limited adaptive histogram equalization) is used to enhance the local details and contrast of the image.

[0033] Furthermore, the image registration and fusion module 2 extracts key feature points and initializes the registration parameters, sets a rough transformation matrix, including rotation and translation for rigid registration, and scaling and shearing for non-rigid registration, to describe the relative position relationship between the two images, and then selects a set of points S from the source image, maps them to the target image coordinate system through the current transformation matrix T, obtains a preliminary corresponding point set S', finds the nearest neighbor point N of each mapped point S' in the target image, forms the target point set N, and calculates the objective function that minimizes the distance between point pairs:

[0034]

[0035] Where E(T) represents the objective function; T represents the transformation matrix; i represents the point set index; n represents the number of points in the point set; S i Represents a point in the point set S; N i Represents a point in the target point set N.

[0036] According to the above error function, the transformation matrix T is optimized. In rigid body registration, the optimal rotation and translation parameters are obtained by solving the singular value decomposition SVD. Then, the transformation matrix T is updated according to the new transformation parameters obtained by minimizing the error function. The above steps are repeated and the optimization is iterated continuously until the termination condition is met, the set maximum number of iterations is reached, and the error is reduced to below the threshold. Finally, multi-source information fusion is performed. After the registration is completed, the breast images of different modal angles are superimposed and fused according to the registration results.

[0037] S2: Convert the 2D slice image stack into a 3D model, and detect and mark the suspicious mass area based on the 3D model.

[0038] Furthermore, the cumulative effect of each beam of light passing through the volume is calculated based on the ray casting model to reflect the transparency and texture characteristics of the tissue. First, from the observer's perspective, a ray is emitted along the line of sight. When passing through the volume, each voxel on the path is sampled to obtain the grayscale value or RGB color value I (x, y, z) at that position. According to the grayscale value of the voxel combined with the pre-defined transfer function, the corresponding absorption rate Opacity, scattering rate and color transformation are determined, which is expressed as:

[0039] Color=(1-Opacity)*Color+Opacity*NewColor

[0040] Among them, Opacity represents the opacity of the current voxel; NewColor represents the new color after conversion by the transfer function.

[0041] The above formula represents the process of accumulating color along the light path. The transfer function is a mapping function that maps the grayscale value of the voxel to opacity, scattering rate, and color space. Common transfer function expressions may include linear interpolation, piecewise functions, or other complex forms, such as:

[0042] Opacity(I)=f(I)

[0043] Color(I)=g9I)

[0044] Among them, f(I) and g(I) are functions of opacity and color changing with grayscale values, respectively, reflecting the texture characteristics of the tissue. The Phong illumination model calculates the light and dark effects. After the above processing, the final pixel color under the influence of all voxels along the line of sight is accumulated.

[0045] Furthermore, the continuous two-dimensional slice images obtained from the medical imaging device are sorted by the depth Z axis to form a stack, and the uneven spacing between the original slices is interpolated to generate continuous spatial sampling based on linear interpolation:

[0046] I(x,y,z)=(1-α)·I(x,y,z0)+α·I(x,y,z1)

[0047] Among them, I(x, y, z) represents the grayscale value of the target position; I(x, y, z0) and I(x, y, z1) are the grayscale values ​​on two adjacent known slices respectively; α represents the interpolation coefficient.

[0048] Each pixel is mapped to a voxel in three-dimensional space, and a three-dimensional array is constructed to represent the entire volume. The internal structure information of each voxel is determined according to the projections on different slices. The maximum grayscale value encountered along the line of sight is taken as the display color through the maximum intensity projection MIP of the three-dimensional reconstruction algorithm to intuitively present the high-density structure inside the mass. The minimum average grayscale value is selected through the minimum intensity projection MinIP and the average intensity projection AveIP. The cumulative effect of each beam of light after passing through the volume is calculated based on the ray casting model to reflect the transparency and texture characteristics of the tissue. Then, surface extraction and rendering are performed. The marching cubes algorithm is used to extract the boundaries of the internal structure of the entity to form a closed surface mesh model. The OpenGL graphics library is used to perform lighting, shadow, and shading rendering operations on the extracted surface model to display the realistic three-dimensional morphology of the mass.

[0049] Furthermore, based on the MarchingCubes algorithm, a three-dimensional surface mesh is generated according to the threshold in the volume data, and a triangular patch is generated on the edge of the cube through a lookup table LUT. Then, threshold segmentation is used on the obtained three-dimensional model to identify the existing mass area. The threshold segmentation is expressed as:

[0050] iff(x,y,z)>T:

[0051] voxe_in_tumor = True

[0052] else:

[0053] voxe_in_tumor = False

[0054] Where T is the set threshold

[0055] Analyze the candidate area, confirm the mass, calculate the shape index, texture features, and size, combine the machine learning classifier to determine whether it is a mass, and then use OpenGL to highlight the identified mass area, through the simplified Phong lighting model:

[0056] iColor=Ambient+Diffuse·max(N·L,0)+Specular·(max(R·V,0)) Shininess

[0057] Among them, Color represents the final pixel color; Ambient and Diffuse represent ambient light and diffuse reflection components respectively; N represents the normal vector; L represents the light source direction vector; R represents the reflection vector, and V represents the observer direction vector; Shininess represents the specular gloss coefficient; by observing the OpenGL rendering results, the mass is identified and marked according to the three-dimensional effect.

[0058] The shape index calculation may be one or more of circularity, aspect ratio, and boundary roughness. Shape-related parameters that are helpful in distinguishing tumors from other tissues may be shape indices. Circularity may be expressed as:

[0059]

[0060] Among them, Area refers to the area of ​​the mass; Perimeter refers to the circumference of the mass.

[0061] Texture features such as gray-level co-occurrence matrix GLCM, local binary pattern LBP and other methods can be used to describe the internal texture information of the mass. A typical statistical feature of GLCM is contrast, which is calculated as follows:

[0062]

[0063] Among them, Contrast represents contrast; P(i,j) represents the joint probability of gray levels i and j in GLCM; then the size of the mass is quantified directly by the number of voxels or the three-dimensional space size, and the volume of the mass is calculated:

[0064]

[0065] Among them, voxel count Indicates the number of voxels contained in the mass; voxel size Represents the spatial size of a single voxel; the shape, texture, size and other features extracted above are used as input, and the trained machine learning model support vector machine is used for classification prediction.

[0066] S3: Separate the tumor area from the background and display the segmented tumor.

[0067] Furthermore, the identified and marked mass areas are segmented. After the mass recognition module determines the suspicious mass area, the threshold segmentation is first used for preliminary segmentation. After the preliminary segmentation, dilation and erosion are used to close the holes and remove noise, and the contour is optimized by the edge detection algorithm. Then, iterative optimization is performed. Based on the segmented mass area, the shape index is calculated, including circularity, aspect ratio, and volume. Circularity calculation:

[0068]

[0069] Aspect ratio calculation:

[0070]

[0071] Among them, Aspect Ratio represents the aspect ratio; MajorAxis Length represents the major axis length; MinorAxisLength represents the minor axis length.

[0072] Long volume calculation:

[0073]

[0074] The texture features reflecting the changes in the internal tissue structure of the mass are extracted based on the texture analysis method gray level co-occurrence matrix GLCM, and a visualization report is generated based on the visualization module 10 in the mass display module 8.

[0075] The mass display module 8 calculates the contrast, energy, entropy, and correlation of multiple texture features of the mass area based on the visualization module 10 using the GLCM algorithm, preprocesses the extracted feature values ​​using Z-score standardization, converts the selected texture features into vector form for input into the classification model, trains the classification model using the supervised learning algorithm support vector machine to distinguish between benign and malignant tumors, divides the training set and the test set, evaluates the model performance through cross-validation, draws a confusion matrix, and intuitively displays the prediction accuracy and misjudgment of each category, including a three-dimensional rendering of the mass, in which color coding represents the classification result, green indicates benign, and red indicates malignant, and generates a heat map and a texture distribution map to display the texture feature values ​​and classification probability distribution of each part of the mass, dynamically adjusts parameters, and switches between different classification models.

[0076] Example 2: The following is an embodiment of the present invention, which provides a method for processing and classifying breast mass images. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0077] The experimental subjects are 2D ultrasound, CT and MRI image data of 30 patients with breast lumps. The main purpose of the experiment is to register and fuse multimodal images, extract the shape and texture features of the lumps through image analysis, and finally use machine learning classifiers for classification prediction to verify the accuracy of classification and the stability of the system.

[0078] The experimental data is obtained through the access module of medical imaging equipment, and the imaging data are all in the standard format that supports the DICOM protocol. The breast imaging data of each patient includes two-dimensional ultrasound, CT and MRI imaging data, which are obtained from the medical imaging equipment in the network through C-FIND and C-MOVE requests. The system uses the DICOM protocol to complete the testing and verification of the communication link, ensure the reliability of data transmission, and perform image preprocessing on the acquired image data.

[0079] The image registration module is used to register two-dimensional ultrasound, CT and MRI image data. The MarchingCubes algorithm is used to generate a three-dimensional surface mesh and fuse image data based on different modalities. In order to accurately describe the location and morphology of breast masses, the system uses the SVD method to optimize the rotation and translation parameters during the registration process, ultimately achieving accurate alignment of images of different modalities.

[0080] The contour and significant feature points of the mass are extracted through edge detection and Sobel operator, and the feature points are extracted using Harris corner detection algorithm. By comparing the features of each modality image, the shape index (circularity, aspect ratio, etc.) and texture features (contrast and entropy based on GLCM gray-level co-occurrence matrix) of the mass are extracted. Then, the system uses the trained support vector machine (SVM) classifier to classify and predict each mass to distinguish between benign and malignant.

[0081] After processing the breast imaging data of each patient, the shape and texture feature values, registration accuracy and final classification results of each modality image are recorded. The following is a partial experimental data record table.

[0082] Table 1 Experimental data record table

[0083]

[0084] It can be seen from the analysis of the experimental data in Table 1 that the multimodal breast image registration and fusion system of the present invention has high accuracy and consistency under different image modalities. Taking shape indexes such as circularity and aspect ratio as an example, the circularity index of MRI modality data is slightly lower due to the low image resolution, while CT and ultrasound data perform better. The mass area and circumference of different modalities are slightly different, but the overall trend is consistent, indicating that the result of image registration is accurate.

[0085] In the analysis of texture features (GLCM contrast), the MRI modality showed higher values ​​in contrast, which is consistent with the high contrast characteristics of MRI data. Due to the influence of noise, the contrast of ultrasound data is relatively low, but the images enhanced by median filtering and CLAHE still show relatively consistent texture features. Combined with the ROC-AUC indicator, it can be seen that the performance of the classification model remains stable in different modal images, and the ROC-AUC values ​​are between 0.89 and 0.94, showing the high classification accuracy of the system for benign and malignant masses.

[0086] Compared with the prior art, the present invention significantly improves the accuracy of mass detection through multimodal fusion and optimized image registration technology. Especially in complex breast mass image analysis, the fusion of multimodal images helps to more comprehensively identify the shape and texture features of the mass, thereby improving the diagnostic ability of the system. At the same time, through the SVD-based registration optimization algorithm, the error between images is significantly reduced, making the registration result more accurate and enhancing the stability and reliability of image analysis.

[0087] Embodiment 3, as Figure 2 As shown, in an exemplary embodiment, a breast mass image processing and classification system is also provided, comprising:

[0088] Data acquisition module 1, real-time and batch acquisition of breast mass image data from medical imaging equipment;

[0089] The data acquisition module 1 is provided with a medical device access module 5 , and the data acquisition module 1 acquires breast mass image data from a medical imaging device through the medical device access module 5 .

[0090] The image registration and fusion module 2 is connected to the data acquisition module 1 and is used to perform spatial registration on breast images of the same patient at different angles and modalities to achieve multi-source information fusion.

[0091] The three-dimensional reconstruction module 3 is connected to the image registration and fusion module 2 and is used to convert the two-dimensional slice image stack into a three-dimensional stereo model;

[0092] The three-dimensional reconstruction module 3 is provided with a three-dimensional reconstruction algorithm introduction module 6 , and the three-dimensional reconstruction module 3 realizes the conversion of a two-dimensional slice image stack into a three-dimensional stereoscopic model through the three-dimensional reconstruction algorithm introduction module 6 .

[0093] The mass identification module 4 is connected to the three-dimensional reconstruction module 3 and is used to detect and mark suspicious mass areas in the breast according to the model reconstructed by the three-dimensional reconstruction module 3 .

[0094] The tumor segmentation module 7 is connected to the tumor recognition module 4 and is used to separate the tumor area from the background according to the tumor recognition module 4;

[0095] The tumor segmentation module 7 includes a morphological feature extraction module 9 , and the tumor segmentation module 7 performs three-dimensional tumor model segmentation based on the morphological feature extraction module 9 .

[0096] The mass display module 8 is connected to the mass segmentation module 7 and is used to perform scaling, rotation and translation display operations on the segmented mass;

[0097] A visualization module 10 is provided in the mass display module 8 .

[0098] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0099] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0100] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0101] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for processing and classifying breast mass images, characterized in that: include: Collect breast mass image data, and perform spatial registration and fusion of breast mass images; Converting a stack of two-dimensional slice images into a three-dimensional model, and detecting and marking suspicious mass areas based on the three-dimensional model; Separate the tumor area from the background and display the segmented tumor.

2. The method for processing and classifying breast mass images as claimed in claim 1, characterized in that: The collecting of breast mass image data includes querying breast images of a specific patient from a medical imaging device to obtain breast mass image data that meets the requirements.

3. The method for processing and classifying breast mass images as claimed in claim 2, characterized in that: After collecting breast mass image data, the method further includes: The breast mass image data is preprocessed and significant feature points are extracted.

4. The method for processing and classifying breast mass images as claimed in claim 3, characterized in that: The breast mass image data includes but is not limited to two-dimensional ultrasound, CT or MRI imaging data; The spatial registration and fusion of the breast mass image includes setting a rough transformation matrix T, selecting a set of point sets S from the source image, mapping them to the target image coordinate system through the current transformation matrix T, obtaining a preliminary corresponding point set S', finding the nearest neighbor point N of each mapped point S' in the target image to form a target point set N, and calculating an objective function that minimizes the distance between point pairs; According to the above objective function, the transformation matrix T is optimized. During rigid body registration, the optimal rotation and translation parameters are obtained. According to the new transformation parameters obtained by minimizing the error function, the transformation matrix T is updated. The transformation matrix T is continuously optimized iteratively until the termination condition is met. After the registration is completed, the breast images of different modal angles are superimposed and fused according to the registration results.

5. The method for processing and classifying breast mass images as claimed in claim 4, characterized in that: The step of converting the two-dimensional slice image stack into a three-dimensional model comprises: The superimposed and fused continuous two-dimensional slice images are sorted according to the depth Z axis to form a stack. Perform grayscale value interpolation to generate continuous spatial sampling, and calculate the grayscale value of each target position based on linear interpolation; Map each pixel to a voxel in three-dimensional space, constructing a three-dimensional array to represent the entire volume; The internal structural information of each voxel is determined by projection on different slices; Applying maximum intensity projection, the maximum gray value encountered along the sight direction is selected as the display color, and using minimum intensity projection and average intensity projection to select the minimum and average gray values; Based on the ray casting model, the cumulative effect of each ray passing through the volume is calculated; The boundaries of the internal structure of the entity are extracted to form a closed surface mesh model, and the extracted surface model is rendered to complete the construction of the three-dimensional model.

6. The method for processing and classifying breast mass images according to claim 5, characterized in that: The detecting and marking of the suspicious mass area includes using threshold segmentation on the three-dimensional model to obtain a candidate area of ​​the existing mass; The candidate area is further analyzed to calculate the shape index, texture features, and size, and combined with the machine learning classifier to determine whether it is a mass, and the identified mass area is highlighted; The shape indexes include but are not limited to circularity, aspect ratio, and boundary roughness.

7. The method for processing and classifying breast mass images according to claim 6, characterized in that: The separation of the mass area from the background includes: after determining the suspicious mass area, using threshold segmentation to perform preliminary segmentation, after the preliminary segmentation, using dilation and erosion to close the holes and remove noise, and optimizing the contour by edge detection algorithm, and then performing iterative optimization to segment the mass area; Based on the segmented mass area, the shape index is calculated, and the texture features reflecting the changes in the internal tissue structure of the mass are extracted.

8. A breast mass image processing and classification system using the method according to any one of claims 1 to 7, characterized in that: include, Data acquisition module, which collects breast mass image data in real time and batches from medical imaging equipment; Image registration and fusion module, which performs spatial registration on breast images of the same patient from different angles and modalities to achieve multi-source information fusion; A 3D reconstruction module converts a 2D slice image stack into a 3D volume model; The mass recognition module detects and marks the suspicious mass area in the breast according to the model reconstructed by the 3D reconstruction module; The mass segmentation module separates the mass area from the background; The tumor display module performs scaling, rotation and translation operations on the segmented tumor.

9. A computer device comprising: A memory and a processor; the memory stores a computer program, wherein the processor implements the steps of the breast mass image processing and classification method as described in any one of claims 1-7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the breast mass image processing and classification method according to any one of claims 1 to 7 are implemented.

Images