Substance composition identification method based on single-energy-level CT image and related equipment

Abstract

The invention provides a substance composition identification method based on a single-energy-level CT image and related equipment, and is applied to the technical field of data processing. The method comprises the following steps: grouping each slice image of a single-energy-level CT image to generate a plurality of single-energy-level CT slice image groups; processing the initial substance component identification model based on the training sample set to generate a target substance component identification model; performing semantic segmentation processing on the CT image of the target object to generate a plurality of image areas and image area types matched with the plurality of image areas; respectively processing each pixel in the plurality of image areas to generate a characteristic parameter value corresponding to each pixel; processing the characteristic parameter value of each pixel in the plurality of image areas and the image area types matched with the plurality of image areas to generate material component information corresponding to each pixel area; and processing the material component information corresponding to each pixel region to generate attribute information of the pixel region.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a material composition identification method based on single-energy-level CT images and related equipment. Background Art

[0002] CT technology has been widely used in various fields, but currently it is mainly used to obtain the three-dimensional structural information inside the object (i.e. structural imaging). Material identification and multi-parameter imaging (such as relative electron density, equivalent atomic number, relative radiation tissue power) are important development directions of CT functional imaging, and there are clear demands in medical diagnosis, safety inspection, radiotherapy and other applications and research. The pixel value of clinical CT images is usually expressed in the form of HU value, which is a physical quantity related to X-ray energy, material density and atomic number. Therefore, in single-energy CT, different materials may show the same or similar HU value due to their different material density and atomic number. By collecting two or more groups of X-rays with different energies, two or more groups of CT images with different energies can be reconstructed respectively. With the help of the different values ​​of HU value in CT images of different energies, the ambiguity that may exist in ordinary CT images can be eliminated to a certain extent, and the material can be identified.

[0003] The core scientific assumption of the traditional material composition modeling method is that it is assumed that the material composition can be approximately expressed by a combination of a pair of basis substances (water and iodine) and a pair of basis functions (approximate photoelectric effect, approximate Compton scattering response), that is, bones are considered to be composed of a certain content of water and a certain content of iodine, or a certain degree of photoelectric effect function and a certain degree of Compton effect function, and air is considered to be composed of zero water and zero iodine. This method has great limitations. For example, the material composition modeling is completely dependent on a priori assumptions, and its objective rationality is questionable. At the same time, the component quantification algorithm is highly dependent on experimental parameters, the reliability of the material composition modeling effect is relatively poor, and the anti-noise performance is relatively poor.

[0004] In recent years, spectral CT based on dual radiation sources, rapid switching of high and low voltage of radiation sources, multi-layer detectors, and photon technology detectors has come into clinical use. Spectral CT can obtain a set (at least two) of single-energy-level CT images corresponding to different energies by finely modeling the X-ray energy spectrum of the radiation source. However, no material composition quantification method based on single-energy-level CT images has been seen so far.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of this application is to provide a material composition modeling method based on single-level CT images and related equipment, which at least to a certain extent overcomes the problems existing in the prior art and achieves the purpose of improving the material composition identification capability and multi-parameter imaging accuracy through feature parameter extraction.

[0007] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present invention.

[0008] According to one aspect of the present application, a material composition modeling method based on a single-energy CT image is provided, comprising: obtaining a CT image of a target object, wherein the CT image of the target object includes at least one group of single-energy-level CT images; grouping and processing each slice image of the single-energy-level CT image to generate a plurality of single-energy-level CT slice image groups; obtaining a training sample set and an initial material composition recognition model, wherein the training sample set includes a single-energy-level CT slice image group corresponding to a known reference substance, the initial material composition recognition model is formed based on the variation trend of CT image values ​​of different reference substances at different energy levels, the initial material composition recognition model includes reference values ​​for characterizing characteristic parameters corresponding to material composition and their corresponding parameter space; based on the training sample set, the initial material composition recognition model is constructed ... The method comprises the steps of: processing the CT image of the target object with a semantic segmentation method to generate a plurality of image regions and image region types matching the plurality of image regions; processing each pixel in the plurality of image regions based on the target material component identification model to generate a characteristic parameter value corresponding to each pixel; processing the characteristic parameter value of each pixel in the plurality of image regions and the image region types matching the plurality of image regions based on the target material component identification model to generate material composition information corresponding to each pixel region; processing the material composition information corresponding to each pixel region to generate attribute information of the pixel region, wherein the attribute information of the pixel region includes relative electron density, equivalent atomic number and relative radiographic tissue power of each pixel.

[0009] In one embodiment of the present application, each pixel in a plurality of image regions is processed based on the target material component recognition model to generate a characteristic parameter value corresponding to each pixel, including:

[0010] The target material component identification model includes a calculation formula for obtaining characteristic parameter values, and the calculation formula is:

[0011] CTHU(keV)=a·e b·keV +c;

[0012] a=ω1·a bone +ω2·a fat ;

[0013] b=ω1·b bone +ω2·b fat ;

[0014] c=ω1·c bone +ω2·c fat ;

[0015] ω1+ω2=1;

[0016] Among them, a, b, and c represent characteristic parameters, e represents the exponential function, keV represents the energy level corresponding to the single-level CT image, x1 and x2 are reference substances, and ω1 and ω2 are the content information of the reference substances.

[0017] In one embodiment of the present application, the characteristic parameter value of each pixel in a plurality of image regions and the image region types matching the plurality of image regions are processed based on the target material composition recognition model to generate material composition information corresponding to each pixel region, including:

[0018] The target material composition identification model includes a calculation formula for obtaining material composition information, and the calculation formula is:

[0019]

[0020] Among them, p x,y,z Represents the material composition information corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, M i represents the i-th reference substance, and n represents the number of reference substances constituting the material composition of the pixel.

[0021] In one embodiment of the present application, the target material composition identification model includes obtaining multi-parameter information such as relative electron density, equivalent atomic number and relative radiation tissue power of the material, and the calculation formulas are:

[0022] The target material component identification model includes a calculation formula for obtaining relative electron density, which is:

[0023]

[0024] Among them, RED x,y,z Represents the relative electron density corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, i represents the i-th reference substance, ρ i represents the density of the i-th reference material, ρ water Represents the density of water, RED irepresents the electron density of the i-th reference substance;

[0025] The target material component identification model includes a calculation formula for obtaining the equivalent atomic number, which is:

[0026]

[0027] Among them, Zeff x,y,z Represents the equivalent atomic number corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, i represents the i-th reference substance, and Z i represents the equivalent atomic number of the i-th reference substance;

[0028] The target material component identification model includes a calculation formula for obtaining relative radiological tissue power, which is:

[0029]

[0030] Among them, SPR stands for relative radiological power and RED stands for the electron density of the reference material.

[0031] In one embodiment of the present application, the initial material component identification model is processed based on a training sample set to generate a target material component identification model, including: preprocessing the training sample set to generate a training sample set with identification information, wherein the training sample set includes information of several standard substances; training the initial material component identification model based on the training sample set with identification information to generate a target material component identification model.

[0032] In one embodiment of the present application, the training sample set is preprocessed to generate a training sample set with identification information, including: extracting features from the training sample set to determine an original feature library; dividing each feature data set according to the original feature library to generate a training data set and a test data set; using a classifier to predict each test data set divided from the original feature library to determine a prediction result; using a preset algorithm to train each training data set divided from the original feature library to obtain a test set class prediction result; and generating a training sample with identification information based on the prediction result and the test set class prediction result.

[0033] In one embodiment of the present application, the feature extraction of the training sample set to determine the original feature library includes: processing the standard material information based on preset processing rules to generate standardized features, wherein the standardized features are features of samples or motifs with clear and known material composition; processing the standardized features based on preset feature screening and dimensionality reduction rules to generate original features; and generating an original feature library from a number of original features.

[0034] Another aspect of the present application is a material composition identification device based on single-energy CT images, characterized in that it includes: an acquisition module, used to acquire CT images of a target object, wherein the CT images of the target object include at least one group of single-energy CT images; acquire a training sample set and an initial material composition identification model, wherein the training sample set includes a single-energy CT slice image group corresponding to a known reference substance, and the initial material composition identification model is composed of a change trend of CT image values ​​of different reference substances at different energy levels, and the initial material composition identification model includes reference values ​​for characterizing characteristic parameters corresponding to material components and their corresponding parameter spaces; a processing module, used to group and process each slice image of the single-energy CT images to generate a plurality of single-energy CT slice image groups; based on the training sample set, the material composition identification model is processed by the processing module; The initial material composition recognition model is processed to generate a target material composition recognition model; the CT image of the target object is semantically segmented to generate a plurality of image regions and image region types matching the plurality of image regions; each pixel in the plurality of image regions is processed based on the target material composition recognition model to generate a characteristic parameter value corresponding to each pixel; the characteristic parameter value of each pixel in the plurality of image regions and the image region types matching the plurality of image regions are processed based on the target material composition recognition model to generate material composition information corresponding to each pixel region; the material composition information corresponding to each pixel region is processed to generate attribute information of the pixel region, wherein the attribute information of the pixel region includes a relative electron density, an equivalent atomic number, and a relative radiographic tissue power of each pixel.

[0035] According to another aspect of the present application, an electronic device is provided, characterized in that it comprises: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to implement the above-mentioned material composition identification method based on single-energy-level CT images by executing the executable instructions.

[0036] According to another aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a second processor, the method for identifying material components based on single-energy-level CT images is implemented.

[0037] According to another aspect of the present application, a computer program product is provided, including a computer program, characterized in that when the computer program is executed by a third processor, the computer program implements the above-mentioned material composition identification method based on single-energy-level CT images.

[0038] The present application provides a material composition identification method based on single-level CT images and related equipment, in which a server obtains and groups single-level CT images of target objects. Then, a group of CT slice images of known substances is used as training samples, combined with an initial material composition identification model, and a target material composition identification model is obtained through machine learning training. Next, the CT image is semantically segmented to identify different regions and their types. The trained model is used to analyze the characteristic parameter values ​​of each pixel to determine the material composition information. Finally, the attribute information of each pixel area is calculated, such as relative electron density, equivalent atomic number, and relative radiation tissue power, to comprehensively describe the material properties. Based on the new material composition modeling method of the single-level CT image formed by spectral CT, the feature parameter extraction is used to improve the material composition identification capability and multi-parameter imaging accuracy.

[0039] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart showing a material composition identification method based on a single-energy-level CT image provided by an embodiment of the present application;

[0041] Figure 2 A schematic diagram of the structure of a material composition identification device based on a single-level CT image provided by an embodiment of the present application is shown;

[0042] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present application is shown;

[0043] Figure 4 A schematic diagram of a storage medium provided by an embodiment of the present application is shown;

[0044] Figure 5 A schematic diagram of partial CTHU>0 material fitting provided by an embodiment of the present application is shown;

[0045] Figure 6 A partial CTHU provided by an embodiment of the present application is shown. < 0Material fitting diagram. DETAILED DESCRIPTION

[0046] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0047] Combine the following Figure 1To describe the material composition identification method based on single-level CT images according to an exemplary embodiment of the present application. It should be noted that the following application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the embodiments of the present application are not limited in this regard. On the contrary, the embodiments of the present application can be applied to any applicable scenario.

[0048] In one embodiment, the present application also proposes a material composition identification method based on single-energy-level CT images and related equipment. Figure 1 The following schematically shows a flow chart of a material composition identification method based on a single-level CT image according to an embodiment of the present application. Figure 1 As shown, the method is applied to a server, comprising:

[0049] S101, acquiring a CT image of a target object.

[0050] In one embodiment, the CT image of the target object includes at least one set of single-energy level CT images, such as through two consecutive acquisitions, Siemens dual source, GE high and low energy rapid alternating switching, Philips double-layer detector, so as to realize high and low energy CT data acquisition, and form a series of single-energy level CT images, wherein the single-energy level CT images include but are not limited to high single-energy level CT images and low single-energy level CT images, respectively simulating CT images under different energies, and the energy span is generally from 50keV to 140keV.

[0051] S102, performing grouping processing on each slice image of the single-energy-level CT image to generate a plurality of single-energy-level CT slice image groups.

[0052] In one embodiment, a preset size value is obtained to process a single-level CT image to generate a single-level CT image of a preset size, a multi-layer slice process is performed on the single-level CT image to generate multiple slice image groups of different depth levels, the slice image group is processed based on the time sequence or attribute information to generate a number of slice images, and each slice image of the single-level CT image is grouped and processed in the order of energy level priority, and corresponding slice images can be generated in sequence according to the order of slices or the type of slices, that is, tomographic images corresponding to the same spatial position. Slice images of different depth levels are obtained based on different scale information, thereby generating multiple slice image groups of different depth levels, and then the same image at different scales is obtained. For example, when viewing the same image from a macroscopic scale or a microscopic scale, the scale from small to large is the process of seeing the details of the objective object from more to less.

[0053] S103, obtaining a training sample set and an initial material composition recognition model.

[0054] In one embodiment, the training sample set includes a group of single-energy CT slice images corresponding to known reference materials, and the initial material component identification model is formed based on the change trend of CT image values ​​of different reference materials at different energy levels. The initial material component identification model includes reference values ​​for characterizing characteristic parameters corresponding to material components and their corresponding parameter spaces. The single-energy CT slice image groups of several groups of reference materials with known components are obtained, and according to the change trend of CT image values ​​(CTHU values) of different reference materials at different energy levels, an initial material component identification model matching the single-energy CT image is established.

[0055] An initial material component identification model and a training sample set for training the initial material component identification model are obtained, wherein the training sample set includes information of several standard materials; the training sample set is preprocessed to generate a training sample set with identification information, wherein the identification information is used to characterize the attributes of various types of material information, and the initial material component identification model is trained based on the training sample set with identification information to generate a target material component identification model. Specifically, the data classification result corresponding to the training data set is obtained, because the classification result of each training data set can be determined in advance, and the data can be directly obtained from the outside. The prediction result is compared with the data classification result to determine the first comparison result; the prediction result of the test set class is compared with the data classification result to determine the second comparison result; when judging whether the second comparison result meets the preset requirements with the first comparison result, when the preset requirements are met, it indicates that the detection result of the current material component identification model is relatively accurate. At this time, the current material component identification model can be used as the target material component identification model, and the material component identification model includes a reference parameter space for characterizing the distribution position of the characteristic parameters corresponding to the material component.

[0056] In addition, the standard material information is samples or models with clear and known material composition, including pure water (H2O), CaCO3 of different densities, fat, muscle, bones of different densities, sodium iodide aqueous solutions of different densities, and common medical implant metals (such as stainless steel and titanium).

[0057] S104: Process the initial material component recognition model based on the training sample set to generate a target material component recognition model.

[0058] In one implementation, the training sample set is preprocessed to generate a training sample set with identification information, wherein the training sample set includes information of several standard substances, and an initial material component recognition model is trained based on the training sample set with identification information to generate a target material component recognition model.

[0059] Specifically, collect information about standard substances, which may include the chemical composition, physical properties, and sources of the substances. These data can be obtained from the National Standard Material Resource Sharing Platform. Data preprocessing is a key step in generating a training sample set, including data cleaning (removing erroneous and inconsistent data), data conversion (converting data into a format suitable for model input), and feature engineering (selecting and constructing features required for model training). In order to improve the generalization ability of the model, the training set is augmented by data augmentation technology. For example, the diversity of samples can be increased by rotation, scaling, color transformation, and so on. In the study of the identification of organic and inorganic substances, the data set is augmented by adding gray edges to the shorter sides, making the picture square, and rotating the angle to ensure that each sample has the correct identification information, which will be used as labels during model training. The annotations should be accurate to avoid the model learning wrong information.

[0060] The dataset is divided into training set, validation set and test set. The training set is used to train the model, the validation set is used for model selection and hyperparameter adjustment, and the test set is used to finally evaluate the model performance. Through training and validation, a target material component recognition model is obtained, which can accurately identify and classify different material components. Remember to monitor and record the performance indicators of the model, such as accuracy, recall rate and F1 score, during the model training process to ensure the effectiveness and reliability of the model.

[0061] In another implementation, feature extraction is performed on the training sample set to determine the original feature library; each feature data set is divided according to the original feature library to generate a training data set and a test data set; a classifier is used to predict each test data set divided from the original feature library to determine a prediction result; a preset algorithm is used to train each training data set divided from the original feature library to obtain a test set class prediction result; and a training sample with identification information is generated based on the prediction result and the test set class prediction result.

[0062] Feature extraction is performed on the training data set to determine the original feature library. Feature extraction includes four features: original features, statistical features, frequency domain features, and time domain features. The commonly used features extracted are specifically introduced as follows: Original features: Original features are the collected data, which are preprocessed and use all the information in each sample data matrix for model training. In order not to lose sample information, each curve of sample Xi is directly unfolded and stretched into a row vector. Statistical features: Statistical features consider the change trend of all data points on the curve. By extracting statistical features, the dimension of data samples can be reduced, which is convenient for analyzing genetic data and accelerating the convergence speed during model training. The extracted statistical features include maximum value, minimum value, mean, variance, and standard deviation. Frequency domain features: Wavelet transform is performed on the data set, and the coefficients obtained by the second-order wavelet transform are used as new features to form frequency domain features. Time domain features: Time domain features are mainly based on the time perspective to invent the changing laws of signals and systems. Time domain features can reflect the information of curve data in time changes. This process mainly extracts the first-order forward difference of the data set to form the first-order difference time domain features, and uses the exponential moving average feature processing method.

[0063] In another embodiment, the standard material information is processed based on preset processing rules to generate standardized features, wherein the standardized features are features of samples or models whose material composition is clear and known, and the standardized features are processed based on preset feature screening and dimensionality reduction rules to generate original features, and a number of original features are used to generate an original feature library.

[0064] The process of processing the standard material information based on the preset processing rules and generating standardized features involves data cleaning, conversion and normalization to ensure the consistency and comparability of the data. Specifically, the data is cleaned, including removing or correcting outliers, missing values ​​and erroneous data to ensure that the information of all substances is up-to-date and accurate. Key features are extracted from the detailed information of the standard material, such as chemical composition, concentration, physical state, spectral characteristics, etc. The features are standardized to have a uniform dimension and range, such as using Z-score standardization or minimum-maximum normalization. Standardized features refer to the features of samples or models with clear and known material composition. According to the preset feature screening rules, the most informative features for material composition identification are selected. Statistical tests, model selection or knowledge-based rules can be used to determine which features should be retained. Dimensionality reduction techniques such as principal component analysis (PCA), linear discriminant analysis (LDA) or autoencoders are applied to reduce the number of features while retaining the key information in the original data as much as possible.

[0065] The filtered and dimensionally reduced features are collected to form an original feature library, which will serve as the basis for training machine learning models. The feature library will be maintained, including regular updates and optimization of feature sets to adapt to new data and discoveries, ensuring the quality and relevance of the feature library to support effective material component identification.

[0066] S105, performing semantic segmentation processing on the CT image of the target object to generate a plurality of image regions and image region types matching the plurality of image regions.

[0067] In one embodiment, HU value threshold segmentation is performed based on 70keV single-level CT images to obtain bone, lung, and soft tissue regions. When performing HU value threshold segmentation of 70keV single-level CT images, it is usually necessary to set appropriate thresholds based on the specific HU value ranges of different human tissues in CT images. The following are the thresholds for bone, lung, and soft tissue region segmentation:

[0068] The CT value of bones is usually higher than 400HU, so for bone segmentation, a higher HU value can be selected as the upper threshold, such as 400HU or higher. The lower threshold can be set lower, such as -1000HU, to ensure that only the bone with the highest density is selected.

[0069] The CT value of the lung is usually low because the lung tissue is full of air. The HU value of the lung is roughly between -1000HU and -200HU. For the segmentation of the lung, you can choose -500HU or -600HU as the lower threshold, and the upper threshold can be set at -200HU or higher to include all lung tissue.

[0070] The CT value of soft tissue is usually between -50HU and 100HU. For the segmentation of soft tissue, a lower lower threshold, such as -50HU, and a moderate upper threshold, such as 100HU, can be selected.

[0071] In actual operation, it may be necessary to fine-tune according to the specific CT image and the required segmentation accuracy. For example, according to research in the literature, for the identification of solid components of subsolid nodules in the lungs, the diagnostic value of the identification is the greatest when the threshold is set to -250HU. For the segmentation of pulmonary nodules, it may be necessary to combine the HU value range of the lungs and the characteristics of the nodules to determine the optimal threshold. In addition, factors such as image noise, artifacts, and partial volume effects may also need to be considered during the segmentation process, which may affect the choice of thresholds and the accuracy of segmentation results. In some cases, it may be necessary to combine multiple segmentation techniques, such as thresholding, region growing, level set segmentation, etc., as well as deep learning methods to improve the accuracy and robustness of segmentation.

[0072] In addition, reference Figure 5 and Figure 6 It can be seen that the pixel value (i.e., CTHU value) of a single-energy CT slice image characterizes the attenuation ability of the material in the pixel to X-rays. Different single-energy CT images represent the distribution of the attenuation ability of the material to X-rays of different energies. In other words, pixels at the same position have different HU values ​​in different single-energy images. The difference in pixel values ​​in the same image pixel area is within a preset threshold, such as the CT scanning site (head, neck, chest, abdomen, limbs), organ type (brain, lungs, intestines, bones, muscles, etc.), and whether there is a contrast agent or implant in the body, which is used to improve the accuracy of material composition identification. This solution does not limit the pixel values ​​of the CT image in each area, and can be set according to the actual needs of the applicant.

[0073] Example 1: If the scanned part is the head, then lung tissue is ruled out; if no contrast agent is used during the scan, then iodine is ruled out.

[0074] Example 2: If the tissue region is lung, the candidate tissues in the region are lung, muscle, blood, and fat; if the tissue region is bone, the candidate tissues in the region are bone, fat, and blood.

[0075] S106, processing each pixel in a plurality of image regions based on the target material component recognition model to generate a characteristic parameter value corresponding to each pixel.

[0076] In one embodiment, the target material component identification model includes a calculation formula for obtaining a characteristic parameter value, and the specific calculation formula is:

[0077] CTHU(keV)=a·e b·keV +c;

[0078] a=ω1·a x1 +ω2·a x2 ;

[0079] b=ω1·b x1 +ω2·b x2 ;

[0080] c=ω1·c x1 +ω2·c x2 ;

[0081] ω1+ω2=1;

[0082] Among them, a, b, and c represent characteristic parameters, e represents the exponential function, keV represents the energy level corresponding to the single-level CT image, x1 and x2 are reference substances, and ω1 and ω2 are the content information of the reference substances.

[0083] The category corresponding to the image area is used as a priori condition to constrain the possible composition of the reference material of the pixels in the area. For example, the possible material composition of the pixels in the bone tissue area should be a weighted combination of bone and fat, and the possible material composition of the pixels in the liver area should be a weighted combination of liver and fat. According to the possible composition of the reference material in different image areas, the bilinear interpolation method is used to obtain the weight wi corresponding to the reference material in each pixel. Taking the bone area as an example, the material composition of a pixel in all bone areas can be characterized by a linear combination of the two reference materials, bone and fat (supplement: the physiological basis is that the real bone is a porous structure, the framework is bone tissue, and the gap is bone marrow. Since bone marrow is similar to fat, fat is used as a substitute)

[0084] a=ω1·a bone +ω2·a fat ;

[0085] b=ω1·b bone +ω2·b fat ;

[0086] c=ω1·c bone +ω2·c fat ;

[0087] ω1+ω2=1.

[0088] When performing HU value threshold segmentation of 70keV single-level CT images, the applicant can calculate the weight corresponding to the reference material in each pixel by bilinear interpolation. This method is used in image processing to estimate pixel values ​​at non-integer coordinates, especially when the image is scaled or rotated. In the segmentation of bone areas, the applicant can use bone and fat as two reference materials, because real bones have a porous structure, whose framework is composed of bone tissue, and the gaps are filled with bone marrow. Since bone marrow is similar to fat, fat can be used as a substitute.

[0089] The calculation process of bilinear interpolation is as follows: first, determine the position of the pixel in the image, which usually involves the scaling ratio of the image, then find the four pixels closest to the position (in the grid of the image), and then perform linear interpolation in the x and y directions respectively. This means that each pair of pixels is first interpolated in the x direction to obtain two temporary values, and then these two temporary values ​​are interpolated in the y direction to obtain the final pixel value. The weight of the interpolation is determined by the distance, which can be determined by calculating the distance function.

[0090] In practical applications, bilinear interpolation can be expressed by the following formula: f(x, y) = (1-u)(1-v)f(x1, y1) + (1-u)vf(x1, y2) + u(1-v)f(x2, y1) + uvf(x2, y2);

[0091] Among them, f(x, y) is the pixel value to be calculated, f(x1, y1), f(x1, y2), f(x2, y1) and f(x2, y2) are four known adjacent pixel values, and u and v are the relative distances from the four known pixels to the pixel to be calculated.

[0092] In the segmentation of the bone region, the applicant can distinguish between bones and fat by adjusting the threshold. For example, the CT value of bones is usually higher than 400HU, while the CT value of fat is usually between -50 and -100HU. By setting a suitable threshold, the applicant can separate bones and fat and use bilinear interpolation to calculate the weight of bones and fat in each pixel, thereby segmenting the bone region more accurately.

[0093] S107, processing the characteristic parameter value of each pixel in the plurality of image regions and the image region types matching the plurality of image regions based on the target material composition recognition model, and generating material composition information corresponding to each pixel region.

[0094] In one embodiment, the material composition information includes the electron density and equivalent atomic number of the material, for example, a certain area is composed of 80% water + 20% fat. Or, it is known that the object under test is pure water and fat, and the corresponding characteristic parameters are (a1, b1, c1) and (a2, b2, c2). For an unknown object, the corresponding characteristic parameters are (a, b, c). If (a, b, c) is between (a1, b1, c1) and (a2, b2, c2), then it can be considered that the unknown substance is a combination of water and fat.

[0095] In another embodiment, the target material composition identification model includes a calculation formula for obtaining material composition information, and the calculation formula is:

[0096]

[0097] Among them, p x,y,z Represents the material composition information corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, M i represents the i-th reference substance, and n represents the number of reference substances constituting the material composition of the pixel.

[0098] This model is used in medical imaging, such as CT or MRI scans, where different tissue types (such as bone, fat, soft tissue, etc.) can be distinguished by their specific density or signal strength. By identifying and quantifying the material composition of each pixel in the image, it can help doctors make more accurate diagnoses. The model provides a way to quantify the proportion of different material components in the image pixels through a mathematical formula. It determines the material composition information of a specific location in the image by weighting multiple reference materials.

[0099] S108, processing the material composition information corresponding to each pixel area to generate attribute information of the pixel area.

[0100] In one implementation, the attribute information of the pixel region includes the relative electron density, equivalent atomic number, and relative radiation tissue power of each pixel.

[0101] The target material component identification model includes a calculation formula for obtaining relative electron density, which is:

[0102]

[0103] Among them, RED x,y,z Represents the relative electron density corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, i represents the i-th reference substance, ρ i represents the density of the i-th reference material, ρ water Represents the density of water, RED i represents the electron density of the i-th reference substance; RED is calculated by multiplying the ratio of the density of each reference substance to the density of water by the RED of the substance, and then taking the weighted sum of all reference substances.

[0104] In one embodiment, the target material component identification model includes a calculation formula for obtaining the equivalent atomic number, and the calculation formula is:

[0105]

[0106] Among them, Zeff x,y,z Represents the equivalent atomic number corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, i represents the i-th reference substance, and Z i Represents the equivalent atomic number of the i-th reference substance; Zeff is calculated by multiplying the 2.94th power of the equivalent atomic number of each reference substance by the content of the substance, and then taking the weighted sum of all reference substances.

[0107] In one embodiment, the target material component identification model includes a calculation formula for obtaining relative radiological tissue power, which is:

[0108]

[0109] Where SPR stands for relative radiological power and RED stands for the electron density of the reference material. The calculation of SPR depends on the value of RED and is calculated using different linear equations for different ranges of RED. The relationship between SPR and RED has different expressions for different ranges of RED values: 0 to 0.9, 0.9 to 1.035, 1.035 to 1.4, and greater than or equal to 1.4.

[0110] This model is used in medical imaging, such as CT scans, where different tissue types can be distinguished by their electron density, atomic number, and radiographic power. These calculations can help identify and distinguish different material components in the image, allowing for more accurate diagnosis and analysis. The model provides a way to quantify the material composition properties of each pixel in the image through mathematical formulas. These properties include relative electron density, equivalent atomic number, and relative radiographic power. In summary, the target material composition recognition model is a pixel material composition analysis method based on physical properties. It identifies and quantifies different material components in an image by calculating RED, Zeff, and SPR for each pixel.

[0111] In the present application, a server obtains a CT image of a target object, wherein the CT image of the target object includes at least one group of single-energy-level CT images; each slice image of the single-energy-level CT image is grouped and processed to generate a plurality of single-energy-level CT slice image groups; a training sample set and an initial material component recognition model are obtained, wherein the training sample set includes a single-energy-level CT slice image group corresponding to a known reference substance, the initial material component recognition model is based on the variation trend of CT image values ​​of different reference substances at different energy levels, and the initial material component recognition model includes reference values ​​for characterizing characteristic parameters corresponding to material components and their corresponding parameter spaces; the initial material component recognition model is processed based on the training sample set to generate a target material component. Identification model; semantic segmentation processing is performed on the CT image of the target object to generate several image regions and image region types that match the several image regions; each pixel in the several image regions is processed based on the target material composition recognition model to generate the characteristic parameter value corresponding to each pixel; the characteristic parameter value of each pixel in the several image regions and the image region types that match the several image regions are processed based on the target material composition recognition model to generate the material composition information corresponding to each pixel region; the material composition information corresponding to each pixel region is processed to generate the attribute information of the pixel region, wherein the attribute information of the pixel region includes the relative electron density, equivalent atomic number, and relative radiation tissue power of each pixel. A new material composition modeling method based on the single-level CT image formed by spectral CT can improve the material composition recognition capability and multi-parameter imaging accuracy through feature parameter extraction.

[0112] Optionally, in another embodiment of the method of the present application, the preprocessing of the training sample set to generate a training sample set with identification information further includes:

[0113] Labeling the samples in the training sample set to obtain a first label for each sample;

[0114] Extracting a single sample from the training sample set, forming a first training data set according to the sample and the corresponding first label, and extracting multiple times to obtain multiple first training data sets;

[0115] Extracting two samples from the training sample set, generating second labels for the two samples according to first labels respectively corresponding to the two samples, forming a second training data set according to the two samples and the corresponding second labels, and extracting multiple times to obtain multiple second training data sets;

[0116] A plurality of training data sets are constructed according to the plurality of first training data sets and the second training data set.

[0117] In one implementation, a preset number of feature information can be randomly extracted with replacement from the multidimensional feature information of each sample each time to form a corresponding sub-sample. Multiple sub-samples constitute a training set. After multiple extractions, multiple training sets are constructed. The preset number can be customized according to actual needs. The training set can be divided into two parts. One part is a single sample x, and it is marked whether the target application will use it next. If so, it can be marked as 1, otherwise it can be marked as 0. The form can be (xi, yi), where yi∈{0, 1}. The other part is a triple, that is, by sampling two samples (xi, xj), if the two sample labels are consistent, they are recorded as 1, and if the labels are inconsistent, they are recorded as -1. The form is (xi, xj, γ), where γ∈{1, -1}.

[0118] In one embodiment, if Figure 2 As shown, the present application also provides a material composition identification device based on a single-level CT image, comprising:

[0119] An acquisition module 201 is used to acquire a CT image of a target object, wherein the CT image of the target object includes at least one set of single-energy-level CT images; acquire a training sample set and an initial material component recognition model, wherein the training sample set includes a single-energy-level CT slice image group corresponding to a known reference material, and the initial material component recognition model is formed based on the change trend of CT image values ​​of different reference materials at different energy levels, and the initial material component recognition model includes reference values ​​for characterizing characteristic parameters corresponding to material components and their corresponding parameter spaces;

[0120] The processing module 202 is used to group each slice image of the single-energy CT image to generate a plurality of single-energy CT slice image groups; process the initial material composition recognition model based on the training sample set to generate a target material composition recognition model; perform semantic segmentation on the CT image of the target object to generate a plurality of image regions and image region types matching the plurality of image regions; process each pixel in the plurality of image regions based on the target material composition recognition model to generate a characteristic parameter value corresponding to each pixel; process the characteristic parameter value of each pixel in the plurality of image regions and the image region types matching the plurality of image regions based on the target material composition recognition model to generate material composition information corresponding to each pixel region; process the material composition information corresponding to each pixel region to generate attribute information of the pixel region, wherein the attribute information of the pixel region includes the relative electron density, equivalent atomic number, and relative radiographic tissue power of each pixel.

[0121] The present application embodiment provides an electronic device, such as Figure 3As shown, the electronic device 3 includes a first processor 300, a memory 301, a bus 302 and a communication interface 303. The first processor 300, the communication interface 303 and the memory 301 are connected via the bus 302; the memory 301 stores a computer program that can be run on the first processor 300, and when the first processor 300 runs the computer program, it executes the material component identification method based on the single-energy-level CT image provided in any of the aforementioned embodiments of the present application.

[0122] The memory 301 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 303 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0123] The bus 302 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 301 is used to store programs, and the first processor 300 executes the programs after receiving the execution instruction. The material component identification method based on the single-energy-level CT image disclosed in any implementation of the embodiment of the present application may be applied to the first processor 300, or implemented by the first processor 300.

[0124] The first processor 300 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the first processor 300. The above-mentioned first processor 300 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a readily available programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be embodied as a hardware decoding processor to be executed, or a combination of hardware and software modules in the decoding processor to be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 301, and the first processor 300 reads the information in the memory 301 and completes the steps of the above method in combination with its hardware.

[0125] The electronic device provided in the above-mentioned embodiments of the present application and the material composition identification method based on single-energy-level CT images provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0126] The present application embodiment provides a computer-readable storage medium, such as Figure 4 As shown, the computer-readable storage medium 401 stores a computer program, and when the computer program is read and executed by the second processor 402, the material composition identification method based on the single-energy-level CT image as described above is implemented.

[0127] The technical solution of the embodiment of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling an electronic device (which may be an air conditioner, a refrigeration device, a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the method described in the embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0128] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the material composition identification method based on single-energy-level CT images provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0129] An embodiment of the present application provides a computer program product, including a computer program, wherein the computer program is executed by a third processor to implement the method described above.

[0130] The computer program product provided in the above-mentioned embodiments of the present application and the material composition identification method based on single-energy-level CT images provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0131] It should be noted that, in this application, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or still includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0132] Each embodiment in the present application is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the evaluation of the material component identification method based on single-energy CT images, electronic devices, electronic devices, and readable storage medium embodiments, since they are basically similar to the above-mentioned material component identification method based on single-energy CT images, the description is relatively simple, and the relevant parts can be referred to the partial description of the above-mentioned material component identification method based on single-energy CT images.

[0133] Although the present application is disclosed as above, the present application is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application shall be subject to the scope defined by the claims.

Claims

1. A material composition identification method based on single-level CT images, characterized in that: include: Acquire a CT image of a target object, wherein the CT image of the target object includes at least one set of single-energy-level CT images; performing grouping processing on each slice image of the single-energy-level CT image to generate a plurality of single-energy-level CT slice image groups; Acquire a training sample set and an initial material component identification model, wherein the training sample set includes a single-energy-level CT slice image group corresponding to a known reference material, and the initial material component identification model is formed based on the change trend of CT image values ​​of different reference materials at different energy levels, and the initial material component identification model includes reference values ​​for characterizing characteristic parameters corresponding to material components and their corresponding parameter spaces; Processing the initial material component recognition model based on the training sample set to generate a target material component recognition model; Performing semantic segmentation processing on the CT image of the target object to generate a plurality of image regions and image region types matching the plurality of image regions; Based on the target material component recognition model, each pixel in a plurality of image regions is processed respectively to generate a characteristic parameter value corresponding to each pixel; Processing the characteristic parameter value of each pixel in the plurality of image regions and the image region types matching the plurality of image regions based on the target material composition recognition model to generate material composition information corresponding to each pixel region; The material composition information corresponding to each pixel area is processed to generate attribute information of the pixel area, wherein the attribute information of the pixel area includes the relative electron density, equivalent atomic number, and relative radiation tissue power of each pixel.

2. The method according to claim 1, characterized in that Based on the target material component recognition model, each pixel in a plurality of image regions is processed respectively to generate a characteristic parameter value corresponding to each pixel, including: The target material component identification model includes a calculation formula for obtaining characteristic parameter values, and the calculation formula is: CTHU(keV)=a e b·keV +c; a=ω1·a x1 +ω2·a x2 ; b=ω1·b x1 +ω2·b x2 ; c=ω1·c x1 +ω2·c x2 ; ω1+ω2=1; Among them, a, b, and c represent characteristic parameters, e represents the exponential function, keV represents the energy level corresponding to the single-level CT image, x1 and x2 are reference substances, and ω1 and ω2 are the content information of the reference substances.

3. The method according to claim 2, characterized in that Based on the target material composition recognition model, the characteristic parameter value of each pixel in the plurality of image regions and the image region types matching the plurality of image regions are processed to generate material composition information corresponding to each pixel region, including: The target material composition identification model includes a calculation formula for obtaining material composition information, and the calculation formula is: Among them, p x,y,z Represents the material composition information corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, M i represents the i-th reference substance, and n represents the number of reference substances constituting the material composition of the pixel.

4. The method according to claim 1, characterized in that The target material composition identification model includes obtaining multi-parameter information such as relative electron density, equivalent atomic number and relative radiation tissue power of the material, and the calculation formulas are: The target material component identification model includes a calculation formula for obtaining relative electron density, which is: Among them, RED x,y,z Represents the relative electron density corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, i represents the i-th reference substance, ρ i represents the density of the i-th reference material, ρ water Represents the density of water, RED i represents the electron density of the i-th reference substance; The target material component identification model includes a calculation formula for obtaining the equivalent atomic number, which is: Among them, Zeff x,y,z Represents the equivalent atomic number corresponding to the pixel with coordinate position x, y, z, ω i represents the content of the i-th reference substance, i represents the i-th reference substance, and Z i represents the equivalent atomic number of the i-th reference substance; The target material component identification model includes a calculation formula for obtaining relative radiological tissue power, which is: Among them, SPR stands for relative radiological power and RED stands for the electron density of the reference material.

5. The method according to claim 1, characterized in that The initial material component recognition model is processed based on the training sample set to generate a target material component recognition model, including: Preprocessing the training sample set to generate a training sample set with identification information, wherein the training sample set includes information of a plurality of standard substances; The initial material component recognition model is trained based on the training sample set with identification information to generate a target material component recognition model.

6. The method according to claim 5, characterized in that Preprocessing the training sample set to generate a training sample set with identification information includes: Extracting features from the training sample set to determine an original feature library; Divide each feature data set according to the original feature library to generate a training data set and a test data set; Use the classifier to divide the original feature library into various test data sets for prediction and determine the prediction results; Use the preset algorithm to divide each training data set in the original feature library for training, and obtain the prediction results of the test set class; Based on the prediction results and the test set class prediction results, training samples with identification information are generated.

7. The method according to claim 6, characterized in that The step of extracting features from the training sample set to determine an original feature library includes: Processing the standard substance information based on preset processing rules to generate standardized features, wherein the standardized features are features of a sample or model with a clear and known material composition; Processing the standardized features based on preset feature screening and dimension reduction rules to generate original features; Generate a primitive feature library from several primitive features.

8. A material composition identification device based on single-level CT images, characterized in that: The device comprises: An acquisition module is used to acquire a CT image of a target object, wherein the CT image of the target object includes at least one group of single-energy-level CT images; acquire a training sample set and an initial material component recognition model, wherein the training sample set includes a group of single-energy-level CT slice images corresponding to a known reference material, and the initial material component recognition model is formed based on the change trend of CT image values ​​of different reference materials at different energy levels, and the initial material component recognition model includes reference values ​​for characterizing characteristic parameters corresponding to material components and their corresponding parameter spaces; The processing module is used to group each slice image of the single-energy CT image to generate a plurality of single-energy CT slice image groups; process the initial material composition recognition model based on the training sample set to generate a target material composition recognition model; perform semantic segmentation on the CT image of the target object to generate a plurality of image regions and image region types matching the plurality of image regions; process each pixel in the plurality of image regions respectively based on the target material composition recognition model to generate a characteristic parameter value corresponding to each pixel; process the characteristic parameter value of each pixel in the plurality of image regions and the image region types matching the plurality of image regions based on the target material composition recognition model to generate material composition information corresponding to each pixel region; process the material composition information corresponding to each pixel region to generate attribute information of the pixel region, wherein the attribute information of the pixel region includes the relative electron density, equivalent atomic number, and relative radiographic tissue power of each pixel.

9. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions of the first processor; Wherein, the first processor is configured to execute the material composition identification method based on single-energy-level CT images as described in any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, the material component identification method based on the single-energy-level CT image according to any one of claims 1 to 7 is implemented.

Images