Model training method and apparatus, method and apparatus for determining elemental density distribution of constituent, and device and medium
By acquiring the scanning energy spectrum characteristic parameters of CT equipment, constructing a sample dataset, and training a fully connected neural network model, the problem of difficulty in distinguishing tissue materials in CT imaging technology was solved, and more accurate determination of element density distribution was achieved.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- CAS ION MEDICAL TECHNOLOGY CO LTD
- Filing Date
- 2025-07-24
- Publication Date
- 2026-06-04
AI Technical Summary
Existing CT imaging technology has difficulty in accurately distinguishing tissue materials, especially when based on tissue materials with the same HU value under a single energy spectrum, leading to deviations in diagnostic results.
By acquiring the energy spectrum characteristic parameters of CT equipment, a sample dataset is constructed. A fully connected neural network is used to train a model to learn the correlation between HU value and element density, and a target model is generated to determine the element density distribution.
It improves the accuracy of CT imaging in distinguishing tissue materials, reduces resource consumption, and enhances the model's robustness to noise and tissue material density distribution.
Smart Images
Figure CN2025110353_04062026_PF_FP_ABST
Abstract
Description
Model training methods, methods for determining the density distribution of material elements, apparatus, equipment, and media.
[0001] This application claims priority to Chinese Patent Application No. 202411738116.5, filed on November 29, 2024, the contents of which are incorporated herein by reference. Technical Field
[0002] This disclosure relates to the fields of physics and computer science, specifically to the application of physics and computer technology in the field of radiomedicine, and more specifically to a model training method, a method for determining the density distribution of material elements, an apparatus, electronic equipment, and a storage medium. Background Technology
[0003] Computed tomography (CT) imaging technology has wide applications in medical diagnostics. CT imaging and image reconstruction techniques can reveal the internal structure of an object being examined. For example, CT images can be used to obtain the elemental density distribution of a substance, which is of great significance in radiological diagnosis and radiotherapy. Summary of the Invention
[0004] This disclosure provides a model training method, a method for determining the density distribution of material elements, an apparatus, an electronic device, a storage medium, and a program product.
[0005] According to the first aspect, this disclosure provides a model training method, comprising: acquiring characteristic parameters of each of M scanning energy spectra of a computed tomography (CT) device; determining M sample CT images of each of the N sample materials under the M scanning energy spectra based on the M characteristic parameters and the elemental density distributions of N sample materials, thereby obtaining M*N sample CT images; constructing a sample dataset based on the M sample CT images of each of the N sample materials and their respective elemental density distributions; and training an initial model using the sample dataset to obtain a target model corresponding to the CT device, wherein the target model is used to determine the elemental density distribution based on the CT images obtained by scanning with the CT device.
[0006] According to a second aspect, this disclosure provides a method for determining the elemental density distribution of a substance, comprising: scanning the substance to be tested with M scanning energy spectra of a CT device to obtain M CT images; and determining the elemental density distribution of the substance to be tested based on the M CT images using an elemental analysis model; wherein the elemental analysis model is a target model trained using the model training method provided in the embodiments of this disclosure.
[0007] According to a third aspect, this disclosure provides a model training apparatus, comprising: an acquisition module for acquiring characteristic parameters of each of M scanning energy spectra of a computed tomography (CT) device; a first determination module for determining M sample CT images of each of the N sample materials under the M scanning energy spectra based on the M characteristic parameters and the elemental density distributions of N sample materials, thereby obtaining M*N sample CT images; a construction module for constructing a sample dataset based on the M sample CT images of each of the N sample materials and their respective elemental density distributions; and a training module for training an initial model using the sample dataset to obtain a target model corresponding to the CT device, wherein the target model is used to determine the elemental density distribution based on the CT images obtained by scanning with the CT device.
[0008] According to a fourth aspect, this disclosure provides an apparatus for determining the elemental density distribution of a substance, comprising: a scanning module for scanning the substance to be tested using M scanning energy spectra of a CT device to obtain M CT images; and a second determining module for determining the elemental density distribution of the substance to be tested based on the M CT images using an elemental analysis model; wherein the elemental analysis model is a target model trained using a model training apparatus as provided in the embodiments of this disclosure.
[0009] According to a fifth aspect, this disclosure provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the model training method and / or the method for determining the density distribution of material elements provided in this disclosure.
[0010] According to a sixth aspect, this disclosure provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the model training method and / or the method for determining the density distribution of material elements provided in this disclosure.
[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description
[0012] Figure 1 is a schematic flowchart of a model training method according to an embodiment of the present disclosure;
[0013] Figure 2 is a schematic diagram of the model training method according to an embodiment of the present disclosure;
[0014] Figure 3 is a schematic diagram of the principle of the target model according to an embodiment of the present disclosure;
[0015] Figure 4 is a flowchart illustrating a method for determining the density distribution of material elements according to an embodiment of the present disclosure;
[0016] Figure 5 is a structural block diagram of a model training apparatus according to an embodiment of the present disclosure;
[0017] Figure 6 is a structural block diagram of a device for determining the density distribution of material elements according to an embodiment of the present disclosure; and
[0018] Figure 7 is a schematic block diagram of an example electronic device used to implement embodiments of the present disclosure. Detailed Implementation
[0019] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0020] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.
[0021] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of data (including but not limited to user personal information) comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.
[0022] Figure 1 is a schematic flowchart of a model training method according to an embodiment of the present disclosure.
[0023] As shown in Figure 1, the model training method 100 of this embodiment may include operations S110 to S140.
[0024] In operation S110, the characteristic parameters of each of the M scanning energy spectra of the computed tomography (CT) device are obtained, where M is a positive integer.
[0025] In this embodiment of the disclosure, the CT device scans the same tissue material based on multiple scanning energy spectra, and the resulting CT images may differ. The CT images include the distribution of HU values of the tissue material, which can reflect the interaction between the material and photons.
[0026] Different scanning energy spectra can be formed based on different X-ray sources, and multiple scanning energy spectra may contain different energy components. When CT equipment scans the same tissue material based on different scanning energy spectra, the HU values may be the same or different.
[0027] For example, dual-energy CT scanners have two scanning energy spectra. When multiple tissue materials are scanned using a single X-ray source, these materials may have different elemental compositions but the same HU value under a single energy spectrum. In such cases, it is impossible to further distinguish between the two tissue materials in the CT images, leading to inaccurate diagnostic results. Therefore, dual-energy CT allows for the comparison of two CT images of two tissue materials scanned using the two energy spectra respectively. When two tissue materials have the same HU value under one energy spectrum, their HU values under the other energy spectrum will often differ, thus enabling further differentiation between the two tissue materials.
[0028] In the embodiments of this disclosure, the characteristic parameters can describe the energy combination of multiple emission sources of a CT device. For example, the emission source can be composed of multiple sub-emission sources with different energies, and the photon energy range emitted by the multiple sub-emission sources can be within 20 to 120 keV. The photon energies emitted by the multiple sub-emission sources form a corresponding scanning energy spectrum.
[0029] For example, among multiple emission sources, some emission sources emit more 40keV photons, some emit more 60keV photons, and some emit more 80keV photons.
[0030] The energy of photons emitted by multiple emission sources can be determined based on the characteristic parameters of the scanned energy spectrum.
[0031] In operation S120, based on M characteristic parameters and the elemental density distribution of N sample materials, M sample CT images of each of the N sample materials are determined under M scanning energy spectra, resulting in M*N sample CT images, where N is a positive integer.
[0032] In this embodiment of the disclosure, the element density distribution of the sample material can be obtained in advance from a database. The element density distribution can describe the types of elements included in the sample material, the density of each element in the sample material, and the distribution of multiple elements in the sample material.
[0033] Based on characteristic parameters, the composition of photon energy emitted by the source corresponding to the scanned energy spectrum can be determined. Based on the elemental density distribution of the sample material, the degree of absorption of photons of different energies by the sample material can be determined, thereby calculating the sample CT image of the sample material.
[0034] Because different CT scanners use different scanning energy spectra, the CT images obtained from scanning the same material tissue will differ. Furthermore, known databases only record relevant parameters of the sample material and cannot obtain real tissue with the same characteristics as the sample material. Therefore, it is impossible to use a CT scanner to scan the real tissue of the sample material to determine the CT image obtained from the CT scan. Therefore, this disclosure simulates a virtual scanning environment of the CT scanner based on characteristic parameters and simulates a virtual tissue based on the elemental density distribution of the sample material. In this case, the virtual scanning environment is used to scan the virtual tissue, resulting in the obtained sample CT image.
[0035] In this embodiment of the disclosure, the sample CT image is calculated based on characteristic parameters of elemental density distribution and scanning energy spectrum. Based on the characteristic parameters, the scanning environment of the CT equipment can be simulated to determine the photon energy emitted by the emission source. Based on the elemental density distribution, the elemental density at different locations in the sample material can be simulated, thereby calculating the absorption capacity of different locations in the sample material for photon energy, and then calculating the HU value at different locations in the sample material based on the absorption capacity to obtain the sample CT image.
[0036] In operation S130, a sample dataset is constructed based on M sample CT images and elemental density distributions of each of the N sample materials.
[0037] In this embodiment of the disclosure, the sample CT image includes multiple HU values, and the elemental density distribution includes the elemental density of the region corresponding to each HU value. For example, each sample material includes multiple voxels. In M sample CT images of each sample material, each voxel corresponds to M HU values. Understandably, the M HU values correspond to the elemental density of the voxel.
[0038] For example, M HU values determined under different scanning energy spectra describe voxels that represent the same elemental density. The sample dataset constructed based on sample CT images and elemental densities includes multiple sets of correspondences between HU values and elemental densities.
[0039] In operation S140, an initial model is trained using the sample dataset to obtain the target model corresponding to the CT device.
[0040] In this embodiment of the disclosure, the initial model can be a fully connected neural network. By training the fully connected neural network using a sample dataset, the correlation between the HU value and element density can be extracted using fully connected operations, enabling the trained target model to determine the element density distribution based on CT images obtained from CT scans.
[0041] For example, a fully connected neural network can extract features of HU value and element density, and perform fully connected operations on the extracted features to determine the correspondence between HU value and element density.
[0042] In the target model, HU values are input data, and elemental densities are output data. Therefore, by inputting CT images of the substance to be tested into the target model, the elemental densities of different regions within the substance can be output, thereby determining the elemental density distribution.
[0043] The following explanation will be based on the scenario shown in Figure 2. Figure 2 is a schematic diagram illustrating the scenario principle of the model training method according to an embodiment of this disclosure.
[0044] As shown in Figure 2, the sample material 200 can be divided into multiple objects. For example, the sample material 200 can be a cubic structure, and the sample material 200 can be divided into multiple objects, each of which is also a cubic structure.
[0045] In this embodiment of the disclosure, each object can be a voxel of the sample material 200, and each object can be equivalent to a point inside the sample material 200. The element density distribution of the sample material 200 can be determined based on the element densities of the multiple objects.
[0046] For example, sample material 200 is divided into 5×3×3 objects. The elemental density distribution can include the density distributions of carbon, hydrogen, oxygen, nitrogen, phosphorus, and calcium. For example, the carbon density distribution of sample material 200 can be represented by a 5×3×3 three-dimensional matrix.
[0047] Based on the characteristic parameters of multiple scanned energy spectra and the elemental density distribution of sample material 200, the HU values of 30 objects can be calculated. Specifically, based on M scanned energy spectra, M HU values can be calculated for each object. Each object's M HU values correspond to the same elemental density.
[0048] For example, based on M scanned energy spectra, M HU values of object 201 can be calculated. These M HU values are then correlated with the elemental densities of object 201. Based on 30 sets of correlated HU values and elemental densities of sample material 200, a sample dataset can be constructed.
[0049] Through the embodiments of this disclosure, a sample dataset for CT equipment can be constructed based on the elemental density distribution of the sample material and the energy spectrum characteristic parameters of the CT equipment. Using the HU values and elemental densities data pairs included in the sample dataset, an initial model is trained, enabling the initial model to learn the correlation between HU values and elemental densities, thus obtaining a target model. Having learned the correlation between HU values and elemental densities under the constraints of the CT equipment's scanning energy spectrum, the target model can output the elemental density distribution of the analyte corresponding to the CT image obtained from the CT scan. Furthermore, the model training method provided in this disclosure can be implemented based on a fully connected neural network, requiring fewer hardware and computational resources for model training, thereby reducing resource overhead.
[0050] In some embodiments, the characteristic parameters of the scanning energy spectrum of the CT device can be determined using a known standard phantom. For example, a reference material is scanned using M scanning energy spectrum distributions of the CT device to obtain M scanning data of the reference material; and the M scanning data are compared with the reference data of the reference material obtained in advance to obtain M characteristic parameters of the M scanning energy spectra.
[0051] In this embodiment of the disclosure, the reference material can be a known standard phantom. For example, information such as the elemental density distribution, total density of the material, and CT images obtained based on known scanning energy spectra of the reference material are all known.
[0052] For example, the reference material could be a Gammex phantom or a CIRS phantom. The Gammex phantom is scanned using a CT scanner based on M scanning energy spectra, yielding M CT images of the Gammex phantom as scan data. Based on the known CT images, elemental density distribution, and total material density of the Gammex phantom, the absorption degree of each voxel in the Gammex phantom for a specific photon energy can be determined. Based on the absorption degree and the scan data, the photon energy emitted by the CT scanner's emission source is calculated, thereby determining the characteristic energy spectrum of the CT scanner.
[0053] For example, the absorption capacity of a voxel of a certain material in a Gammex phantom for rays with a specific energy can be considered constant. Based on known CT images of the Gammex phantom, the correspondence between the HU value and the photon energy of the energy spectrum is determined. Based on this correspondence, and based on the HU values in the scan data, the photon energy of the energy spectrum is calculated. Thus, based on M CT images in the scan data, M characteristic parameters of the scan energy spectrum can be determined.
[0054] In some embodiments, a method for calculating M*N sample CT images obtained by scanning N sample materials with a CT device may include: using M feature parameters and the elemental density distribution of the N sample materials to determine the energy absorption capacity of each of the N sample materials to M scanning energy spectra; and based on the energy absorption capacity, determining M sets of CT values for each of the N sample materials, wherein the M sets of CT values form M CT images.
[0055] In this embodiment of the disclosure, each sample material can be divided into multiple voxels, and the absorption capacity of each voxel for photon energy is determined by the voxel's density and elemental distribution. For example, if the density of calcium and phosphorus in a voxel is high, the voxel may be bone tissue, which has a strong absorption capacity for photon energy and a correspondingly high HU value. If the density of carbon, oxygen, hydrogen, and nitrogen in a voxel is high, the voxel may be soft tissue, which has a weak absorption capacity for photon energy and a correspondingly low HU value.
[0056] Based on the absorption capacity of each voxel in the sample material for photon energy and the correspondence between the HU value and the photon energy of the energy spectrum, the HU value of each voxel in the sample material is calculated, and the CT image of the sample material can be reconstructed based on the HU value.
[0057] Since each voxel in the sample material has a different absorption capacity for different photon energies, different HU values can be calculated for each voxel based on the scanning energy spectrum described by different characteristic parameters. For M characteristic parameters, M HU values can be calculated for each voxel.
[0058] In some embodiments, the method for constructing a sample dataset may include performing the following operations for each of the N sample materials: determining M CT values for each of the multiple sample objects in the sample material based on M sample CT images; obtaining M element densities corresponding to the M CT values based on element density distribution; determining sample data for each of the multiple sample objects based on the M CT values and M element densities; and constructing a sample dataset for the sample materials based on the sample data.
[0059] In this embodiment of the disclosure, the sample object can be a voxel in the sample material, and the CT value is the HU value of each voxel in the CT image. The HU values of multiple voxels in the sample material can be represented in the form of a three-dimensional matrix, and the element density distribution can also be represented in the form of an element density of each voxel in the form of a three-dimensional matrix. In the three-dimensional matrix, the HU value and element density of the same voxel have the same coordinates. In the three-dimensional matrix of HU values of multiple M sample CT images, the M HU values at the same coordinate all correspond to the element density at that coordinate in the three-dimensional matrix of element density distribution.
[0060] M HU values at the same coordinate and their corresponding elemental densities form a data pair for that voxel. This data pair can be used to describe the relationship between the HU value and elemental density of that voxel. Data pairs of voxels from N sample materials form a sample dataset. The sample dataset describes the relationship between the HU value and elemental density of each voxel.
[0061] In some embodiments, data augmentation can also be performed on the sample data of each sample material to improve the accuracy of the sample data. For example, within a preset range, multiple sets of augmenting element densities are generated based on M element densities, each set of augmenting element densities includes M element densities, and the difference between the M element densities and each set of augmenting element densities is within a preset range; and based on the M CT values of each of the multiple sample objects and the multiple sets of augmenting element densities, sample data for each of the multiple sample objects is determined.
[0062] In this embodiment of the disclosure, the accuracy of the element density distribution output by the target model is related to the quality of the sample data. Augmenting the sample data can increase its richness, thereby improving the quality of the target model.
[0063] For example, due to errors and uncertainties in element density and HU values, sample augmentation can be performed on the element density of each voxel in the sample material so that the M HU values of each voxel can correspond to multiple element densities. This increases the number of data pairs used to characterize the correlation between the HU values and element densities of each voxel, thereby reducing the error of the data pairs before data augmentation.
[0064] For example, suppose the elemental density of voxels in a sample material follows a Gaussian distribution. For a given sample material, 100 sets of elemental densities are randomly generated, centered on the elemental density of that voxel and with a standard deviation of 10% of the centered value. Each set of elemental densities includes the densities of multiple elements. Based on these 100 sets of elemental densities and the M HU values of the voxel, 100 data pairs are constructed, thus enhancing the original one set of data pairs into 100 data pairs.
[0065] For example, data augmentation can be performed on the overall density of the sample material, and the overall density of the sample material and CT images can be constructed as data pairs. The data pairs of overall density of the sample material and CT images can describe the correlation between the overall density of the sample material and CT images (all HU values), and can also reflect the influence of the correlation between the HU values of adjacent voxels on the overall density.
[0066] For example, for a sample material, 100 population densities are randomly generated, with the population density of the sample material as the center value and 10% of the center value as the standard deviation. Based on these 100 population densities and M sample CT images of the sample material, 100 data pairs are constructed, thereby enhancing the original set of data pairs into 100 data pairs.
[0067] For example, data augmentation can be performed on the elemental density distribution of the sample material, and the elemental density distribution and CT images can be constructed as data pairs. Constructing elemental density distribution and CT images as data pairs can describe the correlation between the overall elemental density distribution of the sample material and the CT images, and can also reflect the influence of the correlation between the elemental densities of adjacent voxels on the HU value.
[0068] For example, for a sample material, 100 element density distributions are randomly generated, with the element density distribution of the sample material as the center value and 10% of the center value as the standard deviation. Based on these 100 element density distributions and M sample CT images of the sample material, 100 data pairs are constructed, thereby enhancing the original set of data pairs into 100 data pairs.
[0069] In this embodiment of the disclosure, the overall density, elemental density distribution, and elemental density of a single voxel of the sample material can be augmented, and a data pair can be constructed based on at least one of the augmented overall density, augmented elemental density distribution, and augmented elemental density of a single voxel with the HU value of the CT image or a single voxel.
[0070] Based on this, data augmentation can expand the amount of data, reduce data errors, and improve the uncertainty in the data, thereby optimizing the correlation between the two data points in a data pair and increasing the robustness of model training against CT image noise and uncertainties in the elemental density distribution of tissue materials. Furthermore, constructing data pairs based on overall density and elemental density distribution can fully consider the correlation and mutual influence between voxels, avoiding the neglect of inter-data correlations in data pairs constructed based on individual voxels, thus improving the quality of the target model.
[0071] In some embodiments, training an initial model using a sample dataset to obtain a target model corresponding to a CT device may include: determining sample CT images in the sample dataset as input data for the initial model, and determining the element density distribution in the sample dataset as output data for the initial model; using the distribution characteristics of multiple elements in the element density distribution as prompt information; and training the initial model based on the prompt information, input data, and output data to obtain the target model.
[0072] In this embodiment of the disclosure, the sample CT image is used as the input data of the model, and the elemental density of the sample material is used as the output data of the model, so that the model can calculate and output the elemental density distribution of the substance to be tested based on the CT image of the substance to be tested.
[0073] For example, the M HU values of a single voxel can be used as input data, and the elemental density of a single voxel can be used as output data. This allows the model to calculate and output the elemental density of each voxel based on the HU value of each voxel in the substance to be tested, thereby determining the elemental density distribution of the substance to be tested.
[0074] The initial model can learn the correlation between the HU value and element density of a single voxel described by data in the sample dataset, as well as the influence of element density on the HU value between adjacent voxels, the influence of the overall element density distribution of the sample material on the HU value of a single voxel, and the influence of the overall density of the sample material on the HU value of a single voxel.
[0075] In this embodiment of the disclosure, the distribution characteristics of multiple elements in the element density distribution can be the pattern of how the HU value changes with the element density. For example, the trend of element density changing with the HU value differs for different elements. Using the distribution characteristics of each element as prompting information can improve the accuracy of the element density output by the target model based on the HU value.
[0076] For example, based on the HU value of each voxel in the sample material, the relative electron density of each voxel can be calculated. And the effective atomic number Zeff, to determine the relative electron density of multiple voxels. The relationship between effective atomic number Zeff and elemental density.
[0077] For example, relative electron density based on voxels The effective atomic number (Zeff) can determine whether a voxel belongs to bone or soft tissue. For example, if the effective atomic number (Zeff) is greater than 8.2, the material type of the analyte is determined to be bone tissue. If the effective atomic number (Zeff) is less than or equal to 8.2, the material type of the analyte is determined to be soft tissue. For example, the analyte could be animal tissue. Bone tissue can include skeletons, etc., and contains a large amount of calcium and phosphorus, as well as a certain amount of carbon, oxygen, hydrogen, and nitrogen. Soft tissue includes a large amount of fat and protein, etc., and contains a large amount of carbon, oxygen, hydrogen, and nitrogen, as well as a small amount of calcium and phosphorus. The calcium and phosphorus content in soft tissue is usually less than 0.2%, so the calcium and phosphorus content in soft tissue is negligible.
[0078] For example, in bone tissue, the sum of the mass percentages of oxygen and carbon is related to their relative electron density. Both the effective atomic number Zeff and the mass percentage of calcium are negatively correlated with the relative electron density. Both the effective atomic number Zeff and the mass percentage of phosphorus are positively correlated with the relative electron density. Both the mass percentage of carbon and the effective atomic number Zeff are positively correlated with relative electron density. The variation of the effective atomic number Zeff satisfies the constraints of the arctangent function. The mass percentage of nitrogen can be determined by subtracting 100% from the sum of the mass percentages of calcium, phosphorus, carbon, oxygen, and hydrogen. The sum of the mass percentages of carbon, oxygen, hydrogen, nitrogen, calcium, and phosphorus is greater than 99% and less than 100%.
[0079] For soft tissues, the mass percentages of oxygen and carbon are negatively correlated. The mass percentage of carbon decreases as the oxygen content increases.
[0080] Based on the mass percentages of calcium, phosphorus, carbon, oxygen, hydrogen, and nitrogen, the density distributions of these elements can be determined.
[0081] In this embodiment of the disclosure, the mass percentage of each element in the voxel varies with the relative electron density. By studying the variation of effective atomic number Zeff, the model determines the variation of the mass percentage of each element in a voxel with the HU value, and further determines the variation of the elemental density of each element in the voxel with the HU value. The initial model learns this variation in elemental density with the HU value, enabling the target model to improve the accuracy of calculating the elemental density of voxels based on the HU value.
[0082] For example, after the target model calculates the elemental density of multiple voxels based on their HU values, it can optimize and adjust the elemental density of each voxel based on the variation patterns described in the prompt information. The target model can also assist in calculating elemental densities that are difficult to calculate based on HU values, using the variation patterns described in the prompt information and the elemental densities already calculated based on HU values.
[0083] The training process will be illustrated by way of example with reference to Figure 3, which is a schematic diagram of the principle of the target model according to an embodiment of the present disclosure.
[0084] In this embodiment of the disclosure, a CT scanner is used to scan the substance to be tested based on M scanning energy spectra to obtain M CT images. Based on the M CT images, the M HU values HU1, HU2, ..., HUM of each voxel in the substance to be tested are determined.
[0085] The input data 301 of the target model 303 is M HU values HU1, HU2, ..., HUM for each voxel, and the output data 302 is the elemental density for each voxel. The values can be H, C, N, O, P, and Ca, representing the elemental densities of hydrogen, carbon, nitrogen, oxygen, phosphorus, and calcium, respectively.
[0086] In this embodiment, the CT device used to determine the CT image of the substance to be tested and the CT device used to determine the CT image of the sample material are the same CT device. The target model 303 is trained based on the sample CT image simulated using the feature parameters of the CT device, and the target model 303 also needs to perform elemental density distribution calculations based on the CT images scanned by this CT device. Therefore, the target model 303 is associated with the CT device. When the CT device changes, the initial model needs to be retrained to obtain a new target model.
[0087] In this embodiment of the disclosure, the target model 303 is trained based on M CT images of each sample material. This is because, in order to ensure the calculation accuracy of the target model 303, M CT images obtained by M scanning energy spectrum scans based on CT equipment are required as input data 301.
[0088] Figure 4 is a flowchart illustrating a method for determining the density distribution of material elements according to an embodiment of the present disclosure.
[0089] As shown in Figure 4, the method 400 for determining the density distribution of material elements in this embodiment may include operations S410 to S420.
[0090] When operating S410, the substance to be tested is scanned using M scanning energy spectra of the CT equipment to obtain M CT images.
[0091] When operating the S420, the elemental density distribution of the substance to be tested is determined based on M CT images using an elemental analysis model.
[0092] In this embodiment of the disclosure, the elemental analysis model is a target model trained using the model training method 100 provided in this embodiment of the disclosure.
[0093] For example, the HU values from M CT images can be output to an elemental analysis model, which can then output the elemental density distribution of the substance being tested.
[0094] In this embodiment of the disclosure, the process of determining the elemental density distribution of the substance to be tested based on M CT images using an elemental analysis model is similar to the training process described above, and will not be repeated here for the sake of brevity.
[0095] Based on the model training method provided in this disclosure, this disclosure also provides a model training device, which will be described in detail below with reference to Figure 5.
[0096] Figure 5 is a structural block diagram of a model training apparatus according to an embodiment of the present disclosure.
[0097] As shown in FIG5, the model training device 500 of this embodiment may include an acquisition module 510, a first determination module 520, a construction module 530 and a training module 540.
[0098] The acquisition module 510 is used to acquire the characteristic parameters of each of the M scanning energy spectra of the computed tomography (CT) device. In one embodiment, the acquisition module 510 can be used to perform the operation S110 described above, which will not be repeated here.
[0099] The first determining module 520 is used to determine M sample CT images of each of the N sample materials under M scanning energy spectra based on M feature parameters and the elemental density distribution of N sample materials, thereby obtaining M*N sample CT images. In one embodiment, the first determining module 520 is used to perform the operation S120 described above, which will not be repeated here.
[0100] The construction module 530 is used to construct a sample dataset based on M sample CT images and elemental density distributions of each of the N sample materials. In one embodiment, the construction module 530 can be used to perform the operation S130 described above, which will not be repeated here.
[0101] The training module 540 is used to train an initial model using a sample dataset to obtain a target model corresponding to the CT device. The target model is used to determine the element density distribution based on the CT images obtained from the CT scan. In one embodiment, the training module 540 can be used to perform the operation S140 described above, which will not be repeated here.
[0102] According to an embodiment of this disclosure, the acquisition module 510 acquires the characteristic parameters of each of the M scanning energy spectra of the computed tomography (CT) device, including: scanning a reference material using the distribution of the M scanning energy spectra of the CT device to obtain M scanning data of the reference material; and comparing the M scanning data with the reference data of the reference material obtained in advance to obtain the M characteristic parameters of the M scanning energy spectra.
[0103] According to embodiments of this disclosure, the construction module 530 is used to construct a sample dataset based on M sample CT images and element density distributions of each of N sample materials, including: for each of the N sample materials, performing the following operations: based on the M sample CT images, determining M CT values of each of the multiple sample objects in the sample material; based on the element density distribution, obtaining M element densities corresponding to the M CT values; based on the M CT values and M element densities, determining sample data of each of the multiple sample objects; and based on the sample data, constructing a sample dataset of the sample materials.
[0104] According to an embodiment of this disclosure, the construction module 530 is used to determine sample data for each of multiple sample objects based on M CT values and element densities, including: generating multiple sets of enhancement element densities based on the M element densities within a preset range, each set of enhancement element densities including M enhancement element densities, and the difference between the M element densities and each set of enhancement element densities being within a preset range; and determining sample data for each of the multiple sample objects based on the M CT values and multiple sets of enhancement element densities for each of the multiple sample objects.
[0105] According to an embodiment of this disclosure, the first determining module 520 is used to determine M sample CT images of each of the N sample materials under M scanning energy spectra based on M feature parameters and the elemental density distribution of N sample materials, thereby obtaining M*N sample CT images. This includes: using the M feature parameters and the elemental density distribution of the N sample materials to determine the energy absorption capacity of each of the N sample materials to the M scanning energy spectra; and based on the energy absorption capacity, determining M sets of CT values for each of the N sample materials, wherein the M sets of CT values form M CT images.
[0106] According to an embodiment of this disclosure, the training module 540 is used to train an initial model using a sample dataset to obtain a target model corresponding to the CT device. The target model is used to determine the element density distribution based on the CT images obtained by scanning with the CT device, including: determining the sample CT images in the sample dataset as input data of the initial model, and determining the element density distribution in the sample dataset as output data of the initial model; using the distribution characteristics of multiple elements in the element density distribution as prompt information; and training the initial model based on the prompt information, input data, and output data to obtain the target model.
[0107] Figure 6 is a structural block diagram of a device for determining the density distribution of material elements according to an embodiment of the present disclosure.
[0108] As shown in FIG6, the material element density distribution determination device 600 of this embodiment may include a scanning module 610 and a second determination module 620.
[0109] The scanning module 610 is configured to scan the substance to be tested using M scanning energy spectra of the CT device to obtain M CT images. In one embodiment, the scanning module 610 can be used to perform the operation S410 described above, which will not be repeated here.
[0110] The second determining module 620 is configured to use an elemental analysis model to determine the elemental density distribution of the analyte based on M CT images. In one embodiment, the second determining module 620 is used to perform the operation S420 described above, which will not be repeated here.
[0111] In this embodiment of the disclosure, the elemental analysis model is a target model trained using, for example, the model training apparatus 500 provided in this embodiment of the disclosure.
[0112] It should be noted that the collection, storage, use, processing, transmission, provision, disclosure, and application of user personal information in this disclosed technical solution comply with relevant laws and regulations, necessary confidentiality measures have been taken, and it does not violate public order and good morals. In this disclosed technical solution, user authorization or consent has been obtained before acquiring or collecting user personal information.
[0113] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.
[0114] Figure 7 illustrates a schematic block diagram of an example electronic device 700 that can be used to implement the methods of embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0115] As shown in Figure 7, the electronic device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. The RAM 703 can also store various programs and data required for the operation of the electronic device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.
[0116] Multiple components in electronic device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of displays, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows electronic device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0117] The computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as methods for determining the density distribution of material elements and / or model training methods. For example, in some embodiments, the methods for determining the density distribution of material elements and / or model training methods can be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the methods for determining the density distribution of material elements and / or model training methods described above can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured by any other suitable means (e.g., by means of firmware) to perform methods for determining the density distribution of material elements and / or model training.
[0118] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0119] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0120] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0121] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0122] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.
[0123] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is established by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.
[0124] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.
[0125] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A model training method, comprising: Obtain the characteristic parameters of each of the M scanning energy spectra of a computed tomography (CT) device, where M is a positive integer; Based on the M characteristic parameters and the elemental density distribution of N sample materials, M sample CT images of each of the N sample materials are determined under the M scanning energy spectra, resulting in M*N sample CT images, where N is a positive integer. Based on the M sample CT images and element density distributions of the N sample materials, a sample dataset is constructed. as well as An initial model is trained using the sample dataset to obtain a target model corresponding to the CT device. The target model is used to determine the element density distribution based on the CT images obtained by scanning the CT device.
2. The method according to claim 1, wherein, The acquisition of characteristic parameters for each of the M scanning energy spectra of the computed tomography (CT) device includes: Using the M scanning energy spectrum distributions of the CT equipment, a reference material is scanned to obtain M scanning data of the reference material; and The M scan data are compared with the reference data of the reference material obtained in advance to obtain the M characteristic parameters of the M scan energy spectra.
3. The method according to claim 1, wherein, The sample dataset is constructed based on the M sample CT images and element density distributions of each of the N sample materials, including: For each of the N sample materials, perform the following operation: Based on the M sample CT images, determine the M CT values of each of the multiple sample objects in the sample material; Based on the element density distribution, obtain the M element densities corresponding to the M CT values; Based on the M CT values and the M element densities, determine the sample data for each of the multiple sample objects; and Based on the sample data, a sample dataset of the sample material is constructed.
4. The method according to claim 3, wherein, The process of determining the sample data for each of the multiple sample objects based on the M CT values and the elemental density includes: Within a preset range, multiple sets of enhancing element densities are generated based on the M element densities. Each set of enhancing element densities includes M enhancing element densities, and the difference between the M element densities and each set of enhancing element densities lies within the preset range; and Based on the M CT values and the multiple sets of enhancement element densities of each of the multiple sample objects, the sample data of each of the multiple sample objects is determined.
5. The method according to claim 1, wherein, Based on the M characteristic parameters and the elemental density distributions of N sample materials, the method determines M sample CT images of each of the N sample materials under the M scanning energy spectra, resulting in M*N sample CT images, including: Using the M characteristic parameters and the elemental density distributions of the N sample materials, determine the energy absorption capacity of each of the N sample materials for the M scanning energy spectra; and Based on the energy absorption capacity, M sets of CT values are determined for each of the N sample materials, and the M sets of CT values form M CT images.
6. The method according to claim 1, wherein, The step of training an initial model using the sample dataset to obtain the target model corresponding to the CT device includes: The sample CT images in the sample dataset are determined as the input data of the initial model, and the element density distribution in the sample dataset is determined as the output data of the initial model. The distribution characteristics of each element in the aforementioned element density distribution are used as prompting information; and Based on the prompt information, the input data, and the output data, the initial model is trained to obtain the target model.
7. A method for determining the elemental density distribution of a substance, comprising: The substance to be tested is scanned using M scanning energy spectra of the CT device to obtain M CT images, where M is a positive integer. as well as Using an elemental analysis model, the elemental density distribution of the substance to be tested is determined based on the M CT images; The elemental analysis model is a target model trained using the method described in any one of claims 1-6.
8. A model training device, comprising: The acquisition module is used to acquire the characteristic parameters of each of the M scanning energy spectra of the computed tomography (CT) device, where M is a positive integer; The first determining module is used to determine M sample CT images of each of the N sample materials under the M scanning energy spectra based on the M characteristic parameters and the elemental density distribution of the N sample materials, so as to obtain M*N sample CT images, where N is a positive integer. The construction module is used to construct a sample dataset based on the M sample CT images and element density distributions of the N sample materials. as well as The training module is used to train an initial model using the sample dataset to obtain a target model corresponding to the CT device. The target model is used to determine the element density distribution based on the CT images obtained by scanning the CT device.
9. An apparatus for determining the density distribution of elements in a substance, comprising: The scanning module is used to scan the substance to be tested using M scanning energy spectra of the CT equipment to obtain M CT images; as well as The second determining module is used to determine the elemental density distribution of the substance to be tested based on the M CT images using an elemental analysis model. The elemental analysis model is a target model trained using the apparatus as described in claim 8.
10. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.