An image reconstruction method, device, electronic equipment and storage medium
By using a variational autoencoder deep neural network to invert and decode measured data, the problem of large computational load and inflexible utilization of prior information caused by numerous unknowns in image reconstruction is solved, thus achieving efficient image reconstruction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING HUARUI BOSHI MEDICAL IMAGING TECH CO LTD
- Filing Date
- 2023-03-02
- Publication Date
- 2026-08-04
AI Technical Summary
In the process of image reconstruction, the nonlinear ill-conditioned problem that the number of pixels is much greater than the number of data leads to huge computational load and insufficient flexibility in utilizing prior information, making it difficult to effectively reconstruct the internal conductivity or dielectric constant distribution of an object.
A variational autoencoder deep neural network is used to minimize the objective function of the measured data. Through the cooperation of the encoder and decoder, the latent space parameters of the target are obtained and the image is decoded and reconstructed.
It reduces unknowns, improves the computational efficiency and accuracy of image reconstruction, and simplifies the inversion process.
Smart Images
Figure CN117425920B_ABST
Abstract
Description
[0001] Cross-references to related applications
[0002] This disclosure claims priority to Chinese patent application CN202210417793.1, filed on April 20, 2022, entitled “An image reconstruction method, apparatus, electronic device and storage medium”, the entire contents of which are incorporated herein by reference. Technical Field
[0003] This disclosure belongs to the field of image reconstruction technology, specifically relating to an image reconstruction method, apparatus, electronic device, and storage medium. Background Technology
[0004] Electromagnetic data imaging is widely used in fields such as biomedical imaging and non-destructive testing. It collects electromagnetic field data of the object under test through sensors and then reconstructs the distribution of conductivity or dielectric constant inside the object through certain image reconstruction methods.
[0005] In image reconstruction, a common approach is to decompose the inversion domain into pixels and then reconstruct discrete conductivity or dielectric constant by minimizing the residuals between simulated and measured data. Since the number of pixels is typically much larger than the amount of data, this is a nonlinear, ill-conditioned problem that requires prior knowledge for reasonable reconstruction. Summary of the Invention
[0006] This disclosure proposes an image reconstruction method, apparatus, electronic device, and storage medium. The method involves minimizing an objective function derived from measured data using a variational autoencoder deep neural network to obtain the target latent space parameters of the objective function derived from the measured data. The latent space parameters are then decoded using the variational autoencoder deep neural network to obtain the reconstructed image.
[0007] In a first aspect, this disclosure provides an image reconstruction method, comprising: acquiring measured data of a target; constructing a measured data inversion objective function based on the measured data, with the latent space parameters of a variational autoencoder deep neural network as unknowns; minimizing the measured data inversion objective function using the variational autoencoder deep neural network to obtain the target latent space parameters; and decoding the target latent space parameters using the variational autoencoder deep neural network to obtain a reconstructed image of the target.
[0008] In some embodiments, the variational autoencoder deep neural network includes a decoder and an encoder; the step of minimizing the objective function of the measured data using the variational autoencoder deep neural network to obtain the target latent space parameters includes: setting an initial model according to the target; encoding the initial model using the encoder to obtain the encoding of the initial model; decoding and calculating the encoding using the decoder to obtain simulation data; determining whether the difference between the simulation data and the measured data is greater than a first threshold; if the difference is greater than the first threshold, determining the update amount of the encoding; updating the encoding according to the update amount, and continuing to execute the step "decoding and calculating the encoding using the decoder to obtain simulation data"; if the difference is less than or equal to the first threshold, outputting the encoding as the target latent space parameters.
[0009] In some embodiments, the update amount is determined by the following formula: H(v) k )·p k =-g(v k );in, J is the Jacobian matrix, J H Let J be the conjugate transpose of J; I is the identity matrix.
[0010] In some embodiments, training the variational autoencoder deep neural network includes: acquiring training data and constructing a training set based on the training data; constructing a variational autoencoder deep neural network based on the training set; constructing a training function for the variational autoencoder deep neural network based on the training set; and training the variational autoencoder deep neural network using the training function.
[0011] In some embodiments, the training data includes image data. The step of acquiring the training data and constructing a training set based on the training data includes: segmenting the image data to obtain targets of interest in the image data; assigning training parameters to the targets of interest to form an initial training model; and adjusting the initial training model in different orientations to obtain multiple variant training models to obtain a training set.
[0012] In some embodiments, the training function includes: Where Q is the length of the latent space variables, and L is the number of pixels in the model. Let q be the q-th component of the variance vector output by the encoder. Let m be the q-th component of the mean vector output by the encoder. l For the l-th component input to the encoder, Let α be the l-th component output by the decoder, and let α be the regularization coefficient for adjusting the KL divergence of the variational autoencoder.
[0013] In some embodiments, the measured data includes one of the following: timing differential impedance data, absolute impedance data, and microwave data.
[0014] Secondly, this disclosure provides an image reconstruction apparatus, comprising: a first acquisition module configured to acquire measured data of a target; a first construction module configured to construct a measured data inversion objective function with the latent space parameters of a variational autoencoder deep neural network as unknowns based on the measured data; a first execution module configured to minimize the measured data inversion objective function using the variational autoencoder deep neural network to obtain the target latent space parameters; and a second execution module configured to decode the target latent space parameters using the variational autoencoder deep neural network to obtain a reconstructed image.
[0015] Thirdly, this disclosure provides an electronic device including a storage device and a processor, wherein the storage device stores a computer program, and the processor executes the computer program to implement the image reconstruction method described in the first aspect.
[0016] Fourthly, this disclosure provides a storage medium storing a computer program that can be executed by one or more processors to implement the image reconstruction method described in the first aspect.
[0017] This disclosure utilizes a variational autoencoder deep neural network to minimize the objective function derived from measured data, thereby obtaining the target latent space parameters of the objective function derived from the measured data. The variational autoencoder deep neural network is then used to decode these latent space parameters to obtain the reconstructed image. This significantly reduces the number of unknowns in the image reconstruction process and improves computational efficiency. Attached Figure Description
[0018] The scope of this disclosure can be better understood by reading the following detailed description of exemplary embodiments in conjunction with the accompanying drawings. The accompanying drawings are:
[0019] Figure 1 A flowchart illustrating an image reconstruction method provided in this disclosure embodiment; and
[0020] Figure 2 This is a structural block diagram of an image reconstruction apparatus provided in an embodiment of the present disclosure. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of this disclosure clearer, the disclosure will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this disclosure. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.
[0022] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
[0023] If similar descriptions of "first, second, third" appear in this disclosure, the following explanation shall be added: In the following description, the terms "first, second, third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this disclosure described herein can be implemented in an order other than that illustrated or described herein.
[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing embodiments of this disclosure only and is not intended to be limiting of this disclosure.
[0025] In image reconstruction, common methods typically require a much larger number of pixels than the amount of data, making it a non-linear, ill-conditioned problem that relies on prior knowledge for reasonable reconstruction. Since the number of unknowns is often in the thousands or tens of thousands, the computational cost of minimizing the objective function is enormous. Furthermore, the way prior information is utilized during inversion is not flexible enough. The prior knowledge held by interpreters is difficult to describe mathematically, thus failing to constrain the inversion process and posing challenges to model reconstruction and interpretation.
[0026] Example 1
[0027] like Figure 1 As shown, this disclosure provides an image reconstruction method, which is applied to electronic devices such as servers, mobile terminals, computers, and cloud platforms. The functions implemented by the device data processing provided in this disclosure embodiment can be achieved by the processor of the electronic device calling program code, wherein the program code can be stored in a computer storage medium. The image reconstruction method includes steps S1 to S4.
[0028] Step S1: Obtain the measured data of the target.
[0029] The target measured data in this disclosure can be any one of time-series differential impedance data, absolute impedance data, and microwave data.
[0030] Step S2: Construct an objective function based on measured data, with the latent space parameters of the variational autoencoder deep neural network as unknowns.
[0031] A variational autoencoder deep neural network consists of an encoder and a decoder. The encoder outputs latent space parameters, while the decoder is responsible for decoding the latent space parameters output by the encoder.
[0032] Step S3: Minimize the objective function retrieved from the measured data using a variational autoencoder deep neural network to obtain the target latent space parameters.
[0033] In some embodiments, in step S3, minimizing the objective function of the measured data inversion using a variational autoencoder deep neural network to obtain the target latent space parameters includes the following steps S31 to S37.
[0034] Step S31: Set the initial model according to the target.
[0035] Setting up the initial model requires different settings depending on the type of measured data. Different parameters are preset for the initial model. When the measured data is time-series differential impedance data or absolute impedance data, the conductivity is preset for the initial model. When the measured parameters are microwave data, the dielectric constant is preset for the initial model.
[0036] Step S32: Encode the initial model using an encoder to obtain the encoding of the initial model.
[0037] The encoder of the variational autoencoder deep neural network is used to encode the initial model with preset parameters to obtain the encoding of the initial model.
[0038] Step S33: Use a decoder to decode the code and perform calculations to obtain simulation data.
[0039] After encoding is completed, the encoding result is decoded using the decoder of the variational autoencoder deep neural network. After decoding, a preset parameter model is obtained, and then the preset parameter model is calculated to obtain simulation data.
[0040] (1) When the measured data is time-series differential impedance data, the time-series differential impedance data is simulated using the following formula:
[0041] d = A·m,
[0042] Where m is the discrete change in conductivity, d is the time-series differential impedance data calculated by simulation based on the conductivity equation, and A is the forward modeling function, which is represented as a matrix here.
[0043] The discrete conductivity change m can be decoded from the latent space variables by the decoder D of the variational autoencoder:
[0044] m = D(v);
[0045] Therefore, the process of calculating electrical impedance data from implicit space variables can be represented by S:
[0046] d = A·D(v) = S(v).
[0047] (2) When the measured data is absolute impedance data, the absolute impedance data is numerically simulated using the following formula:
[0048] d = F(T(m))
[0049] Where m is the discrete organ conductivity, d is the absolute electrical impedance data calculated by simulation based on Maxwell's equations, T is the mapping function used to map the organ conductivity onto a triangular finite element mesh to generate a thoracic cavity conductivity model, and F is the forward modeling function.
[0050] The organ conductivity m can be obtained by decoding the latent space variable v using the decoder D of the variational autoencoder:
[0051] m = D(v)
[0052] Therefore, the process of calculating voltage simulation data from the implicit space variable v can be represented by S:
[0053] d = F(T(D(v))) = S(v).
[0054] (3) In microwave data inversion, electromagnetic scattering data is numerically simulated using the following formula:
[0055] d = F(m);
[0056] Where m is the discrete dielectric constant, d is the scattered field data calculated based on Maxwell's equations, and F is the forward modeling function.
[0057] The discrete dielectric constant m can be decoded from the latent space variables by the decoder D of the variational autoencoder:
[0058] m = D(v);
[0059] Therefore, the process of calculating electromagnetic data from implicit space variables can be represented by S:
[0060] d = F(D(v)) = S(v).
[0061] Step S34: Determine whether the difference between the simulation data and the measured data is greater than the first threshold.
[0062] The final result is determined by comparing the difference between the simulation data and the measured data to determine whether it converges to the first threshold.
[0063] In deterministic inversion, the inversion problem can be equated to finding the optimal parameter v that minimizes the following objective function:
[0064]
[0065] Where |||| represents the L2 norm, d obs For measurement data, λ is the regularization coefficient of the objective function, used to stabilize the optimization process.
[0066] Step S35: When the difference is greater than the first threshold, determine the amount of code update.
[0067] When the result fails to converge to the first threshold, the encoding needs to be updated. When updating the encoding, the amount of encoding to be updated needs to be determined.
[0068] When determining the amount of code updates, the update amount needs to be determined using the following formula:
[0069] H(v k )·p k =-g(v k ).
[0070] in,
[0071]
[0072]
[0073] J is the Jacobian matrix, J H Let J be the conjugate transpose of J. Let I be the identity matrix.
[0074] Step S36: Update the encoding according to the update amount, and use the decoder to decode and calculate the encoding using the continue execution step to obtain simulation data.
[0075] After determining the update amount, the code is updated according to the update amount, and step S33 is then performed on the updated code.
[0076] After determining the update amount, the encoding is updated according to the following formula:
[0077] v k+1 =v k +p k .
[0078] Step S37: When the difference is less than or equal to the first threshold, the encoding is output as the target latent space parameter.
[0079] When the difference between the simulation data and the measured data determines that the final result converges to the first threshold, the encoding corresponding to the simulation data is output as the target latent space parameter.
[0080] Step S4: Use a variational autoencoder deep neural network to decode the latent space parameters of the target to obtain the reconstructed image of the target.
[0081] After obtaining the target latent space parameters, the latent space parameters are decoded using the decoder of the variational autoencoder deep neural network to obtain the preset parameter model. Finally, the target reconstructed image is obtained based on the preset parameter model.
[0082] After obtaining v k+1 Then, by decoding it using decoder D in the variational autoencoder, the dielectric constant model / conductivity model can be obtained as follows:
[0083] m k+1 =D(v) k+1 ).
[0084] This disclosure utilizes a variational autoencoder deep neural network to minimize the objective function derived from measured data, thereby obtaining the target latent space parameters of the objective function derived from the measured data. The variational autoencoder deep neural network is then used to decode these latent space parameters to obtain the reconstructed image. This significantly reduces the number of unknowns in the image reconstruction process and improves computational efficiency.
[0085] In some embodiments, the training required when training a variational autoencoder deep neural network includes steps S51 to S54.
[0086] Step S51: Obtain training data and construct a training set based on the training data.
[0087] The training data comes from image data obtained by other imaging methods, such as CT scan data, or 3D image data obtained from 3D scan data. The training set is constructed based on the obtained image data.
[0088] In some embodiments, in step S51, acquiring training data and constructing a training set based on the training data includes steps S511 to S513.
[0089] Step S511: Segment the image data to obtain the target of interest in the image data.
[0090] Step S512: Assign training parameters to the target of interest to form an initial training model.
[0091] For example, when performing electrical impedance image reconstruction, the assigned training parameter is the conductivity of the target, which forms the initial training model.
[0092] Step S513: Adjust the different orientations of the initial training model to obtain multiple deformed training models, thus obtaining a training set.
[0093] To increase the amount of data used to train the model, the shape of the target can be modified.
[0094] For example, when training a variational autoencoder deep neural network using lung images, lung images obtained through other detection methods are first acquired. Since different detection methods yield different properties—for instance, lung images obtained through CT scans do not possess conductivity or microwave data—the conductivity of the acquired images needs to be preset. This preset conductivity forms an initial training model of lung electrical impedance. Then, data augmentation is performed on this initial training model by randomly removing lung lobes from the images. Because the lobes are removed, the electrical impedance in the entire initial training model changes, resulting in multiple variant training models. These variant training models ultimately form the training set.
[0095] Step S52: Construct a variational autoencoder deep neural network based on the training set.
[0096] In this embodiment (temporal differential impedance imaging example), the designed variational autoencoder deep neural network includes an encoder and a decoder. The encoder's input and output sizes are 32×48×48. In the encoder, convolution and activation are applied three times alternately to output a tensor with a size of 4×6×6×32. After rearranging the tensor into a vector, two dense layers are applied to generate the mean and variance vectors. A Gaussian distribution is sampled using a reparameterization method.
[0097] In this example (temporal differential impedance imaging), the length of the latent space variable is 32, achieving a compression ratio of 32 / 73728 = 0.043%. Except for upsampling via transposed convolutional layers, the decoder structure is symmetrical to the encoder structure. The network uses the Rectified Linear Unit (ReLU) function for nonlinear activation.
[0098] Step S53: Construct the training function for the variational autoencoder deep neural network based on the training set.
[0099] In some embodiments, the training function includes:
[0100]
[0101] Where Q is the length of the latent space variables, and L is the number of pixels in the model. Let q be the q-th component of the variance vector output by the encoder. Let m be the q-th component of the mean vector output by the encoder. lFor the l-th component input to the encoder, Let α be the l-th component output by the decoder, and let α be the regularization coefficient for adjusting the KL divergence of the variational autoencoder.
[0102] Step S54: Train the variational autoencoder deep neural network using the training function.
[0103] Finally, the designed variational autoencoder deep neural network is trained using the Adam optimization algorithm and training function. After training, the optimal parameters of the variational autoencoder deep neural network can be obtained.
[0104] This disclosure utilizes a variational autoencoder deep neural network to minimize the objective function derived from measured data, thereby obtaining the target latent space parameters of the objective function derived from the measured data. The variational autoencoder deep neural network is then used to decode these latent space parameters to obtain the reconstructed image. This significantly reduces the number of unknowns in the image reconstruction process and improves computational efficiency.
[0105] Example 2
[0106] Based on the foregoing embodiments, this disclosure provides an image reconstruction apparatus. The various modules and units included in the apparatus can be implemented by a processor in a computer device; of course, they can also be implemented by logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.
[0107] As shown in Figure 5, the second aspect provides an image reconstruction device, including: a first acquisition module 1, a first construction module 2, a first execution module 3, and a second execution module 4.
[0108] The first acquisition module 1 is configured to acquire the measured data of the target. The first construction module 2 is configured to construct a measured data inversion objective function based on the measured data, with the latent space parameters of the variational autoencoder deep neural network as unknowns. The first execution module 3 is configured to minimize the measured data inversion objective function using the variational autoencoder deep neural network to obtain the target latent space parameters. The second execution module 4 is configured to decode the target latent space parameters using the variational autoencoder deep neural network to obtain the reconstructed image.
[0109] In some embodiments, the first execution module 3 includes: a first setting module, a third execution module, a fourth execution module, a first determining module, a second determining module, and a first output module.
[0110] The first setting module is configured to set an initial model based on the target. The third execution module is configured to encode the initial model using an encoder to obtain the initial model's encoding. The fourth execution module is configured to decode the encoding using a decoder and perform calculations to obtain simulation data. The first determination module is configured to determine whether the difference between the simulation data and the measured data is greater than a first threshold. The second determination module is configured to determine the amount of encoding update when the difference is greater than the first threshold. The first output module is configured to output the encoding as the target latent space parameter when the difference is less than or equal to the first threshold.
[0111] In some embodiments, the image reconstruction apparatus further includes a second acquisition module, a fifth execution module, a sixth execution module, and a seventh execution module.
[0112] The second acquisition module is configured to acquire training data and construct a training set based on the training data. The fifth execution module is configured to construct a variational autoencoder deep neural network based on the training set. The sixth execution module is configured to construct a training function for the variational autoencoder deep neural network based on the training set. The seventh execution module is configured to train the variational autoencoder deep neural network using the training function.
[0113] In some embodiments, the fifth execution module includes an eighth execution module, a ninth execution module, and a tenth execution module.
[0114] The eighth execution module is configured to segment the image data to obtain the target of interest (ROI) within the image data. The ninth execution module is configured to assign training parameters to the ROI to form an initial training model. The tenth execution module adjusts the initial training model in different orientations to obtain multiple variant training models, thus forming a training set.
[0115] The modules in the aforementioned image reconstruction apparatus can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor within the device in hardware form, or stored in the memory of the processing device in software form, so that the processor can call and execute the operations corresponding to each module. It should be noted that the module division in this embodiment is illustrative and represents only a logical functional division; in actual implementation, other division methods may be used.
[0116] Example 3
[0117] The third aspect provides an electronic device including a storage device and a processor, the storage device storing a computer program, the processor executing the computer program to implement the steps of an image reconstruction method.
[0118] Example 4
[0119] The fourth aspect provides a storage medium storing a computer program that can be executed by one or more processors, the computer program being able to implement the steps of any of the image reconstruction methods in the first aspect.
[0120] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.
[0121] It should be understood that the phrases "one embodiment" or "some embodiments" throughout the specification mean that a specific feature, structure, or characteristic related to an embodiment is included in at least one embodiment of this disclosure. Therefore, "in one embodiment" or "in one embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this disclosure, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this disclosure. The sequence numbers of the above-described embodiments are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0122] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0123] In the several embodiments provided in this disclosure, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.
[0124] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.
[0125] In addition, each functional unit in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.
[0126] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.
[0127] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this disclosure, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a controller to execute all or part of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.
[0128] The above description is merely an embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.
Claims
1. An image reconstruction method, comprising: Obtain the measured data of the target; Based on the measured data, construct the measured data inversion objective function with the latent space parameters of the variational autoencoder deep neural network as unknowns; The target latent space parameters are obtained by minimizing the objective function retrieved from the measured data using the variational autoencoder deep neural network. as well as The latent space parameters of the target are decoded using the variational autoencoder deep neural network to obtain the target reconstructed image; The variational autoencoder deep neural network includes a decoder and an encoder; the step of minimizing the objective function retrieved from the measured data using the variational autoencoder deep neural network to obtain the target latent space parameters includes: Set an initial model based on the objectives; The initial model is encoded using the encoder to obtain the encoding of the initial model; The decoder is used to decode and calculate the encoded data to obtain simulation data; Determine whether the difference between the simulation data and the measured data is greater than a first threshold. If the difference is greater than a first threshold, the amount of code update is determined; The encoding is updated according to the update amount, and the execution continues with the step "decoding the encoding using the decoder and calculating to obtain simulation data"; and If the difference is less than or equal to the first threshold, the encoding is output as the target latent space parameter; The update amount is determined by the following formula: ； in, ; ； ； It is a Jacobian matrix. for The conjugate transpose of; It is an identity matrix.
2. The image reconstruction method according to claim 1, wherein, The training of the variational autoencoder deep neural network includes: Acquire training data and construct a training set based on the training data; Construct a variational autoencoder deep neural network based on the training set; The training function for constructing the variational autoencoder deep neural network is based on the training set; and The variational autoencoder deep neural network is trained using the training function.
3. The image reconstruction method according to claim 2, wherein, The training data includes: image data. The process of acquiring the training data and constructing a training set based on the training data includes: The image data is segmented to obtain the target of interest in the image data; Assign training parameters to the target of interest to form an initial training model; and The initial training model is adjusted in different orientations to obtain multiple deformed training models, thus obtaining a training set.
4. The image reconstruction method according to claim 2, wherein, The training function includes: ； Where Q is the length of the latent space variables, and L is the number of pixels in the model. The first variance vector of the encoder output One portion, The first value of the mean vector output by the encoder. One portion, The first input to the encoder One portion, The first output of the decoder One portion, To adjust the regularization coefficient of the KL divergence of the variational autoencoder.
5. The image reconstruction method according to claim 1, wherein, The measured data includes one of the following: timing differential impedance data, absolute impedance data, and microwave data.
6. An image reconstruction apparatus for implementing the image reconstruction method according to any one of claims 1-5, comprising: The first acquisition module is configured to acquire the measured data of the target; The first construction module is configured to construct a measured data inversion objective function based on the measured data, with the latent space parameters of the variational autoencoder deep neural network as unknowns; The first execution module is configured to minimize the objective function retrieved from the measured data using the variational autoencoder deep neural network to obtain the target latent space parameters. as well as The second execution module is configured to decode the latent space parameters of the target using the variational autoencoder deep neural network to obtain a reconstructed image.
7. An electronic device, comprising: A memory and a processor, wherein the memory stores a computer program that, when executed by the processor, performs the image reconstruction method as described in any one of claims 1 to 5.
8. A storage medium storing a computer program that can be executed by one or more processors to implement the image reconstruction method as described in any one of claims 1 to 5.