Reconstruction method and device for diffused light tomography, medium and near-infrared brain function imaging system

Abstract

The invention discloses a reconstruction method and device for diffusion light tomography, a medium and a near-infrared brain function imaging system.The reconstruction method for diffusion light tomography comprises the steps that a processor is used for obtaining a reconstruction result based on measurement data of all measurement positions of a high-density measurement array arranged on the surface of a to-be-measured body; forming a one-dimensional measurement data sequence with a predetermined measurement bit sequence; performing rough reconstruction based on a one-dimensional measurement data sequence of a predetermined measurement bit order to obtain 3D input data, and aligning the 3D input data with the spatial resolution of the ground truth value reconstruction image; and taking the 3D input data as input and a constraint condition, and utilizing a conditional variational auto-encoder to reconstruct the spatial distribution of the optical parameters in the to-be-measured body.

Description

Technical Field

[0001] The present application belongs to the field of medical artificial intelligence, and in particular, relates to a reconstruction method, device, medium and near-infrared brain function imaging system for diffuse light tomography. Background Art

[0002] Early diagnosis and treatment evaluation of brain diseases are particularly important. Brain functional imaging technologies, including functional magnetic resonance imaging (fMRI), electrophysiological imaging (EEG), magnetoencephalography (MEG), functional near-infrared spectroscopy (fNIRS), etc., can quantitatively image the spatiotemporal state of brain activity, providing an important reference for the early diagnosis and treatment of brain diseases. Among them, fNIRS, as an emerging brain functional imaging technology, has the advantages of low cost, high portability, non-invasiveness, and no ionizing radiation. It is suitable for large-scale clinical screening and daily status monitoring and evaluation of patients with brain diseases.

[0003] However, the development of near-infrared brain functional imaging technology is still restricted by some factors. On the one hand, the layered structure of the brain makes the detection of the state of the cerebral cortex susceptible to physiological interference from shallow tissues; on the other hand, the strong scattering of tissues inside the brain makes it easy for photons to migrate randomly during transmission inside the tissues, resulting in serious ill-posedness of brain functional imaging problems. Even tiny noises will cause large fluctuations in the imaging results, affecting the credibility of the imaging results. In order to solve the above problems, diffuse optical tomography (DOT) eliminates the physiological interference of shallow tissues by designing alternating detection channels and has depth-resolved information, which can alleviate the ill-posedness of the inverse problem to a certain extent, but the accuracy of its imaging still needs to be improved. Summary of the invention

[0004] In response to the above-mentioned technical problems existing in the prior art, the present application provides a reconstruction method, device, medium and near-infrared brain function imaging system for diffuse optical tomography, which can establish a mapping relationship from the sensor domain to the reconstructed image domain through a conditional autoencoder, and realize a gradually refined reconstruction method by forming a one-dimensional measurement data sequence with a predetermined measurement bit order based on the measurement data, roughly reconstructing 3D input data, and adding constraints in the model reasoning stage, thereby improving the accuracy of the reconstructed spatial distribution of optical parameters in the body to be measured.

[0005] According to the first scheme of the present application, a reconstruction method for diffuse optical tomography is provided, the reconstruction method comprising: using a processor: based on the measurement data of each measurement position of a high-density measurement array arranged on the surface of the object to be measured, forming a one-dimensional measurement data sequence with a predetermined measurement position order; based on the one-dimensional measurement data sequence with the predetermined measurement position order, performing a rough reconstruction to obtain 3D input data so that it is aligned with the spatial resolution of the ground truth reconstructed image; using the 3D input data as input and as a constraint condition, using a conditional variational autoencoder to reconstruct the spatial distribution of optical parameters in the object to be measured.

[0006] According to a second scheme of the present application, a reconstruction device for diffuse optical tomography is provided, the reconstruction device comprising at least a processor and a memory, the memory storing computer executable instructions, and the processor executing the above-mentioned reconstruction method for diffuse optical tomography when executing the computer executable instructions.

[0007] According to the third scheme of the present application, a near-infrared brain function imaging system is provided, comprising a near-infrared optical data acquisition device and a reconstruction device for diffuse light tomography; wherein the near-infrared optical data acquisition device is used to acquire near-infrared data of an object; and the reconstruction device is used to perform the above-mentioned reconstruction method for diffuse light tomography using the near-infrared data.

[0008] According to a fourth aspect of the present application, there is provided a non-transitory computer-readable storage medium storing a program, wherein the program causes a processor to execute the operations of the above-mentioned reconstruction method for diffuse optical tomography.

[0009] Compared with the prior art, the beneficial effects of the embodiments of the present application are: The reconstruction method for diffuse light tomography provided by each embodiment of the present application integrates the measurement data of each measurement position of the high-density measurement array arranged on the surface of the object to be measured to form a one-dimensional measurement data sequence with a predetermined measurement position order, so that the conditional variational autoencoder used in the subsequent model reasoning stage can extract the spatial distribution semantics that characterize the relative positional relationship between the measurement position and the object to be measured, so that in the model reasoning stage, more abundant features can be extracted in combination with the arrangement of each transmitting position and receiving position in the high-density measurement array, and the extracted spatial distribution semantics is also strongly related to the reconstruction process, which can alleviate the ill-posed problem of the solution caused by the lack of feature semantics in the inverse problem of diffuse light tomography reconstruction, and improve the accuracy of the reconstructed spatial distribution of optical parameters in the object to be measured. Furthermore, compared with directly reconstructing the one-dimensional measurement data sequence, by first roughly reconstructing the one-dimensional measurement data sequence into 3D input data that matches the spatial resolution of the ground truth reconstructed image, the feature semantics input to the conditional variational autoencoder can be further enriched, the ill-posed problem of the solution during reconstruction by the conditional variational autoencoder can be alleviated, and the accuracy of the reconstructed spatial distribution of optical parameters in the body to be measured can be improved. In the model inference stage, a mapping relationship from the sensor domain to the reconstructed image domain is established by using the conditional autoencoder, and the accuracy of the reconstruction results obtained by the conditional autoencoder is improved by adding constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0011] Figure 1 A flowchart showing a reconstruction method for diffuse optical tomography provided by an embodiment of the present application; Figure 2 A schematic diagram showing a reconstruction method for diffuse optical tomography provided by an embodiment of the present application, wherein a diagram showing measurement of the surface of an object to be measured using a high-density measurement array is shown; Figure 3 A schematic diagram showing a high-density measurement array in a reconstruction method for diffuse optical tomography provided in an embodiment of the present application; Figure 4 A schematic diagram showing a training process in a reconstruction method for diffuse optical tomography provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] In order to enable those skilled in the art to better understand the technical solution of the present application, the present application is described in detail below in conjunction with the accompanying drawings and specific implementation examples. The embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and specific implementation examples, but are not intended to limit the present application.

[0013] The words "first", "second" and similar words used in this application do not indicate any order, quantity or importance, but are only used to distinguish. The words "include" or "comprises" and similar words used in this application mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of covering other elements. In this application, the arrows shown in the figures of each step are only examples of the execution order, not limitations. The technical solution of this application is not limited to the execution order described in the embodiments. The various steps in the execution order can be combined, decomposed, or the order can be changed, as long as the logical relationship of the execution content is not affected.

[0014] All terms (including technical terms or scientific terms) used in this application have the same meaning as those understood by ordinary technicians in the field to which this application belongs, unless otherwise specifically defined. It should also be understood that terms defined in general dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined here. Techniques and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, the techniques and equipment should be considered as part of the specification.

[0015] In the present application, the arrows shown in the figures of each step are merely examples of the execution order, not limitations. The technical solution of the present application is not limited to the execution order described in the embodiments. The various steps in the execution order can be combined for execution, can be decomposed for execution, and can be reordered, as long as the logical relationship of the execution content is not affected.

[0016] See also Figure 1 FIG. 1 is a flowchart of a reconstruction method for diffuse optical tomography provided by an embodiment of the present disclosure. The reconstruction method includes executing S101 to S103 by a processor, wherein: S101: Based on the measurement data of each measurement position of the high-density measurement array arranged on the surface of the object to be measured, a one-dimensional measurement data sequence with a predetermined measurement position order is formed.

[0017] S102: Based on the one-dimensional measurement data sequence in a predetermined measurement position order, perform a rough reconstruction to obtain 3D input data so that the 3D input data is aligned with the spatial resolution of the ground truth reconstructed image.

[0018] S103: Using the 3D input data as input and as constraints, a conditional variational autoencoder is used to reconstruct the spatial distribution of optical parameters in the body to be tested.

[0019] The following is a detailed description of the above steps.

[0020] With respect to S101, the object to be measured may be a living thing, a specific organ of a living thing, or a non-living thing, for example, the head of a subject, the abdomen of a subject, etc. The high-density measurement array may be a high-density arrangement measurement array, and the high-density arrangement may be understood as a measurement array that satisfies a predetermined high-density arrangement condition, and the high-density arrangement condition may be, for example, that the interval between each measurement position in the measurement array is less than a predetermined interval, and the number of measurement positions is greater than a predetermined number, etc.

[0021] In some embodiments, the measurement data of the measurement position can be collected by a measurement device deployed at the measurement position. The measurement device has a probe capable of emitting or receiving near-infrared light. The measurement data can be collected by deploying the probe of the measurement device at each measurement position of a high-density measurement array.

[0022] It should be noted that the "one dimension" in the one-dimensional measurement data sequence represents a data dimension, which can be understood as the properties of the near-infrared light collected at the corresponding measurement position, such as the light intensity change data of the near-infrared light, that is, the "one dimension" in the one-dimensional measurement data sequence does not represent a spatial dimension. For example, the one-dimensional measurement data sequence can be represented as the light intensity change data corresponding to each measurement position arranged in sequence.

[0023] For example, a schematic diagram of using a high-density measurement array to measure the surface of an object to be measured can be as follows: Figure 2 As shown, Figure 2 In the embodiment, the object to be measured is the head of the subject, and a high-density measurement array is deployed on the head of the subject, so as to achieve high-density acquisition of measurement data. After the measurement data are acquired, a one-dimensional measurement data sequence with a predetermined measurement bit order can be formed based on the acquired measurement data, so as to reconstruct the spatial distribution of optical parameters in the object to be measured through subsequent model reasoning steps.

[0024] In some embodiments, the high-density measurement array may include transmission bits and receiving bits arranged in sequence; the one-dimensional measurement data sequence of the predetermined measurement bit order includes measurement data corresponding to multiple transmission bits spliced ​​in sequence, each transmission bit has at least one target receiving bit and forms a measurement channel therewith, and the measurement data corresponding to each transmission bit includes the measurement data collected by each target receiving bit.

[0025] Among them, the transmitting position can be used to deploy a transmitting probe of the measuring device, and the receiving position can be used to deploy a receiving probe of the measuring device. The transmitting probe can transmit near-infrared light to the object to be measured, and the receiving probe can receive the near-infrared light emitted from the object to be measured.

[0026] Exemplarily, the schematic diagram of the high-density measurement array can be as follows: Figure 3 As shown, Figure 3 In the figure, the 13 red dots represent the 13 transmitting positions in the high-density measurement array, and the 12 green dots represent the 12 receiving positions in the high-density measurement array. Taking the transmitting position A in the 1st row and the 1st column and the receiving position B in the 1st row and the 2nd column as examples, the measurement data corresponding to the transmitting position A includes the data collected by its receiving position B for the transmitting position A, that is, the near-infrared light emitted by the transmitting probe of the measuring device at the transmitting position A, after being emitted from the object to be measured and collected by the receiving probe deployed at the receiving position B, the collected measurement data can be used as the measurement data of the measurement channel AB corresponding to the transmitting position A (or receiving position B).

[0027] In some other embodiments, the one-dimensional measurement data sequence of the predetermined measurement bit order includes measurement data corresponding to multiple receiving bits spliced ​​in sequence, that is, the one-dimensional measurement data sequence can be obtained by splicing measurement data corresponding to multiple receiving bits in sequence.

[0028] In some embodiments, the target receiving bit of each transmitting bit includes several receiving bits whose distances from it meet preset distance conditions; the measurement data corresponding to each transmitting bit is obtained by sequentially splicing the measurement data collected by each target receiving bit according to the predetermined serial number of each target receiving bit in the high-density measurement array.

[0029] Among them, for any transmission bit, the receiving bits in the high-density measurement array can be divided into several categories according to the distance from the transmission bit, and the target receiving bit can be some of the receiving bits, such as the closest category 2 or category 3 receiving bits.

[0030] For example, taking the target receiving bit as the three types of receiving bits closest to the transmitting bit as an example, Figure 3 The transmitting bit U in the example can be divided into six categories according to the distance between each receiving bit and the transmitting bit U, which are receiving bits P and V, receiving bits L and R, receiving bits F and X, receiving bits H and N, receiving bits B and T, and receiving bits D and J, so that the three types of receiving bits with the closest distance can be selected as the target receiving bits, that is, the receiving bits P and V, receiving bits L and R, and receiving bits F and X can be selected as the target receiving bits corresponding to the transmitting bit U.

[0031] In some embodiments, the measurement data in the one-dimensional measurement data sequence of the predetermined measurement bit order are arranged following the predetermined measurement bit order.

[0032] For example, the transmitting probes deployed at the transmitting positions can be lit up in sequence, and the receiving probes deployed at the target receiving positions corresponding to the transmitting probes can collect measurement data at the corresponding time. According to the predetermined sequence number of each target receiving position in the high-density measurement array, the measurement data collected by the receiving probes deployed at each target receiving position can be spliced ​​in sequence to obtain a one-dimensional measurement data sequence.

[0033] Still Figure 3 Taking the high-density measurement array in as an example, the predetermined serial numbers corresponding to the measurement positions can be Figure 3 As shown in A to Y, when the measurement data are spliced ​​in sequence, the measurement data of each target receiving position can be spliced ​​in the order of the serial number, and the measurement data sequence corresponding to the transmission position U is FLPRVX (that is, the measurement data corresponding to each target receiving position). The one-dimensional measurement data sequence can be obtained by splicing the measurement data sequences of each transmission position, for example, according to the measurement position order of ACEGIKMOQSUWY, the measurement data corresponding to each transmission position are spliced ​​in sequence.

[0034] In some other embodiments, in a one-dimensional measurement data sequence with a predetermined measurement position order, the measurement data may also be arranged in association with its position indication information, where the position indication information indicates the measurement position order.

[0035] Among them, the position indication information can be, for example, the coordinates, serial number, etc. of the corresponding measurement position. By directly adding the position indication information to the one-dimensional measurement data sequence, the relative position relationship between each measurement position and the object to be measured can be explicitly expressed, so that the corresponding feature semantics can be extracted more directly in the subsequent model reasoning stage.

[0036] In this way, the measurement data with spatial distribution semantics are spliced ​​in the above manner to obtain a one-dimensional measurement data sequence with a predetermined measurement position order, or position indication information is directly added to the one-dimensional measurement data sequence, so that the conditional variational autoencoder used in the subsequent model reasoning stage can extract the spatial distribution semantics that characterize the relevant positional relationship between the measurement position and the object to be measured, so that in the model reasoning stage, the arrangement of each transmitting position and the receiving position in the high-density measurement array can be combined to extract richer features, and the extracted spatial distribution semantics is also strongly related to the reconstruction process, which can alleviate the ill-posed problem of the solution caused by the lack of feature semantics in the inverse problem of diffuse optical tomography reconstruction, and improve the accuracy of the reconstructed spatial distribution of optical parameters in the object to be measured.

[0037] In some embodiments, when verifying the splicing effect of a one-dimensional measurement data sequence, the following test groups may be set: Group 1: According to the randomly generated measurement position order, the measurement data of each measurement position are spliced ​​to generate a one-dimensional measurement data sequence.

[0038] Group 2: According to the predetermined measurement position order described in the above embodiment, the measurement data of each measurement position are spliced ​​to generate a one-dimensional measurement data sequence.

[0039] Group 3: According to the predetermined measurement position sequence described in the above embodiment, the measurement data of the target receiving positions that meet the preset distance condition are selected for splicing to generate a one-dimensional measurement data sequence.

[0040] When selecting indicators for evaluating the image reconstruction effect, you can select indicators such as SSIM (Structural Similarity Index), PSNR (Peak Signal-to-Noise Ratio), CNR (Contrast-to-Noise Ratio), R value (Pearson Correlation Coefficient, also known as Pearson correlation coefficient), Jaccard index (Jaccard index), MAE (Mean Absolute Error), etc. The definitions, calculation methods and meanings of the above indicators are known to those skilled in the art and are not repeated here.

[0041] In some embodiments, a phantom with multiple targets scattered in a medium with a depth of more than 3 cm or even more than 5 cm is used as a reconstruction object, and the optical parameter data at various locations in the space is used as the ground truth, and the same high-density measurement array is used to verify the reconstruction effect of the image obtained after each group of one-dimensional measurement data sequence is input into the same model. On the one hand, it can be verified that the predetermined measurement position sequence that can contain spatial distribution semantics selected in group 3 can provide a better reconstruction effect than the measurement position sequence containing spatial distribution semantics randomly generated in group 1; on the other hand, it can also be verified that the measurement data of the target receiving positions that meet the preset distance conditions are selected in group 3 for splicing, which can provide a better reconstruction effect by increasing the feature semantic density and reducing noise input compared to the measurement data of all measurement positions selected for splicing in group 2.

[0042] For S102, the rough reconstruction can be understood as the process of converting a one-dimensional measurement data sequence into 3D input data with the same spatial resolution as the ground truth reconstructed image, which can be achieved through a pre-trained U-Net, GAN or other model with upsampling function, or can also be achieved through an interpolation algorithm, a deconvolution algorithm or other algorithm with upsampling function. The ground truth reconstructed image represents the real spatial distribution of optical parameters in the body to be measured, that is, the desired reconstruction target of diffuse optical tomography. The spatial resolution of the ground truth reconstructed image can be, for example, 2mm, 3mm, 5mm, etc. The size of the 3D input data obtained after the rough reconstruction also matches the size of the ground truth reconstructed image, so as to facilitate the subsequent use of the conditional variational autoencoder for corresponding reconstruction processing.

[0043] In some embodiments, when a one-dimensional measurement data sequence based on a predetermined measurement bit order is roughly reconstructed, the one-dimensional measurement data sequence can be roughly reconstructed using a pre-trained model with an upsampling module to obtain 3D input data.

[0044] In this way, compared with directly reconstructing the one-dimensional measurement data sequence, by first roughly reconstructing the one-dimensional measurement data sequence into 3D input data that matches the spatial resolution of the ground truth reconstructed image, it is possible to enrich the feature semantics input to the conditional variational autoencoder, alleviate the problem of the ill-posedness of the solution during reconstruction of the conditional variational autoencoder, and improve the accuracy of the reconstructed spatial distribution of the optical parameters in the body to be measured.

[0045] For S103, the spatial distribution of optical parameters in the object to be tested can be understood as optical parameters at various locations in the object to be tested space, such as near-infrared light related parameters at various locations in the object to be tested space. The model architecture of the conditional variational autoencoder includes an encoder and a decoder.

[0046] Among them, the architecture of the encoder and decoder can be a multi-layer perceptron MLP including a fully connected layer, a convolutional neural network CNN including a convolutional layer, a Transformer with serialized data processing capabilities, etc.

[0047] In some embodiments, when the 3D input data is used as input and as a constraint condition, and a conditional variational autoencoder is used to reconstruct the spatial distribution of optical parameters in the body to be measured, the following steps A1 to A3 may be performed: A1: Based on the 3D input data, an encoder is used to extract measurement domain feature information, and sampling feature information in a latent space that satisfies a first preset distribution is extracted.

[0048] Here, the first preset distribution may be a Gaussian distribution.

[0049] In some embodiments, when extracting sampling feature information in a latent space that satisfies a first preset distribution, the sampling feature information in a latent space that satisfies a multi-dimensional Gaussian distribution can be extracted based on the 3D input data using a trained priori network.

[0050] like Figure 2 As shown, after rough reconstruction, the one-dimensional measurement data sequence can obtain 3D input data. The 3D input data is input into the trained prior network, and the sampling feature information in the latent space satisfying the multi-dimensional Gaussian distribution can be extracted through the prior network.

[0051] Among them, the prior network can be a deep learning model including multiple continuously stacked convolution modules, which can extract feature representations of multiple dimensions layer by layer through the continuously stacked convolution modules, so that the sampling feature information in the latent space that satisfies the multi-dimensional Gaussian distribution can be extracted through the prior network.

[0052] In some embodiments, the number of dimensions in the multidimensional Gaussian distribution may be 2 to 6. When the multidimensional Gaussian distribution includes at least 4 dimensions, the attributes corresponding to each dimension in the multidimensional Gaussian distribution include the position of the activation area, the shape of the activation area, the number of activation areas, and the activation degree of the activation area.

[0053] In this way, the activation area is described by four dimensions: the position of the activation area, the shape of the activation area, the number of the activation area, and the activation degree of the activation area, so that the feature semantics of the sampled feature information in the latent space can fully express the properties of the activation area; on the other hand, the Gaussian distribution dimension of 2-6 also avoids the poor reconstruction effect caused by too few dimensions and the more complicated model training and use caused by too many dimensions, taking into account both reconstruction effect and reconstruction efficiency.

[0054] In some embodiments, the encoder may include multiple continuously stacked convolution modules, and an attention mechanism may be introduced in the convolution module. By introducing the attention mechanism, an attention weight matrix that is strongly correlated with tissue depth may be obtained, so that the measurement domain feature information extracted by the encoder contains feature semantics that are strongly correlated with tissue depth. This can be used as a constraint condition during the decoding process to improve the accuracy of the final reconstruction result.

[0055] A2: Fusing the measurement domain feature information and the sampling feature information to obtain fused feature information.

[0056] In some embodiments, when the measurement domain feature information and the sampling feature information are fused, the measurement domain feature information and the sampling feature information may be cascaded to obtain fused feature information.

[0057] Wherein, during the cascading, the measurement domain characteristic information and the sampling characteristic information may be cascaded according to the channel dimension.

[0058] In some embodiments, when the measurement domain characteristic information and the sampling characteristic information are fused, the fusion of the measurement domain characteristic information and the sampling characteristic information can also be achieved by weighted summation or the like. The embodiments of the present disclosure do not limit the specific fusion method.

[0059] A3: The fused feature information is decoded into the spatial distribution of optical parameters in the body to be tested through a decoder.

[0060] Here, the fused characteristic information obtained by fusing the measurement domain characteristic information and the sampling characteristic information can be input into the decoder to obtain the spatial distribution of optical parameters in the body to be measured output by the decoder.

[0061] When the decoder decodes the fused feature information, the measurement domain feature information in the fused feature information may be used as a constraint condition to constrain the decoder's decoding of the sampling feature information in the fused feature information.

[0062] In this way, by introducing the measurement domain feature information extracted by the encoder from the 3D input data, the sampling feature information in the latent space that satisfies the multidimensional Gaussian distribution is constrained, which can not only give full play to the feature semantic expression ability of the multidimensional Gaussian distribution, but also take into account the accuracy of reconstructing the spatial distribution of optical parameters in the body to be measured.

[0063] In some embodiments, the conditional variational autoencoder is trained via the following first training process, which may include the following steps B1-B2: B1: Using the posterior network, we extract embedded features from the ground truth reconstructed image.

[0064] Here, the posterior network may have the same network architecture as the a priori network, or may include a more complex network architecture than the a priori network, that is, the network performance of the posterior network may be the same as the network performance of the a priori network, or may be stronger than the network performance of the a priori network.

[0065] B2: Utilize the embedded features to adjust parameters of the encoder and the decoder, and coordinate with the 3D input data to adjust parameters of the prior network.

[0066] For example, the schematic diagram of the training process can be as follows: Figure 4 As shown, Figure 4In the above example, the posterior network can extract embedded features from the ground truth reconstructed image and use the embedded features to participate in subsequent parameter adjustment. The specific parameter adjustment method will be introduced below and will not be described in detail here.

[0067] In some embodiments, the prior network is trained via the following second training process, which may include the following steps C1-C3: C1: Roughly reconstruct the sample one-dimensional measurement data sequence to obtain the sample 3D input data.

[0068] C2: The sample 3D input data and the ground truth reconstructed image are concatenated along the channel dimension and fed into the posterior network to extract the baseline sampling feature information containing the embedded features of the ground truth reconstructed image.

[0069] Here, by cascading the sample 3D input data and the ground truth reconstructed image along the channel dimension, the input content containing both the measured data feature semantics and the reconstruction result feature semantics can be obtained. After feeding the corresponding input content to the posterior network, the benchmark sampling feature information containing both the measured data feature semantics and the reconstruction result feature semantics can be extracted.

[0070] In this way, since the benchmark sampling feature information contains the embedded features of the ground truth reconstructed image, that is, it contains the feature semantics of the reconstruction result corresponding to the reconstruction target, it can be used as a reference target in the training process.

[0071] C3: Utilizing the sample sampling feature information corresponding to the sample one-dimensional measurement data sequence and the reference sampling feature information corresponding to the ground truth reconstructed image, the parameters of the prior network are adjusted.

[0072] In some embodiments, when adjusting the parameters of the prior network, the KL loss function can be used to determine the distribution deviation between the sample sampling feature information and the reference sampling feature information, and the parameters of the prior network can be adjusted according to the distribution deviation.

[0073] Here, since the baseline sampling feature information contains the feature semantics of the reconstruction result corresponding to the reconstruction target, the parameters of the prior network are adjusted according to the distribution deviation between the sample sampling feature information and the baseline sampling feature information, so that the sampling feature information output by the prior network can be closer to the reconstruction target, thereby improving the accuracy of the spatial distribution of optical parameters subsequently reconstructed by the decoder.

[0074] Furthermore, when the parameters of the a priori network are adjusted, the parameters of the a posteriori network may also be adjusted synchronously, that is, the parameters of the a priori network and the a posteriori network may be adjusted simultaneously.

[0075] The posterior network may be a pre-trained deep learning model, and when adjusting the parameters of the prior network, the network parameters of the posterior network may also be fine-tuned.

[0076] In some embodiments, both the prior network and the posterior network are untrained deep learning models, and in the second training process, parameters of the prior network and the posterior network can be adjusted simultaneously.

[0077] In some embodiments, the training process of the encoder and the decoder may include the following steps D1 to D5: D1: Roughly reconstruct the sample one-dimensional measurement data sequence to obtain the sample 3D input data.

[0078] Here, the sample one-dimensional measurement data sequence can be obtained by collecting the sample to be tested through a high-density measurement array. The spatial distribution of optical parameters in the sample to be tested is pre-set. Therefore, the pre-set spatial distribution of optical parameters in the sample to be tested can be used as a ground truth reconstructed image (reconstruction target).

[0079] D2: Based on the sample 3D input data, an encoder is used to extract sample measurement domain feature information, and sample sampling feature information in a latent space that satisfies a first preset distribution is extracted.

[0080] Here, the first preset distribution may be a multi-dimensional Gaussian distribution. For other related descriptions about the first preset distribution, reference may be made to the above related content and will not be repeated here.

[0081] D3: Fusing the sample measurement domain feature information and the sample sampling feature information to obtain sample fusion feature information.

[0082] Here, when the sample measurement domain feature information and the sample sampling feature information are fused, the sample measurement domain feature information and the sample sampling feature information may be cascaded according to a channel dimension to obtain sample fusion feature information.

[0083] D4: The sample fusion feature information is decoded into the spatial distribution of the sample optical parameters in the body to be tested through a decoder.

[0084] Here, the sample fusion characteristic information obtained by fusing the sample measurement domain characteristic information and the sample sampling characteristic information can be input into the decoder to obtain the spatial distribution of the sample optical parameters in the body to be measured output by the decoder.

[0085] When the decoder decodes the sample fusion feature information, the sample measurement domain feature information in the sample fusion feature information may be used as a constraint condition to constrain the decoder's decoding of the sample sampling feature information in the sample fusion feature information.

[0086] D5: Using the MSE loss function to determine the deviation between the spatial distribution of the sample optical parameters and the ground truth reconstructed image, and adjusting the parameters of the encoder and the decoder according to the deviation.

[0087] Here, the ground truth reconstructed image is the reconstruction target, and the spatial distribution of the sample optical parameters is the reconstruction result output by the decoder. The deviation between the current reconstruction result and the reconstruction target is determined by the MSE loss function, and the parameters of the encoder and the decoder can be adjusted according to the deviation to reduce the deviation between the reconstruction result and the reconstruction target, thereby continuously improving the reconstruction effect.

[0088] The reconstruction method for diffuse light tomography provided by each embodiment of the present application integrates the measurement data of each measurement position of the high-density measurement array arranged on the surface of the object to be measured to form a one-dimensional measurement data sequence with a predetermined measurement position order, so that the conditional variational autoencoder used in the subsequent model reasoning stage can extract the spatial distribution semantics that characterize the relative positional relationship between the measurement position and the object to be measured, so that in the model reasoning stage, more abundant features can be extracted in combination with the arrangement of each transmitting position and receiving position in the high-density measurement array, and the extracted spatial distribution semantics is also strongly related to the reconstruction process, which can alleviate the ill-posed problem of the solution caused by the lack of feature semantics in the inverse problem of diffuse light tomography reconstruction, and improve the accuracy of the reconstructed spatial distribution of optical parameters in the object to be measured. Furthermore, compared with directly reconstructing the one-dimensional measurement data sequence, by first roughly reconstructing the one-dimensional measurement data sequence into 3D input data that matches the spatial resolution of the ground truth reconstructed image, the feature semantics input to the conditional variational autoencoder can be further enriched, the ill-posed problem of the solution during reconstruction by the conditional variational autoencoder can be alleviated, and the accuracy of the reconstructed spatial distribution of optical parameters in the body to be measured can be improved. In the model inference stage, a mapping relationship from the sensor domain to the reconstructed image domain is established by using the conditional autoencoder, and the accuracy of the reconstruction results obtained by the conditional autoencoder is improved by adding constraints.

[0089] In some embodiments of the present application, a reconstruction device for diffuse optical tomography is provided, the reconstruction device comprising at least a processor and a memory, the memory storing computer executable instructions, and the processor executing the reconstruction method for diffuse optical tomography described in any embodiment of the present application when executing the computer executable instructions.

[0090] The processor may be a processing device including one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), etc. More specifically, the processor may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor running other instruction sets, or a processor running a combination of instruction sets. The processor may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), etc.

[0091] In some embodiments of the present application, a near-infrared brain function imaging system is provided, comprising a near-infrared optical data acquisition device and a reconstruction device for diffuse light tomography; wherein the near-infrared optical data acquisition device is used to acquire near-infrared data of an object; and the reconstruction device is used to use the near-infrared data to execute the reconstruction method for diffuse light tomography described in any embodiment of the present application.

[0092] This application describes various operations or functions that may be implemented as software codes or instructions or defined as software codes or instructions. Such content may be source code or differential code ("incremental" or "patch" code) that can be directly executed ("object" or "executable" form). Software codes or instructions may be stored in a computer-readable storage medium, and when executed, may cause a machine to perform the described functions or operations, and include any mechanism for storing information in a form accessible to a machine (e.g., a computing device, an electronic system, etc.), such as recordable or non-recordable media (e.g., read-only memory (ROM), random access memory (RAM), magnetic disk storage media, optical storage media, flash memory devices, etc.).

[0093] The exemplary methods described in this application may be implemented at least in part by a machine or computer. In some embodiments, a computer-readable storage medium stores computer program instructions, which, when executed by a processor, cause the processor to perform the reconstruction method for diffuse optical tomography described in various embodiments of this application.

[0094] The implementation of such a method may include software code, such as microcode, assembly language code, high-level language code, etc. Various software programming techniques may be used to create various programs or program modules. For example, a program portion or program module may be designed with or by means of Java, Python, C, C++, assembly language, or any known programming language. One or more of such software portions or modules may be integrated into a computer system and / or a computer-readable medium. Such software code may include computer-readable instructions for executing various methods. The software code may form a part of a computer program product or a computer program module. In addition, in an example, the software code may be tangibly stored on one or more volatile, non-temporary or non-volatile tangible computer-readable media, such as during execution or at other times. Examples of these tangible computer-readable media may include, but are not limited to, hard disks, removable disks, removable optical disks (such as optical disks and digital video disks), cassettes, memory cards or memory sticks, random access memories (RAMs), read-only memories (ROMs), etc.

[0095] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present application with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims are to be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the practice of this application, and the examples are to be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the following claims and the full scope of their equivalents.

[0096] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more schemes thereof) may be used in combination with each other. For example, a person of ordinary skill in the art may use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features may be grouped together to simplify the present application. This should not be interpreted as an intention that a disclosed feature that is not required to be protected is necessary for any claim. On the contrary, the subject matter of the present application may be less than all the features of a particular disclosed embodiment. Thus, the claims are incorporated herein as examples or embodiments in the specific embodiments, wherein each claim is independently a separate embodiment, and it is considered that these embodiments may be combined with each other in various combinations or arrangements. The scope of the present application should be determined with reference to the attached claims and the full scope of equivalent forms granted by these claims.

[0097] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A reconstruction method for diffuse optical tomography, characterized in that: The reconstruction method comprises, using a processor: Based on the measurement data of each measurement position of the high-density measurement array arranged on the surface of the object to be measured, a one-dimensional measurement data sequence of a predetermined measurement position order is formed; Based on a one-dimensional measurement data sequence with a predetermined measurement position order, a coarse reconstruction is performed to obtain 3D input data so that the spatial resolution is aligned with the ground truth reconstructed image; The 3D input data is used as input and as a constraint condition, and a conditional variational autoencoder is used to reconstruct the spatial distribution of optical parameters in the body to be tested.

2. The reconstruction method according to claim 1, characterized in that: The high-density measurement array includes transmission bits and receiving bits arranged in sequence and at intervals; The one-dimensional measurement data sequence of the predetermined measurement bit order includes measurement data corresponding to a plurality of transmission bits spliced ​​in sequence, each transmission bit has at least one target receiving bit and forms a measurement channel therewith, and the measurement data corresponding to each transmission bit includes the measurement data collected by each target receiving bit thereof.

3. The reconstruction method according to claim 2, characterized in that: The target receiving position of each transmitting position includes a plurality of receiving positions whose distances satisfy a preset distance condition; The measurement data corresponding to each transmitting position is obtained by sequentially splicing the measurement data collected by each target receiving position according to the predetermined sequence number of each target receiving position in the high-density measurement array.

4. The reconstruction method according to claim 1, characterized in that: The 3D input data is used as input and as a constraint condition, and a conditional variational autoencoder is used to reconstruct the spatial distribution of optical parameters in the body to be tested, specifically including: Based on the 3D input data, using an encoder to extract measurement domain feature information, and extracting sampling feature information in a latent space that satisfies a first preset distribution; Fusing the measurement domain feature information and the sampling feature information to obtain fused feature information; The fused feature information is decoded into the spatial distribution of optical parameters in the body to be tested by a decoder.

5. The reconstruction method according to claim 4, characterized in that: Based on the 3D input data, extracting sampling feature information in the latent space that satisfies the first preset distribution specifically includes: Based on the 3D input data, the trained prior network is used to extract sampling feature information in a latent space that satisfies a multi-dimensional Gaussian distribution.

6. The reconstruction method according to claim 5, characterized in that: The number of dimensions in the multidimensional Gaussian distribution is 2-6; When the multidimensional Gaussian distribution includes at least 4 dimensions, the attributes corresponding to each dimension in the multidimensional Gaussian distribution include: The location of the activation area, the shape of the activation area, the number of activation areas, and the degree of activation of the activation areas.

7. The reconstruction method according to claim 5, characterized in that: The conditional variational autoencoder is trained through the following first training process, which includes: Using the posterior network, embedding features from the ground truth reconstructed image are extracted; The encoder and the decoder are parameterized using the embedded features, and the prior network is parameterized in coordination with the 3D input data.

8. The reconstruction method according to claim 7, characterized in that: The prior network is trained through the following second training process, and the second training process specifically includes: Roughly reconstruct the sample one-dimensional measurement data sequence to obtain the sample 3D input data; Concatenating the sample 3D input data and the ground truth reconstructed image along the channel dimension and feeding them to the posterior network to extract reference sampled feature information including embedded features of the ground truth reconstructed image; The parameters of the prior network are adjusted using sample sampling feature information corresponding to the sample one-dimensional measurement data sequence and reference sampling feature information corresponding to the ground truth reconstructed image.

9. The reconstruction method according to claim 8, characterized in that: The parameters of the prior network are adjusted, specifically including: The KL loss function is used to determine the distribution deviation between the sample sampling feature information and the reference sampling feature information, and the parameters of the prior network are adjusted according to the distribution deviation.

10. The reconstruction method according to claim 7, characterized in that: The training process of the encoder and the decoder includes: Roughly reconstruct the sample one-dimensional measurement data sequence to obtain the sample 3D input data; Based on the sample 3D input data, extracting sample measurement domain feature information using an encoder, and extracting sample sampling feature information in a latent space that satisfies a first preset distribution; Fusion of the sample measurement domain feature information and the sample sampling feature information to obtain sample fusion feature information; The sample fusion feature information is decoded into the spatial distribution of the sample optical parameters in the body to be tested by a decoder; The MSE loss function is used to determine the deviation between the spatial distribution of the sample optical parameters and the ground truth reconstructed image, and the parameters of the encoder and the decoder are adjusted according to the deviation.

11. The reconstruction method according to claim 1, characterized in that: In a one-dimensional measurement data sequence with a predetermined measurement bit order: the measurement data is arranged following the predetermined measurement bit order; or, the measurement data is arranged in association with its position indication information, wherein the position indication information indicates the measurement bit order.

12. A reconstruction device for diffuse optical tomography, characterized in that: The reconstruction device at least comprises a processor and a memory, wherein the memory stores computer executable instructions, and the processor executes the reconstruction method for diffuse optical tomography according to any one of claims 1 to 11 when executing the computer executable instructions.

13. A near-infrared brain function imaging system, characterized in that: It includes a near-infrared optical data acquisition device and a reconstruction device for diffuse light tomography; wherein, The near-infrared optical data acquisition device is used to collect near-infrared data of the object; The reconstruction device is used to perform the reconstruction method for diffuse optical tomography according to any one of claims 1 to 11 using the near-infrared data.

14. A non-transitory computer-readable storage medium storing a program, characterized in that: The program causes the processor to perform the operations of the reconstruction method for diffuse optical tomography according to any one of claims 1-11.

Images