CTP image generation model construction method and device, CTP image generation method and device, equipment and medium

Abstract

The invention relates to the field of medical image processing, and provides a CTP image generation model construction method, a generation method, a device, equipment and a medium, and the model construction method comprises the steps: splicing a sample NCCT image and a sample CTA image in a channel dimension, and obtaining a combined image; performing feature extraction on the combined image to obtain combined image features, and performing feature refinement and enhancement on the combined image features to obtain refined features; performing feature decoding of the corresponding tasks on the refined features to obtain a reconstructed CTP image corresponding to each task; determining a physical consistency loss based on a physical parameter relationship between the reconstructed CTP images and the tag CTP image; and performing parameter iteration on the initial model based on the physical consistency loss, and taking the initial model after parameter iteration as a CTP image generation model. According to the model construction method and device, the generation method and device, the equipment and the medium provided by the invention, the generation quality of the CTP image can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a CTP image generation model construction method, a generation method, a device, equipment and a medium. Background Art

[0002] One-stop stroke CT examination for suspected acute ischemic stroke (AIS), including non-contrast computed tomography (NCCT), computed tomography angiography (CTA) and computed tomography perfusion (CTP).

[0003] CTP can quantitatively evaluate the ischemic penumbra, which is important for the treatment of AIS patients. However, CTP imaging requires a large number of back-and-forth scans, which results in a relatively large amount of radiation exposure to the patient. Generating CTP through other images can reduce the patient's radiation dose and imaging examination time. Since NCCT does not have contrast agent injection, it is difficult to obtain vascular information, and it is difficult to generate CTP through NCCT. CTA only requires one scan to obtain vascular information, so it is relatively feasible to generate CTP through NCCT and CTA.

[0004] In the related art, the image quality and image effect of CTP generated by NCCT and CTA are still poor. Summary of the invention

[0005] The present invention provides a CTP image generation model construction method, a generation method, a device, a equipment and a medium, which are used to solve the defects of poor image quality and image effect of CTP generated by NCCT and CTA in the prior art. The present invention provides a CTP image generation model construction method, comprising: Acquire a sample NCCT image, a sample CTA image and a corresponding label CTP image, and splice the sample NCCT image and the sample CTA image in the channel dimension to obtain a joint image; Based on the shared encoder in the initial model, feature extraction is performed on the joint image to obtain joint image features, and based on the feature refinement module in the initial model, feature refinement and enhancement are performed on the joint image features to obtain refined features; Based on the multi-task decoder in the initial model, the refined features are respectively decoded for the corresponding tasks to obtain a reconstructed CTP image corresponding to each task; Determining a physical consistency loss based on a physical parameter relationship between the reconstructed CTP images and the labeled CTP image; Based on the physical consistency loss, perform parameter iteration on the initial model, and use the initial model after parameter iteration as the CTP image generation model.

[0006] According to the CTP image generation model construction method provided by the present invention, determining the physical consistency loss based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP images includes: Based on the physical parameter relationship between the reconstructed CBF image and the reconstructed MTT image in the reconstructed CTP image, determine the calculated CBV image; Based on the difference between the calculated CBV image and the labeled CBV image in the labeled CTP image, determine the physical consistency loss.

[0007] According to the CTP image generation model construction method provided by the present invention, performing parameter iteration on the initial model based on the physical consistency loss includes: Based on the difference between the reconstructed images of each modality in the reconstructed CTP image and the labeled images of the corresponding modalities in the labeled CTP image, determine the image similarity loss; Stitch the reconstructed images of each modality in the reconstructed CTP image, and based on the discriminator, discriminate the stitched reconstructed CTP image, and determine the generative adversarial loss based on the discrimination result; Based on the physical consistency loss, the image similarity loss, and the generative adversarial loss, perform parameter iteration on the initial model.

[0008] According to the CTP image generation model construction method provided by the present invention, refining and enhancing the joint image features based on the feature refinement module in the initial model to obtain refined features includes: Based on the residual module in the feature refinement module, extract the residual features in the joint image features; Based on the self-attention module in the feature refinement module, extract the self-attention features in the joint image features; Fuse the residual features and the self-attention features to obtain the refined features.

[0009] According to the CTP image generation model construction method provided by the present invention, obtaining the sample NCCT image, the sample CTA image, and the corresponding labeled CTP image includes: Obtain the original sample NCCT image, the original sample CTA image, and the corresponding original labeled CTP image; Perform image preprocessing on the original sample NCCT image, the original sample CTA image, and the corresponding original label CTP image to obtain the sample NCCT image, the sample CTA image, and the corresponding label CTP image. The image preprocessing includes at least one of resampling, normalization, image slicing, and image screening.

[0010] The present invention also provides a CTP image generation method, including: Obtain the NCCT image and the CTA image of the CTP image to be generated; Input the NCCT image and the CTA image into the trained CTP image generation model to obtain the CTP images corresponding to each modality output by the CTP image generation model; The CTP image generation model is constructed based on the CTP image generation model construction method described in any one of claims 1 to 5.

[0011] The present invention also provides a CTP image generation model construction device, including: The first image acquisition unit is used to acquire the sample NCCT image, the sample CTA image, and the corresponding label CTP image, and splice the sample NCCT image and the sample CTA image in the channel dimension to obtain a combined image; The feature extraction unit is used to extract features from the combined image based on the shared encoder in the initial model to obtain combined image features, and refine and enhance the combined image features based on the feature refinement module in the initial model to obtain refined features; The feature decoding unit is used to perform feature decoding of corresponding tasks on the refined features based on the multi-task decoder in the initial model to obtain the reconstructed CTP images corresponding to each task; The loss determination unit is used to determine the physical consistency loss based on the physical parameter relationship between the reconstructed CTP images and the label CTP image; The parameter iteration unit is used to perform parameter iteration on the initial model based on the physical consistency loss, and use the initial model after parameter iteration as the CTP image generation model.

[0012] The present invention also provides a CTP image generation device, including: The second image acquisition unit is used to acquire the NCCT image and the CTA image of the CTP image to be generated; The image generation unit is used to input the NCCT image and the CTA image into the trained CTP image generation model to obtain the CTP images corresponding to each modality output by the CTP image generation model; The CTP image generation model is constructed based on the CTP image generation model construction method described in any one of claims 1 to 5.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the CTP image generation model construction method or the CTP image generation method described in any one of the above.

[0014] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the CTP image generation model construction method or the CTP image generation method described in any one of the above.

[0015] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the CTP image generation model construction method or the CTP image generation method described in any one of the above.

[0016] In the CTP image generation model construction method, generation method, device, equipment, and medium provided by the present invention, during the model training process, the physical parameter relationship between the reconstructed CTP images is fully considered, and the parameters of the initial model are iterated based on the determined physical consistency loss. Thus, the obtained CTP image generation model can generate CTP images with higher quality and better effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 is one of the flow schematic diagrams of the CTP image generation model construction method provided by the present invention.

[0019] Figure 2 is the second flow schematic diagram of the CTP image generation model construction method provided by the present invention.

[0020] Figure 3 is the schematic diagram of the generated CTP image provided by the present invention.

[0021] Figure 4 is the flow schematic diagram of the CTP image generation method provided by the present invention.

[0022] Figure 5It is a schematic structural diagram of the CTP image generation model construction device provided by the present invention.

[0023] Figure 6 It is a schematic structural diagram of the CTP image generation device provided by the present invention.

[0024] Figure 7 It is a schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0025] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] Aiming at the problem that the image quality and image effect of generating CTP through NCCT and CTA in the prior art are poor, an embodiment of the present invention proposes a method for constructing a CTP image generation model. In this method, first, a sample NCCT image, a sample CTA image, and a corresponding labeled CTP image are obtained, and the sample NCCT image and the sample CTA image are spliced in the channel dimension to obtain a combined image; based on the shared encoder in the initial model, feature extraction is performed on the combined image to obtain combined image features, and based on the feature refinement module in the initial model, the combined image features are refined and enhanced to obtain refined features; based on the multi-task decoder in the initial model, feature decoding for corresponding tasks is respectively performed on the refined features to obtain reconstructed CTP images corresponding to each task respectively; based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image, a physical consistency loss is determined; based on the physical consistency loss, parameter iteration is performed on the initial model, and the initial model after parameter iteration is used as the CTP image generation model.

[0027] In the process of model training in the embodiment of the present invention, the physical parameter relationship between the reconstructed CTP images is fully considered, and parameter iteration is performed based on the determined physical consistency loss. The CTP image generation model obtained thereby can generate CTP images with higher quality and better effect.

[0028] Embodiments of the present invention can be applied to scenarios where it is necessary to generate CTP images by combining NCCT images and CTA images. The execution subject of this method can be an electronic device such as a terminal device, a computer, a server, a server cluster, or a specially designed CTP image generation model construction device, or it can be a CTP image generation model construction device set in the electronic device, and the CTP image generation model construction device can be implemented by software, hardware, or a combination of both.

[0029] In the description of the embodiments of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the embodiments of the present invention, the meaning of "plurality" is two or more, unless otherwise specifically defined.

[0030] Figure 1 is one of the flow schematic diagrams of the CTP image generation model construction method provided by the present invention. As Figure 1 shown, the method includes the following steps: Step 110, obtain a sample NCCT image, a sample CTA image, and a corresponding labeled CTP image, splice the sample NCCT image and the sample CTA image in the channel dimension to obtain a combined image; Step 120, based on the shared encoder in the initial model, extract features from the combined image to obtain combined image features, and based on the feature refinement module in the initial model, refine and enhance the combined image features to obtain refined features; Step 130, based on the multi-task decoder in the initial model, perform feature decoding for corresponding tasks on the refined features to obtain reconstructed CTP images corresponding to each task respectively; Step 140, determine the physical consistency loss based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image; Step 150, perform parameter iteration on the initial model based on the physical consistency loss, and use the initial model after parameter iteration as the CTP image generation model.

[0031] Specifically, the sample NCCT image, the sample CTA image, and the corresponding labeled CTP image refer to the training data for model training. The NCCT image refers to the image obtained by non-contrast-enhanced CT technology without injecting contrast agent, the CTA image refers to the CT imaging technology obtained by injecting contrast agent to display blood vessels, and the CTP image refers to the CT imaging technology obtained by injecting contrast agent to evaluate the cerebral tissue blood perfusion.

[0032] CTP images may include parametric maps of multiple modalities. By performing continuous dynamic scans on selected interested levels, the time-density curves of each pixel in the selected levels are obtained, and various parametric maps are obtained through mathematical model processing.

[0033] Each parametric map respectively includes: cerebral blood volume image (CBV), cerebral blood flow image (CBF), mean transit time image (MTT), and time to maximum image (TMAX).

[0034] In some embodiments, obtaining the sample NCCT image, the sample CTA image, and the corresponding labeled CTP image in step 110 specifically includes: Step 111, obtaining the original sample NCCT image, the original sample CTA image, and the corresponding original labeled CTP image; Step 112, performing image preprocessing on the original sample NCCT image, the original sample CTA image, and the corresponding original labeled CTP image to obtain the sample NCCT image, the sample CTA image, and the corresponding labeled CTP image. The image preprocessing includes at least one of resampling, normalization, image slicing, and image screening.

[0035] Specifically, the original sample NCCT image, the original sample CTA image, and the corresponding original labeled CTP image may be the original images directly obtained by scanning. After obtaining the original images, image preprocessing operations need to be performed.

[0036] The resampling here refers to resampling the four-modal images of the original labeled CTP image to align it with the original sample NCCT image and the original sample CTA image. The original sample NCCT image and the original sample CTA image are already aligned, so only the original labeled CTP image needs to be aligned with the original sample NCCT image. Assume that the original sample NCCT image after removing the skull is , and the original sample CTP image is . The alignment operation can be expressed as:

[0037] Among them, is the original labeled CTP image after resampling. represents resampling the CTP image to the space of the NCCT image.

[0038] Normalization refers to performing normalization operations on the original sample NCCT images, original sample CTA images, and resampled original label CTP images. Since NCCT, CTA, and CTP images of different modalities have different image features, gray-scale ranges, and physical meanings, the brightness and contrast of the images are adjusted by using window level and window width parameters to highlight important anatomical structures or lesions. In this embodiment, the window level and window width of the NCCT images are set to 40 and 80 respectively, and the window level and window width of the CTA images are set to 100 and 200 respectively. The window levels of CBF, CBV, TMAX, and MTT in the CTP images are set to 40, 10, 10, and 20 respectively, and the window widths are set to 80, 20, 20, and 40 respectively.

[0039] Preferably, perform slicing operations on the NCCT, CTA, and CTP three-dimensional images, and delete unqualified two-dimensional slice images through doctor inspection.

[0040] Divide the preprocessed images. According to a ratio (for example, it can be 8:1:1), divide the corresponding NCCT, CTA, and CTP images of four modalities into a training set, a validation set, and a test set, thereby obtaining sample NCCT images, sample CTA images, and corresponding label CTP images.

[0041] Figure 2 It is the second flow schematic diagram of the method for constructing a CTP image generation model provided by the present invention. As Figure 2 shown, after obtaining the sample NCCT images and sample CTA images, immediately splice the sample NCCT images and sample CTA images in the channel dimension to obtain a combined image. Assume the sample NCCT image is and the sample CTA image is . The image combination formula can be expressed as:

[0042] where is the combined image, and represents the splicing operation of the two images in the channel dimension.

[0043] Step 120: Based on the shared encoder in the initial model, extract features from the combined image to obtain combined image features. Based on the feature refinement module in the initial model, refine and enhance the combined image features to obtain refined features.

[0044] Specifically, assume the shared encoder is Encoder. The formula for extracting features from the combined image can be expressed as:

[0045] where The joint image features extracted by the shared encoder.

[0046] Here, in step 120, based on the feature refinement module in the initial model, the joint image features are refined and enhanced to obtain refined features, which specifically may include: Step 121, based on the residual module in the feature refinement module, extract the residual features in the joint image features; Step 122, based on the self-attention module in the feature refinement module, extract the self-attention features in the joint image features; Step 123, fuse the residual features and the self-attention features to obtain refined features.

[0047] Specifically, the feature refinement module adopts a parallel structure of a residual module and a self-attention module. Assuming the joint image features are , the extraction of the residual features in the joint image features can be expressed as:

[0048] Among them, contains a 3×3 convolution, InstanceNorm2d normalization, and a ReLU activation function, is the residual feature after passing through the residual block.

[0049] Based on the self-attention module in the feature refinement module, the extraction of the self-attention features in the joint image features can be expressed as:

[0050] Among them, is the self-attention mechanism, is the feed-forward network, which contains two linear layers and a ReLU activation function. is layer normalization. is the self-attention feature after passing through the self-attention module. Here, the self-attention module can be, for example, a transformer block.

[0051] The refinement module fuses the outputs of the residual module and the self-attention module through a parallel structure, which can be expressed as:

[0052] Among them, is the feature after passing through the refinement module, that is, the refined feature; represents a 1×1 convolution operation, represents an InstanceNorm2d layer, represents a ReLU activation function. represents 4 residual modules, represents 2 transformer blocks, Indicates the concatenation operation of features in the channel dimension.

[0053] Step 130: Based on the multi-task decoder in the initial model, perform feature decoding for the corresponding tasks on the refined features respectively to obtain the reconstructed CTP images corresponding to each task.

[0054] Specifically, the reconstructed CTP images corresponding to each task here can be the four modalities corresponding to the reconstructed CTP images, that is, the four parameter maps, namely the reconstructed CBF image, the image CBV image, the reconstructed TMAX image, and the reconstructed MTT image.

[0055] Assume that the decoders for generating the reconstructed CBF, CBV, TMAX, and MTT images are respectively , , and . Therefore, the reconstruction of virtual CTP images of different modalities can be expressed as:

[0056]

[0057]

[0058]

[0059] Among them, , , and are the reconstructed CBF image, the reconstructed CBV image, the reconstructed TMAX image, and the reconstructed MTT image respectively, which are reconstructed by the four specific task decoders.

[0060] Considering that the physical parameter relationship between the reconstructed CTP images is not considered in the related technology, resulting in poor quality and effect of the generated CTP images, the embodiments of the present invention consider the physical parameter relationship between the reconstructed CTP images during model training and perform parameter iteration based on the determined physical consistency loss, so as to obtain the CTP image generation model, thereby improving the quality and effect of the generated CTP images.

[0061] In some embodiments, based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP images, determine the physical consistency loss, that is, step 140 specifically includes: Step 141: Based on the physical parameter relationship between the reconstructed CBF image and the reconstructed MTT image in the reconstructed CTP image, determine the calculated CBV image; Step 142: Based on the difference between the calculated CBV image and the labeled CBV image in the labeled CTP image, determine the physical consistency loss.

[0062] Specifically, there is a close physical relationship among the three modalities of CBF, CBV, and MTT images, and their relationship can be expressed by the formula:

[0063] Among them, CBF is cerebral blood flow, CBV is cerebral blood volume, and MTT is mean transit time. Therefore, the physical consistency loss can be expressed as:

[0064] In the formula, is the reconstructed CBF image, is the reconstructed MTT image, is the calculated CBV image obtained, is the labeled CBV image.

[0065] In some embodiments, the parameter iteration of the initial model based on the physical consistency loss in step 150 specifically includes: Step 151, determining the image similarity loss based on the difference between the reconstructed images of each modality in the reconstructed CTP image and the labeled images of the corresponding modalities in the labeled CTP image; Step 152, splicing the reconstructed images of each modality in the reconstructed CTP image, discriminating the spliced reconstructed CTP image based on the discriminator, and determining the generative adversarial loss based on the discrimination result; Step 153, performing parameter iteration on the initial model based on the physical consistency loss, the image similarity loss, and the generative adversarial loss.

[0066] Specifically, in order to further improve the generation effect of the CTP image generation model, the image similarity loss and the generative adversarial loss are also introduced in this embodiment during model training.

[0067] The image similarity loss can be expressed as:

[0068]

[0069] Here, , , and respectively represent the reconstructed images of each modality in the reconstructed CTP image, , , and respectively represent the labeled images of the corresponding modalities in the labeled CTP image.

[0070] In addition, to further improve the quality of the generated CTP images, the initial model can also be trained adversarially in combination with a discriminator. The adversarial loss of the generator can be expressed as:

[0071] where is the binary cross-entropy loss, is the discriminator. denotes that the four different generated CTP modality images are concatenated in the channel dimension and then input into the discriminator for joint discrimination.

[0072] Therefore, the total loss of the initial model can be expressed as:

[0073] and are weight coefficients, which are set to 10 and 0.1 respectively.

[0074] Here, the adversarial loss of the discriminator is expressed as:

[0075] According to the description of the above embodiments, it can be seen that in the process of model training of the present invention, the physical parameter relationship between the reconstructed CTP images is fully considered, and parameter iteration is performed based on the determined physical consistency loss. The CTP image generation model obtained thereby can generate CTP images with higher quality and better effects.

[0076] To verify the performance of the multi-task generation model in the present invention, structural similarity (SSIM) is used as a quantitative index to evaluate the model.

[0077]

[0078] where is the real CTP image, is the virtual CTP image generated by the model. and are the means of the images and respectively, and are the variances of the images and respectively, is the covariance between the images and respectively, and are two constants.

[0079] Table 1

[0080] As shown in Table 1 above, Table 1 shows the comparison between the present invention and traditional single-task generation models. According to the results in Table 1, it can be concluded that the SSIM result of the present invention has a significant improvement in single-task generation. Figure 3 It is a schematic diagram of generating a CTP image provided by the present invention. From Figure 3 it can be seen that the visual effect of the present invention is also better than that of traditional single-task generation models.

[0081] Based on any of the above embodiments, Figure 4 It is a schematic flowchart of the CTP image generation method provided by the present invention. As Figure 4 shown, the method includes: Step 410, obtaining the NCCT image and CTA image of the CTP image to be generated; Step 420, inputting the NCCT image and CTA image into the trained CTP image generation model to obtain the CTP images corresponding to each modality output by the CTP image generation model; the CTP image generation model is constructed based on the above CTP image generation model construction method.

[0082] Specifically, after constructing the CTP image generation model according to the method described in the above embodiment, the CTP image can be generated based on the CTP image generation model.

[0083] First, obtain the NCCT image and CTA image for generating the CTP image. It can be understood that the NCCT image and CTA image here are aligned.

[0084] Then, input the NCCT image and CTA image into the trained CTP image generation model to obtain the CTP images corresponding to each modality output by the CTP image generation model. The CTP images corresponding to each modality here may include , , and the parametric images of these four modalities.

[0085] Next, the CTP image generation model construction device provided by the present invention will be described. The CTP image generation model construction device described below can be correspondingly referred to the CTP image generation model construction method described above.

[0086] Based on any of the above embodiments, Figure 5 It is a schematic structural diagram of the CTP image generation model construction device provided by the present invention. As Figure 5 shown, the device includes: The first image acquisition unit 510 is configured to acquire a sample NCCT image, a sample CTA image, and a corresponding labeled CTP image, and splice the sample NCCT image and the sample CTA image in the channel dimension to obtain a combined image; The feature extraction unit 520 is configured to extract features from the combined image based on the shared encoder in the initial model to obtain combined image features, and refine and enhance the combined image features based on the feature refinement module in the initial model to obtain refined features; The feature decoding unit 530 is configured to perform feature decoding for corresponding tasks on the refined features based on the multi-task decoder in the initial model to obtain reconstructed CTP images corresponding to each task respectively; The loss determination unit 540 is configured to determine a physical consistency loss based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image; The parameter iteration unit 550 is configured to perform parameter iteration on the initial model based on the physical consistency loss, and use the initial model after parameter iteration as the CTP image generation model.

[0087] In the device provided by the embodiment of the present invention, during the model training process, the physical parameter relationship between the reconstructed CTP images is fully considered, and parameter iteration is performed based on the determined physical consistency loss. Therefore, the CTP image generation model obtained can generate CTP images with higher quality and better effects.

[0088] Based on any of the above embodiments, the loss determination unit is specifically configured to: Determine a calculated CBV image based on the physical parameter relationship between the reconstructed CBF image and the reconstructed MTT image in the reconstructed CTP image; Determine the physical consistency loss based on the difference between the calculated CBV image and the labeled CBV image in the labeled CTP image.

[0089] Based on any of the above embodiments, the parameter iteration unit is specifically configured to: Determine an image similarity loss based on the difference between the reconstructed images of each modality in the reconstructed CTP image and the labeled images of the corresponding modalities in the labeled CTP image; Splice the reconstructed images of each modality in the reconstructed CTP image, and perform discrimination on the spliced reconstructed CTP image based on a discriminator, and determine a generative adversarial loss based on the discrimination result; Perform parameter iteration on the initial model based on the physical consistency loss, the image similarity loss, and the generative adversarial loss.

[0090] Based on any of the above embodiments, the feature extraction unit is specifically configured to: Extract the residual features in the joint image features based on the residual module in the feature refinement module; Extract the self-attention features in the joint image features based on the self-attention module in the feature refinement module; Perform feature fusion on the residual features and the self-attention features to obtain the refined features.

[0091] Based on any of the above embodiments, the first image acquisition unit is specifically configured to: Acquire the original sample NCCT image, the original sample CTA image, and the corresponding original label CTP image; Perform image preprocessing on the original sample NCCT image, the original sample CTA image, and the corresponding original label CTP image to obtain the sample NCCT image, the sample CTA image, and the corresponding label CTP image, where the image preprocessing includes at least one of resampling, normalization, image slicing, and image screening.

[0092] Based on any of the above embodiments, Figure 6 is a schematic structural diagram of a CTP image generation device provided by the present invention, as Figure 6 shown, the device includes: A second image acquisition unit 610, configured to acquire the NCCT image and the CTA image of the CTP image to be generated; An image generation unit 620, configured to input the NCCT image and the CTA image into a trained CTP image generation model to obtain the CTP images corresponding to each modality output by the CTP image generation model; The CTP image generation model is constructed based on the CTP image generation model construction method.

[0093] Figure 7 Illustrates a schematic structural diagram of an electronic device, as Figure 7As shown in the figure, the electronic device may include: a processor 710, a communications interface 720, a memory 730, and a communication bus 740. Among them, the processor 710, the communications interface 720, and the memory 730 complete communication with each other through the communication bus 740. The processor 710 may call the logical instructions in the memory 730 to execute the CTP image generation model construction method, and the method includes: obtaining a sample NCCT image, a sample CTA image, and a corresponding labeled CTP image, splicing the sample NCCT image and the sample CTA image in the channel dimension to obtain a combined image; based on the shared encoder in the initial model, extracting features from the combined image to obtain combined image features, and based on the feature refinement module in the initial model, refining and enhancing the combined image features to obtain refined features; based on the multi-task decoder in the initial model, respectively performing feature decoding of corresponding tasks on the refined features to obtain reconstructed CTP images corresponding to each task respectively; based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image, determining the physical consistency loss; based on the physical consistency loss, performing parameter iteration on the initial model, and using the initial model after parameter iteration as the CTP image generation model.

[0094] The processor may call the logical instructions in the memory to execute the CTP image generation method, and the method includes: obtaining an NCCT image and a CTA image of the CTP image to be generated; inputting the NCCT image and the CTA image into the trained CTP image generation model to obtain CTP images corresponding to each modality output by the CTP image generation model; the CTP image generation model is constructed based on the CTP image generation model construction method.

[0095] In addition, when the logical instructions in the above-mentioned memory 730 are implemented in the form of software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc that can store program codes.

[0096] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the CTP image generation model construction method provided by each of the above methods. The method includes: obtaining a sample NCCT image, a sample CTA image, and a corresponding labeled CTP image, splicing the sample NCCT image and the sample CTA image in the channel dimension to obtain a combined image; based on the shared encoder in the initial model, extracting features from the combined image to obtain combined image features, and based on the feature refinement module in the initial model, refining and enhancing the combined image features to obtain refined features; based on the multi-task decoder in the initial model, respectively performing feature decoding for corresponding tasks on the refined features to obtain reconstructed CTP images corresponding to each task respectively; based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image, determining a physical consistency loss; based on the physical consistency loss, performing parameter iteration on the initial model, and using the initial model after parameter iteration as the CTP image generation model.

[0097] When the computer program is executed by a processor, the computer can execute the CTP image generation method provided by each of the above methods. The method includes: obtaining an NCCT image and a CTA image of the CTP image to be generated; inputting the NCCT image and the CTA image into the trained CTP image generation model to obtain CTP images corresponding to each modality output by the CTP image generation model; the CTP image generation model is constructed based on the CTP image generation model construction method.

[0098] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the CTP image generation model construction method provided by the above-mentioned various methods. The method includes: obtaining a sample NCCT image, a sample CTA image, and a corresponding labeled CTP image; splicing the sample NCCT image and the sample CTA image in the channel dimension to obtain a combined image; based on a shared encoder in an initial model, extracting features from the combined image to obtain combined image features, and based on a feature refinement module in the initial model, refining and enhancing the combined image features to obtain refined features; based on a multi-task decoder in the initial model, respectively performing feature decoding for corresponding tasks on the refined features to obtain reconstructed CTP images respectively corresponding to each task; based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image, determining a physical consistency loss; based on the physical consistency loss, performing parameter iteration on the initial model, and taking the initial model after parameter iteration as the CTP image generation model.

[0099] When the computer program is executed by a processor, the computer can execute the CTP image generation method provided by the above-mentioned various methods. The method includes: obtaining an NCCT image and a CTA image of a CTP image to be generated; inputting the NCCT image and the CTA image into a trained CTP image generation model to obtain CTP images respectively corresponding to each modality output by the CTP image generation model; the CTP image generation model is constructed based on the CTP image generation model construction method.

[0100] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0101] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence, or the parts that contribute to the prior art can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a CTP image generation model, characterized in that Including: Obtain a sample NCCT image, a sample CTA image, and a corresponding labeled CTP image, and splice the sample NCCT image and the sample CTA image in the channel dimension to obtain a combined image; Based on the shared encoder in the initial model, extract features from the combined image to obtain combined image features, and based on the feature refinement module in the initial model, refine and enhance the combined image features to obtain refined features; Based on the multi-task decoder in the initial model, perform feature decoding for corresponding tasks on the refined features to obtain reconstructed CTP images corresponding to each task respectively; Based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image, determine the physical consistency loss; Based on the physical consistency loss, perform parameter iteration on the initial model, and use the initial model after parameter iteration as the CTP image generation model.

2. The method for constructing a CTP image generation model according to claim 1, wherein The determining the physical consistency loss based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image includes: Based on the physical parameter relationship between the reconstructed CBF image and the reconstructed MTT image in the reconstructed CTP image, determine the calculated CBV image; Based on the difference between the calculated CBV image and the labeled CBV image in the labeled CTP image, determine the physical consistency loss.

3. The method for constructing a CTP image generation model according to claim 1, wherein The performing parameter iteration on the initial model based on the physical consistency loss includes: Based on the difference between the reconstructed images of each modality in the reconstructed CTP image and the labeled images of the corresponding modalities in the labeled CTP image, determine the image similarity loss; Splice the reconstructed images of each modality in the reconstructed CTP image, and based on a discriminator, discriminate the spliced reconstructed CTP image, and determine the generative adversarial loss based on the discrimination result; Based on the physical consistency loss, the image similarity loss, and the generative adversarial loss, perform parameter iteration on the initial model.

4. The method for constructing a CTP image generation model according to claim 1, wherein The refining and enhancing the combined image features based on the feature refinement module in the initial model to obtain refined features includes: Based on the residual module in the feature refinement module, extract the residual features in the combined image features; Based on the self-attention module in the feature refinement module, extract the self-attention features in the combined image features; Fuse the residual features and the self-attention features to obtain the refined features.

5. The method for constructing a CTP image generation model according to any one of claims 1 to 4, characterized in that The obtaining the sample NCCT image, the sample CTA image, and the corresponding labeled CTP image includes: Obtain the original sample NCCT image, the original sample CTA image, and the corresponding original labeled CTP image; Perform image preprocessing on the original sample NCCT image, the original sample CTA image, and the corresponding original labeled CTP image to obtain the sample NCCT image, the sample CTA image, and the corresponding labeled CTP image, and the image preprocessing includes at least one of resampling, normalization, image slicing, and image screening.

6. A method for generating a CTP image, characterized in that, Including: Obtain the NCCT image and CTA image of the CTP image to be generated; Input the NCCT image and CTA image into the trained CTP image generation model to obtain the CTP images corresponding to each modality output by the CTP image generation model. The CTP image generation model is constructed based on the CTP image generation model construction method described in any one of claims 1 to 5.

7. An apparatus for constructing a CTP image generation model, characterized in that, It includes: A first image acquisition unit, configured to acquire a sample NCCT image, a sample CTA image, and a corresponding labeled CTP image, splice the sample NCCT image and the sample CTA image in the channel dimension to obtain a combined image; A feature extraction unit, configured to extract features from the combined image based on the shared encoder in the initial model to obtain combined image features, and refine and enhance the combined image features based on the feature refinement module in the initial model to obtain refined features; A feature decoding unit, configured to perform feature decoding of corresponding tasks on the refined features respectively based on the multi-task decoder in the initial model to obtain the reconstructed CTP images corresponding to each task; A loss determination unit, configured to determine the physical consistency loss based on the physical parameter relationship between the reconstructed CTP images and the labeled CTP image; A parameter iteration unit, configured to perform parameter iteration on the initial model based on the physical consistency loss, and use the initial model after parameter iteration as the CTP image generation model.

8. A CTP image generation device, characterized in that, It includes: A second image acquisition unit, configured to acquire the NCCT image and CTA image of the CTP image to be generated; An image generation unit, configured to input the NCCT image and CTA image into the trained CTP image generation model to obtain the CTP images corresponding to each modality output by the CTP image generation model; The CTP image generation model is constructed based on the CTP image generation model construction method described in any one of claims 1 to 5.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the CTP image generation model construction method described in any one of claims 1 to 5, or the CTP image generation method described in claim 6.

10. A non-transitory computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the CTP image generation model construction method described in any one of claims 1 to 5, or the CTP image generation method described in claim 6.

Images