Parameter image generation method and device for shortening dynamic PET scanning time

Abstract

According to the parameter image generation method and device for shortening the dynamic PET scanning time, the electronic equipment and the storage medium, the 3D image of the time frame of the first 25 minutes of dynamic PET scanning is obtained, feature extraction is conducted on the 3D image, the multi-scale feature map is generated through the network encoder part, the multi-scale feature map is mapped through the network decoder part, and the dynamic PET scanning time is shortened. According to the method, a Ki parameter image for network prediction is generated, the Ki parameter image for network prediction is optimized, the optimized Ki parameter image serves as input of next iteration, the steps are repeated until the set convergence condition is met, and the Ki parameter image meeting the requirement is finally obtained. And the quality of the generated Ki parameter image is beneficial to auxiliary diagnosis, so that the clinical examination efficiency and the comfort of a patient are remarkably improved, motion artifacts which are easy to occur during dynamic PET scanning are reduced as much as possible, and the quality of the generated Ki image and the diagnosis accuracy of a doctor are further ensured.

Description

Technical Field

[0001] The present application relates to the technical field of medical positron emission tomography (PET) imaging, and particularly relates to a method for generating parametric images for shortening dynamic PET scan time, an imaging device, an electronic device, and a storage medium. Background Art

[0002] Positron emission tomography / computed tomography (PET / CT) combined with 18F-fluorodeoxyglucose (18F-FDG) plays a crucial role in the management of cancer patients. This imaging technique provides information on the distribution of tracer concentration over time through dynamic PET scans, and can accurately quantify the glucose metabolic rate (Ki parameter), providing key support for the early diagnosis, staging, and treatment monitoring of cancer. Compared with traditional static PET, the core advantage of dynamic PET technology is its ability to capture the dynamic distribution process of the tracer in the human body, thereby generating the Ki parameter that reflects the tissue metabolic rate. This parameter can not only more accurately describe the biological behavior of tumors, but also provide clinicians with quantitative diagnostic evidence. For example, in the diagnosis and treatment of lung cancer, breast cancer, and lymphoma, the Ki parameter has been proven to be able to distinguish benign and malignant lesions, evaluate treatment responses, and predict prognosis, becoming an important tool for precision medicine.

[0003] The core advantages of dynamic PET and the Ki parameter are mainly reflected in the following aspects. First, by continuously collecting time-series data, dynamic PET can directly reflect the metabolic kinetics process of the tracer in tissues. Its core parameter Ki (net influx rate) is calculated through the Patlak model, which directly quantifies the rate of glucose entering tissues from the blood. This parameter has higher biological significance compared to the standardized uptake value (SUV) of static PET. SUV only reflects the tracer uptake at a certain time point and is easily affected by factors such as blood glucose level, injection time difference, and patient body size, resulting in a relatively high variability in measurement results. The Ki parameter can eliminate these interfering factors through the integration of dynamic data, improving specificity and reducing the occurrence of false positive results.

[0004] Secondly, by providing objective and reproducible quantitative data, the Ki parameter significantly reduces the subjective differences in clinical film reading. Traditional PET images rely on doctors' visual interpretation of SUV values, and differences may occur among different film readers due to differences in experience or judgment criteria. This problem is particularly prominent in complex cases (such as the differentiation of small lesions or inflammation and tumors). The Ki image is generated through mathematical modeling of dynamic data, and its value has a strict physical meaning, providing a standardized basis for diagnosis.

[0005] In addition, the Ki parameter has significant advantages in lesion detection capabilities. Because dynamic PET can capture subtle changes in metabolic rate, Ki images are superior to traditional SUV images in terms of target-to-background ratio (TBR) and contrast-to-noise ratio (CNR). This feature enables it to detect tiny lesions less than 1 cm in diameter earlier. For example, in early screening for liver cancer, the Ki parameter can identify tiny metastases that are difficult to detect with traditional SUV images, buying valuable time for clinical intervention. This improvement in early detection capabilities is of great significance to improving patient prognosis, as early treatment of tumors is often associated with higher survival rates and lower risks of recurrence.

[0006] However, the traditional method of generating Ki parameters relies on dynamic PET scan data of up to 60 minutes, which has caused multiple problems in actual clinical applications. First, patients need to remain still for a long time during the scan, which is extremely difficult for many patients (especially elderly patients or those who are sensitive to pain). According to a clinical statistics in the Journal of Nuclear Medicine in 2021, about 35% of patients move their bodies because they cannot tolerate the 60-minute scan time, resulting in motion artifacts in the image. These artifacts can significantly reduce the calculation accuracy of Ki parameters and even require repeated scans, further exacerbating patients' discomfort and anxiety. Secondly, the occupation of medical resources by long-term scanning cannot be ignored. A single dynamic PET scan takes more than 1 hour to use the equipment, resulting in a more than 50% decrease in the hospital's daily inspection throughput. In areas with tight medical resources, this problem directly prolongs the patient's appointment waiting time and delays the timeliness of diagnosis and treatment. Finally, attempts to shorten the scan time face technical bottlenecks. The traditional Patlak model requires a complete 60-minute time-activity curve (TAC) data to accurately calculate the Ki parameter. If the scanning time is shortened to 25 minutes, the loss of TAC tail data will lead to a significant increase in the model fitting error.

[0007] Dynamic PET and Ki parameters have irreplaceable advantages in cancer diagnosis and treatment, but their clinical application is limited by the reliance of traditional methods on long-term scanning. How to break through this bottleneck through technological innovation and significantly shorten the scanning time while ensuring the accuracy of Ki images has become a key issue that needs to be solved urgently. Summary of the invention

[0008] In view of this, the present invention provides a parameter image generation method, an imaging device, an electronic device and a storage medium for shortening the dynamic PET scanning time, which greatly shortens the scanning time while ensuring the accuracy of the Ki image.

[0009] To solve the above problems, this application adopts the following technical solutions:

[0010] One of the objectives of the present application is to provide a method for generating parametric images to shorten the dynamic PET scanning time. The method includes the following steps:

[0011] Obtain a 3D image of the first 25-minute time frame of the dynamic PET scan;

[0012] Use a network encoder to extract features from the 3D image to generate a multi-scale feature map;

[0013] Map the multi-scale feature map through a network decoder part to generate a network-predicted Ki parametric image;

[0014] Optimize the network-predicted Ki parametric image, and use the optimized Ki parametric image as the input for the next iteration;

[0015] Repeat the above steps until the set convergence condition is met;

[0016] Obtain a Ki parametric image that meets the requirements.

[0017] In some embodiments, in the step of using a network encoder to extract features from the 3D image to generate a multi-scale feature map, it specifically includes:

[0018] The network encoder is an encoder with a single-channel 3D U-Net architecture. The encoder consists of multiple 3D convolutional layers and residual modules. Each residual module contains two 3×3×3 convolutional layers, with batch normalization BN and ReLU activation functions inserted in the middle. Downsampling is achieved through a convolutional layer with a stride of 2. The input slice group D is 3, and the number of channels of all convolutional layers is uniformly set to 20, which is specifically expressed as:

[0019] f j+1 = Conv3D(f j )

[0020] where f j is the feature map of the j-th layer, and high-level features are gradually extracted through stacked residual modules.

[0021] In some embodiments, in the step of mapping the multi-scale feature map through a network decoder part to generate a network-predicted Ki parametric image, it specifically includes:

[0022] The network decoder consists of deconvolution layers and skip connections. Each upsampling stage uses a 3×3×3 deconvolution kernel to expand the spatial dimension, and channel splicing is performed in combination with the feature map of the corresponding layer of the encoder. Each decoding module contains a deconvolution layer, a BN layer, and a ReLU activation function. Finally, a network Ki parametric image is generated through a 1×1×1 convolution. The decoding process of the network decoder is expressed as:

[0023]

[0024] Among them, f skip is the skip connection feature of the encoder, indicating channel concatenation.

[0025] In some embodiments, in the step of optimizing the Ki parameter image predicted by the network and using the optimized Ki parameter image as the input for the next iteration, it specifically includes:

[0026] Adopt a weighted combination of mean square error and mean absolute error as the loss function, and the loss function is as follows:

[0027]

[0028] Among them, G(x i ) is the Ki image predicted by the network, y i is the ground truth generated by the Patlak model, and λ1 and λ2 are set to 1.0;

[0029] Optimize the Ki parameter image predicted by the network according to the loss function.

[0030] In some embodiments, in the step of repeating the above steps until the set convergence condition is met, it specifically includes:

[0031] Use the Adam optimizer in conjunction with the ReduceLROnPlateau learning rate scheduler to monitor the validation loss. If the loss has not improved for 4 consecutive epochs, multiply the learning rate by 0.8 until it improves. The initial learning rate of the Adam optimizer is 0.0002, and the momentum parameters are β1 = 0.5 and β2 = 0.999.

[0032] Another object of the present application is to provide a parameter image generation device for shortening the dynamic PET scan time, and the device includes:

[0033] An image acquisition unit for acquiring 3D images of the first 25-minute time frames of a dynamic PET scan;

[0034] A feature extraction unit that uses a network encoder to extract features from the 3D images to generate multi-scale feature maps;

[0035] A parameter acquisition unit for mapping the multi-scale feature maps through a network decoder part to generate a Ki parameter image predicted by the network;

[0036] An image optimization unit for optimizing the Ki parameter image predicted by the network and using the optimized Ki parameter image as the input for the next iteration;

[0037] An iterative unit for repeating the above steps until a set convergence condition is met;

[0038] An image output unit for obtaining a Ki parameter image that meets the requirements.

[0039] A third object of the present application also provides an electronic device, including one or more processors, one or more memories, one or more communication interfaces, and one or more programs; the one or more programs are stored in the memory and are configured to be executed by the one or more processors;

[0040] When the computer program is executed by the processor, it implements the parameter image generation method for shortening the dynamic PET scan time described above.

[0041] A fourth object of the present application also provides a storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, it implements the parameter image generation method for shortening the dynamic PET scan time described in any one of the above.

[0042] The present application adopts the above technical solutions, and the beneficial effects are as follows:

[0043] The parameter image generation method, imaging device, electronic device and storage medium for shortening the dynamic PET scan time provided by the present application obtain 3D images of the first 25-minute time frames of the dynamic PET scan, extract features from the 3D images, generate multi-scale feature maps using the network encoder part, map the multi-scale feature maps through the network decoder part to generate a network-predicted Ki parameter image, optimize the network-predicted Ki parameter image, use the optimized Ki parameter image as the input for the next iteration, repeat the above steps until a set convergence condition is met, and obtain a finally obtained Ki parameter image that meets the requirements. The parameter image generation method for shortening the dynamic PET scan time provided by the present application shortens the dynamic PET scan time from the traditional 60 minutes to within 25 minutes. The quality of the generated Ki parameter image helps with auxiliary diagnosis, not only significantly improving the clinical examination efficiency and patient comfort, but also minimizing the motion artifacts that are prone to occur during dynamic PET scans, thereby ensuring the quality of the generated Ki image and the diagnostic accuracy of doctors; in addition, different from the cross-modal method that needs to combine low-dose CT, the parameter image generation method for shortening the dynamic PET scan time provided by the present application completely relies on dynamic PET data to generate Ki parameters, avoiding additional radiation dose exposure for patients; furthermore, the parameter image generation method for shortening the dynamic PET scan time provided by the present application improves the dynamic metabolic feature modeling ability by capturing the metabolic rate changes of the first 25 frames of dynamic PET data (the time frames cover the early to mid-metabolic stages). Description of the Drawings

[0044] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings required for describing the embodiments of the present application or the prior art. Obviously, the following described accompanying drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0045] Figure 1 It is a flowchart of the steps of a method for generating a parametric image to shorten the dynamic PET scan time provided in Embodiment 1 of the present application.

[0046] Figure 2 It is a schematic diagram of the principle of a method for generating a parametric image to shorten the dynamic PET scan time provided in Embodiment 1 of the present application.

[0047] Figure 3 It is a Ki parametric image predicted by the network provided in Embodiment 1 of the present application.

[0048] Figure 4 It is a Ki parametric image generated after deep learning of a 25-minute dynamic PET image provided in Embodiment 1 of the present application.

[0049] Figure 5 It is a schematic structural diagram of a device for generating a parametric image to shorten the dynamic PET scan time provided in Embodiment 2 of the present application.

[0050] Figure 6 It is a schematic structural diagram of an electronic device provided in Embodiment 3 of the present application. Detailed Embodiments

[0051] The following will describe in detail the embodiments of the present application. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, but should not be construed as limiting the present application.

[0052] In the description of the present application, it should be understood that the orientation or positional relationship indicated by the terms "upper", "lower", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the present application.

[0053] In addition, the terms "first" and "second" are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this application, "a plurality of" means two or more, unless otherwise specifically defined.

[0054] In order to make the objectives, technical solutions and advantages of this application more clear and understandable, the following further details this application in conjunction with the accompanying drawings and embodiments.

[0055] Embodiment 1

[0056] Please refer to Figure 1 and Figure 2 , which is the flowchart of the steps of the method for generating parametric images for shortening the dynamic PET scan time provided in Embodiment 1 of this application, including the following steps S110 to S160. The following details the implemented technical solution.

[0057] Step S110: Obtain a 3D image of the first 25-minute time frame of the dynamic PET scan.

[0058] In this embodiment, the dynamic PET scan time is compressed from 60 minutes to within 25 minutes to generate a high-quality Ki image that meets the clinical diagnosis requirements. This not only significantly improves the clinical examination efficiency and patient comfort, but also minimizes the motion artifacts that are prone to occur during the dynamic PET scan, thereby ensuring the quality of the generated Ki image and the diagnostic accuracy of the doctor.

[0059] Step S120: Use a network encoder to extract features from the 3D image to generate a multi-scale feature map.

[0060] In this embodiment, the network encoder adopts an encoder with a single-channel 3D U-Net architecture. The encoder consists of multiple 3D convolutional layers and residual modules. Each residual module contains two 3×3×3 convolutional layers, with batch normalization (BN) and ReLU activation functions inserted in the middle. Downsampling is achieved through a convolutional layer with a stride (stride = 2). The spatial dimension of the feature map is halved, but the depth dimension remains D = 3 for the input slice group. The number of channels of all convolutional layers is uniformly set to 20 to save memory. The input of the encoder is a 3D image block (256×256×3) of the dynamic PET time frame, and the output is a multi-scale feature map, specifically expressed as:

[0061] f j+1 = Conv 3D(f j )

[0062] where f jis the feature map of the j-th layer, and high-level features are gradually extracted through the stacked residual module.

[0063] It should be noted that since the input is dynamic PET data of consecutive time frames (the first 25 frames), in this embodiment, the spatio-temporal feature alignment is achieved through the grouped slicing strategy. Each time frame contains 71 slices. To reduce the computational load, three adjacent slices are grouped as a set (D = 3) to form an input block of 256×256×3. During the encoding process, the feature maps of different levels are passed to the subsequent network decoder through skip connections to ensure the alignment and fusion of time and space information. This process implicitly realizes the feature consistency constraint across time frames.

[0064] Step S130: Map the multi-scale feature maps through the network decoder part to generate the Ki parameter image predicted by the network.

[0065] In this embodiment, the network decoder is composed of deconvolution layers and skip connections. Each upsampling stage uses a 3×3×3 deconvolution kernel (stride = 2) to expand the spatial dimension and performs channel splicing with the feature maps of the corresponding levels of the encoder. Each decoding module contains a deconvolution layer, a BN layer, and a ReLU activation function, and finally generates the Ki parameter image (256×256×3) through a 1×1×1 convolution. The decoding process is expressed as:

[0066]

[0067] where f skip is the skip connection feature of the encoder, represents channel splicing.

[0068] It can be understood that in this embodiment, mapping is performed through the network decoder part, and affine transformation is used to achieve the spatial alignment of multi-time-frame metabolic information, solving the problem of insufficient temporal resolution of short-time scan data.

[0069] Step S140: Optimize the Ki parameter image predicted by the network, and use the optimized Ki parameter image as the input for the next iteration.

[0070] In this embodiment, a weighted combination of mean squared error and mean absolute error is used as the loss function, and the loss function is as follows:

[0071]

[0072] where G(x i ) is the Ki image predicted by the network, y i is the ground truth generated by the Patlak model, and λ1 and λ2 are set to 1.0;

[0073] Optimize the Ki parameter image predicted by the network according to the loss function.

[0074] Step S150: Repeat the above steps until the set convergence condition is met.

[0075] In this embodiment, in the step of repeating the above steps until the set convergence condition is met, it specifically includes: using the Adam optimizer in cooperation with the ReduceLROnPlateau learning rate scheduler to monitor the validation loss, multiplying the learning rate by 0.8 if it has not improved for 4 consecutive epochs until it improves. The initial learning rate of the Adam optimizer is 0.0002, and the momentum parameters are β_1 = 0.5 and β_2 = 0.999.

[0076] Step S160: Obtain the Ki parameter image that meets the requirements.

[0077] In this embodiment, the deep learning model directly maps the dynamic PET data within 25 minutes to the Ki image, without relying on the 60-minute complete time-activity curve (TAC) required by the traditional Patlak model.

[0078] Please refer to Figure 3 and Figure 4 , which are the Ki parameter image predicted by the network provided in Embodiment 1 of the present application and the Ki parameter image generated after deep learning of the 25-minute dynamic PET image respectively.

[0079] In this embodiment, the training batch size is 8, with a total of 100 epochs. The data set is divided into a training set, a validation set, and a test set according to 70:15:15. The above method for generating the parameter image for shortening the dynamic PET scanning time is used to input the dynamic PET images of the first 20 time frames (each time frame is 256×256×71, and 3 adjacent slices are taken as 256×256×3), and the required parameter Ki image (256×256×3, supervised by the Patlak model true value) is output. After training, the model can directly generate the Ki image from the dynamic SUV image, evaluate the performance on an independent test set (15 patients, 1065 slices), and compare with baseline models such as 3D / 2DU-Net and CycleGAN. The results show the advantages of the present application in terms of computational efficiency and accuracy.

[0080] The parametric image generation method for shortening the dynamic PET scan time provided by this application shortens the dynamic PET scan time from the traditional 60 minutes to within 25 minutes. The quality of the generated Ki parametric image helps with auxiliary diagnosis, not only significantly improving the clinical examination efficiency and patient comfort, but also minimizing the motion artifacts that are prone to occur during dynamic PET scans as much as possible, thereby ensuring the quality of the generated Ki image and the diagnostic accuracy of doctors. In addition, different from the cross-modal method that requires the combination of low-dose CT, the parametric image generation method for shortening the dynamic PET scan time provided by this application completely relies on dynamic PET data to generate Ki parameters, avoiding the patient from additionally receiving radiation dose. Moreover, the parametric image generation method for shortening the dynamic PET scan time provided by this application improves the dynamic metabolic feature modeling ability by capturing the metabolic rate changes of the first 25 frames of dynamic PET data (the time frames cover the early to mid-metabolic stages) through an encoder.

[0081] Embodiment 2

[0082] Please refer to Figure 5 , which is the parametric image generation device for shortening the dynamic PET scan time provided by this embodiment. The device includes:

[0083] An image acquisition unit 110, configured to acquire 3D images of the first 25-minute time frames of a dynamic PET scan.

[0084] In this embodiment, the dynamic PET scan time is compressed from 60 minutes to within 25 minutes to generate high-quality Ki images that meet the clinical diagnosis requirements, not only significantly improving the clinical examination efficiency and patient comfort, but also minimizing the motion artifacts that are prone to occur during dynamic PET scans as much as possible, thereby ensuring the quality of the generated Ki image and the diagnostic accuracy of doctors.

[0085] A feature extraction unit 120, which uses a network encoder to extract features from the 3D images to generate multi-scale feature maps.

[0086] In this embodiment, the network encoder adopts an encoder with a single-channel 3D U-Net architecture. The encoder consists of multiple 3D convolutional layers and residual modules. Each residual module contains two 3×3×3 convolutional layers, with batch normalization (BN) and ReLU activation functions inserted in the middle. Downsampling is achieved through convolutional layers with a stride (stride = 2), and the spatial size of the feature map is halved, but the depth dimension remains D = 3 for the input slice group. The number of channels of all convolutional layers is uniformly set to 20 to save memory. The input of the encoder is a 3D image block (256×256×3) of dynamic PET time frames, and the output is a multi-scale feature map, specifically expressed as:

[0087] f j+1 = Conv 3D(f j )

[0088] Among them, f j is the feature map of the j-th layer, and high-level features are gradually extracted through the stacked residual module.

[0089] It should be noted that: since the input is dynamic PET data of consecutive time frames (the first 25 frames), in this embodiment, the spatio-temporal feature alignment is achieved through the grouped slicing strategy. Each time frame contains 71 slices. To reduce the computational amount, adjacent 3 slices are taken as a group (D = 3) to form an input block of 256×256×3. During the encoding process, the feature maps of different levels are passed to the subsequent network decoder through skip connections to ensure the alignment and fusion of time and space information. This process implicitly realizes the feature consistency constraint across time frames.

[0090] The parameter acquisition unit 130 is used to map the multi-scale feature map through the network decoder part to generate the Ki parameter image predicted by the network.

[0091] In this embodiment, the network decoder is composed of a transposed convolution layer and skip connections. Each upsampling stage uses a 3×3×3 transposed convolution kernel (stride = 2) to expand the spatial dimension, and channel splicing is performed in combination with the feature map of the corresponding level of the encoder. Each decoding module contains a transposed convolution layer, a BN layer, and a ReLU activation function, and finally generates the Ki parameter image (256×256×3) through a 1×1×1 convolution. The decoding process is expressed as:

[0092]

[0093] Among them, f skip is the skip connection feature of the encoder, represents channel splicing.

[0094] It can be understood that in this embodiment, mapping is performed through the network decoder part, and the spatial alignment of the metabolic information of multiple time frames is achieved by using affine transformation, so as to solve the problem of insufficient time resolution of short-time scan data.

[0095] The image optimization unit 140 is used to perform optimization processing on the Ki parameter image predicted by the network, and use the optimized Ki parameter image as the input for the next iteration.

[0096] In this embodiment, a weighted combination of mean square error and mean absolute error is used as the loss function, and the loss function is as follows:

[0097]

[0098] Among them, G(x i ) is the Ki image predicted by the network, and y iThe ground truth generated for the Patlak model, with λ1 and λ2 set to 1.0;

[0099] Optimize the Ki parameter image predicted by the network according to the loss function.

[0100] Iteration unit 150 is used to repeat the above steps until the set convergence condition is met.

[0101] In this embodiment, in the step of repeating the above steps until the set convergence condition is met, it specifically includes: using the Adam optimizer in conjunction with the ReduceLROnPlateau learning rate scheduler to monitor the validation loss. If it has not improved for 4 consecutive epochs, the learning rate is multiplied by 0.8 until improvement. The initial learning rate of the Adam optimizer is 0.0002, and the momentum parameters are β_1 = 0.5 and β_2 = 0.999.

[0102] Image output unit 160 is used to obtain the Ki parameter image that meets the requirements.

[0103] In this embodiment, the deep learning model directly maps the dynamic PET data within 25 minutes to the Ki image without relying on the 60 - minute complete time - activity curve (TAC) required by the traditional Patlak model.

[0104] The parameter image generation device for shortening the dynamic PET scan time provided by this application shortens the dynamic PET scan time from the traditional 60 minutes to within 25 minutes. The quality of the generated Ki parameter image helps with auxiliary diagnosis, not only significantly improving the clinical examination efficiency and patient comfort, but also minimizing the motion artifacts that are prone to occur during dynamic PET scanning, thereby ensuring the quality of the generated Ki image and the doctor's diagnostic accuracy. In addition, different from the cross - modal method that requires combining low - dose CT, the parameter image generation method for shortening the dynamic PET scan time provided by this application completely relies on dynamic PET data to generate Ki parameters, avoiding the patient from additionally receiving radiation dose. Moreover, the parameter image generation method for shortening the dynamic PET scan time provided by this application improves the dynamic metabolic feature modeling ability by using the encoder to capture the metabolic rate changes of the first 25 frames of dynamic PET data (the time frames cover the early to mid - metabolic stages).

[0105] Embodiment 3

[0106] Please refer to Figure 6 , Figure 6 which is a schematic structural diagram of an electronic device provided by an embodiment of this application. The electronic device includes: one or more processors, one or more memories, one or more communication interfaces, and one or more programs; the one or more programs are stored in the memory and are configured to be executed by the one or more processors.

[0107] The above program includes instructions for performing the following steps:

[0108] Obtain a 3D image of the first 25-minute time frame of a dynamic PET scan;

[0109] Use a network encoder to extract features from the 3D image to generate a multi-scale feature map;

[0110] Map the multi-scale feature map through a network decoder part to generate a Ki parameter image predicted by the network;

[0111] Perform optimization processing on the Ki parameter image predicted by the network, and use the optimized Ki parameter image as the input for the next iteration;

[0112] Repeat the above steps until the set convergence condition is met;

[0113] Obtain a Ki parameter image that meets the requirements.

[0114] Among them, all relevant contents of each scenario involved in the above method embodiments can be cited in the function descriptions of the corresponding functional modules, and will not be elaborated here.

[0115] It should be understood that the above memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0116] In the embodiments of the present application, the processor of the above device may be a central processing unit (CPU), and this processor may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or this processor may also be any conventional processor, etc.

[0117] It should be understood that the "at least one" involved in the embodiments of the present application refers to one or more, and the "multiple" refers to two or more. "And / or" describes the association relationship of associated objects and indicates that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one (item)" or its similar expression refers to any combination of these items, including any combination of single item (item) or plural items (items). For example, at least one (item) of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0118] In addition, unless otherwise stated, the ordinal numbers such as "first" and "second" mentioned in the embodiments of the present application are used to distinguish multiple objects and are not used to limit the order, time sequence, priority, or importance of multiple objects. For example, the first information and the second information are only used to distinguish different information, rather than indicating differences in the content, priority, sending order, or importance of these two types of information.

[0119] In the implementation process, each step of the above method can be completed by the integrated logic circuit in the hardware of the processor or the instructions in the form of software. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as being executed and completed by the hardware processor, or executed and completed by the combination of the hardware and software units in the processor. The software unit can be located in a mature storage medium in the art such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, register, etc. This storage medium is located in the memory, and the processor executes the instructions in the memory and combines its hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0120] The embodiments of the present application also provide a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, and the computer program enables the computer to execute some or all of the steps of any method recorded in the above method embodiments.

[0121] The embodiments of the present application also provide a computer program product. The above computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the above computer program is operable to enable the computer to execute some or all of the steps of any method recorded in the above method embodiments. The computer program product can be a software installation package.

[0122] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0123] In the above embodiments, the descriptions of the respective embodiments have their own focuses. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0124] In the several embodiments provided by this application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the above division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical or other form.

[0125] The units described as separate components above may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments of this application.

[0126] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0127] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or TRP, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0128] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing the relevant hardware through a program, and the program may be stored in a computer-readable memory, which may include a flash drive, ROM, RAM, a magnetic disk or an optical disk, etc.

[0129] The above are only preferred embodiments of the present application, and only specifically describe the technical principles of the present application. These descriptions are only for explaining the principles of the present application and cannot be interpreted as limiting the scope of protection of the present application in any way. Based on the explanation here, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application, and other specific implementation methods of the present application that can be associated with the technicians in this field without creative work, should be included in the scope of protection of the present application.

Claims

1. A method for generating parametric images to shorten the dynamic PET scan time, characterized in that The method includes the following steps: Obtain 3D images of the first 25-minute time frame of a dynamic PET scan; Use a network encoder to extract features from the 3D images to generate multi-scale feature maps; Map the multi-scale feature maps through a network decoder part to generate a network-predicted Ki parameter image; Perform optimization processing on the network-predicted Ki parameter image and use the optimized Ki parameter image as the input for the next iteration; Repeat the above steps until the set convergence condition is met; Obtain a Ki parameter image that meets the requirements.

2. The parameter image generation method for shortening the dynamic PET scan time according to claim 1, wherein, In the step of using a network encoder to extract features from the 3D images to generate multi-scale feature maps, it specifically includes: The network encoder is an encoder with a single-channel 3D U-Net architecture. The encoder consists of multiple 3D convolutional layers and residual modules. Each residual module contains two 3×3×3 convolutional layers, with batch normalization BN and ReLU activation functions inserted in the middle. Downsampling is achieved through a convolutional layer with a stride of 2. The input slice group D is 3, and the number of channels of all convolutional layers is uniformly set to 20, which is specifically expressed as: f j+1 = Conv 3D(f j ) Among them, f j is the feature map of the j-th layer, and high-level features are gradually extracted through the stacked residual module.

3. The method for generating a parametric image for shortening the dynamic PET scan time according to claim 1, wherein In the step of mapping the multi-scale feature maps through a network decoder part to generate a network-predicted Ki parameter image, it specifically includes: The network decoder consists of deconvolution layers and skip connections. Each upsampling stage uses a 3×3×3 deconvolution kernel to expand the spatial dimension and combines the feature maps of the corresponding levels of the encoder for channel splicing. Each decoding module contains a deconvolution layer, a BN layer, and a ReLU activation function. Finally, a network Ki parameter image is generated through a 1×1×1 convolution. The decoding process of the network decoder is expressed as: Among them, f skip is the skip connection feature of the encoder, represents channel concatenation.

4. The method for generating a parametric image for shortening the dynamic PET scan time according to claim 1, characterized in that, In the step of performing optimization processing on the network-predicted Ki parameter image and using the optimized Ki parameter image as the input for the next iteration, it specifically includes: Adopt a weighted combination of mean squared error and mean absolute error as the loss function. The loss function is as follows: Among them, G(x i ) is the Ki image predicted by the network, y i is the true value generated by the Patlak model, and λ1 and λ2 are set to 1.0; Optimize the network-predicted Ki parameter image according to the loss function.

5. The method for generating a parametric image for shortening the dynamic PET scan time according to claim 4, wherein, In the step of repeating the above steps until the set convergence condition is met, it specifically includes: Use an Adam optimizer in conjunction with a ReduceLROnPlateau learning rate scheduler to monitor the validation loss. If it has not improved for 4 consecutive epochs, multiply the learning rate by 0.8 until it improves. The initial learning rate of the Adam optimizer is 0.0002, and the momentum parameters are β1 = 0.5 and β2 = 0.

999.

6. A parametric image generation device for shortening the dynamic PET scan time, characterized in that, The device includes: An image acquisition unit for obtaining 3D images of the first 25-minute time frame of a dynamic PET scan; A feature extraction unit for using a network encoder to extract features from the 3D images to generate multi-scale feature maps; A parameter acquisition unit for mapping the multi-scale feature maps through a network decoder part to generate a network-predicted Ki parameter image; An image optimization unit for performing optimization processing on the network-predicted Ki parameter image and using the optimized Ki parameter image as the input for the next iteration; An iteration unit for repeating the above steps until the set convergence condition is met; An image output unit for obtaining a Ki parameter image that meets the requirements.

7. An electronic device, characterized in that, It includes at least one processor, at least one memory, and at least one communication bus. Among them, a computer program is stored on the memory, and the processor reads the computer program in the memory through the communication bus; When the computer program is executed by the processor, it implements the parameter image generation method for shortening the dynamic PET scan time according to any one of claims 1 to 5.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the parameter image generation method for shortening the dynamic PET scan time according to any one of claims 1 to 5.

Images