ROI image reconstruction method and system based on SPECT data

Abstract

The invention relates to the technical field of nuclear medicine diagnosis, and particularly provides an ROI image reconstruction method and system based on SPECT data, and the method comprises the steps: reconstructing a coarse scanning image according to large-view coarse scanning projection data; generating ROI image estimation according to the coarse scanning image; performing loop iteration updating on the ROI image estimation until a preset convergence condition is met; the loop iteration updating process comprises the following steps: generating estimated projection data according to ROI image estimation of current iteration; obtaining projection deviation according to the estimated projection data and the small-view fine scanning projection data; obtaining an image state index according to ROI image estimation of the current iteration and the previous iteration; determining a coarse scanning data contribution weight according to the image state index; updating the ROI image estimation of the current iteration according to the projection deviation, the small-view fine scanning projection data, the coarse scanning data contribution weight and the large-view coarse scanning projection data; according to the method, the iterative reconstruction process can adapt to the dynamic change of the image state.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nuclear medicine diagnosis, in particular to a ROI image reconstruction method and system based on SPECT data. BACKGROUND

[0002] In the field of nuclear medicine diagnosis, single photon emission computed tomography (SPECT) is widely used to obtain functional metabolic images, especially in the accurate diagnosis of specific organs such as the heart or tumors. Since large field of view coarse scanning has the advantage of low dose and the disadvantage of low data resolution and inability to provide ROI detail information, while small field of view fine scanning has the advantages of high data resolution and the ability to provide ROI detail information and the disadvantage of high dose, the prior art combines the modes of large field of view coarse scanning and small field of view fine scanning to achieve the control of radioactive dose and the shortening of scanning time while improving the resolution and signal-to-noise ratio of the target region image. The development trend in this field focuses on optimizing the data fusion strategy to reconstruct clearer images to meet the needs of clinical diagnosis.

[0003] However, the prior art uses a fixed weight factor for iterative reconstruction, and it is difficult for the fixed weight factor to adapt to the dynamic changes of the image state during the iterative reconstruction process. Specifically, at the beginning of the iteration, the fixed weight factor can effectively suppress macroscopic artifacts, but as the iteration deepens, the focus of the iterative reconstruction shifts from image quality improvement to detail recovery, and the fixed weight factor will cause the low resolution characteristics of the coarse scanning data to excessively smooth the internal details of the target region, such as small perfusion defect regions being blurred, while introducing statistical noise, reducing the signal-to-noise ratio, and damaging the clarity and quantitative accuracy of the image. As can be seen, the prior art has the problem of being unable to achieve the best balance between artifact suppression and detail fidelity due to the fixed weight factor being difficult to simultaneously meet the effective suppression of truncation artifacts at the beginning of the iteration and the high fidelity requirement of ROI internal fine structure recovery and radioactive count value accuracy after the iteration, thereby leading to a decrease in the reliability of the diagnostic results.

[0004] At present, there is no effective technical solution to the above problems. It should be noted that the above information disclosed in this part is only for understanding the background of the inventive concept, and therefore can contain information that does not constitute the prior art. SUMMARY

[0005] The purpose of the present application is to provide a ROI image reconstruction method and system based on SPECT data, which can adaptively adjust the contribution of large field of view coarse scanning data to the estimation of ROI image during the iterative reconstruction process.

[0006] In a first aspect, the present application provides a ROI image reconstruction method based on SPECT data, comprising the following steps: S1, acquire large field of view coarse scanning projection data, and acquire small field of view fine scanning projection data based on ROI; S2, reconstruct a coarse scanning image according to the large field of view coarse scanning projection data; S3, generate an initial ROI image estimate according to the coarse scanning image; S4, cyclically and iteratively update the ROI image estimate until a preset convergence condition is met; The process of cyclically and iteratively updating the ROI image estimate comprises: A1, generate estimated projection data according to the ROI image estimate of the current iteration; A2, acquire projection deviation according to the estimated projection data and the small field of view fine scanning projection data; A3, acquire an image state index according to the ROI image estimate of the current iteration and the ROI image estimate of the previous iteration; A4, determine a coarse scanning data contribution weight according to the image state index; A5, update the ROI image estimate of the current iteration according to the projection deviation, the small field of view fine scanning projection data, the coarse scanning data contribution weight, and the large field of view coarse scanning projection data.

[0007] In a second aspect, the present application further provides a ROI image reconstruction system based on SPECT data, comprising: A data acquisition module, configured to acquire large field of view coarse scanning projection data, and acquire small field of view fine scanning projection data based on ROI; A coarse scanning image reconstruction module, configured to reconstruct a coarse scanning image according to the large field of view coarse scanning projection data; An ROI image estimate generation module, configured to generate an initial ROI image estimate according to the coarse scanning image; An iteration update module, configured to cyclically and iteratively update the ROI image estimate until a preset convergence condition is met; The process of cyclically and iteratively updating the ROI image estimate comprises: A1, generate estimated projection data according to the ROI image estimate of the current iteration; A2, acquire projection deviation according to the estimated projection data and the small field of view fine scanning projection data; A3, acquire an image state index according to the ROI image estimate of the current iteration and the ROI image estimate of the previous iteration; A4, determine a coarse scanning data contribution weight according to the image state index; A5, update the ROI image estimate of the current iteration according to the projection deviation, the small field of view fine scanning projection data, the coarse scanning data contribution weight, and the large field of view coarse scanning projection data.

[0008] From the above, the ROI image reconstruction method and system based on SPECT data provided by the application can realize self-adaptive adjustment of the contribution degree of the large field of view rough scanning data to the ROI image estimation in the iterative reconstruction process by introducing an image state index and dynamically determining the rough scanning data contribution weight according to the index, that is, the application can make the iterative reconstruction process adapt to the dynamic change of the image state, so as to effectively suppress the truncation artifact in the early stage of iteration and restore the fine structure inside the ROI in high fidelity in the later stage of iteration. Therefore, the application can effectively solve the problem that the fixed weight factor cannot simultaneously meet the effective suppression requirement of the truncation artifact in the early stage of iteration and the high-fidelity requirement of the fine structure inside the ROI and the accuracy of the radioactivity count value in the later stage of iteration, so as to effectively improve the reliability of the diagnostic result. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 A flowchart of the ROI image reconstruction method based on SPECT data provided by the embodiment of the application.

[0010] Figure 2 A structure schematic diagram of the ROI image reconstruction system based on SPECT data provided by the embodiment of the application.

[0011] The drawings show that: 1, data acquisition module; 2, rough scanning image reconstruction module; 3, ROI image estimation generation module; 4, iterative update module. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. The components of the embodiments of the application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the application.

[0013] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the application, the terms "first", "second" and the like are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0014] In a first aspect, as Figure 1As shown, this application provides a method for ROI image reconstruction based on SPECT data, which includes the following steps: S1. Obtain large field-of-view coarse scan projection data and obtain small field-of-view fine scan projection data based on ROI; S2. Reconstruct the coarse scan image based on the large field-of-view coarse scan projection data; S3. Generate initial ROI image estimates based on the coarse scan image; S4. Iterate and update the ROI image estimation until the preset convergence condition is met. The process of iteratively updating the ROI image estimation includes: A1. Generate estimated projection data based on the ROI image of the current iteration; A2. Obtain the projection deviation based on the estimated projection data and the small field-of-view fine scan projection data; A3. Obtain image state indicators based on the ROI image estimation of the current iteration and the ROI image estimation of the previous iteration; A4. Determine the contribution weight of coarse scan data based on image status indicators; A5. Update the ROI image estimation for the current iteration based on projection bias, contribution weights of small field-of-view fine scan projection data and coarse scan data, and large field-of-view coarse scan projection data.

[0015] Acquiring large-field-of-view (FOP) coarse-scan projection data refers to performing a low-resolution scan of the entire target area using a SPECT scanner to obtain FOP projection data containing background information and overall structure. This embodiment can achieve FOP coarse-scan projection data acquisition using low-dose, short-duration, or sparse angle sampling methods. For example, using a low dose of radioactive tracer for a whole-body scan, or completing a coarse projection data acquisition in a short time. Acquiring small-field-of-view (FOV) fine-scan projection data refers to performing a high-resolution scan of the region of interest (ROI) to obtain fine structure and high signal-to-noise ratio projection data within the ROI. This embodiment can achieve FAV fine-scan projection data acquisition using high-dose, long-duration, or dense angle sampling methods. For example, increasing the local concentration of radioactive tracer within the ROI, extending the scan time in that region to obtain more photon counts, or performing denser angle sampling around the ROI. The acquisition of the two types of projection data can be performed in parallel or sequentially, and they together provide complementary information for subsequent image reconstruction: FOP coarse-scan projection data is used for global background and artifact suppression, while FAV fine-scan projection data is used for ROI detail restoration.

[0016] Reconstructing a coarse scan image refers to generating a low-resolution image covering the entire scan area using image reconstruction algorithms based on large-field-of-view coarse scan projection data. This coarse scan image provides global contextual information for subsequent ROI localization and initial estimation, which helps suppress artifacts caused by truncation of small-field-of-view fine scan data in the early stages of iteration. This embodiment can use filtered back projection (FBP) algorithms, iterative reconstruction algorithms (such as OSEM, MLEM), or deep learning-based reconstruction methods to reconstruct the coarse scan image. For example, the FBP algorithm can be used to quickly generate a preliminary coarse scan image based on large-field-of-view coarse scan projection data, or the OSEM algorithm with a small number of iterations can be used to generate a coarse scan image with a certain degree of smoothness based on large-field-of-view coarse scan projection data.

[0017] Generating an initial ROI image estimate refers to extracting or inferring a preliminary image of the region of interest from a coarse-scanned image. This embodiment can achieve the initial ROI image estimate using methods such as threshold segmentation, region growing, morphological operations, or template matching based on prior knowledge. Alternatively, this embodiment can obtain the initial ROI image estimate by directly cropping or interpolating the portion of the coarse-scanned image corresponding to the small field-of-view fine-scanned region to the ROI size. Since the initial ROI image estimate is the starting point for subsequent iterative updates, this embodiment is equivalent to using the global information of the coarse-scanned image to initially define and estimate the shape and location of the ROI, laying the foundation for fine-grained reconstruction and avoiding blind iteration from scratch, thereby effectively improving convergence efficiency.

[0018] Iterative updates refer to the process of gradually optimizing the ROI image estimate by repeatedly executing a series of computational steps to make it gradually approach the true ROI image. This embodiment can use iterative algorithms based on the expectation-maximization principle (such as maximum likelihood expectation-maximization (MLEM) or ordered subset expectation-maximization (OSEM)) or gradient descent-based optimization algorithms to implement iterative updates. A preset convergence condition refers to the criterion for determining whether the iteration process has ended. In this embodiment, the preset convergence condition can be that the change in the ROI image estimate in several consecutive iterations is less than a preset threshold, the image's signal-to-noise ratio and contrast reach preset values, or the number of iterations reaches a preset maximum number of iterations. For example, the preset convergence condition can be set to stop when the mean square error (MSE) of the images in two adjacent iterations is less than 0.001, or when the number of iterations reaches 50.

[0019] Generating estimated projection data refers to simulating the projection data generated during SPECT scanning based on the ROI image of the current iteration using a forward projection operation. This embodiment can employ ray tracing algorithms, Fourier transform-based methods, or GPU-accelerated parallel computing to generate estimated projection data. For example, it can simulate the attenuation, scattering, and detector response of X-rays in image voxels, mapping the intensity values ​​in the image domain to the projection domain. This embodiment can map the features estimated by the ROI image of the current iteration to the projection space by generating estimated projection data, so as to compare it with the actually acquired projection data and provide a basis for subsequent error calculation.

[0020] The projection bias refers to the difference between the calculated estimated projection data and the actual small field-of-view fine scan projection data. This embodiment can use various error measurement methods to obtain the projection bias. For example, it can calculate the Euclidean distance, mean square error (MSE), relative entropy, or Poisson likelihood function difference between the estimated projection data and the small field-of-view fine scan projection data. The projection bias reflects the inconsistency between the current ROI image estimate and the real ROI in the projection domain. The projection bias is the main error signal that drives iterative updates and image corrections to ensure the consistency between the reconstruction result and the high-resolution fine scan data.

[0021] Obtaining image state metrics refers to dynamically quantifying the changing trend, detail recovery degree, or noise level of an image during the iterative process by comparing the ROI image estimates of the current iteration with those of the previous iteration. This embodiment can obtain image state metrics by calculating the structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), image gradient change, texture features, or local variance between the current and previous ROI image estimates. This embodiment can also obtain image state metrics by calculating the local gradient magnitude change to reflect the detail recovery and by calculating the image smoothness index to reflect the noise level.

[0022] Determining the contribution weight of coarse scan data refers to dynamically adjusting the influence of large-field-of-view coarse scan data in the image update process based on the acquired image state indicators. This embodiment can use methods such as lookup tables, piecewise functions, fuzzy logic, or neural networks to determine the contribution weight of coarse scan data. For example, the lookup table is set as follows: when the image state indicators show that there are many image artifacts or blurred details, the coarse scan data is given a higher weight to utilize its global information to suppress artifacts; when the image state indicators show that the image details are gradually becoming clearer and noise begins to appear, the weight of the coarse scan data is reduced to avoid its low-resolution characteristics from excessively smoothing details and to allow the fine scan data to take advantage of its high resolution. This embodiment can solve the key problem of fixed weight factors in the prior art by dynamically determining the contribution weight of coarse scan data based on image state indicators, so that the system can adaptively adjust the reconstruction strategy according to the real-time state of the image, thereby achieving the best balance between artifact suppression and detail preservation.

[0023] Updating the current iteration's ROI image estimate refers to the process of adjusting the current ROI image estimate based on projection bias, the contribution weight of coarse scan data, and large-field-of-view coarse scan projection data. This embodiment can employ a modified iterative reconstruction algorithm to update the current iteration's ROI image estimate. For example, a weighting factor can be introduced into the update formula of OSEM or MLEM, which adjusts the contribution of coarse scan data to image updates based on its contribution weight. This embodiment can progressively optimize the ROI image by adaptively fusing large-field-of-view coarse scan projection data and small-field-of-view fine scan projection data according to projection bias and coarse scan data contribution weights, thereby improving the clarity and accuracy of local details while maintaining overall consistency.

[0024] The core innovation of this application lies in introducing an image state index and dynamically determining the contribution weight of coarse scan data based on this index to adaptively adjust the contribution of large-field-of-view coarse scan data to ROI image estimation during iterative reconstruction. In other words, this application enables the iterative reconstruction process to adapt to the dynamic changes in image state, effectively suppressing truncation artifacts in the early stage of iteration and restoring the fine structure inside the ROI with high fidelity in the later stage of iteration. Therefore, this application can effectively solve the problem that it is difficult to achieve the optimal balance between artifact suppression and detail fidelity because fixed weight factors cannot simultaneously meet the requirements of effective suppression of truncation artifacts in the early stage of iteration and the high fidelity requirements of restoration of the fine structure inside the ROI and accuracy of radioactivity count values ​​in the later stage of iteration. This effectively improves the reliability of diagnostic results.

[0025] Specifically, this method first acquires large-field-of-view coarse-scan projection data and small-field-of-view fine-scan projection data. The large-field-of-view coarse-scan data provides global background information, while the small-field-of-view fine-scan data provides high-resolution details of the ROI region. Next, a coarse-scan image is reconstructed based on the large-field-of-view coarse-scan projection data to generate a preliminary image encompassing the entire scanned area, leveraging the global nature of the coarse-scan data. Then, an initial ROI image estimate is generated based on the coarse-scan image to reduce artifacts in the initial stage and provide a starting point for focusing subsequent fine-iterative reconstruction. Subsequently, the ROI image estimate is iteratively updated until a preset convergence condition is met. During the iterative update process, estimated projection data is first generated by simulating forward projection of the current iteration's ROI image estimate to produce theoretical estimated projection data. Secondly, the projection deviation is obtained by comparing the simulated estimated projection data with the actually acquired small-field-of-view fine-scan projection data to quantify the difference between the current image estimate and the high-resolution real data. This projection deviation serves as the error signal driving the iterative update. Next, image state indicators are obtained by analyzing the changes between the current iteration's ROI image estimate and the previous iteration's ROI image estimate, thus providing an indicator that reflects the current state of the ROI image estimate. Then, the contribution weight of coarse scan data (the degree of contribution of large-field-of-view coarse scan data in the current iteration) is determined based on the image state indicators. For example, when the image state indicators show a large change in the image, the contribution weight of coarse scan data is appropriately increased to enhance its influence on image updates; conversely, when the image state indicators show a small change in the image, the contribution weight of coarse scan data is appropriately decreased to reduce its influence on image updates. Finally, the current iteration's ROI image estimate is updated based on projection bias, coarse scan data contribution weights, and large-field-of-view coarse scan projection data. This involves weighted fusion of accurate error information from small-field-of-view fine scan data and global context information from large-field-of-view coarse scan data, based on dynamically determined coarse scan data contribution weights. This ensures that the ROI image can effectively utilize high-resolution fine scan data for detail restoration during iteration, while also suppressing artifacts and maintaining global consistency by adaptively introducing coarse scan data.

[0026] As a preferred embodiment, the solution of this application is implemented as follows: In the data acquisition stage, this embodiment uses a dual-probe SPECT system to simultaneously acquire large-field-of-view coarse scanning projection data and high-resolution small-field-of-view fine scanning projection data. During coarse scanning image reconstruction, this embodiment employs a filtered back-projection (FBP) algorithm or an iterative reconstruction algorithm such as the expectation-maximization (EM) algorithm. The generation of the initial ROI image estimate includes cropping the region in the coarse scanning image corresponding to the fine scanning field of view and normalizing its intensity. During the iterative update process, the generation of the estimated projection data can utilize a ray tracing model to simulate the attenuation and scattering effects of rays in the tissue. The projection deviation can be obtained by pixel-by-pixel subtraction between the estimated projection data and the fine scanning projection data. The acquisition of image state indicators can specifically involve calculating the L2 norm of the pixel value difference between the current iteration's ROI image estimate and the previous iteration's ROI image estimate, or calculating the local gradient change rate of the image. The determination of the contribution weight of coarse scan data can be based on a preset nonlinear function that maps image state indicators to a dynamically changing weight value. For example, when the image state indicator indicates the presence of macroscopic artifacts, the coarse scan data is given a higher weight; when the image state indicator indicates stable image details, the weight of the coarse scan data is reduced. The final ROI image estimation update can employ a modified iterative reconstruction algorithm. For instance, in the update formula, the contribution of the fine scan data remains constant, while the contribution of the coarse scan data dynamically changes according to its contribution weight. This method optimizes the reconstructed image at different iteration stages by dynamically adjusting the contribution of the coarse scan data during the iteration process, thereby improving image clarity, signal-to-noise ratio, and quantitative accuracy, and ultimately enhancing the reliability of the diagnostic results.

[0027] In some preferred embodiments, step A3 includes: A31. Obtain the gradient map of the ROI image estimation in the current iteration; A32. Using a clustering algorithm, the ROI image of the current iteration is estimated to be divided into multiple gradient regions based on the gradient map; A33. For each gradient region, obtain the image change information of the ROI image estimate in the current iteration and the ROI image estimate in the previous iteration for that gradient region, and then perform weighted processing on the image change information according to the gradient features corresponding to that gradient region. A34. Determine the image status index based on all weighted image change information.

[0028] A gradient map represents the rate of change of pixel intensity in an image. It reflects the edge, texture, and detail information of the estimated Region of Interest (ROI). This embodiment can obtain the gradient map by using edge detection algorithms such as the Sobel operator, Prewitt operator, Roberts operator, or Canny operator to calculate the gradient information of the ROI image. A clustering algorithm is an unsupervised learning method used to divide objects in a dataset into different groups or clusters based on their similarity. This clustering algorithm can be K-means, DBSCAN, hierarchical clustering, or Gaussian mixture models. A gradient region refers to a sub-region of the image with similar gradient characteristics, defined based on the image gradient features. Image change information refers to the difference in pixel values ​​or features between the current iteration image and the previous iteration image. This image change information can be indices such as pixel difference, mean squared error, structural similarity index, or mutual information. Gradient features are quantitative indicators describing the local gradient characteristics of an image. These gradient features can include gradient magnitude, gradient direction, gradient variance, or gradient histogram. Weighting refers to adjusting data according to specific weights to reflect the importance of different data points or regions. This weighting can be linear weighting, nonlinear weighting, feature-based adaptive weighting, or dynamic weighting based on the iteration stage. Image state indicators are numerical or vector values ​​used to quantify the convergence degree, detail recovery, or noise level of the current image during the iteration process. An image state indicator can be a single scalar value or a vector containing information across multiple dimensions.

[0029] This scheme introduces a gradient-based region partitioning and weighted processing approach, enabling image state indicators to accurately reflect the local convergence status and detailed changes in different regions of the image. This provides a detailed basis for the adaptive adjustment of the contribution weights of subsequent coarse scan data. Specifically, during the iterative update of ROI image estimation, the gradient map of the current iteration's image estimation is first obtained. This gradient map captures the structural features and degree of change of the image. Next, a clustering algorithm is used to divide the image estimation into multiple gradient regions based on the gradient map, such as flat regions, edge regions, and texture regions, to distinguish local image features and avoid generalizing to all regions. Then, for each gradient region, the image change information between the current iteration and the previous iteration is obtained. This image change information reflects the local convergence or degree of change in the region. Furthermore, the image change information is weighted according to the gradient features corresponding to the gradient region. This means that regions with different gradient features contribute differently to the image state indicators; for example, change information in edge regions is more important than change information in flat regions. This weighted processing allows the image state indicators to sensitively reflect changes in key image regions, ensuring that attention is paid to the detailed structural recovery of these regions in later iterations. Finally, the image change information after weighted processing of all gradient regions is synthesized to determine the final image state index. This comprehensive index integrates multi-dimensional information of local image features and change sensitivity. Therefore, the image state index in this embodiment can accurately and comprehensively reflect the overall convergence state and detail evolution trend of the current iterative image, providing a reliable and detailed basis for the adaptive adjustment of the contribution weights of subsequent coarse scan data. In this way, the acquisition of the image state index is no longer a simple global difference, but considers the local features and detail evolution of different regions within the image, so that the adjustment of the contribution weights of coarse scan data can be more refined and adaptive, thereby achieving a balance between artifact suppression and detail preservation during the iteration process.

[0030] As a specific implementation method, the scheme of this application is implemented as follows: After obtaining the ROI image estimate of the current iteration, the Sobel operator is used to process the image estimate to generate its gradient magnitude map (gradient map). Then, the K-means clustering algorithm is used to divide the pixels in the gradient map into three gradient regions based on their gradient magnitude, for example, a low gradient region (corresponding to a flat region), a medium gradient region (corresponding to a textured region), and a high gradient region (corresponding to an edge region). Then, for each divided gradient region, the average absolute difference between the pixel values ​​of the current iteration's ROI image estimate and the previous iteration's ROI image estimate within that region is calculated (image change information). Simultaneously, a weighting factor is determined based on the average gradient magnitude of each gradient region; for example, the weighting factor for the high gradient region is higher than that for the low gradient region. Subsequently, the image change information is weighted by multiplying the image change information of each region by its corresponding weighting factor. Finally, the weighted image change information of all gradient regions is accumulated to obtain the final image state index. Through the above scheme, this embodiment can enable the image state index to accurately reflect the local convergence state and detail changes of different regions of the image, so as to provide a detailed basis for the adaptive adjustment of the contribution weight of subsequent coarse scan data. It also helps to achieve a balance between artifact suppression and detail preservation during the iteration process. Therefore, this embodiment can restore the detailed structure inside the ROI while avoiding the introduction of statistical noise, thereby effectively improving the quality of the reconstructed image.

[0031] In some preferred embodiments, step A33 includes: A331. Obtain the global convergence state parameters based on the overall difference between the current ROI image estimate and the initial ROI image estimate. A332. For each gradient region, obtain the image change information of that gradient region in the current iteration of ROI image estimation and the previous iteration of ROI image estimation; A333. For each gradient region, determine the weighting weights based on the current iteration stage, global convergence state parameters, and corresponding gradient features, and then perform weighted processing on the image change information based on the weighting weights.

[0032] The global convergence state parameter refers to a numerical value reflecting the overall difference between the current iteration's ROI image estimate and the initial ROI image estimate. This embodiment can obtain the global convergence state parameter by calculating indicators such as the norm of the pixel value difference between the current iteration's ROI image estimate and the initial ROI image estimate, structural similarity index, or peak signal-to-noise ratio. This global convergence state parameter can evaluate the overall convergence process of the image from a macroscopic perspective, providing global guidance for subsequent local weighted processing. The current iteration stage refers to the specific period or progress of the iteration process. This embodiment can obtain the current iteration stage based on the ratio of the current iteration number to the total number of iterations or a preset iteration stage division (e.g., early, middle, late). This embodiment can dynamically adjust the weighting weights according to the progress of the iteration by determining the weighting based on the current iteration stage, adapting to the different emphases of image processing at different iteration stages. The purpose of determining the weighting weights is to intelligently adjust the contribution of image change information in different gradient regions to the final image state index, thereby optimizing the image reconstruction effect. In this embodiment, the weighting weights can be determined by querying a pre-built mapping table of iteration stages, convergence parameters, gradient features, and weighting weights based on the current iteration stage, global convergence state parameters, and gradient features. This mapping table stores the weighting weights corresponding to different combinations of iteration stages, convergence parameters, and gradient features. For example, for high gradient regions, the weighting weight is 0.5 in the later stages of iteration when the convergence parameters are large, and 1.1 in the later stages of iteration when the convergence parameters are small.

[0033] Specifically, this embodiment calculates and obtains a global convergence state parameter based on the overall difference between the current iteration's ROI image estimate and the initial ROI image estimate. This parameter provides a macroscopic perspective on the overall convergence degree of the image. Subsequently, for each gradient region segmented in the image, this embodiment obtains image change information for that region between the current iteration's ROI image estimate and the previous iteration's ROI image estimate. This image change information reflects the microscopic changes of the gradient region in adjacent iterations. Then, this embodiment determines weighting weights based on the current iteration stage, the global convergence state parameter, and the gradient characteristics of the gradient region itself. That is, the weighting weights in this embodiment are no longer fixed values. Specifically, the information from the current iteration stage allows the weighting weights to be dynamically adjusted as the iteration progresses. For example, in the early stages of iteration, the focus is on macroscopic artifact suppression, while in later stages, the focus is on detail preservation. The global convergence state parameter correlates the changes in local regions with the overall convergence of the image, ensuring that the weighting process both promotes overall convergence and avoids over-smoothing. The introduction of gradient characteristics ensures sensitivity to local image structure and edge information. Finally, this embodiment performs weighted processing on the image change information based on the weighting weights. This embodiment enables image state indicators to more accurately reflect the true convergence status and detailed changes in ROI image estimation by dynamically determining weighting weights based on the current iteration stage, global convergence state parameters, and gradient features. This effectively improves the accuracy and reliability of the contribution weights from coarse scan data. Through the above scheme, this application addresses the problem that simply weighting image change information based on gradient features alone cannot fully reflect the global convergence state and characteristics of the image throughout the entire iteration process by comprehensively considering the current iteration stage, global convergence state parameters, and local gradient features during the acquisition of image state indicators.

[0034] In some preferred embodiments, step A4 includes: A41. Determine the initial contribution weights based on the image state indicators of the current iteration; A42. Calculate the rate of change of the image status index based on the current iteration's image status index and the previous iteration's image status index; A43. Adjust the initial contribution weights based on the rate of change of the indicators to obtain the contribution weights of the coarse scan data.

[0035] First, this embodiment provides an initial estimate of the coarse scan data contribution weight based on the current image state index by determining the initial contribution weight according to the current iteration's image state index. This ensures that the coarse scan data contribution weight can respond to the real-time image characteristics during the iteration process. This embodiment determines the initial contribution weight by querying a pre-built mapping table of image state indices and contribution weights based on the current iteration's image state index. This mapping table stores the contribution weights corresponding to different image state indices; for example, the higher the image entropy value, the higher the initial contribution weight. Second, this embodiment quantifies the speed or trend of image state index changes between consecutive iterations by calculating the rate of change of the index based on the current iteration's image state index and the previous iteration's image state index. This allows the system to perceive the dynamic evolution of the image. For example, a rapid rate of change indicates rapid improvement of the image or rapid suppression of macroscopic artifacts in the early stages of iteration, while a slow rate of change may indicate that the image is close to convergence, at which point more refined detail recovery is needed. Finally, the initial contribution weight is adjusted according to the rate of change of the index to obtain the final coarse scan data contribution weight. This adjustment process allows the contribution weight of coarse scan data to not only depend on the current image state indicators but also adaptively respond to the convergence speed and detail recovery requirements of the image. For example, in the early stages of iteration, when the rate of change of image state indicators is large, the initial contribution weight can be further increased to more effectively utilize coarse scan data to suppress truncation artifacts. In the later stages of iteration, when the rate of change of image state indicators tends to level off, the initial contribution weight can be further reduced to decrease the excessive smoothing of fine structures by coarse scan data, thereby better recovering the internal details of the ROI and maintaining quantitative accuracy. Through this dynamic adjustment based on the trend of image state changes, this scheme can more intelligently balance the contribution of coarse scan data at different iteration stages, effectively improving the flexibility of coarse scan data contribution adjustment, thus ensuring that both artifact suppression and detail preservation requirements are met simultaneously.

[0036] In some preferred embodiments, step A5 includes: A51. Determine the contribution weight of the fine scan data based on the projection deviation; A52. Normalize the contribution weights of fine scan data and coarse scan data; A53. Update the ROI image estimate for the current iteration based on the contribution weights of the fine scan data, the small field-of-view fine scan projection data, the coarse scan data contribution weights, and the large field-of-view coarse scan projection data.

[0037] The contribution weight of fine-scanned data refers to a parameter used to adjust the influence of small-field-of-view fine-scanned projection data on the ROI image estimation correction during image iterative update. It can be a scalar, a vector, or a function, and its value can be dynamically adjusted according to the projection deviation. Specifically, the projection deviation reflects the difference between the current estimated image (estimated projection data) and the actual fine-scanned projection data. Since this embodiment uses the projection deviation to determine the contribution weight of fine-scanned data, it can assign a suitable contribution weight to the small-field-of-view fine-scanned projection data based on the current estimated image and the actual fine-scanned projection data. For example, when the projection deviation is small, the contribution weight of the small-field-of-view fine-scanned projection data is increased; when the projection deviation is large, the contribution weight of the small-field-of-view fine-scanned projection data is decreased. Step A52 normalizes the contribution weights of fine-scanned data and coarse-scanned data to make the sum of the contribution weights of fine-scanned data and coarse-scanned data equal to 1. For example, if the original contribution weight of fine-scanned data is 2 and the original contribution weight of coarse-scanned data is 1, the normalized contribution weight of fine-scanned data is 0.67, and the normalized contribution weight of coarse-scanned data is 0.33. Since step A53 updates the ROI image estimation of the current iteration based on the contribution weights of fine scan data, small field-of-view fine scan projection data, coarse scan data, and large field-of-view coarse scan projection data, this embodiment can adaptively balance the contributions from large field-of-view coarse scan data and small field-of-view fine scan data according to the current image state and the characteristics of the two types of data. This allows the iterative process to better adapt to the dynamic changes of the image from macroscopic correction to detail restoration, thereby continuously optimizing image quality throughout the reconstruction process and achieving the best balance between artifact suppression and detail preservation.

[0038] In some preferred embodiments, step S2 includes: S21. Reconstruct the image from the large field-of-view coarse scan projection data to obtain a preliminary image; S22. Divide the preliminary image into multiple coarse scan regions; S23. For each coarse-scanned region, obtain local texture features, edge features, and noise distribution features, and then determine the smoothing intensity parameter of the coarse-scanned region based on the local texture features, edge features, and noise distribution features. S24. Smooth the preliminary image according to all smoothing intensity parameters to obtain a coarse scan image.

[0039] The coarse-scanned region refers to multiple sub-regions obtained by segmenting the initial image. Each sub-region can be independently analyzed for features and its parameters determined. This embodiment can use a fixed-size grid, adaptive segmentation based on image content (such as edges and textures), or clustering algorithms to divide the initial image into multiple coarse-scanned regions. Local texture features refer to attributes describing the pixel intensity variation patterns within the coarse-scanned region. Local texture features reflect the level of detail and structural complexity of the coarse-scanned region. This embodiment can use Gray-Level Co-occurrence Matrix (GLCM) features, Local Binary Pattern (LBP), Gabor filter responses, or wavelet transform coefficients to obtain local texture features. Edge features refer to attributes describing abrupt changes in pixel intensity within the coarse-scanned region. Edge features indicate the structural boundaries and contours of the image. This embodiment can use gradient magnitude, Sobel operator, Canny operator, Laplacian operator, or LoG (Laplacian of Gaussian) operator to obtain edge features. Noise distribution characteristics refer to attributes describing the noise level and statistical properties within a coarse-scanned region. These characteristics quantify the noise intensity of that region. This embodiment can obtain noise distribution characteristics by calculating local standard deviation, the relationship between local mean and variance, noise power spectral density, or wavelet domain-based noise estimation methods. The smoothing intensity parameter is a numerical value used to control the degree of image smoothing. A larger smoothing intensity parameter results in a stronger smoothing effect. This parameter can be the standard deviation of a Gaussian filter, the spatial and intensity domain parameters of a bilateral filter, the similarity weight parameter of a nonlocal mean filter, or the window size of an adaptive median filter. Smoothing refers to the process of improving the visual quality of an image by reducing the intensity of high-frequency components (such as noise and sharp edges). This embodiment can achieve smoothing using Gaussian filtering, mean filtering, median filtering, bilateral filtering, nonlocal mean (NLM) filtering, or wavelet domain denoising methods.

[0040] First, this embodiment obtains a preliminary image by reconstructing the large-field-of-view coarse-scan projection data, converting the original projection data into an image spatial representation and laying the foundation for subsequent fine-tuning. Next, since the features (such as texture, edges, and noise levels) of different regions in the preliminary image may vary, a uniform processing method is unlikely to achieve the desired effect. Therefore, this embodiment divides the preliminary image into multiple coarse-scan regions to lay the foundation for subsequent adaptive processing of different regions of the preliminary image. Then, for each coarse-scan region, its local texture features, edge features, and noise distribution features are acquired. These features are key indicators for evaluating the local quality and content of the image. Specifically, local texture features reflect the level of detail within the region, edge features indicate the structural boundaries of the image, and noise distribution features quantify the noise level of the region. By acquiring these multi-dimensional features, the system can understand the image characteristics of each region, providing a basis for determining subsequent adaptive smoothing parameters. After acquiring these features, the smoothing intensity parameter for the coarse-scanned region is determined based on local texture features, edge features, and noise distribution features. For example, in regions with rich texture or edges, the smoothing intensity parameter can be set to a smaller value to avoid blurring details; while in regions with significant noise but lacking structure, the smoothing intensity parameter can be set to a larger value to effectively suppress noise. This parameter determination method based on local features ensures the accuracy and effectiveness of the smoothing process, avoiding detail loss due to over-smoothing or noise residue due to insufficient smoothing. Finally, the preliminary image is smoothed according to all determined smoothing intensity parameters to obtain the coarse-scanned image. By applying the adaptive smoothing intensity parameter for each region to the preliminary image, the system can generate a coarse-scanned image that effectively suppresses noise and preserves image details across the entire range. Compared to the unadapted image, this embodiment can effectively improve the signal-to-noise ratio and structure fidelity of the coarse-scanned image, providing accurate input for subsequent initial ROI image estimation, thereby effectively improving the accuracy of the subsequently generated initial ROI image estimation, and thus effectively accelerating the convergence process and improving the reconstruction quality and diagnostic reliability of the final ROI image.

[0041] As a preferred implementation, the solution of this application is specifically implemented as follows: First, the Ordered Subset Expectation-Maximization (OSEM) algorithm is used to reconstruct the image from the large field-of-view coarse-scan projection data to obtain a preliminary image. Next, the preliminary image is divided into multiple coarse-scan regions using a fixed-size grid, for example, dividing the image into 16x16 pixel non-overlapping regions. Then, for each coarse-scan region, local texture features are obtained by calculating its contrast, energy, and homogeneity using the Gray-Level Co-occurrence Matrix (GLCM), edge features are obtained by detecting the edge intensity and direction within the region using the Canny operator, and noise distribution features are obtained by calculating the local standard deviation of the pixel intensity within the region. After obtaining these features, a preset rule or lookup table is used to determine the smoothing intensity parameter of the coarse-scan region. For example, the lookup table could be set as follows: if the local texture is rich and the edge intensity is high, while the noise level is low, the smoothing intensity parameter can be set to a smaller value to preserve details; if the local texture is flat and the edge intensity is low, while the noise level is high, the smoothing intensity parameter can be set to a larger value to effectively suppress noise. Finally, the preliminary image is smoothed according to all determined smoothing intensity parameters to obtain a coarse-scanned image. This embodiment can use a bilateral filter for smoothing, with the spatial and intensity domain parameters of the bilateral filter adaptively adjusted based on the smoothing intensity parameters determined for each coarse-scanned region. For example, the smoothing intensity parameters can be directly mapped to the spatial and intensity standard deviations of the bilateral filter, thereby achieving differentiated smoothing for different regions. This embodiment effectively solves the problem of failing to achieve a balance between noise suppression and detail preservation in the preliminary image due to simple image reconstruction of the coarse-scanned projection data. It also avoids situations where excessive noise in the preliminary image affects the accuracy of subsequent ROI image estimation and iterative updates, and where over-smoothing leads to the loss of structural details in the preliminary image.

[0042] In some preferred embodiments, step S3 includes: S31. Take the region in the coarse scan image that corresponds to the small field-of-view fine scan projection data as the initial target region. S32. Obtain image feature information of the initial target region based on the coarse scan image. The image feature information includes the geometric center and intensity distribution characteristics of the initial target region. S33. Obtain projection feature information based on small field-of-view fine scan projection data. The projection feature information includes the geometric center of the projection and the projection intensity distribution characteristics. S34. Obtain the spatial transformation parameters between the initial target area and the small field-of-view fine scan projection data based on the image feature information and projection feature information; S35. Perform spatial correction processing on the initial target region in the coarse scan image according to the spatial transformation parameters to obtain the spatially corrected ROI image region, and use the ROI image region as the initial ROI image estimate.

[0043] The specific process of step S31 can be as follows: Analyze the field of view of the coarse scan image and the acquisition range of the fine scan projection data in the small field of view, determine the overlapping area between the two in space, and take the overlapping area as the initial target area. Step S32 can obtain the geometric center of the initial target area by calculating the average value of all pixel coordinates in the initial target area. Step S32 can obtain the intensity distribution characteristics of the initial target area by first constructing the gray-level histogram of the initial target area, and then calculating its statistical moments (mean, variance, skewness, and kurtosis) or extracting texture features (gray-level co-occurrence matrix features). Step S33 can obtain the geometric center of the projection by calculating the centroid or center point of the projection curve of the fine scan projection data at each angle, and then averaging or fitting these center points or centroids in the projection domain. Step S33 can obtain the projection intensity distribution characteristics by analyzing the intensity histogram of the projection data, calculating its statistical moments, or extracting projection texture features. Step S34 can obtain spatial transformation parameters by comparing the image feature information obtained in step S32 with the projection feature information obtained in step S33. For example, it can compare the similarity of the positional deviation of the geometric center and the intensity distribution, and use image registration algorithms (such as algorithms based on mutual information, normalized cross-correlation, or feature point matching) to calculate the spatial transformation parameters such as translation, rotation, and scaling required to transform the initial target region in the image domain to the projection domain or transform the projection domain data to the image domain based on this similarity. The specific process of step S35 can be as follows: apply the spatial transformation parameters to the initial target region in the coarse scan image, and reposition and assign pixels through image resampling techniques (such as bilinear interpolation, bicubic interpolation, or nearest neighbor interpolation) to generate a ROI image region that is spatially aligned with the small field-of-view fine scan projection data. This embodiment can make the initial ROI image and the small field-of-view fine scan projection data highly consistent in space by spatially correcting the initial ROI image estimation. This provides a starting point with high spatial consistency for subsequent iterative reconstruction and reduces errors caused by misalignment between the initial ROI image and the small field-of-view fine scan projection data. Therefore, this embodiment can effectively improve the artifact suppression and detail recovery effect in the iterative update process, thereby effectively improving the detail fidelity and quantitative accuracy of the reconstructed image.

[0044] In some preferred embodiments, step A1 includes: A11. Simulate the attenuation effect of rays in tissue based on the ROI image of the current iteration and the preset attenuation correction parameters; A12. Simulate the scattering effect of rays in tissue based on the ROI image of the current iteration and the preset scattering correction parameters; A13. Based on the current iteration of the ROI image, estimate and simulate the detector's response characteristics to rays using the preset detector response function; A14. The attenuation effect, scattering effect, and response characteristics are fused to obtain the estimated projection data.

[0045] Attenuation correction parameters are physical quantities used to describe the degree of intensity attenuation of radiation as it passes through tissue. They can be implemented using attenuation coefficient maps, attenuation spectra, or attenuation correction factors generated from CT images. Scattering correction parameters are physical quantities used to describe the scattering effect of radiation in tissue. They can be implemented using scattering point spread functions, scattering fraction maps, or scattering distribution models obtained from Monte Carlo simulations. Detector response functions are mathematical models describing the detector's response characteristics to incident radiation. They can be characterized using point spread functions, line spread functions, or modulation transfer functions.

[0046] Specifically, this scheme improves the accuracy of estimated projection data by comprehensively simulating the physical processes of SPECT imaging when generating the estimated projection data. First, this embodiment simulates the attenuation effect of rays in tissue based on the current iteration's ROI image estimation and preset attenuation correction parameters. This simulation considers energy loss due to absorption and scattering of rays as they penetrate biological tissue, ensuring that the estimated projection data reflects the reduction in ray intensity during actual measurement. Second, this embodiment simulates the scattering effect of rays in tissue based on the current iteration's ROI image estimation and preset scattering correction parameters. This simulates the phenomenon where rays change direction after Compton scattering in tissue and are recorded by the detector, ensuring that the estimated projection data includes scattering components and is closer to the actual measurement data. Third, it simulates the detector's response characteristics to rays based on the current iteration's ROI image estimation and preset detector response function. This considers the detector's inherent spatial resolution limitations and energy broadening, ensuring that the estimated projection data reflects the blurring effect introduced by the detector. Finally, the attenuation effect, scattering effect, and detector response characteristics obtained from the above simulations are integrated to form the final estimated projection data. This fusion ensures that the estimated projection data comprehensively and accurately reflects all major physical processes from the current ROI image estimate to the actual projection data, thereby effectively improving the accuracy of the projection data assessment. In this way, the projection bias obtained in step A2 can more precisely indicate the difference between the current ROI image estimate and the real image, rather than simply the bias caused by inaccuracies in the physical model. This provides a more reliable correction direction when updating the ROI image estimate in the current iteration, thereby optimizing the convergence of the entire iterative reconstruction process and ultimately contributing to the reconstruction of higher quality and more accurate SPECT ROI images. Combined with the overall framework of iteratively updating the ROI image estimate, this approach makes the correction direction more precise in each iteration, effectively avoiding artifacts, blurring, or quantitative inaccuracies in the reconstructed images caused by insufficient simulation of physical effects, and significantly improving the clinical diagnostic value of the reconstructed images.

[0047] In some preferred embodiments, the preset convergence condition is that the cumulative number of iterations reaches a preset number of iterations and the image state index reaches a preset index. The preset index refers to a target threshold set for the image state index. When the image state index reaches or meets this threshold, it indicates that the image has become stable or has reached the expected quality. It can be set based on experience, experiments, or specific application requirements.

[0048] Secondly, such as Figure 2 As shown, this application also provides a ROI image reconstruction system based on SPECT data, which includes: Data acquisition module 1 is used to acquire large field-of-view coarse scan projection data and acquire small field-of-view fine scan projection data based on ROI; Coarse scan image reconstruction module 2 is used to reconstruct coarse scan images based on large field-of-view coarse scan projection data; ROI image estimation generation module 3 is used to generate an initial ROI image estimate based on the coarse scan image; Iterative update module 4 is used to iteratively update the ROI image estimate until the preset convergence condition is met. The process of iteratively updating the ROI image estimation includes: A1. Generate estimated projection data based on the ROI image of the current iteration; A2. Obtain the projection deviation based on the estimated projection data and the small field-of-view fine scan projection data; A3. Obtain image state indicators based on the ROI image estimation of the current iteration and the ROI image estimation of the previous iteration; A4. Determine the contribution weight of coarse scan data based on image status indicators; A5. Update the ROI image estimation for the current iteration based on projection bias, contribution weights of small field-of-view fine scan projection data and coarse scan data, and large field-of-view coarse scan projection data.

[0049] This application provides a ROI image reconstruction system based on SPECT data, comprising a data acquisition module 1, a coarse scan image reconstruction module 2, an ROI image estimation generation module 3, and an iterative update module 4. The ROI image reconstruction system based on SPECT data provided in this embodiment is used to perform the steps in the ROI image reconstruction method based on SPECT data provided in the first aspect above. The principle of the ROI image reconstruction system based on SPECT data provided in this embodiment is the same as that of the ROI image reconstruction method based on SPECT data provided in the first aspect above, and will not be discussed in detail here.

[0050] As can be seen from the above, the ROI image reconstruction method and system based on SPECT data provided in this application, by introducing an image state index and dynamically determining the contribution weight of coarse scan data according to the index, adaptively adjusts the contribution of large field-of-view coarse scan data to ROI image estimation during the iterative reconstruction process. That is, this application enables the iterative reconstruction process to adapt to the dynamic changes in image state, so as to effectively suppress truncation artifacts in the early stage of iteration and restore the fine structure inside the ROI with high fidelity in the later stage of iteration. Therefore, this application can effectively solve the problem that it is difficult to achieve the best balance between artifact suppression and detail fidelity because fixed weight factors cannot simultaneously meet the requirements of effective suppression of truncation artifacts in the early stage of iteration and the high fidelity requirements of restoration of fine structure inside the ROI and accuracy of radioactivity count values ​​in the later stage of iteration, thereby effectively improving the reliability of diagnostic results.

[0051] In the embodiments provided in this application, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of the above units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple units or components may be combined or integrated into another robot, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0052] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0053] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0054] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0055] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for ROI image reconstruction based on SPECT data, characterized in that, The ROI image reconstruction method based on SPECT data includes: S1. Obtain large field-of-view coarse scan projection data and obtain small field-of-view fine scan projection data based on ROI; S2. Reconstruct the coarse scan image based on the large field-of-view coarse scan projection data; S3. Generate an initial ROI image estimate based on the coarse scan image; S4. The ROI image estimation is iteratively updated until the preset convergence condition is met; The process of iteratively updating the ROI image estimate includes: A1. Generate estimated projection data based on the ROI image of the current iteration; A2. Obtain the projection deviation based on the estimated projection data and the small field-of-view fine scan projection data; A3. Obtain image state indicators based on the ROI image estimation of the current iteration and the ROI image estimation of the previous iteration; A4. Determine the contribution weight of the coarse scan data based on the image status indicators; A5. Update the ROI image estimation for the current iteration based on the projection deviation, the small field-of-view fine scan projection data, the contribution weight of the coarse scan data, and the large field-of-view coarse scan projection data.

2. The ROI image reconstruction method based on SPECT data according to claim 1, characterized in that, Step A3 includes: A31. Obtain the gradient map of the ROI image estimation in the current iteration; A32. Using a clustering algorithm, the ROI image of the current iteration is estimated to be divided into multiple gradient regions based on the gradient map; A33. For each gradient region, obtain the image change information of the gradient region in the current iteration of ROI image estimation and the previous iteration of ROI image estimation, and then perform weighted processing on the image change information according to the gradient features corresponding to the gradient region. A34. Determine the image status index based on all weighted image change information.

3. The ROI image reconstruction method based on SPECT data according to claim 2, characterized in that, Step A33 includes: A331. Obtain the global convergence state parameters based on the overall difference between the current ROI image estimate and the initial ROI image estimate. A332. For each gradient region, obtain the image change information of the gradient region in the current iteration of ROI image estimation and the previous iteration of ROI image estimation; A333. For each gradient region, a weighting weight is determined based on the current iteration stage, the global convergence state parameter, and the corresponding gradient feature. Then, the image change information is weighted according to the weighting weight.

4. The ROI image reconstruction method based on SPECT data according to claim 1, characterized in that, Step A4 includes: A41. Determine the initial contribution weights based on the image state indicators of the current iteration; A42. Calculate the rate of change of the image status index based on the current iteration's image status index and the previous iteration's image status index; A43. Adjust the preliminary contribution weight according to the rate of change of the indicator to obtain the contribution weight of the coarse scan data.

5. The ROI image reconstruction method based on SPECT data according to claim 1, characterized in that, Step A5 includes: A51. Determine the contribution weight of the fine scan data based on the projection deviation; A52. Normalize the contribution weights of the fine scan data and the coarse scan data. A53. Update the ROI image estimation for the current iteration based on the contribution weight of the fine scan data, the small field-of-view fine scan projection data, the contribution weight of the coarse scan data, and the large field-of-view coarse scan projection data.

6. The ROI image reconstruction method based on SPECT data according to claim 1, characterized in that, Step S2 includes: S21. Perform image reconstruction on the large field-of-view coarse scan projection data to obtain a preliminary image; S22. Divide the preliminary image into multiple coarse scan regions; S23. For each of the coarse-scanned regions, obtain local texture features, edge features, and noise distribution features, and then determine the smoothing intensity parameter of the coarse-scanned region based on the local texture features, the edge features, and the noise distribution features; S24. Smooth the preliminary image according to all the smoothing intensity parameters to obtain a coarse scan image.

7. The ROI image reconstruction method based on SPECT data according to claim 1, characterized in that, Step S3 includes: S31. Take the region in the coarse scan image that corresponds to the small field-of-view fine scan projection data as the initial target region. S32. Obtain image feature information of the initial target region based on the coarse scan image, wherein the image feature information includes the geometric center and intensity distribution features of the initial target region; S33. Obtain projection feature information based on the small field-of-view fine scan projection data, wherein the projection feature information includes the geometric center of the projection and the projection intensity distribution characteristics; S34. Obtain the spatial transformation parameters between the initial target region and the small field-of-view fine scan projection data based on the image feature information and the projection feature information; S35. Perform spatial correction processing on the initial target region in the coarse scan image according to the spatial transformation parameters to obtain the spatially corrected ROI image region, and use the ROI image region as the initial ROI image estimate.

8. The ROI image reconstruction method based on SPECT data according to claim 1, characterized in that, Step A1 includes: A11. Simulate the attenuation effect of rays in tissue based on the ROI image of the current iteration and the preset attenuation correction parameters; A12. Simulate the scattering effect of the rays in the tissue based on the ROI image estimation of the current iteration and the preset scattering correction parameters; A13. Simulate the detector's response characteristics to the ray based on the ROI image estimation of the current iteration and the preset detector response function; A14. The attenuation effect, the scattering effect, and the response characteristics are fused to obtain estimated projection data.

9. The ROI image reconstruction method based on SPECT data according to claim 1, characterized in that, The preset convergence condition is that the cumulative number of iterations reaches a preset number of iterations and the image state index reaches a preset index.

10. A ROI image reconstruction system based on SPECT data, characterized in that, The ROI image reconstruction system based on SPECT data includes: The data acquisition module is used to acquire large field-of-view coarse scan projection data and small field-of-view fine scan projection data based on ROI; The coarse scan image reconstruction module is used to reconstruct the coarse scan image based on the large field-of-view coarse scan projection data; The ROI image estimation generation module is used to generate an initial ROI image estimate based on the coarse scan image. The iterative update module is used to iteratively update the ROI image estimate until a preset convergence condition is met. The process of iteratively updating the ROI image estimate includes: A1. Generate estimated projection data based on the ROI image of the current iteration; A2. Obtain the projection deviation based on the estimated projection data and the small field-of-view fine scan projection data; A3. Obtain image state indicators based on the ROI image estimation of the current iteration and the ROI image estimation of the previous iteration; A4. Determine the contribution weight of the coarse scan data based on the image status indicators; A5. Update the ROI image estimation for the current iteration based on the projection deviation, the small field-of-view fine scan projection data, the contribution weight of the coarse scan data, and the large field-of-view coarse scan projection data.

Images