PET probe extensive field image multi-model adaptive deformation and FPGA curing method
By employing a multi-model adaptive deformation method, local optimization and spillover risk control are performed on the PET probe over-field image, solving the problems of insufficient local optimization and uncontrollable spillover risk in existing technologies, and realizing the construction of an efficient and accurate crystal position lookup table.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGSU SINOGRAM MEDICAL TECH CO LTD
- Filing Date
- 2026-02-11
- Publication Date
- 2026-05-19
AI Technical Summary
Existing PET probe over-field image preprocessing methods suffer from insufficient local optimization capabilities, poor controllability of event spillover risk, and difficulty in balancing processing efficiency and accuracy.
A multi-model adaptive deformation method is adopted. The original over-field image is subjected to coordinate traversal transformation through a pre-built multi-model transformation library. The optimal model and parameters are selected using multiple evaluation indicators, and the enhanced image is generated and solidified into the FPGA to achieve local precise optimization and overflow risk control.
It significantly improves the distinguishability of crystal event clusters, ensures no event loss, has high processing efficiency, adapts to different probe defect modes, and is suitable for high-precision LUT construction of various PET detectors.
Smart Images
Figure CN122066748A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of positron emission tomography (PET) technology, and more particularly to a method for multi-model adaptive deformation and FPGA curing of PET probe over-field images. Background Technology
[0002] The over-field image of a PET detector is essentially a statistical result of the integer pixel coordinates of a large number of gamma-photon events. By counting millions of events and accumulating the counts of the same integer coordinate positions, an event density integer pixel map is formed. The bright areas in the image represent the event sphere of influence of the crystal, and the clarity of their boundaries directly affects the calibration accuracy of the crystal position lookup table (LUT).
[0003] Existing technologies mainly use a global isometric stretching method to process Floodimages, but this method has the following drawbacks: 1) Lack of local optimization: It is impossible to specifically widen the gap between the bright areas of adjacent crystals that are too close due to crystal defects, which can easily lead to large errors in determining the coordinates of the crystal edge range.
[0004] 2) Imbalance between efficiency and accuracy: Global stretching causes redundant computing power, and direct operation based on integer coordinates is prone to loss of accuracy, making it impossible to achieve local precision enhancement and controllable global computing power.
[0005] 3) Poor process fit: The characteristics of digital output integer coordinates, the need to retain intermediate precision during processing, and the need to adapt to statistics in the final stage are not fully considered, resulting in a mismatch between the distribution of image events after processing and the physical response.
[0006] 4) Uncontrollable overflow risk: The stretching coefficient setting lacks correlation with accuracy requirements, which can easily lead to the loss of effective events that exceed the image field of view. Furthermore, there is a lack of effective quantitative evaluation indicators to avoid this risk.
[0007] Therefore, a Floodimage preprocessing method is needed that can specifically optimize the distribution of local events, control computing power and overflow risks, and deeply fit the actual processing flow. Summary of the Invention
[0008] Technical problems to be solved In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a multi-model adaptive deformation and FPGA solidification method for PET probe over-field images, which solves the technical problems of insufficient local optimization capability, poor controllability of event overflow risk, and difficulty in balancing processing efficiency and accuracy in existing PET over-field image preprocessing methods.
[0009] Technical solution To achieve the above objectives, the main technical solutions adopted by the present invention include: In a first aspect, the present invention provides a method for multi-model adaptive deformation and FPGA solidification of PET probe over-field images, comprising the following steps: Step 1: Collect single-lift event data from the PET probe; Step 2: Generate the original overlay image based on the single-event data; Step 3: Using the image deformation models in the pre-set multi-model transformation library, perform coordinate traversal transformations on the original overlay image respectively; Step 4: Using multiple preset evaluation indicators, perform offline analysis on each model and its parameters in the multi-model transformation library, and select the optimal model and optimal parameters. Step 5: Transform the original overlay image using the optimal model and optimal parameters to generate an enhanced image and a crystal position lookup table, and then save it to the FPGA.
[0010] As a further improvement to the method of the present invention, in step 2, the preset multi-model transformation library includes one or more image coordinate transformation models with different deformation characteristics, and these models constitute an offline preferred candidate set.
[0011] As a further improvement to the method of the present invention, the image coordinate transformation model is specifically as follows: The first model is a model that performs a global linear scaling transformation on the pixel coordinates of the overlay image, used to achieve global isotropic / anisotropic linear stretching. The second model is a model that performs radial concave transformation on the pixel coordinates of the overlay image, used for nonlinear stretching of events that are stuck in the left and right regions of the image. The third model is a power-law transformation of the pixel coordinates of the overlay image, used to enhance and stretch the image to address the problem of dense events in the four corner regions.
[0012] As a further improvement to the method of the present invention, the evaluation index in step 4 includes a combination of one or more of the following: event loss rate L, identification result of adjacent crystal centroid proximity pairs, and distance increase rate G of proximity pairs.
[0013] As a further improvement to the method of the present invention, the event loss rate L, the identification result of adjacent crystal centroid proximity pairs, and the distance increase rate G of proximity pairs are specifically as follows: The event loss rate L is calculated as follows: ;in, This represents the total event count on the original overlay image. This represents the total event count after the transformation; The identification results of adjacent crystal centroid proximity pairs are used to calculate the distance between all adjacent crystal centroids in the original overlay image, and crystal centroid point pairs whose distance is less than a set threshold M are marked as proximity point pairs. The distance increase rate G of the proximity pair is calculated as follows: ;in , These represent the distances before and after the transformation, respectively.
[0014] As a further improvement to the method of the present invention, the model optimization process in step 4 is as follows: by traversing all models and their parameter combinations in the preset multi-model library through preset evaluation indicators, the optimal model and corresponding optimal parameter combination that meet the evaluation indicator constraints are selected.
[0015] As a further improvement to the method of the present invention, in step 5, the specific way to solidify the optimal model and optimal parameters into a lookup table is as follows: the optimized optimal model and optimal parameters are applied to transform the original overlay image to generate an enhanced overlay image, and a crystal position lookup table is generated based on this enhanced image. The lookup table provides crystal position information corresponding to the x and y coordinate ranges according to the overlay image. The optimal image coordinate transformation model equations, corresponding optimal parameters, and the generated crystal position lookup table are written into the FPGA in the form of a lookup table.
[0016] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements the PET probe over-field image multi-model adaptive deformation and FPGA solidification method described in any of the first aspects above.
[0017] Thirdly, the present invention provides a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements the PET probe over-field image multi-model adaptive deformation and FPGA solidification method described in any of the first aspects above.
[0018] Beneficial effects The beneficial effects of this invention are: 1. Precise Local Optimization: For the coordinate spacing of crystal events that are too close (adhesive) in the Floodimage, a nonlinear transformation model (such as radial concavity and four-corner power transformation) is used to stretch the locally adhered areas, which can significantly improve the distinguishability of crystal event clusters and fundamentally improve the LUT calibration accuracy.
[0019] 2. Strictly controllable overflow: During the optimization process, the event loss rate is used as a hard constraint indicator for offline selection to ensure that no events are lost, while also taking into account processing efficiency. This ensures that no changes will lead to the loss of valid events, thus guaranteeing the statistical integrity of the data.
[0020] 3. Extremely high processing efficiency: The most time-consuming model optimization and parameter calculation are completed offline. In the online stage, the FPGA only needs to perform table lookup operations, resulting in low resource consumption and extremely short processing latency, which meets real-time requirements.
[0021] 4. Strong adaptability: Through multiple model libraries and offline optimization mechanisms, it can automatically adapt to different probes and different defect patterns (such as edge adhesion and dense corners), making it highly versatile.
[0022] 5. Optimize algorithm deployment for high efficiency: Deploy optimization algorithms efficiently to achieve low latency and low resource consumption in online processing, providing support for high-precision LUT construction.
[0023] 6. An iterative technological closed loop has been formed: the model library is scalable, and the optimization metrics are adjustable. As data accumulates, more effective models can be continuously added to the library, or evaluation metrics can be optimized, thereby continuously improving system performance. Attached Figure Description
[0024] Figure 1 This is a flowchart of a method for multi-model adaptive deformation and FPGA curing of PET probe over-field images provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the original Floodimage in an embodiment of the present invention; Figure 3 This is a schematic diagram of the Floodimage after global linear transformation in an embodiment of the present invention; Figure 4 This is a schematic diagram of the Floodimage after radial concave transformation in an embodiment of the present invention; Figure 5 This is a schematic diagram of the Floodimage after four-corner power transformation in an embodiment of the present invention; Figure 6 This is a comparison chart showing the distribution of the distance increase rate of three image coordinate transformation models in this invention for close point pairs with a vertical distance of less than 16 pixels. Detailed Implementation
[0025] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
[0026] This invention provides a method for multi-model adaptive deformation and FPGA (Field-Programmable Gate Array) solidification of PET probe floodimages. This method is used for floodimage preprocessing and event localization in PET detectors and is applicable to PET detector crystal arrays of various sizes. Its core lies in: establishing a library containing multiple image coordinate transformation models; determining the optimal model and parameters through offline analysis; and solidifying the final determined transformation logic into the FPGA for execution. This moves the time-consuming model selection and parameter optimization process offline, and solidifies the lightweight optimal model application process through hardware solidification. This allows the PET system detector image, even with some pixel adhesion, to more closely approximate the ideal situation of segmenting the crystal strip influence area, achieving efficient online processing.
[0027] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.
[0028] In this embodiment, a PET probe module containing a 15×15 crystal array is used for illustration, with an original over-field image resolution of 512×512 pixels. This image has an inherent structural characteristic: the top two and bottom two rows of crystals are too close together vertically, making them prone to event response adhesion. The evaluation is based on the lateral and vertical distances between the centroid coordinates of adjacent crystal strips, with an event loss rate L=0 and crystal strip adhesion pixels set to M<16 pixels.
[0029] Firstly, such as Figure 1 As shown, this embodiment of the invention provides a method for multi-model adaptive deformation and FPGA solidification of PET probe over-field images, including the following steps: Step 1: Collect single-lift event data from the PET probe; Step 2: Generate the original floodimage based on the single-event data; Step 3: Using multiple image deformation models from the pre-set multi-model transformation library, perform coordinate traversal transformations on the original Floodimage respectively; Specifically, the pre-set multi-model transformation library includes image coordinate transformation models with different deformation characteristics, and can seamlessly add high-order polynomial models, elastic deformation models based on radial basis functions (RBF), etc., to the model transformation library as needed. These models constitute a candidate set for offline optimization. Specifically, it includes the following three image coordinate transformation models: The first model is used to achieve globally isotropic / anisotropic linear stretching. The second model is used for nonlinear stretching of events that cause adhesion between the left and right regions of an image. The third model is used to enhance the stretching of images for the problem of dense events in the four corner regions.
[0030] The first model is used to perform a global linear scaling transformation on the pixel coordinates of the overlay image. Specifically, it is a global linear transformation model, the expression of which is: ; In the formula, and Here, x and y are the coordinates after the global linear transformation, respectively, while x and y are the original coordinates. and These are the linear stretching coefficients in the x and y directions, respectively.
[0031] The second model is used to perform radial concave transformation on the pixel coordinates of the overlay image. Specifically, it is a radial concave transformation model, the expression of which is: ; In the formula, and The coordinates are after radial concavity transformation. These are the enhancement coefficients in the y-direction, and These are the basic tensile coefficients in the x and y directions, respectively; 2 d The pixel side length of the Floodimage is matched logically with the coordinate transformation of the center point (0,0) to normalize the radial distance to a reasonable range; β for y Directional weighting coefficient, used for adjustment x and y The contribution of coordinates to radial distance.
[0032] The third model is used to perform a power transformation on the pixel coordinates of the overlay image, specifically a four-corner power transformation model, the expression of which is: ; In the formula, and The coordinates are after the four-angle power transformation. , These are the enhancement coefficients in the x and y directions, respectively. , These are the power parameters in the x and y directions, respectively. and These are the basic stretching coefficients in the x and y directions, respectively; d is half the side length of the Floodimage pixel, which is logically matched with the coordinate transformation of the center point (0,0) and is used to normalize the radial distance to a reasonable range; β is the weighting coefficient in the y direction, which is used to adjust the contribution of the x and y coordinates to the radial distance.
[0033] Step 4: Using preset evaluation indicators, perform offline analysis on each model and its parameters in the multi-model transformation library, and select the optimal model and optimal parameters; Among them, the original Floodimage generated from single-event data is analyzed offline using evaluation metrics to analyze the models in the multi-model transformation library. The evaluation metrics include one or more of the following: event loss rate L, identification results of adjacent crystal centroid proximity pairs, and distance increase rate G of proximity pairs. The evaluation metrics include the event loss rate L, the identification results of adjacent crystal centroid proximity pairs, and the distance increase rate G of proximity pairs. The event loss rate L, the identification results of adjacent crystal centroid proximity pairs, and the distance increase rate G of proximity pairs are specifically as follows: The event loss rate L (overflow control) is calculated as follows: ;in, This represents the total event count on the original Floodimage. This represents the total event count after the transformation; and L ≤ 1 / 100000 (configurable) to ensure that the transformed events do not exceed the original field of view.
[0034] The identification results of adjacent crystal centroid proximity point pairs (adhesion identification) are used to calculate the distance between all adjacent crystal centroids in the original Floodimage, and crystal centroid point pairs whose distance is less than the set crystal adhesion threshold M (e.g., 16 pixels) are marked as proximity point pairs; The rate of increase in distance G (improvement assessment) for the proximity point pairs is calculated as follows: ;in, , These represent the distances before and after the transformation, respectively.
[0035] In this embodiment, the combination of the above three evaluation indicators is used to perform offline analysis on the image coordinate transformation model. The model optimization process is as follows: traverse all models and their parameters in the preset multi-model library; firstly, screen all candidate combinations that meet the event loss rate L (L meets the standard); then, from the candidate combinations of event loss rate L, screen the combinations that can meet the recognition result constraint of adjacent crystal centroid proximity point pairs (the distance between all adjacent points is greater than M); finally, from the combinations that meet the first two conditions, screen the combination that maximizes the distance increase rate G of proximity point pairs as the optimal parameter.
[0036] Step 5: Transform the original Floodimage using the optimal model and optimal parameters to generate an enhanced image and a crystal location lookup table, and then save it to the FPGA.
[0037] Specifically, the optimal model and parameters are solidified into a lookup table as follows: the selected optimal model and parameters are applied to transform the original Flood image to generate an enhanced Flood image, and a high-precision crystal position lookup table is generated based on this enhanced Flood image. This lookup table provides crystal position information corresponding to the x and y coordinate ranges based on the Flood image.
[0038] The optimal image coordinate transformation model equations, corresponding optimal parameters, and generated crystal position lookup table are written into the FPGA in the form of a lookup table.
[0039] During online processing, the FPGA directly calls the fixed lookup table for coordinate mapping and event localization, achieving hardware acceleration, which can be used for floodimage preprocessing and event localization of PET detectors.
[0040] like Figure 2 The image shown is the original 512×512 integer coordinate diagram of sensor BK0_4. The first and last rows of the diagram are close to points A and B, with a centroid distance of 14 pixels and 16 pixels respectively, which are used for subsequent deformation comparison.
[0041] In this embodiment, 8 PET probes were used for testing, and only 2 of the 8 probes had a small lateral distance in the original Floodimage, while the distance between adjacent crystals was 16 pixels for 4 points.
[0042] Table 1 shows the optimal parameters for each of the three image coordinate transformation models obtained through multi-index evaluation, and the changes in the horizontal and vertical distances of the Floodimage reanalysis points after applying the optimal parameters.
[0043] Table 1. Data results of PET probes used in the tests
[0044] In Table 1, M0 represents the original distance between adjacent points, M1, M2, and M3 correspond to three transformation methods: global linear transformation, radial concave transformation, and four-corner power transformation, respectively, and G1-G3 correspond to the amplification rate.
[0045] like Figures 3 to 5 The figures shown are 512×512 integer coordinate graphs after deformations of M1, M2, and M3, respectively. The changes in the distance between points A and B near each other are shown in the figures, which can be used to visually compare the stretching effect of the three models.
[0046] The original Floodimage used in this embodiment has 158 vertically smaller points (less than 16 pixels) due to its inherent structural characteristics. The top two and bottom two rows are close together vertically, and the distance between them increases significantly after transformation using three image coordinate transformation models.
[0047] Figure 6 shows the distribution of the distance increase rate G for close point pairs with a vertical distance of <16 pixels. The horizontal axis represents the probe number, and the vertical axis represents G (%). The three curves correspond to M1, M2, and M3, respectively. As can be seen from the figure, M3 has the highest mean G value among all 158 close point pairs. Based on this, M3 can be automatically selected as the default image coordinate transformation model for this batch of PET probes.
[0048] The present invention provides a multi-model adaptive deformation and FPGA solidification method for PET probe over-field images. Through a three-level architecture of multi-model deformation library, multi-index offline optimization and FPGA online solidification, it significantly increases the spacing between close point pairs between crystals without destroying the statistical characteristics of integer coordinates, thereby improving the crystal sphere differentiation. It has no event loss, low processing latency, and low resource consumption, and is suitable for large-scale online calibration and production deployment of various PET detectors.
[0049] In a second aspect, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements the PET probe over-field image multi-model adaptive deformation and FPGA solidification method described in any of the first aspects above.
[0050] Thirdly, embodiments of the present invention provide a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements the PET probe over-field image multi-model adaptive deformation and FPGA solidification method described in any of the first aspects above.
[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0052] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.
[0053] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A method for multi-model adaptive deformation and FPGA-based curing of PET probe over-field images, characterized in that, Includes the following steps: Step 1: Collect single-lift event data from the PET probe; Step 2: Generate the original overlay image based on the single-event data; Step 3: Using the image deformation models in the pre-set multi-model transformation library, perform coordinate traversal transformations on the original overlay image respectively; Step 4: Using preset evaluation indicators, perform offline analysis on each model and its parameters in the multi-model transformation library, and select the optimal model and optimal parameters; Step 5: Transform the original overlay image using the optimal model and optimal parameters to generate an enhanced image and a crystal position lookup table, and then save it to the FPGA.
2. The PET probe over-field image multi-model adaptive deformation and FPGA solidification method according to claim 1, characterized in that, In step 2, the preset multi-model transformation library includes one or more image coordinate transformation models with different deformation characteristics, and these models constitute a candidate set for offline optimization.
3. The PET probe over-field image multi-model adaptive deformation and FPGA solidification method according to claim 2, characterized in that, The image coordinate transformation model is specifically as follows: The first model is a model that performs a global linear scaling transformation on the pixel coordinates of the overlay image, used to achieve global isotropic / anisotropic linear stretching. The second model is a model that performs radial concave transformation on the pixel coordinates of the overlay image, used for nonlinear stretching of events that are stuck in the left and right regions of the image. The third model is a power-law transformation of the pixel coordinates of the overlay image, used to enhance and stretch the image to address the problem of dense events in the four corner regions.
4. The method for multi-model adaptive deformation and FPGA curing of PET probe over-field images according to claim 1, characterized in that, The evaluation metrics in step 4 include a combination of one or more of the following metrics: event loss rate L, identification results of adjacent crystal centroid proximity pairs, and distance increase rate G of proximity pairs.
5. The PET probe over-field image multi-model adaptive deformation and FPGA solidification method according to claim 4, characterized in that, The event loss rate L, the identification results of adjacent crystal centroid proximity pairs, and the distance increase rate G of proximity pairs are specifically as follows: The event loss rate L is calculated as L = (N0 - N1) / N0; where N0 represents the total event count on the original overlay image and N1 represents the total event count after transformation. The identification results of adjacent crystal centroid proximity pairs are used to calculate the distance between all adjacent crystal centroids in the original overlay image, and crystal centroid point pairs whose distance is less than a set threshold M are marked as proximity point pairs. The distance increase rate G of the proximity pair is calculated as G = (M1 - M0) / M0; where M0 and M1 are the distances before and after the transformation, respectively.
6. The method for multi-model adaptive deformation and FPGA curing of PET probe over-field images according to claim 1, characterized in that, The model optimization process in step 4 is as follows: by traversing all models and their parameter combinations in the pre-set multi-model library through preset evaluation indicators, the optimal model and corresponding optimal parameter combination that meet the evaluation indicator constraints are selected.
7. The method for multi-model adaptive deformation and FPGA curing of PET probe over-field images according to claim 1, characterized in that, In step 5, the specific method of solidifying the optimal model and optimal parameters into a lookup table is as follows: the optimized optimal model and optimal parameters are applied to transform the original overlay image to generate an enhanced overlay image, and a crystal position lookup table is generated based on this enhanced image. The lookup table provides crystal position information corresponding to the x and y coordinate ranges according to the overlay image. The optimal image coordinate transformation model equations, corresponding optimal parameters, and the generated crystal position lookup table are written into the FPGA in the form of a lookup table.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the PET probe over-field image multi-model adaptive deformation and FPGA solidification method according to any one of claims 1 to 7.
9. A storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, characterized in that, When the processor executes the computer program, it implements the PET probe over-field image multi-model adaptive deformation and FPGA solidification method according to any one of claims 1 to 7.