CT-guided brain pet image spatial standardization system, method and device for the elderly

Abstract

The application discloses a CT-guided old brain PET image spatial standardization system, method and device, the system comprises a PET / CT image acquisition and format conversion module, a PET / CT preprocessing module, a PET / CT image coarse standardization module, a PET / CT image fine standardization module and an automatic SUV extraction module, the old brain PET molecular image spatial standardization is carried out by using a low-dose CT brain structure image as an auxiliary, and the SUV of a region of interest is automatically extracted by using a standard space brain atlas partitioning, the application adopts a two-step strategy from coarse to fine, and by means of an optimized process, an accurate, stable and user-friendly spatial standardization system is provided for the old brain PET image, and the application has important clinical application value and scientific research significance.

Description

Technical Field

[0001] This invention relates to the field of PET image registration, and more particularly to a CT-guided spatial standardization system, method, and apparatus for PET images of the brains of the elderly. Background Technology

[0002] Positron emission tomography (PET) is a cutting-edge molecular imaging technique that uses radiolabeled tracers to visualize the distribution of specific molecules in vivo, quantitatively reflecting physiological and pathological processes. PET technology plays a crucial role in scientific research and clinical diagnosis of common brain diseases in the elderly, such as Alzheimer's and Parkinson's diseases. However, current PET imaging diagnosis relies heavily on physician visual assessment, which is not only highly subjective but also prone to missed diagnoses and misdiagnoses.

[0003] To overcome these limitations, the automated extraction of standardized uptake values ​​(SUVs) or standardized uptake value ratios (SUVRs) of brain regions becomes crucial. This not only helps physicians make objective diagnoses and reduce reliance on subjective experience, but also enables the exploration of physiological and pathological changes in different brain regions in scientific research. The key to achieving this goal lies in the spatial standardization of PET images, that is, registering individual images to a standard brain template space.

[0004] Currently, there are two main methods for spatial normalization of PET images: brain structural image-assisted spatial normalization and PET brain template-based spatial normalization. Brain structural images include magnetic resonance imaging (MRI) structural images (such as T1-weighted and T2-weighted images) and computed tomography (CT) structural images. The brain structural image-assisted method typically involves three steps: first, rigidly registering the individual PET image to the individual's structural brain image; second, nonlinearly registering the individual's structural brain image to a standard brain structural template; and finally, applying the nonlinear deformation field generated in the previous step to register the PET image to a standard space. The PET brain template-based method involves constructing a standard PET brain template for a specific population, and then nonlinearly registering the individual PET image to the template.

[0005] High-resolution MRI provides precise information about brain anatomy and is considered the gold standard for PET normalization. While PET-based brain template methods are computationally fast, they require constructing standard PET brain templates for specific populations and can introduce significant registration errors when processing highly variable PET images. On the other hand, CT-based PET normalization methods, due to the resolution limitations of CT images in depicting brain anatomy, typically require further optimization to approximate the results of MRI.

[0006] Previous research has proposed a brain PET normalization method based on low-dose CT (Source: Presotto, Luca et al. "Low-dose CT for the spatial normalization of PET images: A validation procedure for amyloid-PET semi-quantification." NeuroImage. Clinical vol. 20153-160. 19 Jul. 2018, doi:10.1016 / j.nicl.2018.07.013). This method follows the idea of ​​structure-based registration, finding the optimal transformation from low-dose CT images to standard space to achieve spatial normalization of PET images registered with individual CT scans. This method improves the normalization effect of low-dose CT images through a "cleaning" procedure and an optimized SPM12 (Statistical Parametric Mapping 12) segmentation algorithm. Specifically, the method first performs a "cleaning" procedure, setting all values ​​below -300 HU to -1024 HU to avoid the influence of low-density structures outside the head on the algorithm. The preprocessed CT images are then loaded into the optimized SPM12 segmentation algorithm. In SPM12 segmentation, the optimized parameters include disabling skew field correction and using only one Gaussian distribution for the Gaussian mixture model of gray and white matter; for the other four tissue types (cerebrospinal fluid, bone, external tissue, and air), two Gaussian distributions are used. A spatially normalized deformation field is generated from CT tissue segmentation and applied to the PET image registered with CT, thus achieving spatial normalization of the PET image. However, this method may experience decreased registration accuracy when processing PET images of elderly patients due to the use of tissue probability maps (TPM) from healthy young adults, especially in elderly patients with significant brain atrophy. Furthermore, directly performing tissue segmentation on the individual CT image to obtain a normalized deformation field may also produce unstable results.

[0007] Therefore, in PET / CT scanning systems, for elderly brain patients with only low-dose CT structural images, developing an accurate and stable spatial normalization method for PET images of the elderly brain is particularly urgent. This method utilizes low-dose CT images, not only avoiding the need for additional MRI scans but also improving the accuracy and stability of the normalization process, which is of great significance for both clinical practice and scientific research.

[0008] Current PET standardization techniques have several limitations, which are particularly evident in clinical applications. First, standardization methods based on MRI structural brain images require patients to undergo additional high-quality MRI scans within a short period. This not only increases the financial burden on patients but also, because PET and MRI scans are typically not performed on the same equipment when PET / MRI is unavailable, this non-integrated scanning approach may reduce the accuracy of registration between PET images and individual MRI images, thus affecting the reliability of the standardization results. Furthermore, in clinical practice, it is common to encounter PET images without matching high-resolution MRI images, which limits the use of MRI structural image-assisted spatial standardization procedures.

[0009] Standardization methods based on PET brain templates require specific standard PET brain templates. This process necessitates collecting a large number of PET images of similar probes to construct PET brain templates for a specific population, a requirement that is often difficult to achieve. Furthermore, since PET images of the same probe vary significantly among patients in different disease states, there may be substantial differences between individual PET images and the constructed standard brain templates, which limits the accuracy and stability of this method in spatial standardization.

[0010] Existing studies have proposed low-dose CT-assisted PET registration methods, demonstrating that low-dose CT can also assist in spatial normalization of PET images, but these methods still have limitations. In particular, these methods use a tissue probability map (TPM) template based on healthy young adults. This template space differs significantly from the brain structure of elderly patients with significant lobar atrophy, easily leading to decreased registration accuracy and making it unsuitable for processing images of such elderly patients. Furthermore, one-step spatial normalization methods that directly perform tissue segmentation and registration on CT images in individual spatial dimensions are prone to getting trapped in local optima during algorithmic optimization to find the optimal registration nonlinear deformation field, leading to spatial normalization failure and affecting the consistency and reliability of the PET image normalization process. Summary of the Invention

[0011] The purpose of this invention is to address the shortcomings of existing technologies by proposing a CT-guided spatial standardization system and method for PET images of the elderly brain. This invention utilizes low-dose CT brain structure images acquired in the same bed position within a PET / CT scanning system to achieve accurate and stable spatial standardization of PET images of elderly brains exhibiting ventricular enlargement and lobar atrophy. This technique does not require MRI; therefore, MRI scans are unnecessary for annotation of PET images acquired on a PET / CT scanner. It also eliminates the need to construct population-specific PET brain templates, overcoming the problem of insufficient specificity in PET brain templates constructed with small case volumes. The low-dose CT images used in this technique can be scanned in the same bed position as the PET images within the PET / CT scanning system, ensuring the accuracy of initial individual spatial registration between PET and CT images. By employing an improved standardization method, this invention effectively overcomes the problems of spatial standardization accuracy and stability caused by insufficient contrast in CT brain tissue images. Especially for elderly patients, this technology can adapt to the special circumstances of ventricular enlargement and lobar atrophy, providing a more accurate and stable spatial standardization solution for PET images.

[0012] In summary, this invention aims to solve the following technical problems:

[0013] 1. Provide accurate and stable spatial normalization for PET images of the brains of elderly patients with enlarged ventricles and atrophied cerebral lobes.

[0014] 2. Using low-dose CT images to guide PET image standardization avoids the need for additional MRI scans and reduces the burden on patients.

[0015] 3. Overcome the limitations of the PET standard brain template method in cases with a small number of cases, and provide a PET spatial standardization system with high universality.

[0016] 4. To address the issue of insufficient accuracy and stability in the standardization process caused by low contrast in brain tissue from CT images, the process is optimized to improve the accuracy and stability of standardization.

[0017] The objective of this invention is achieved through the following technical solution: Firstly, this invention proposes a CT-guided spatial normalization system for PET images of the brains of the elderly, the system comprising:

[0018] The PET / CT image acquisition and format conversion module is used to acquire paired PET and CT images of elderly patients and perform format conversion processing.

[0019] The PET / CT preprocessing module is used to perform rigid transformation preprocessing on PET and CT images after format conversion.

[0020] The PET / CT image coarse normalization module is used to optimize the nonlinear deformation field through nonlinear registration and apply it to rigidly transformed PET and CT images to generate coarsely normalized PET and CT images.

[0021] The PET / CT image fine normalization module is used to classify tissues in coarsely registered CT images and align them with the tissue probability map in the standard space of the elderly brain. It updates the voxel classification probability, optimizes the iterative nonlinear deformation field, and applies it to coarsely registered CT and PET images to generate finely normalized CT and PET images.

[0022] The automated SUV extraction module is used to extract SUV values ​​of cortical and subcortical ROIs from finely normalized PET images using MIITRA spatial brain atlases, for subsequent SUVR calculations and disease diagnostic analysis.

[0023] Furthermore, after PET / CT image acquisition, the images are converted to NIFTI format, and the standardized uptake value (SUV) is calculated for each voxel of the PET data.

[0024] Furthermore, the preprocessing process includes PET image resampling to match the voxel resolution of the CT images, and rigid transformation of the CT images and the resampled PET images.

[0025] Furthermore, the PET / CT image coarse normalization module uses the Old Normalize method to normalize the rigidly transformed CT image onto the CT brain template image, and then performs nonlinear registration. Through optimization and regularization, the nonlinear deformation field is obtained.

[0026] Furthermore, the PET / CT image fine normalization module uses a Gaussian mixture model to classify tissues in coarsely registered CT images, updates voxel classification probabilities based on spatial prior information and CT image segmentation probabilities, and iterates the nonlinear deformation field through optimization and regularization.

[0027] Furthermore, the automated SUV extraction module uses a brain atlas in the MIITRA space to segment the finely normalized PET image into multiple brain regions, and calculates the average SUV value for each brain region separately.

[0028] Secondly, the present invention also provides a CT-guided spatial normalization method for PET images of the brains of the elderly, the method comprising the following steps:

[0029] (1) Acquire paired PET and CT images of elderly patients and perform format conversion processing;

[0030] (2) Perform rigid transformation preprocessing on the converted PET and CT images;

[0031] (3) Optimize the nonlinear deformation field through nonlinear registration and apply it to the rigidly transformed PET and CT images to generate coarsely normalized PET and CT images;

[0032] (4) Perform tissue classification on the coarsely registered CT images and align them with the tissue probability map of the standard space of the elderly brain. Update the voxel classification probability, and then optimize the iterative nonlinear deformation field. Apply it to the coarsely registered CT and PET images to generate finely standardized CT and PET images.

[0033] (5) For finely standardized PET images, brain atlases in MIITRA space are used to extract SUV values ​​of cortical and subcortical ROIs for subsequent SUVR calculation and disease diagnosis analysis.

[0034] Thirdly, the present invention also provides a CT-guided spatial standardization device for PET images of the brain in the elderly, including a memory and one or more processors. The memory stores executable code, and when the processor executes the executable code, it implements the CT-guided spatial standardization method for PET images of the brain in the elderly.

[0035] Fourthly, the present invention also provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the aforementioned CT-guided spatial standardization method for PET images of the elderly brain.

[0036] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned CT-guided spatial standardization method for PET images of the elderly brain.

[0037] The beneficial effects of this invention are:

[0038] 1) Simplified scanning process and reduced patient burden: Utilizing low-dose CT images to guide the standardization of PET images avoids additional MRI scans, simplifying the scanning process and reducing the economic and physical burden on patients. 2) Improved stability and accuracy of the standardization process: This invention, through an optimized standardization method guided by low-dose CT images, improves registration accuracy, overcoming the registration accuracy problems caused by PET and MRI scans being performed on different beds. It effectively solves the accuracy and stability problems in the standardization process caused by low contrast of brain tissue in CT images, especially in elderly patients with enlarged ventricles and lobar atrophy.

[0039] 3) Solving the problem of small sample size: This invention avoids the problem of insufficient specificity of PET standard brain templates due to small case size, and provides a standardization method that is still effective when the sample size is limited.

[0040] 4) Highly adaptable and suitable for elderly patients: This invention is specifically designed for the characteristics of brain PET images of elderly patients. It is applicable to various types of brain PET probes, and is not limited to a single probe. It can adapt to special cases of ventricular enlargement and cerebral lobe atrophy, providing a more accurate and stable spatial standardization solution. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a structural framework diagram of a low-dose CT-based spatial normalization system for PET images of the elderly brain, according to an embodiment of the present invention.

[0043] Figure 2 This is a flowchart illustrating the implementation of a spatial normalization system for PET images of the elderly brain based on low-dose CT, according to an embodiment of the present invention.

[0044] Figure 3 This is a schematic diagram of rigid registration in a PET / CT image preprocessing module according to an embodiment of the present invention.

[0045] Figure 4 This is a schematic diagram of coarse normalization of PET / CT images according to an embodiment of the present invention.

[0046] Figure 5 This is a schematic diagram of fine normalization of a PET / CT image according to an embodiment of the present invention.

[0047] Figure 6 This is a schematic diagram of SUV automated calculation according to an embodiment of the present invention.

[0048] Figure 7 This is a correlation graph between the results of one embodiment of the present invention and other different standardization methods, and the gold standard. The horizontal axis represents the SUV values ​​of brain regions obtained by the spatial standardization gold standard method based on high-quality MRI, and the vertical axis represents the SUV values ​​of the same brain region extracted by three PET workflows. Here, CT represents the workflow system proposed in this invention; Template represents the standardization workflow based on PET brain templates; and CT_MNI_TPM represents replacing the tissue probability map with MNI spatial TPM in the fine standardization module based on this system. The closer the slope of the scatter plot is to 1, and the closer the Pearson correlation coefficient r is to 1, the closer the method is to the gold standard method.

[0049] Figure 8This is a comparison of the results and accuracy of different spatial normalization methods for a single PET scan in one embodiment of the present invention. The images in this example are derived from PET scans of a patient with multisystem atrophy who has significant striatal dopaminergic loss. Image a shows the gold standard method for spatial normalization based on high-quality MRI, image b shows the workflow system proposed in this invention, image c shows the method based on PET brain templates, and image d shows the replacement of MIITRA spatial TPM with MNI spatial TPM in fine spatial normalization. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are merely illustrative and not intended to limit the invention.

[0051] like Figure 1 and 2 This invention proposes a CT-guided spatial normalization method for PET images of the elderly brain. This method employs a two-step strategy, from coarse to fine, to guide PET image registration to a standard space using CT images, thereby achieving fully automated and accurate extraction of the SUV (Radio Functional Area) of the region of interest (ROI). The system comprises five modules: 1. PET / CT image acquisition and format conversion module; 2. PET / CT preprocessing module; 3. PET / CT image coarse normalization module; 4. PET / CT image fine normalization module; and 5. Automated SUV extraction module.

[0052] I. PET / CT Image Acquisition and Format Conversion Module

[0053] The image acquisition and format conversion module is used to acquire paired PET and CT images of elderly patients and convert them into NIFTI data format suitable for subsequent processing.

[0054] Implementation steps:

[0055] 1. Three-dimensional head PET molecular images and low-dose CT images of elderly patients were acquired using a PET / CT scanner. During acquisition, subjects were kept in the same bed position to ensure accurate registration of PET and CT images;

[0056] 2. PET and CT images acquired by PET / CT scanners are typically stored in DICOM (Digital Imaging and Communications in Medicine) format. For easier subsequent processing, DICOM images are converted to NIFTI format.

[0057] 3. Calculate the standardized uptake value (SUV) for each voxel of PET data, using the following formula:

[0058]

[0059] II. PET / CT Preprocessing Module

[0060] The PET / CT preprocessing module is used to preprocess SUV PET and CT images converted to NIFTI format.

[0061] Implementation steps:

[0062] 1. Resample the PET images to match the voxel resolution of the CT images. Assume the voxel resolution of the CT images is Δx. CT The voxel resolution of the PET image is Δx PET The resampling process ensures Δx PET =Δx CT ;

[0063] 2. For example Figure 3 As shown, using a CT template image as a reference, rigid transformations are performed step-by-step on the CT image and the resampled PET image. Rigid transformations include rotation, translation, and scaling, and their mathematical representation is as follows:

[0064] x′=R·x+t

[0065] Where x is the original coordinate, x′ is the transformed coordinate, R is the rotation matrix, and t is the translation vector.

[0066] The standard CT template image is from a published brain template (Rorden, Christopher et al. “Age-specific CT and MRI templates for spatial normalization.” NeuroImage vol. 61, 4(2012): 957-65. doi:10.1016 / j.neuroimage.2012.03.020).

[0067] III. PET / CT Image Coarse Normalization Module

[0068] like Figure 4As shown, the PET / CT image coarse normalization module is used to coarsely normalize the rigidly transformed PET and CT images. This process yields coarsely normalized CT and PET images, which are used as input for the next step of fine normalization. At this point, the coarsely normalized CT image has better similarity to the standard space to be registered compared to the original CT image, thus providing a stable iterative starting point for registration optimization. This facilitates finding the global optimum during registration optimization and prevents getting trapped in local optima during the fine normalization process. This module is implemented based on the Old Normalize algorithm in the MATLAB brain image processing tool SPM12 (Statistical Parametric Mapping 12). The specific steps and principles are as follows:

[0069] Implementation steps:

[0070] 1. Using the Old Normalize method, the rigidly transformed CT images are normalized to the CT brain template images, and further nonlinear registration is performed:

[0071] x″=x′+φ(x′)

[0072] Where φ(x′) is the nonlinear deformation field used to describe more refined local deformation; x′ is the coordinate after rigid transformation, and x″ is the coordinate after coarse normalization.

[0073] 2. Optimization and Regularization: The nonlinear transformation parameters are solved using the Levenberg-Marquardt optimization algorithm to maximize the match between the registered image I(x″) and the CT standard template image T(x). The objective function can be expressed as:

[0074]

[0075] Where R(φ) is the regularization term used to constrain the smoothness of the nonlinear deformation field, and λ is the regularization parameter;

[0076] 3. Apply the generated nonlinear deformation field to the rigidly transformed PET and CT images to generate coarsely normalized PET and CT images:

[0077] I PET,norm (x)=I PET (x″)

[0078] I CT,norm (x)=I CT (x″)

[0079] Among them, I PET,norm (x) and I CT,norm (x) are the PET and CT images obtained after coarse normalization, respectively.PET and I CT These are PET and CT images with rigid transformations, respectively.

[0080] IV. PET / CT Image Fine Normalization Module

[0081] like Figure 5 As shown, the PET / CT image fine normalization module performs a more refined normalization on the coarsely registered PET and CT images. This module is implemented based on the New Segment algorithm in the MATLAB brain imaging toolkit SPM12. The input image to be segmented and registered in this step comes from the coarsely normalized CT image from the previous step. This step outputs a finely normalized spatial nonlinear deformation field. This deformation field is applied to the coarsely normalized CT and PET images, thereby generating finely normalized CT and PET images registered to the MIITRA standard space of the aged brain.

[0082] The fine standardization employed the MIITRA standard space for aged brains, improving the similarity between the source and reference images during registration. The two-step standardization strategy, from coarse to fine, avoided getting trapped in local optima during iterative optimization; the coarse standardization result provided a stable starting point for the fine standardization iteration, optimizing the registration process. These factors collectively enhanced the stability of spatial standardization for images of aged brains with brain atrophy.

[0083] Implementation steps:

[0084] 1. Tissue classification is performed on coarsely registered CT images using optimized segmentation parameters. Due to the low contrast of CT scans for brain tissue, a Gaussian mixture model is employed. One Gaussian distribution is used for gray and white matter, while two Gaussian distributions are used for cerebrospinal fluid, bone, external tissues, and air.

[0085]

[0086] in, μ represents a Gaussian distribution. k and σ kThese are the mean and standard deviation of brain tissue category k, respectively. Specifically, the choice of Gaussian parameters is based on the following: In CT imaging, gray matter and white matter each have only one average intensity value, therefore one Gaussian function is used. For cerebrospinal fluid, since cerebrospinal fluid near bone may have a higher intensity due to spillage effects, two Gaussian functions are used. Because bone has a very wide value range in CT, two Gaussian functions are also used. External tissue can be considered as consisting of regions such as low-intensity fat and high-intensity muscle, therefore two Gaussian functions are also used. In the air category, one Gaussian function is used to model most of the -1024 HU, and another Gaussian function is used to model other low-intensity pixels with intensities between approximately -50 and -300 HU.

[0087] 2. Align the coarsely registered CT images with the tissue probability map (TPM) of the Multichannel Illinois Institute of Technology and Rush University Aging (MIITRA) standard space for aged brains, and update the voxel classification probabilities. This process incorporates spatial prior information P. k (x) and the segmentation probability p(k|I(x)) of the CT image;

[0088] p(k|I(x),x)∝p(I(x)|k)·P k (x)

[0089] Among them, P k (x) is the spatial prior probability of category k. The tissue probability map TPM of the standard spatial MIITRA of the elderly brain is derived from the published study (Niaz, Mohammad Rakeen et al. "Development and evaluation of a high resolution 0.5mm isotropic T1-weighted template of the older adult brain." NeuroImage vol.248(2022):118869.doi:10.1016 / j.neuroimage.2021.118869);

[0090] 3. Optimize the nonlinear deformation field φ(x) by minimizing the objective function:

[0091]

[0092] Among them, T k(x+φ(x)) is the probability image of class k in the template image, R(φ) is the regularization term used to constrain the smoothness of the deformation field, and λ is the regularization parameter;

[0093] 4. Regularization term R(φ): The regularization term is used to constrain the smoothness and continuity of the deformable field. The regularization form is:

[0094]

[0095] This regularization term ensures the continuity and smoothness of the deformation field by suppressing the magnitude of the deformation field gradient.

[0096] 5. Using the Levenberg-Marquardt optimization algorithm, the deformation field is iteratively optimized.

[0097]

[0098] Where α is the learning rate. It is the gradient of the objective function with respect to the deformation field;

[0099] 6. Apply the finely normalized nonlinear deformation field to the coarsely registered CT and PET images to generate finely normalized CT and PET images:

[0100] I PET,fine (x)=I PET,norm (x+φ(x))

[0101] I CT,fine (x)=I CT,norm (x+φ(x))

[0102] V. Automated SUV Extraction Module

[0103] like Figure 6 As shown, the automated SUV extraction module is used to extract SUV values ​​of cortical and subcortical ROIs from finely standardized PET images using MIITRA spatial brain atlases, for subsequent SUVR calculations and disease diagnostic analysis.

[0104] Implementation steps:

[0105] 1. Using MIITRA spatial brain mapping to refine the normalized PET images PET,fine Divided into multiple brain regions;

[0106] 2. For each brain region, calculate the average SUV value:

[0107]

[0108] Where N is the number of voxels in the brain region, SUV i This is the SUV value for each voxel.

[0109] Example

[0110] In one embodiment of this system, spatial normalization of dopamine transporter (DAT) PET / CT images of elderly patients with multiple system atrophy is achieved. Specifically, in the PET / CT image acquisition and format conversion module, images of 50 elderly patients with multiple system atrophy are acquired using a PET / CT scanner. 18 FN-(3-fluoropropyl)-2β-methyl ester-3β-(4′-iodophenyl)desmethyltropane (FP-β-CIT)- 18 F-FP-β-CIT PET / CT images, using a tracer for DAT. The PET images are acquired at 400×400×148 voxels with a resolution of 1×1×1.5 mm. 3 CT images of the same patient were acquired at the same bed position. The image size was 512×512×148 voxels, and the resolution was 0.586×0.586×1.5mm. 3 After the image set is completed, the original DICOM format CT images are converted to NIFTI format for easier subsequent analysis. Based on the patient scan parameters, weight, and injection dose information in the DICOM header file of the PET images, NIFTI format SUV PET images are calculated and generated. In the PET / CT image preprocessing module, linear interpolation is performed on the PET and CT images with inconsistent resolutions to ensure that the two images match in matrix size and resolution. Subsequently, rigid transformations are performed on the PET and CT images to align them with the CT brain template. In the PET / CT coarse normalization module, the OldNormalize tool in SPM12 is used to non-linearly normalize the CT images to the CT brain template. The annotated image size is selected as 181×217×181 voxels, and the resolution is 1×1×1 mm. 3 To encompass the entire brain and reduce unwanted background, a nonlinear deformation field was generated and applied to the PET images. The images were then processed in the PET / CT fine normalization module, using the Segment tool of SPM12 to select segmentation parameters suitable for low-dose CT. A single Gaussian distribution was used for gray and white matter, while a double Gaussian distribution was used for cerebrospinal fluid, bone, external tissues, and air. Simultaneously, the tissue probability map TPM of the standard spatial MIITRA model of the aged brain was selected as the reference template for segmentation. Tissue segmentation was performed on the coarsely registered CT images, and a forward deformation field was generated. This field was then applied to the coarsely normalized PET and CT images to obtain finely normalized PET and CT images with a size of 181×217×181 voxels and a resolution of 1×1×1 mm. 3Finally, the automated SUV extraction module is used, employing a 1×1×1mm space within the MIITRA area. 3 Brain atlases analyze finely normalized PET images to extract SUV values ​​for specific brain regions (such as the striatum and occipital cortex). These SUV values ​​will be used for subsequent SUVR calculations and disease diagnostic analysis.

[0111] To fully evaluate the beneficial effects of the system of the present invention, the present invention will be compared with that based on 18 A comparative analysis was conducted on traditional brain template methods using F-FP-β-CIT PET. Simultaneously, the suitability of the MIITRA standard space used in this invention was assessed; specifically, the tissue probability map was replaced with a tissue probability map from the Montreal Neurological Institute (MNI) space within the fine normalization module, and a comparative analysis was performed. The comparative analysis of this invention used high-quality MRI-based spatial normalization methods as the gold standard. In the PET-based brain template method, the brain template was constructed using an average of 50 MRI-guided spatially normalized DAT-PET images. In the comparative analysis, given that DAT-PET image analysis focuses on the putamen and caudate nucleus of the striatum, and often uses the occipital cortex as a reference region to calculate SUVR, we paid particular attention to the accuracy of SUV value extraction from these key brain regions. Figure 7 As shown, the correlation plot reveals the linear relationship between each method and the gold standard. The closer the slope of the line is to 1, the closer the Pearson correlation coefficient is to 1, indicating a better fit between the method and the gold standard. The comparative results clearly show that the system proposed in this invention performs best in terms of correlation with the gold standard, with the smallest error. Specifically, as... Figure 8 As shown, when dealing with individuals with significant striatal dopaminergic impairment, the method of this invention maintains an accuracy as high as 0.95, while the accuracy of the template-based method drops to 0.78. This significant difference can be attributed to the fact that in individuals with substantial striatal dopaminergic loss, the changes in brain structure reflected by CT images are not significant, while the differences between DAT-PET images and DAT-PET brain templates become particularly significant. During the iterative calculation of the nonlinear deformation field, the template-based method is prone to calculation errors due to the large differences between the source and reference images. Furthermore, after replacing the MIITRA TPM with the MNI TPM, the accuracy of SUV extraction also decreased to 0.86, indicating that the use of a standard space suitable for elderly brain images in the fine spatial normalization of this system is beneficial to improving the accuracy of spatial normalization.

[0112] Corresponding to the aforementioned embodiment of a CT-guided spatial normalization system for PET images of the brain in the elderly, this invention also provides an embodiment of a CT-guided spatial normalization method for PET images of the brain in the elderly. For the specific implementation of each step of this method, please refer to the specific process of the aforementioned embodiment of a CT-guided spatial normalization system for PET images of the brain in the elderly. The steps of this method are as follows:

[0113] (1) Acquire paired PET and CT images of elderly patients and perform format conversion processing;

[0114] (2) Perform rigid transformation preprocessing on the converted PET and CT images;

[0115] (3) Optimize the nonlinear deformation field through nonlinear registration and apply it to the rigidly transformed PET and CT images to generate coarsely normalized PET and CT images;

[0116] (4) Perform tissue classification on the coarsely registered CT images and align them with the tissue probability map of the standard space of the elderly brain. Update the voxel classification probability, and then optimize the iterative nonlinear deformation field. Apply it to the coarsely registered CT and PET images to generate finely standardized CT and PET images.

[0117] (5) For finely standardized PET images, brain atlases in MIITRA space are used to extract SUV values ​​of cortical and subcortical ROIs for subsequent SUVR calculation and disease diagnosis analysis.

[0118] Corresponding to the aforementioned embodiment of a CT-guided spatial normalization method for PET images of the brain in the elderly, the present invention also provides an embodiment of a CT-guided spatial normalization device for PET images of the brain in the elderly.

[0119] This invention provides a CT-guided spatial standardization device for PET images of the brain in the elderly, comprising a memory and one or more processors. The memory stores executable code, and when the processor executes the executable code, it implements a CT-guided spatial standardization method for PET images of the brain in the elderly as described in the above embodiment.

[0120] The embodiment of the CT-guided spatial standardization device for PET images of the elderly brain provided by this invention can be applied to any device with data processing capabilities, such as a computer. The device embodiment can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of the device with data processing capabilities reading the corresponding computer program instructions from non-volatile memory into memory and executing them. From a hardware perspective, the hardware structure diagram of any device with data processing capabilities for the CT-guided spatial standardization device for the elderly brain PET images provided by this invention includes, in addition to the processor, memory, network interface, and non-volatile memory, other hardware components may also be included depending on the actual function of the device, which will not be elaborated further.

[0121] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0122] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0123] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements a CT-guided spatial normalization method for PET images of the elderly brain in the above embodiments.

[0124] The computer-readable storage medium can be an internal storage unit of any data processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device of any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of any data processing device. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0125] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned CT-guided spatial standardization method for PET images of the elderly brain.

[0126] The above embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A CT-guided spatial normalization system for PET images of the brain in the elderly, characterized in that, The system includes: The PET / CT image acquisition and format conversion module is used to acquire paired PET images and low-dose CT images of elderly patients and perform format conversion processing. The PET / CT preprocessing module is used to perform rigid transformation preprocessing on PET and CT images after format conversion. The PET / CT image coarse normalization module is used to optimize the nonlinear deformation field through nonlinear registration and apply it to rigidly transformed PET and CT images to generate coarsely normalized PET and CT images. A fine-normalization module for PET / CT images is used to classify tissues in coarsely registered CT images using a Gaussian mixture model. One Gaussian distribution is used for gray and white matter, while two Gaussian distributions are used for cerebrospinal fluid, bone, external tissues, and air. The module aligns with the tissue probability map in the standard space of the aged brain. Based on spatial prior information and the segmentation probability of the CT image, the voxel classification probability is updated. Then, through optimization and regularization of the iterative nonlinear deformation field, it is applied to the coarsely registered CT and PET images to generate fine-normalized CT and PET images registered to the MIITRA standard space of the aged brain. This improves the similarity between the source and reference images during registration and enhances the stability of spatial normalization in images of the aged brain with brain atrophy. Specifically, the implementation is as follows: 1) Tissue classification is performed on coarsely registered CT images using optimized segmentation parameters: in, Indicates a Gaussian distribution. and These are brain tissue categories The mean and standard deviation of the Gaussian parameters are used. Specifically, the selection of the Gaussian parameters is based on the following: In CT imaging, gray matter and white matter each have only one average intensity value, so one Gaussian function is used; for cerebrospinal fluid, two Gaussian functions are used because cerebrospinal fluid near bones may have a higher intensity due to the overflow effect; two Gaussian functions are also used because the value range of bones in CT is very wide; external tissues are considered to be composed of fat and muscle regions, so two Gaussian functions are also used; in the air class, one Gaussian function is used to model most of the -1024 HU, and another Gaussian function is used to model other pixels with intensities between -50 and -300 HU. 2) Align the coarsely registered CT images with the tissue probability map (TPM) of the standard spatial MIITRA model of the elderly brain, and update the voxel classification probabilities. This process incorporates spatial prior information. Segmentation probability of CT images ; in, It is a category Spatial prior probability; 3) Optimize the nonlinear deformation field by minimizing the objective function. : in, It is a category in the template image The probability graph, It is a regularization term used to constrain the smoothness of the deformation field. It is a regularization parameter; 4) Regularization term The regularization term is used to constrain the smoothness and continuity of the deformable field; the regularization form is: This regularization term ensures the continuity and smoothness of the deformation field by suppressing the magnitude of the deformation field gradient. 5) Using an optimization algorithm, the Levenberg-Marquardt iterative optimization of the deformation field: in, It's the learning rate. It is the gradient of the objective function with respect to the deformation field; 6) Apply the finely normalized nonlinear deformation field to the coarsely registered CT and PET images to generate finely normalized CT and PET images: The automated SUV extraction module is used to extract SUV values ​​of cortical and subcortical ROIs from finely normalized PET images using MIITRA spatial brain atlases, for subsequent SUVR calculations and disease diagnostic analysis.

2. The spatial normalization system for CT-guided PET images of the brain in the elderly according to claim 1, characterized in that, After PET / CT images are acquired, they are converted to NIFTI format, and the normalized uptake value (SUV) is calculated for each voxel of the PET data.

3. The spatial normalization system for CT-guided PET images of the brain in the elderly according to claim 1, characterized in that, The preprocessing process includes PET image resampling to match the voxel resolution of the CT images and rigid transformation of the CT images and the resampled PET images.

4. The spatial normalization system for CT-guided PET images of the brain in the elderly according to claim 1, characterized in that, The PET / CT image coarse normalization module uses the Old Normalize method to normalize the rigidly transformed CT image onto the CT brain template image, and then performs nonlinear registration. The nonlinear deformation field is obtained through optimization and regularization.

5. A CT-guided spatial normalization system for PET images of the brain in the elderly according to claim 1, characterized in that, The automated SUV extraction module uses a brain atlas in MIITRA space to segment finely normalized PET images into multiple brain regions and calculates the average SUV value for each brain region.

6. A method for spatial normalization of PET images of the elderly brain based on the CT-guided spatial normalization system for PET images of the elderly as described in any one of claims 1-5, characterized in that, The method includes the following steps: (1) Acquire paired PET and CT images of elderly patients and perform format conversion processing; (2) Perform rigid transformation preprocessing on the converted PET and CT images; (3) The nonlinear deformation field is optimized by nonlinear registration and applied to the rigidly transformed PET and CT images to generate coarsely normalized PET and CT images; (4) Perform tissue classification on the coarsely registered CT images and align them with the tissue probability map of the standard space of the elderly brain. Update the voxel classification probability, and then optimize the iterative nonlinear deformation field. Apply it to the coarsely registered CT and PET images to generate finely standardized CT and PET images. (5) For finely standardized PET images, brain atlases in MIITRA space are used to extract SUV values ​​of cortical and subcortical ROIs for subsequent SUVR calculation and disease diagnosis analysis.

7. A CT-guided spatial normalization device for PET images of the brain in the elderly, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that, When the processor executes the executable code, it implements the spatial standardization method for CT-guided PET images of the elderly brain as described in claim 6.

8. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the spatial standardization method for CT-guided PET images of the elderly brain as described in claim 6.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the spatial standardization method for CT-guided PET images of the elderly brain as described in claim 6.

Images