Nuclear medicine functional imaging analysis method and system for neurological diseases

Abstract

The application relates to the technical field of medical image processing, in particular to a nuclear medical functional imaging analysis method and system for nervous system diseases. The method comprises the following steps: acquiring a three-dimensional volume image; dividing the three-dimensional volume image to obtain initial ROI blocks and non-initial ROI blocks; expanding the initial ROI region according to the change of the SUV value of the voxel points between the initial ROI blocks and the surrounding non-initial ROI blocks to obtain a target region; obtaining an SUV distribution factor according to the SUV value of the voxel points in each target region, the distance of the voxel points to the minimum enclosing volume and the gradient value; obtaining a consistency factor by combining the joint trend of the deviation of the SUV value of the voxel points in each target region relative to the whole and the deviation of the CT value relative to the whole, the dispersion of the SUV value of the voxel points and the dispersion of the CT value, and screening a suspected abnormal region in combination with the SUV distribution factor. The application improves the accuracy of the screening result of the suspected abnormal region.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to a nuclear medical functional imaging analysis method and system for nervous system diseases. BACKGROUND

[0002] Nervous system diseases (such as brain tumors, Alzheimer's disease, etc.) seriously threaten human health, and the pathogenesis is complex, the disease course is occult, and is often accompanied by complex metabolic abnormalities, neurotransmitter disorders and abnormal activity of brain function areas. Nuclear medical functional imaging (positron emission tomography (PET) and single photon emission computed tomography (SPECT)) can reflect brain metabolism, blood perfusion and neurotransmitter distribution in real time and dynamically through radioactive tracers, and can realize the positioning, activity determination and metabolic heterogeneity analysis of various brain lesions, and is an important tool for early screening, pathological mechanism research, efficacy evaluation and prognosis prediction of nervous system diseases.

[0003] Although nuclear medical imaging is widely used in the field of nervous system diseases at present, there are still some problems in the imaging data analysis method and lesion information extraction, for example, in 18F-FDG PET-CT brain imaging, postoperative brain tissue, inflammation, and reactive lesions after radiotherapy often have metabolic increase phenomenon, which shows similar high uptake to tumor active area in the SUV value level. The traditional method cannot realize reliable distinction only through SUV value, resulting in certain differences between the suspected abnormal regions screened and the real abnormal regions, which may affect the doctor's diagnosis and treatment results. SUMMARY

[0004] In order to solve the problem that the existing method has certain differences between the screening results and the real situation when screening the suspected abnormal regions of the head of the person to be analyzed, the purpose of the present application is to provide a nuclear medical functional imaging analysis method and system for nervous system diseases, and the technical scheme adopted is as follows:

[0005] In the first aspect, the present application provides a nuclear medical functional imaging analysis method for nervous system diseases, which comprises the following steps:

[0006] Obtaining the PET three-dimensional volume image and the CT three-dimensional volume image of the head of the person to be analyzed;

[0007] Based on the SUV value, CT value and position coordinates of each voxel point in the PET three-dimensional volume image or the CT three-dimensional volume image, the initial ROI block and the non-initial ROI block are obtained; according to the distance between each voxel point on the surface of the initial ROI block and the corresponding minimum enclosing volume, and the change characteristics of the SUV value of the voxel points between the initial ROI block and its surrounding non-initial ROI block, the initial ROI block is corrected to obtain the target region;

[0008] According to the distribution characteristics of the SUV values of the voxel points in each target region, the distances between the surface voxel points of each target region and the corresponding minimum bounding volume, and the gradient values of the surface voxel points, an SUV distribution factor of each target region is obtained; in combination with the joint trend of the deviation of the SUV values of the voxel points of each target region from the whole and the deviation of the CT values from the whole, the dispersion of the SUV values of the voxel points, and the dispersion of the CT values, a metabolic-structural consistency factor of each target region is obtained.

[0009] The SUV distribution factor and the metabolic-structural consistency factor are combined to screen a suspected abnormal region.

[0010] Preferably, the SUV values, CT values, and position coordinates of each voxel point in the PET three-dimensional volume image or the CT three-dimensional volume image are used to obtain initial ROI blocks and non-initial ROI blocks, including:

[0011] The DBSCAN algorithm is used to cluster all the voxel points based on the Euclidean distances between the feature vectors of different voxel points, and a plurality of blocks are obtained; the feature vector is composed of the SUV value, the CT value, and the coordinate;

[0012] The block with a mean SUV value greater than an SUV threshold value is determined as an initial ROI block, and the block with a mean SUV value less than or equal to the SUV threshold value is determined as a non-initial ROI block.

[0013] Preferably, the initial ROI block is corrected according to the distances between the surface voxel points of the initial ROI block and the corresponding minimum bounding volume, and the variation characteristics of the SUV values of the voxel points between the initial ROI block and its surrounding non-initial ROI blocks, and a target region is obtained, including:

[0014] The Euclidean distances between the spatial coordinates of each surface voxel point of the initial ROI block and the nearest point on the surface of the minimum bounding volume of the initial ROI block are recorded as the first distances corresponding to the surface voxel points of the initial ROI block;

[0015] The first distances corresponding to all the surface voxel points of the initial ROI block are arranged in ascending order to obtain a first distance sequence;

[0016] The difference between two adjacent elements in the first distance sequence is calculated, and the minimum value of the first distances corresponding to the two voxel points when the difference is maximized is recorded as a reference distance; the voxel points of the initial ROI block whose first distances are less than or equal to the reference distance are determined as the first voxel points corresponding to the initial ROI block;

[0017] other blocks in which the voxel points located in the non-initial ROI blocks and not in the minimum enclosing volume of the initial ROI block are taken as the peripheral edema feature blocks;

[0018] For any peripheral edema feature block, the voxel points adjacent to the initial ROI block in the any peripheral edema feature block are recorded as first neighboring points; and all the voxel points of the any peripheral edema feature block on the extension line of the connecting line between the center point of the initial ROI block and the first neighboring points in the direction of the any peripheral edema feature block are arranged in the extension direction to form a voxel point sequence;

[0019] The initial ROI block is corrected based on the change of the CT value of the voxel point in the voxel point sequence to obtain a target region.

[0020] Preferably, the correction of the initial ROI block based on the change of the CT value of the voxel point in the voxel point sequence to obtain a target region comprises:

[0021] The normalized result of the mean value of the difference of the CT value of all adjacent voxel points in each voxel point sequence is taken as the CT feature value of each voxel point sequence;

[0022] The mean value of the CT feature value of all the voxel point sequences corresponding to the any peripheral edema feature block is determined as the outward permeation and expansion feature coefficient of the any peripheral edema feature block;

[0023] The peripheral edema feature block corresponding to the maximum outward permeation and expansion feature coefficient is taken as the peripheral edema metabolic heterogeneous block;

[0024] If the mean value of the SUV value of all the voxel points in the peripheral edema metabolic heterogeneous block is greater than the mean value of the SUV value of all the voxel points in the minimum enclosing frame of the initial ROI block, the region composed of the peripheral edema metabolic heterogeneous block and the minimum enclosing frame of the initial ROI block is taken as the target region.

[0025] Preferably, the SUV distribution factor of each target region is obtained according to the distribution characteristics of the SUV value of the voxel point in each target region, the distance between the surface voxel point of each target region and the corresponding minimum enclosing volume, and the gradient value of the surface voxel point, comprising:

[0026] For any target region:

[0027] The principal axis of the edge line in the any target region is obtained by using the PCA algorithm, and the included angle between the principal axis and a preset plane is obtained;

[0028] calculating a variance of the included angle corresponding to the principal axis of all edge lines in the any target region; and determining a product of the variance and information entropy of the SUV value of all voxel points in the any target region as a texture feature coefficient of the any target region;

[0029] obtaining a boundary blur feature value of the any target region according to the gradient value of the surface voxel point of the any target region and the corresponding first distance;

[0030] determining a product of the texture feature coefficient and the boundary blur feature value of the any target region as an SUV distribution factor of the any target region.

[0031] Preferably, the obtaining of the boundary blur feature value of the any target region according to the gradient value of the surface voxel point of the any target region and the corresponding first distance comprises:

[0032] obtaining the boundary blur feature value of the any target region according to a difference between a gradient mean value of the surface voxel point of the any target region and an average value of the first distance corresponding to each voxel point of the surface of the any target region and the first distance corresponding to all voxel points of the surface of the any target region, the gradient mean value and the difference between the first distance corresponding to each voxel point and the average value of the first distance corresponding to all voxel points being positively correlated with the boundary blur feature value.

[0033] Preferably, the obtaining of the metabolic-structural consistency factor of each target region according to the combined trend of the deviation of the SUV value of each voxel point of the target region from the whole and the deviation of the CT value from the whole, the dispersion of the SUV value of the voxel point and the dispersion of the CT value comprises:

[0034] for any target region:

[0035] taking a normalized result of the SUV value of each voxel point in the any target region as a first feature value of each voxel point, and taking a normalized result of the CT value of each voxel point in the any target region as a second feature value of each voxel point;

[0036] obtaining a metabolic-structural consistency factor of the any target region according to the first feature value and the second feature value.

[0037] Preferably, the obtaining of the metabolic-structural consistency factor of the any target region according to the first feature value and the second feature value comprises:

[0038] calculating a covariance of the first feature value and the second feature value of all voxel points in the any target region;

[0039] According to the covariance, a standard deviation of the first eigenvalue of all voxel points in the any target region, and a standard deviation of the second eigenvalue of all voxel points in the any target region, a metabolic-structural consistency factor of the any target region is obtained, the covariance is in positive correlation with the metabolic-structural consistency factor, and the standard deviation of the first eigenvalue and the standard deviation of the second eigenvalue are both in negative correlation with the metabolic-structural consistency factor.

[0040] Preferably, the SUV distribution factor and the metabolic-structural consistency factor are comprehensively used to screen a suspected abnormal region, and the method comprises the following steps of:

[0041] According to the SUV distribution factor and the metabolic-structural consistency factor of each target region, an abnormality index of each target region is obtained, the SUV distribution factor is in positive correlation with the abnormality index, and the metabolic-structural consistency factor is in negative correlation with the abnormality index.

[0042] If the abnormality index is greater than a preset abnormal threshold, the corresponding target region is determined as a suspected abnormal region.

[0043] In a second aspect, the present application provides a nuclear medical functional imaging analysis system for nervous system diseases, which is used to execute the method described above, and the system comprises the following components:

[0044] A data acquisition module is configured to acquire a PET three-dimensional volume image and a CT three-dimensional volume image of a head of a person to be analyzed.

[0045] A target region extraction module is configured to obtain an initial ROI block and a non-initial ROI block based on an SUV value, a CT value and a position coordinate of each voxel point in the PET three-dimensional volume image or the CT three-dimensional volume image, correct the initial ROI block based on a distance between each voxel point on a surface of the initial ROI block and a corresponding minimum enclosing volume and a variation feature of the SUV value of a voxel point between the initial ROI block and a surrounding non-initial ROI block, and obtain a target region.

[0046] An evaluation module is configured to obtain an SUV distribution factor of each target region based on a distribution feature of an SUV value of each voxel point in the target region, a distance between a surface voxel point of the target region and a corresponding minimum enclosing volume, and a gradient value of the surface voxel point, and obtain a metabolic-structural consistency factor of each target region based on a joint trend of a deviation of an SUV value of each voxel point relative to a whole and a deviation of a CT value relative to the whole, a dispersion of the SUV value of the voxel point, and a dispersion of the CT value.

[0047] A screening module is configured to comprehensively use the SUV distribution factor and the metabolic-structural consistency factor to screen a suspected abnormal region.

[0048] The present application has at least the following beneficial effects:

[0049] The present application firstly fuses the PET-CT three-dimensional volume data of the head of the person to be analyzed, realizes the block processing of the image, and combines the SUV value features of the voxel points in the image to preliminarily screen each block, obtains the initial ROI block and the non-initial ROI block, and combines the brain tumor structure features to expand the initial ROI block, and obtains the target region; analyzes the distribution features of the SUV values of the voxel points in each target region, realizes the evaluation of the texture and edge features of the brain tumor, evaluates the brain tumor SUV distribution features of each ROI region; combines the CT value data change features of the brain tumor, obtains the SUV distribution factor of each target region; then, combines the SUV value distribution features, the CT value distribution features and the metabolic-structure consistent features in each target region, screens out the suspected abnormal regions, and doctors can focus on these regions in subsequent analysis, which provides a good auxiliary role for the doctors to analyze the nuclear medical functional imaging, and can effectively improve the positioning accuracy of the brain tumor. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0051] Figure 1 The flowchart of the nuclear medical functional imaging analysis method for nervous system diseases provided by the embodiments of the present application. DETAILED DESCRIPTION

[0052] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following will combine the drawings and the preferred embodiments to specifically describe the nuclear medical functional imaging analysis method and system for nervous system diseases according to the present application as follows.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0054] The specific scheme of the nuclear medical functional imaging analysis method and system for nervous system diseases provided by the present application will be specifically described below in combination with the drawings.

[0055] Embodiment of the nuclear medical functional imaging analysis method for nervous system diseases:

[0056] The specific scenario targeted by the present embodiment is: brain tumor as a common intracranial space-occupying lesion in central nervous system, various types, prognosis difference is significant, its clinical diagnosis and treatment depends on the accurate positioning of the lesion, metabolic activity evaluation and internal heterogeneity characteristic analysis. Although MRI performs excellently in the structural imaging of brain tumor, it has certain limitations in the identification of tumor active area, infiltration boundary and post-treatment recurrence lesion. PET-CT fusion metabolic function and anatomical structure image, especially based on 18F-FDG, 11C-MET, 18F-FET and other radioactive tracers, can sensitively reflect tumor metabolic activity, assist in accurate positioning of suspected abnormal areas, screen preoperative active areas, and provide reference basis for subsequent analysis of doctors. The present embodiment will collect PET three-dimensional volume images and CT three-dimensional volume images of the head of the person to be analyzed, and comprehensively analyze the PET three-dimensional volume images and the CT three-dimensional volume images to screen the suspected abnormal areas.

[0057] The present embodiment proposes a nuclear medicine functional imaging analysis method for nervous system diseases, as shown in Figure 1 The nuclear medicine functional imaging analysis method for nervous system diseases of the present embodiment includes the following steps:

[0058] Step S1, acquiring PET three-dimensional volume images and CT three-dimensional volume images of the head of the person to be analyzed.

[0059] At present, most of the PET-CT examinations (especially tumor diagnosis) use a single tracer, mainly 18F-FDG (fluorodeoxyglucose). As a glucose analogue, it can reflect tissue metabolic activity and is suitable for the initial screening and staging of various tumors. Because of its fast metabolism and low side effects, it has become the most commonly used tracer in clinical practice. The present embodiment collects PET-CT data of the head of the person to be analyzed by using 18F-FDG as a tracer.

[0060] Specifically, 18F-FDG is used as a radioactive tracer, the dose is set according to the weight and disease type, the tracer is injected intravenously, the injection time is recorded to ensure the accuracy of the acquisition time window; a certain period of time (such as 45-60 min) is waited after the tracer injection to allow distribution in the brain tissue, and attention is paid to the rest state to prevent muscle uptake of the tracer from interfering with the results; a PET-CT integrated device is used, and a positioning CT scan is first performed to obtain the anatomical structure of the head: the CT scan parameters are 120 kV, 100-200 mAs, the pitch is 0.8-1.2, the layer thickness is 1 mm, and the travel speed is 0.5 s / revolution; the reconstruction matrix is 512x512, the reconstruction layer thickness is 1 mm, and the field of view (FOV) is 250 mm; a CT three-dimensional volume image is reconstructed, and the voxel value unit is HU, representing the tissue density value. PET scanning is performed to collect the radioactive emission signals in the head range to generate original sinogram data; a 3D mode is used for acquisition, the matrix is 256x256, and the slice thickness is 3 mm; an OSEM iterative algorithm is used to reconstruct the PET original sinogram data, and time attenuation correction, scatter correction, and attenuation correction are embedded to obtain a preliminary reconstructed PET three-dimensional volume image.

[0061] Based on the rigid registration algorithm, the PET three-dimensional volume image and the CT three-dimensional volume image are spatially aligned to realize PET-CT registration and ensure the consistency of lesion positioning. A filtering algorithm (such as non-local mean filtering) is used to perform noise reduction processing on the data to obtain pre-processed PET three-dimensional volume images and CT three-dimensional volume images. The PET three-dimensional volume image and the CT three-dimensional volume image are corresponding three-dimensional medical images in the same spatial coordinate system. It should be noted that the PET three-dimensional volume image and the CT three-dimensional volume image mentioned later are filtered images.

[0062] At this point, the PET three-dimensional volume image and the CT three-dimensional volume image of the head of the person to be analyzed are obtained, the SUV value of each voxel point in the PET three-dimensional volume image is used to reflect the uptake intensity of the local tissue to the tracer, and the CT value of each voxel point in the CT three-dimensional volume image is used to reflect the density of the tissue.

[0063] In step S2, based on the SUV value, CT value, and position coordinates of each voxel point in the PET three-dimensional volume image or the CT three-dimensional volume image, an initial ROI block and a non-initial ROI block are obtained; the initial ROI block is corrected according to the distance between each voxel point on the surface of the initial ROI block and the corresponding minimum enclosing volume, and the variation characteristics of the SUV value of the voxel points between the initial ROI block and its surrounding non-initial ROI blocks, to obtain a target region.

[0064] 18F-FDG as a glucose analog, its uptake mechanism depends on glucose transporter (such as GLUT1), but this mechanism is not unique to tumor cells. Normal brain gray matter neuronal activity, inflammatory response, radiation injury, etc. can all uptake tracers through the same pathway, resulting in uptake signals that are not tumor-specific. Due to the broad-spectrum glucose metabolism mechanism, it is difficult to clearly present the metabolic heterogeneity of brain tumors (such as active areas, boundaries, and grading characteristics) under the interference of high background signals in normal brain gray matter. It is necessary to reveal the uneven distribution of lesion area metabolism through texture features and heterogeneity indicators to assist in precise positioning of brain tumor lesions.

[0065] The number of voxels of brain PET-CT fusion 3D volume data is usually large, and the tumor area only accounts for a small part of the volume data. Direct lesion detection analysis of the whole image is low in efficiency, and brain tumors account for a very small part of the whole brain volume. Direct analysis of the whole brain will be disturbed by a large number of non-lesion areas with high metabolism physiological structures (basal ganglia, cortex, thalamus, etc.); Local ROI-based analysis and block-level processing can significantly improve the efficiency of brain PET-CT lesion detection and positioning, especially in the case of strong brain tumor heterogeneity, small volume, and fuzzy edges. Local feature analysis and extraction based on block partitioning are much better than whole image operation. Therefore, the image will be first divided into blocks in the following embodiment. The purpose of block partitioning is to accommodate a small lesion in each block, and there is appropriate overlap between blocks to avoid loss of lesion boundary segmentation. The block partitioning method based on clustering algorithm is used to partition the three-dimensional volume image. In addition, brain tumors appear as local high-metabolism areas in PET images, which is reflected in the image as SUV values significantly higher than normal brain cortex values. After the three-dimensional volume image is partitioned, each block is preliminarily screened.

[0066] Since the voxel points in PET three-dimensional volume images and CT three-dimensional volume images have a one-to-one correspondence, the subsequent block partitioning results are the same, so any one of the PET three-dimensional volume image or the CT three-dimensional volume image can be used as an example for block partitioning, and the other image is divided according to the division result.

[0067] In this embodiment, the PET three-dimensional volume image is first partitioned, specifically, a feature vector of each voxel point is constructed, which is composed of the SUV value, CT value and coordinates of the voxel point, and each voxel point has a corresponding feature vector. It should be noted that each element in the feature vector is a dimensionless value.

[0068] Based on the Euclidean distance between the feature vectors of different voxel points, DBSCAN algorithm is used to cluster all the voxel points to obtain multiple clusters, and each cluster is a sub-block. When using DBSCAN algorithm for clustering, the neighborhood radius parameter can be obtained by k-distance algorithm, and the minimum neighborhood point number parameter is selected by adding one to the dimension of the clustering data according to the existing guiding principle, which is 6 in this embodiment. DBSCAN algorithm is a prior art, which will not be described in detail here.

[0069] The mean value of the SUV value of all voxel points in each sub-block is calculated respectively, and the sub-block with a mean SUV value greater than the SUV threshold is determined as an initial ROI sub-block, and the sub-block with a mean SUV value less than or equal to the SUV threshold is determined as a non-initial ROI sub-block. Thus, multiple initial ROI sub-blocks and multiple non-initial ROI sub-blocks are obtained in this embodiment. It should be noted that the SUV threshold can be set by the implementer according to the specific situation, or it can be obtained by using the Otsu threshold algorithm. In this embodiment, it is the SUV threshold obtained by using the Otsu threshold algorithm.

[0070] After obtaining the initial ROI sub-block, in high-grade brain tumors such as glioblastoma GBM, there are usually metabolic heterogeneity structures inside the lesion, that is, different regions in the same lesion show different metabolic intensities. This metabolic heterogeneity is not randomly distributed, but has obvious spatial histological dependence. Therefore, the initial ROI sub-block obtained directly by clustering may not fully express the characteristics of a certain type of tumor, and further expansion of the initial ROI sub-block is needed. There are high and low metabolic mixed areas inside this type of tumor, and the typical feature is low uptake in the central necrotic area, which is specifically manifested as extremely low glucose metabolism, and the PET signal (SUV value) is close to or slightly higher than the cerebrospinal fluid area. The periphery of the high metabolic activity area has high uptake, which is specifically manifested as vigorous metabolism of cells in the high metabolic activity area, and the SUV is significantly increased. The metabolic rate of the peripheral edema zone is slightly higher than that of the normal brain parenchyma, but lower than that of the high metabolic activity area. These regions show a ring-like nested pattern in spatial structure, which can reflect the metabolic complexity and metabolic heterogeneity inside the lesion. The sub-block obtained by DBSCAN clustering usually retains the geometric features of point sets with relatively complex shapes, so the above situation may be characterized by the mixed intersection of initial ROI sub-blocks and non-initial ROI sub-blocks in the data space. In addition, the brain tumor edema area is usually extracellular brain edema, and the water accumulation is in the intercellular space, which expands outward in a finger-like manner, and the CT value also shows the characteristic of outward expansion weakening.

[0071] Next, this embodiment will be described taking one initial ROI sub-block as an example. The method provided in this embodiment can be used to process other initial ROI sub-blocks.

[0072] Specifically, for any initial ROI patch, first, the minimum bounding volume of the initial ROI patch is obtained, which is a cuboid in this embodiment; the Euclidean distance between each voxel point on the surface of the initial ROI patch and the point on the surface of the minimum bounding volume of the initial ROI patch closest to the voxel point is recorded as the first distance corresponding to the voxel point. The first distances corresponding to all the voxel points on the surface of the initial ROI patch are arranged in ascending order to obtain a first distance sequence. The difference between adjacent two elements in the first distance sequence is calculated respectively, and it should be noted that the difference between adjacent two elements in this embodiment is obtained by subtracting the first element from the second element; the minimum value of the two first distances corresponding to the two voxel points at which the maximum value of the difference is taken is recorded as the reference distance; the voxel points on the surface of the initial ROI patch whose first distances are less than or equal to the reference distance are determined as the first voxel points corresponding to the initial ROI patch.

[0073] Other patches in which the voxel points adjacent to the first voxel points corresponding to the initial ROI patch are located, and which are not located in the minimum bounding volume of the initial ROI patch, are taken as peripheral edema feature patches. For any peripheral edema feature patch, the voxel points in the peripheral edema feature patch adjacent to the initial ROI patch are recorded as first neighboring points, that is, there can be multiple first neighboring points; the extension line of the line connecting the center point of the initial ROI patch and the first neighboring points in the direction of the peripheral edema feature patch is extended to all the voxel points of the peripheral edema feature patch in the direction of the extension, and a voxel point sequence is constructed according to the extension direction, and each first neighboring point has a corresponding voxel point sequence. It should be noted that the line is a straight line. The normalized result of the average of the difference between the CT values of adjacent voxel points in each voxel point sequence is taken as the CT feature value of each voxel point sequence; wherein the difference between the CT values of adjacent voxel points in the voxel point sequence is obtained by subtracting the CT value of the first voxel point from the CT value of the second voxel point. The normalization method of the average is to normalize the average by using a sigmoid function, so that the normalized result is valued between 0 and 1. Further, the average of the CT feature values of all the voxel point sequences corresponding to the peripheral edema feature patch is determined as the outward permeation and expansion feature coefficient of the peripheral edema feature patch, and the larger the value is, the more the current peripheral edema feature patch has the finger-like outward permeation and expansion feature of the brain tumor edema area.

[0074] The peripheral edema metabolic heterogeneous sub-block corresponding to the maximum outward permeation expansion feature coefficient is taken as the peripheral edema metabolic heterogeneous sub-block. If the mean value of the SUV values of all voxel points in the peripheral edema metabolic heterogeneous sub-block is greater than the mean value of the SUV values of all voxel points in the minimum bounding box of the initial ROI sub-block, the region composed of the peripheral edema metabolic heterogeneous sub-block and the minimum bounding box of the initial ROI sub-block is taken as the target region.

[0075] The initial ROI region is corrected by using the above method, and a plurality of target regions are obtained.

[0076] In step S3, the SUV distribution factor of each target region is obtained according to the distribution characteristics of the SUV values of the voxel points in each target region, the distance between the surface voxel points of each target region and the corresponding minimum bounding volume, and the gradient value of the surface voxel points; and the metabolic-structural consistency factor of each target region is obtained in combination with the joint trend of the deviation of the SUV values of the voxel points and the deviation of the CT values of the voxel points relative to the whole, the dispersion of the SUV values of the voxel points, and the dispersion of the CT values.

[0077] In 18F-FDG PET-CT brain imaging, the regions of postoperative brain tissue, inflammation, and post-radiation reactive lesions often have metabolic elevation, which shows high uptake similar to the active tumor area at the SUV value level, and cannot be reliably distinguished by only the SUV value, so it is necessary to further extract the three-dimensional spatial texture features, metabolic dynamic mode, morphological characteristics, and spatial position coupling relationship in the region, and further analyze the data characteristics of each ROI region.

[0078] First, the tumor region and the lung tumor region are distinguished by PET scan data. The metabolic heterogeneity of tumor tissue is derived from the complexity of the tumor microenvironment: cell density, angiogenesis, hypoxia necrosis, local necrosis, and interlaced proliferation, which easily produces complex local metabolic gradient; and the tissue structure of the non-tumor high metabolic region such as inflammation is relatively simple, the spatial distribution is more homogeneous, the blood vessels are uniformly expanded, and the cell infiltration has no tumor-like spatial complex structure, resulting in more homogeneous metabolic distribution and simple texture structure.

[0079] In addition to the difference in texture quantity, the geometric shape of the PET high uptake area can reflect the spatial expansion mode and biological heterogeneity of the internal pathological process of the tissue. The brain tumor expands in an infiltrative manner, which easily leads to rough, blurred, and irregular characteristics; and for the non-tumor high metabolic region such as inflammation, it is mostly acute lesions or local reactions, and the boundary of the high metabolic value region is relatively regular and sharp, and the SUV value has a significant mutation. In combination with the above characteristics, the texture characteristics of the brain tumor are analyzed in the PET data space by the distribution characteristics of the SUV value.

[0080] Next, this embodiment takes a target region as an example for illustration, and the method provided in this embodiment can be used to process other target regions.

[0081] Specifically, for any target region:

[0082] All edge lines of the target region are acquired by using three-dimensional Canny edge detection, the principal axes of the edge lines in the target region are acquired by using a PCA algorithm, an included angle between the principal axes and a preset plane is acquired, and the included angle is recorded as an included angle corresponding to the edge line. A variance of the included angles corresponding to the principal axes of all edge lines in the target region is calculated. The greater the variance, the more complex the texture structure inside the target region is. In this embodiment, the preset plane is the XOY plane. Then, a product of the variance and an information entropy of SUV values of all voxel points in the target region is determined as a texture feature coefficient of the target region.

[0083] Further, a boundary blur feature value of the target region is obtained according to a gradient mean value of the surface voxel points of the target region and a difference between the first distance corresponding to each voxel point of the target region and the average value of the first distances corresponding to all voxel points. The gradient mean value and the difference between the first distance corresponding to each voxel point of the target region and the average value of the first distances corresponding to all voxel points are in positive correlation with the boundary blur feature value.

[0084] In this embodiment, a specific calculation formula of the boundary blur feature value is given, and the boundary blur feature value of the target region can be represented as:

[0085] ;

[0086] wherein, the boundary blur feature value of the target region, the number of surface voxel points of the target region, the average value of gradient values of all surface voxel points of the target region, the first distance corresponding to the i-th voxel point on the surface of the target region, the average value of the first distances corresponding to all surface voxel points of the target region, the absolute value symbol, the normalization function.

[0087] the difference between the first distance corresponding to the i-th voxel point on the surface of the target region and the average value of the first distances corresponding to all voxel points, the greater the value, the more the first distance corresponding to the i-th voxel point deviates from the average level; in this embodiment, the normalized result of the average value of the gradient values of all surface voxel points of the target region is used as a weight to amplify the boundary irregular feature of the voxel points with clear boundaries in the target region.

[0088] Next, the product of the texture feature coefficient of the target region and the boundary blur feature value of the target region is determined as the SUV distribution factor of the target region.

[0089] By using the above method, the SUV distribution factor of each target region can be obtained.

[0090] However, some problems may still occur in the evaluation only by the distribution characteristics of the SUV value: in PET-CT fusion imaging, in the ideal state, the high metabolic region (high SUV value) should correspond to the high-density tumor parenchymal region (high CT value), and the low metabolic region corresponds to the necrosis / cystic change (low CT value); however, the heterogeneity specific to malignant tumors will lead to metabolic-structural mismatch, such as necrotic or hypoxic regions, enhanced glycolysis leading to high SUV, but structural liquefaction, low CT value; or calcified regions, high CT, but no metabolic activity, low SUV. This metabolic-structural mismatch may cause the SUV texture of the tumor region to appear alienation characteristics. Considering that the non-tumor high metabolic region such as inflammation basically corresponds to the edema or parenchymal thickening characteristics, lacks mismatch phenomenon, and the metabolic and structural density are consistent, the overall relationship between SUV and CT is linear coupling; in order to further improve the positioning analysis accuracy, the SUV value and the CT value of each target region also need to be comprehensively analyzed.

[0091] Next, this embodiment still takes one target region as an example for description, and the method provided in this embodiment can be used for processing of other target regions.

[0092] For any target region:

[0093] The normalized result of the SUV value of each voxel point in the target region is recorded as the first feature value of each voxel point, and the normalized result of the CT value of each voxel point in the target region is recorded as the second feature value of each voxel point; in this embodiment, the maximum and minimum value normalization method is used for processing when the SUV value and the CT value are normalized, which is a prior art, and will not be described in detail here. The covariance of the first feature value and the second feature value of all voxel points in the target region is calculated, which is used to reflect the joint trend when the CT value and the SUV value deviate from the respective mean values simultaneously, and if the CT value and the SUV value change in the same direction, the value is positive, indicating that the coupling is strong; according to the covariance, the standard deviation of the first feature value of all voxel points in the target region, and the standard deviation of the second feature value of all voxel points in the target region, the metabolic-structural consistency factor of the target region is obtained, the covariance is positively correlated with the metabolic-structural consistency factor, and the standard deviation of the first feature value and the standard deviation of the second feature value are both negatively correlated with the metabolic-structural consistency factor.

[0094] In the embodiment, a specific calculation formula of the metabolic-structural consistency factor is given, and the metabolic-structural consistency factor of the target region can be represented as:

[0095] ;

[0096] wherein, represents the metabolic-structural consistency factor of the target region, represents a data sequence composed of the first eigenvalues of all the voxel points in the target region, represents a data sequence composed of the second eigenvalues of all the voxel points in the target region, represents a covariance of the first eigenvalues of all the voxel points in the target region and the second eigenvalues of all the voxel points in the target region, represents a standard deviation of the first eigenvalues of all the voxel points in the target region, represents a standard deviation of the second eigenvalues of all the voxel points in the target region, represents a preset first adjustment parameter.

[0097] The preset first adjustment parameter is introduced into the calculation formula of the metabolic-structural consistency factor in the embodiment to prevent the denominator from being 0. In the embodiment, the preset first adjustment parameter is 0.01, which can be set according to specific conditions in specific applications. The standard deviation of the first eigenvalues is used to reflect the dispersion of the SUV values, and the standard deviation of the second eigenvalues is used to reflect the dispersion of the CT values. When the covariance of the first eigenvalues of all the voxel points in the target region and the second eigenvalues of all the voxel points in the target region is larger, the standard deviation of the first eigenvalues of all the voxel points in the target region is smaller, and the standard deviation of the second eigenvalues of all the voxel points in the target region is also smaller, it indicates that the metabolic-structural consistency of the target region is higher, and the possibility of metabolic-structural mismatch is lower, that is, the metabolic-structural consistency factor of the target region is larger.

[0098] By using the above method, the metabolic-structural consistency factor of each target region can be obtained.

[0099] In step S4, the SUV distribution factor and the metabolic-structural consistency factor are integrated to screen the suspected abnormal region.

[0100] In the embodiment, the SUV distribution factor and the metabolic-structural consistency factor of each target region have been obtained, and next, the SUV distribution factor and the metabolic-structural consistency factor are integrated to screen the suspected abnormal region.

[0101] Specifically, the abnormality index of each target region is obtained according to the SUV distribution factor of each target region and the metabolic-structural consistency factor of each target region, respectively, the SUV distribution factor is positively correlated with the abnormality index, and the metabolic-structural consistency factor is negatively correlated with the abnormality index.

[0102] In this embodiment, for any target region, the sum of the metabolic-structural consistency factor of the target region and the constant 1 is calculated, and then the normalization result of the ratio between the SUV distribution factor of the target region and the sum value is determined as the abnormality index of the target region. In this embodiment, the method for normalizing the ratio is: substituting the ratio into the hyperbolic tangent function, that is, taking the ratio as the independent variable in the hyperbolic tangent function, and taking the hyperbolic tangent function value as the abnormality index of the target region, that is, taking as the abnormality index of the target region, wherein x represents the ratio between the SUV distribution factor of the target region and the sum value, and tanh represents the hyperbolic tangent function.

[0103] By using this method, the abnormality index of each target region can be obtained, and the greater the abnormality index, the more likely the corresponding target region has an abnormality. Therefore, if the abnormality index is greater than a preset abnormality threshold, it is determined that the corresponding target region is a suspected abnormal region. In this embodiment, the preset abnormality threshold is 0.45, and in specific applications, the implementer can set it according to the specific situation.

[0104] After obtaining the suspected abnormal region, the suspected abnormal region is marked in the PET three-dimensional volume image and the CT three-dimensional volume image, and the doctor can focus on these regions in subsequent analysis, thereby improving the analysis efficiency and the accuracy of the judgment result.

[0105] This embodiment first fuses the PET-CT three-dimensional volume data of the head of the person to be analyzed, realizes the block processing of the image, and preliminarily screens each block in combination with the SUV value features of the voxel points in the image, obtains the initial ROI block and the non-initial ROI block, and expands the initial ROI block in combination with the brain tumor structure features, thereby obtaining the target region; analyzes the distribution features of the SUV values of the voxel points in each target region, realizes the evaluation of the texture and edge features of the brain tumor, evaluates the brain tumor SUV distribution features of each ROI region; in combination with the CT value data variation features of the brain tumor, the SUV distribution factor of each target region is obtained; then, in combination with the SUV value distribution features, the CT value distribution features and the metabolic-structural consistency features in each target region, the suspected abnormal region is screened out, the doctor can focus on these regions in subsequent analysis, thereby providing a good auxiliary role for the doctor's nuclear medicine functional imaging analysis, and effectively improving the positioning accuracy of the brain tumor.

[0106] A nuclear medicine functional imaging analysis system for nervous system diseases

[0107] The nuclear medicine functional imaging analysis system for nervous system diseases provided by one embodiment of the present application can comprise a data acquisition module, a target region extraction module, an evaluation module and a screening module.

[0108] The data acquisition module is configured to acquire a PET three-dimensional volume image and a CT three-dimensional volume image of a head of a person to be analyzed.

[0109] The target region extraction module is configured to obtain an initial ROI block and a non-initial ROI block based on SUV values, CT values and position coordinates of each voxel in the PET three-dimensional volume image or the CT three-dimensional volume image; correct the initial ROI block according to distances between each voxel on a surface of the initial ROI block and a corresponding minimum enclosing volume, and a variation feature of SUV values of voxels between the initial ROI block and surrounding non-initial ROI blocks, to obtain a target region.

[0110] The evaluation module is configured to obtain an SUV distribution factor of each target region according to a distribution feature of SUV values of voxels in each target region, distances between surface voxels of each target region and a corresponding minimum enclosing volume, and gradient values of the surface voxels; and obtain a metabolic-structural consistency factor of each target region according to a joint trend of a deviation of SUV values of voxels of each target region from a whole and a deviation of CT values of the voxels from the whole, a dispersion of SUV values of the voxels and a dispersion of CT values.

[0111] The screening module is configured to screen a suspected abnormal region by comprehensively considering the SUV distribution factor and the metabolic-structural consistency factor.

[0112] It should be appreciated that the structural diagram of the nuclear medicine functional imaging analysis system for nervous system diseases and its modules can be implemented in various ways. For example, in some embodiments, the system and its modules can be implemented by hardware, software, or a combination of software and hardware. Among them, the hardware part can be implemented by using special logic; the software part can be stored in the memory and executed by a suitable instruction execution system, such as a microprocessor or a specially designed hardware. Those skilled in the art can understand that the above-mentioned method and system can be implemented by using computer executable instructions and / or contained in processor control code, for example, such code is provided on a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The system and its modules of the present specification can not only have hardware circuit implementation such as very large scale integrated circuit or gate array, semiconductor such as logic chip, transistor, or programmable hardware device such as field programmable gate array, programmable logic device, etc., but also can be implemented by software executed by various types of processors, and also can be implemented by a combination of the above hardware circuit and software (for example, firmware).

[0113] More details about each of the above modules can be referred to other places in the specification, and will not be described here.

[0114] It should be noted that the above is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for nuclear medical functional imaging analysis of nervous system diseases, characterized by, The method comprises the following steps: Obtain PET three-dimensional volume images and CT three-dimensional volume images of the head of the person to be analyzed; Based on the SUV value, CT value and position coordinates of each voxel point in the PET three-dimensional volume image or the CT three-dimensional volume image, obtain an initial ROI block and a non-initial ROI block; according to the distance between each voxel point on the surface of the initial ROI block and the corresponding minimum enclosing volume, and the variation characteristics of the SUV value of the voxel point between the initial ROI block and the surrounding non-initial ROI block, correct the initial ROI block to obtain a target region; According to the distribution characteristics of the SUV value of the voxel points in each target region, the distance between the surface voxel points of each target region and the corresponding minimum enclosing volume, and the gradient value of the surface voxel points, obtain the SUV distribution factor of each target region; in combination with the joint trend of the deviation of the SUV value of the voxel points of each target region relative to the whole and the deviation of the CT value relative to the whole, the dispersion of the SUV value of the voxel points and the dispersion of the CT value, obtain the metabolic-structural consistency factor of each target region; Integrate the SUV distribution factor and the metabolic-structural consistency factor to screen a suspected abnormal region; The target region is obtained, comprising: Respectively, the Euclidean distance between each voxel point on the surface of the initial ROI block and the nearest point on the surface of the minimum enclosing volume of the initial ROI block is recorded as the first distance corresponding to each voxel point on the surface of the initial ROI block; Arrange the first distances corresponding to all voxel points on the surface of the initial ROI block in ascending order to obtain a first distance sequence; Respectively, calculate the difference between two adjacent elements in the first distance sequence, and take the minimum value of the first distance corresponding to the two voxel points when the maximum value of the difference is obtained as the reference distance; the voxel points with a first distance less than or equal to the reference distance among all voxel points on the surface of the initial ROI block are determined as the first voxel points corresponding to the initial ROI block; Other blocks in which the voxel points adjacent to the first voxel points corresponding to the initial ROI block are located in the non-initial ROI block and not in the minimum enclosing volume of the initial ROI block are taken as peripheral edema feature blocks; For any peripheral edema feature block, the voxel points adjacent to the initial ROI block in the any peripheral edema feature block are recorded as first adjacent points; the extension line of the connecting line between the center point of the initial ROI block and the first adjacent point in the direction of the any peripheral edema feature block is extended to all voxel points of the any peripheral edema feature block, and a voxel point sequence is constructed according to the extension direction; Based on the variation of the CT value of the voxel points in the voxel point sequence, the initial ROI block is corrected to obtain a target region; The SUV distribution factor of each target region is obtained, comprising: For any target region: The principal axis of the edge line in the any target region is obtained by using the PCA algorithm, and the included angle between the principal axis and a preset plane is obtained. Calculate a variance of the included angle corresponding to the principal axis of all edge lines in the any target region; and determine a product of the variance and an information entropy of the SUV value of all voxel points in the any target region as a texture feature coefficient of the any target region; Obtain a boundary blur feature value of the any target region according to the gradient value and the corresponding first distance of the surface voxel point of the any target region; Determine a product of the texture feature coefficient and the boundary blur feature value of the any target region as an SUV distribution factor of the any target region; The method for obtaining the metabolic-structural consistency factor of each target region comprises: For any target region: Record a normalized result of the SUV value of each voxel point in the any target region as a first feature value of each voxel point, and record a normalized result of the CT value of each voxel point in the any target region as a second feature value of each voxel point; Obtain a metabolic-structural consistency factor of the any target region according to the first feature value and the second feature value; The method for obtaining the metabolic-structural consistency factor of the any target region according to the first feature value and the second feature value comprises: Calculate a covariance of the first feature value and the second feature value of all voxel points in the any target region; Obtain a metabolic-structural consistency factor of the any target region according to the covariance, a standard deviation of the first feature value of all voxel points in the any target region, and a standard deviation of the second feature value of all voxel points in the any target region, wherein the covariance is positively correlated with the metabolic-structural consistency factor, and the standard deviation of the first feature value and the standard deviation of the second feature value are negatively correlated with the metabolic-structural consistency factor; The method for screening a suspected abnormal region by comprehensively considering the SUV distribution factor and the metabolic-structural consistency factor comprises: Obtain an abnormal index of each target region according to the SUV distribution factor and the metabolic-structural consistency factor of each target region, wherein the SUV distribution factor is positively correlated with the abnormal index, and the metabolic-structural consistency factor is negatively correlated with the abnormal index; If the abnormal index is greater than a preset abnormal threshold, determine that the corresponding target region is a suspected abnormal region.

2. The method for nuclear medicine functional imaging analysis of nervous system diseases according to claim 1, characterized in that, The method for obtaining the initial ROI block and the non-initial ROI block based on the SUV value, the CT value and the position coordinates of each voxel point in the PET three-dimensional volume image or the CT three-dimensional volume image comprises: Clustering all voxel points by using a DBSCAN algorithm based on the Euclidean distance between feature vectors of different voxel points, to obtain a plurality of blocks; wherein the feature vector is composed of the SUV value, the CT value and the coordinates; Determine a block with a mean SUV value greater than an SUV threshold as an initial ROI block, and determine a block with a mean SUV value less than or equal to the SUV threshold as a non-initial ROI block.

3. The method for nuclear medicine functional imaging analysis of nervous system diseases according to claim 1, characterized in that, The method for correcting the initial ROI block based on the change of the CT value of the voxel point in the voxel point sequence to obtain the target region comprises: The normalized result of the mean of the difference of the CT values of all adjacent voxels in each voxel sequence is taken as the CT characteristic value of each voxel sequence; The mean of the CT characteristic values of all voxel sequences corresponding to any peripheral edema feature block is determined as the outward permeation and expansion characteristic coefficient of any peripheral edema feature block; The peripheral edema feature block corresponding to the maximum outward permeation and expansion characteristic coefficient is taken as the peripheral edema metabolic heterogeneity block; If the mean of the SUV values of all voxels in the peripheral edema metabolic heterogeneity block is greater than the mean of the SUV values of all voxels in the minimum bounding box of the initial ROI block, the region composed of the peripheral edema metabolic heterogeneity block and the minimum bounding box of the initial ROI block is taken as the target region.

4. The method for nuclear medicine functional imaging analysis of nervous system diseases according to claim 1, characterized in that, The boundary blur characteristic value of any target region is obtained according to the gradient value of the surface voxel of the target region and the corresponding first distance, and the method comprises the following steps: The boundary blur characteristic value of any target region is obtained according to the gradient mean value of the surface voxel of the target region, the difference between the first distance corresponding to each voxel on the surface of the target region and the average value of the first distance corresponding to all voxels on the surface of the target region, and the difference between the gradient mean value and the average value of the first distance corresponding to all voxels on the surface of the target region, wherein the gradient mean value and the average value of the first distance corresponding to all voxels on the surface of the target region are positively correlated with the boundary blur characteristic value.

5. A nuclear medicine functional imaging analysis system for neurological disorders, said system being for performing the method of claim 1, characterized in that, The system comprises: A data acquisition module is configured to acquire a PET three-dimensional volume image and a CT three-dimensional volume image of a head of a person to be analyzed; A target region extraction module is configured to obtain an initial ROI block and a non-initial ROI block based on the SUV value, the CT value and the position coordinates of each voxel in the PET three-dimensional volume image or the CT three-dimensional volume image, correct the initial ROI block based on the distance between each voxel on the surface of the initial ROI block and the corresponding minimum bounding volume and the variation characteristic of the SUV value of the voxel between the initial ROI block and the surrounding non-initial ROI block, and obtain a target region; An evaluation module is configured to obtain an SUV distribution factor of each target region based on the distribution characteristic of the SUV value of each voxel in the target region, the distance between each voxel on the surface of the target region and the corresponding minimum bounding volume and the gradient value of the surface voxel, and obtain a metabolic-structural consistency factor of each target region based on the joint trend of the deviation of the SUV value of each voxel in the target region relative to the whole and the deviation of the CT value relative to the whole, the discrete condition of the SUV value of the voxel and the discrete condition of the CT value; A screening module is configured to screen a suspected abnormal region by comprehensively considering the SUV distribution factor and the metabolic-structural consistency factor.

Images