CN111047523A - Method and device for processing PET image and computer storage medium - Google Patents
Method and device for processing PET image and computer storage medium Download PDFInfo
- Publication number
- CN111047523A CN111047523A CN201911095433.9A CN201911095433A CN111047523A CN 111047523 A CN111047523 A CN 111047523A CN 201911095433 A CN201911095433 A CN 201911095433A CN 111047523 A CN111047523 A CN 111047523A
- Authority
- CN
- China
- Prior art keywords
- pet image
- data
- pet
- unit
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012545 processing Methods 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 51
- 230000009466 transformation Effects 0.000 claims abstract description 37
- 238000005070 sampling Methods 0.000 claims abstract description 33
- 239000000654 additive Substances 0.000 claims abstract description 17
- 230000000996 additive effect Effects 0.000 claims abstract description 17
- 238000013501 data transformation Methods 0.000 claims abstract description 8
- 238000004891 communication Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 6
- 238000011084 recovery Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 239000013598 vector Substances 0.000 claims description 5
- 230000001131 transforming effect Effects 0.000 claims description 2
- 238000002372 labelling Methods 0.000 claims 1
- 238000004422 calculation algorithm Methods 0.000 abstract description 3
- 238000002600 positron emission tomography Methods 0.000 description 127
- 230000006870 function Effects 0.000 description 10
- 238000001914 filtration Methods 0.000 description 6
- 206010028980 Neoplasm Diseases 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000009792 diffusion process Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 230000002503 metabolic effect Effects 0.000 description 2
- 230000004060 metabolic process Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000010187 selection method Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 description 1
- 206010061218 Inflammation Diseases 0.000 description 1
- 208000012902 Nervous system disease Diseases 0.000 description 1
- 208000037273 Pathologic Processes Diseases 0.000 description 1
- 238000009098 adjuvant therapy Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 230000019522 cellular metabolic process Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 235000014113 dietary fatty acids Nutrition 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 229930195729 fatty acid Natural products 0.000 description 1
- 239000000194 fatty acid Substances 0.000 description 1
- 150000004665 fatty acids Chemical class 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 239000008103 glucose Substances 0.000 description 1
- 239000012216 imaging agent Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000000338 in vitro Methods 0.000 description 1
- 208000015181 infectious disease Diseases 0.000 description 1
- 230000004054 inflammatory process Effects 0.000 description 1
- 239000003446 ligand Substances 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 239000002207 metabolite Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 102000039446 nucleic acids Human genes 0.000 description 1
- 108020004707 nucleic acids Proteins 0.000 description 1
- 150000007523 nucleic acids Chemical class 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 230000009054 pathological process Effects 0.000 description 1
- 230000035790 physiological processes and functions Effects 0.000 description 1
- 102000004169 proteins and genes Human genes 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000002285 radioactive effect Effects 0.000 description 1
- 239000000700 radioactive tracer Substances 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10104—Positron emission tomography [PET]
Abstract
The invention discloses a method, a device and a computer storage medium for processing PET images, wherein the method comprises the following steps of S1: acquiring original data of a PET image; step S2: modeling the original data, wherein the model comprises additive noise and Poisson noise; step S3: performing a first transformation on the model; step S4: and performing inverse transformation on the data subjected to the first transformation. The method further comprises step T1: acquiring original data of a PET image; step T2: randomly searching a window image with a certain size; step T3: sampling voxels; step T4: and restoring the actual value of the voxel by adopting a region mean value method. The device comprises a data processing unit, a data modeling unit, a data transformation unit and an inverse transformation unit. According to the invention, the mixed Poisson-Gaussian model is adopted to improve the assumption of an early pure Poisson noise model, and then the noise removal after the PET image reconstruction is completed through the denoising algorithm based on the region mean value, so that the quality of the PET image under the low counting condition can be improved.
Description
Technical Field
The present invention relates to a method and apparatus for processing medical images, and more particularly, to a method and apparatus for processing PET images and a computer storage medium.
Background
Positron Emission Tomography (PET) is an imaging device capable of reflecting the genetic, molecular, metabolic and functional states of lesions. PET uses radionuclide with positron to mark human body compounds or metabolites, such as glucose, protein, nucleic acid, fatty acid, receptor ligand and water, etc. as imaging agent, after being injected into the body of the examined person, the positron nuclide is accompanied by the marked substance to participate in the cell metabolism activity of human body tissue and organ in the course of physiological and pathological processes, and then is gathered and distributed again, and the PET detector arranged in vitro is provided for information acquisition, so that the information of human body metabolism activity molecule level can be obtained, and the biological metabolism information of diseases can be provided for clinical treatment.
Currently, PET is widely used in the adjuvant treatment of various diseases, including nervous system diseases, infections, inflammations, tumors, and the like. For example, in radiation oncology applications, PET aids in the localization and staging of tumors. By quantitative indicators based on PET images, it is helpful to delineate the boundaries of the lesion, e.g., disease severity and treatment response assessment requires calculation of metabolic tumor volume over the delineated region. The signal intensity indexes such as maximum and average standard uptake values based on the PET images can be routinely used for cancer staging, tumor characterization, treatment response assessment and the like.
PET image reconstruction includes inherent noise generated by the original as well as other noise. Therefore, noise cancellation methods (i.e., denoising) are of paramount importance, which help to enhance the quantitative indicators and visual quality of PET images, thereby aiding in making better diagnostic decisions. Noise in PET images follows a non-gaussian distribution and due to distortion of the PET image appearance, noise reduces the sensitivity of quantitative metrics based on PET images. Therefore, denoising is a necessary step to improve the quantitative evaluation of PET images. Many effective denoising solutions have been proposed by many doctors and researchers, and these denoising solutions can be broadly classified into filter-based and statistical-based denoising methods at present. The filter-based denoising method mainly includes gaussian smoothing, adaptive diffusion filtering and filtering in an image transformation domain under the support of anatomical information, but the filter-based denoising methods have certain limitations, for example, gaussian filtering often causes information loss due to excessive blurring; adaptive diffusion filtering is not very effective for PET images because the necessary structural information of the diffusion process is limited by the low resolution of PET; filtering in the image transform domain in combination with anatomical information (by corresponding CT or MRI tissue segmentation) may create artifacts since anatomical functions are not always applicable to all voxel locations. Statistical-based denoising methods establish a deterministic relationship between noise and statistical measures, however, strong image structures such as ridges, edges, and textures can negatively impact statistical estimation.
For the processing of PET images, the most important detection index in analyzing and reconstructing a concentration distribution image of a radionuclide is pixel count, and the pixel count value of a certain part in a body can accurately reflect the concentration of a drug of the radionuclide absorbed by the part. The detector is used for detecting biological tissues injected with the radionuclide, the coincidence count of PET is acquired, and the condition that the average radioactive concentration of the radionuclide tracer is 50counts/pixel (counting value/pixel) is a low counting rate condition and 200 counts/pixel is a high counting rate condition is generally considered. In a PET image under a high count condition, the noise distribution can be well approximated by a gaussian distribution based on the central limit theorem. However, for the PET image under the low count condition, the poisson and PVC enhanced additive noise can be observed, so that the noise removal after the PET image reconstruction cannot be well completed by the denoising method in the prior art.
Disclosure of Invention
The invention aims to provide a method, a device and a computer storage medium for processing a PET image, thereby solving the problem that the prior art cannot remove Poisson noise and additive noise in the PET image under the condition of low counting.
The invention provides a method for processing a PET image, which comprises the following steps:
step S1: acquiring original data of a PET image;
step S2: modeling raw data of the PET image in the step S1, the model including additive noise and poisson noise;
step S3: performing a first transformation on the model in the step S2, the first transformation being implemented by:
wherein G represents a function of the first transformation, x represents the intensity of the PET image, α represents a scale term for the relationship between pixel data of the PET image and the Poisson noise, μ and σ represent the expected and standard deviation, respectively, of the additive noise;
step S4: inverse transforming the data that has undergone the first transformation, the inverse transformation being:
IG(y)＝∫G(x)p(x|λ,α)dx
where IG represents the inverse transform and p (x | λ, α) is a probability density function of the variable x associated with the poisson noise.
According to one embodiment of the invention, the raw data of the PET image includes raw count data acquired from PET detectors and voltage, time, position, energy information of the corresponding pulse signals of gamma photons.
According to an embodiment of the invention, in the step S2, the model is represented by the intensity x of the PET image:
x＝αp+n，
where p represents the distribution of poisson noise with a potentially expected value λ and n represents gaussian noise with an expected value μ and standard deviation σ.
According to an embodiment of the present invention, in the step S3, the parameters α are randomly selected and form an a priori database.
According to another embodiment of the present invention, there is provided another method of processing a PET image, the method including the steps of:
step T1: acquiring original data of a PET image;
step T2: randomly searching a window image with a certain size in the PET image;
step T3: sampling voxels within a windowed image region;
step T4: and restoring the actual value of the voxel by adopting a region mean value method, wherein the region mean value method is realized by the following formula:
wherein RM represents a region mean value, Ω is a sampling region of the PET image in the step T3, w is a weight coefficient,
wherein | Au-Av‖2Representing the distance between two intensity vectors of two blocks centered on voxels u and v, and z (u) is a normalization constant.
According to one embodiment of the invention, the raw data of the PET image includes raw count data acquired from PET detectors and voltage, time, position, energy information of the corresponding pulse signals of gamma photons.
According to an embodiment of the present invention, in the step T2, the size of the window image is denoted by M × M, M is a natural number, the intensity of the PET image corresponding to the voxel u is denoted by x (u), the intensity of the PET image corresponding to the voxel v is denoted by x (v), and then different window images are sorted and marked, and if x (u) > x (v), L (u) > L (v) is obtained, where L denotes a number for marking the window image.
According to an embodiment of the present invention, in the step T3, the following rule is followed when the voxels are sampled:
first, randomly sample min { M ] in the region with class label L3D (l (u)) } voxels, D denotes the number of voxels having the same label as voxel u.
Second, when L (u)>1, randomly sampling min { M } in a first neighboring region with a class label L (u) -13D (L (u) -1) } voxel,
third, when L (u)<maxL (u), randomly sample min { M } in the second neighboring region with the category label L (u) +13And D (L (u) +1) } voxel.
The invention provides a device for processing PET images, comprising: the PET image processing device comprises a data processing unit, a data modeling unit, a data transformation unit and an inverse transformation unit, wherein the data processing unit is used for acquiring original data of a PET image; the data modeling unit is in communication connection with the data processing unit and performs modeling processing on the acquired raw data of the PET image, wherein the model comprises additive noise and Poisson noise; the data transformation unit is connected with the data modeling unit in a communication mode and conducts first transformation on the model in the data modeling unit, and the first transformation is achieved through the following modes:
wherein G represents a function of the first transform, x represents an intensity of the PET image, α represents a scale term for a relationship between pixel data of the PET image and the Poisson noise, μ and σ represent an expectation and a standard deviation, respectively, of the additive noise, the inverse transform unit is communicatively connected to the data transform unit and inverse transforms the data subjected to the first transform:
IG(y)＝∫G(x)p(x|λ,α)dx
where IG represents the inverse transform and p (x | λ, α) is a probability density function of the variable x associated with the poisson noise.
The device for processing the PET image, provided by the invention, can further comprise a data processing unit, a data searching unit, a voxel sampling unit and a voxel recovery unit, wherein the data processing unit is used for acquiring original data of the PET image; the data searching unit is in communication connection with the data processing unit and is used for randomly searching the window image in the acquired raw data of the PET image; the voxel sampling unit is in communication connection with the data searching unit and performs voxel sampling on a window image at a searching position; the voxel recovery unit is connected with the voxel sampling unit in a communication way and recovers the actual value of the voxel by adopting a region mean value, wherein the region mean value is realized by the following formula:
wherein RM represents a region mean value, Ω is a sampling region of the PET image in the voxel sampling unit, w is a weight coefficient,
wherein | Au-Av‖2Representing the distance between two intensity vectors of two blocks centered on voxels u and v, and z (u) is a normalization constant.
The invention provides a computer storage medium having stored thereon a computer program which, when executed, performs the steps of the method of any one of the above.
According to the method, the device and the computer storage medium for processing the PET image, provided by the invention, the early pure Poisson noise model assumption is improved by adopting the mixed Poisson-Gaussian model, and then the noise removal after the PET image reconstruction is completed by the denoising algorithm based on the region mean value, so that the quality of the PET image under the low counting condition can be improved, and the method, the device and the computer storage medium are also suitable for the PET image obtained under other counting conditions. In addition, the present invention is very effective for cost control that increases computational complexity.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic illustration of the steps of a method of processing a PET image according to one embodiment of the invention;
FIG. 2 is a schematic illustration of the steps of a method of processing a PET image according to another embodiment of the invention;
FIG. 3 is a schematic diagram of an apparatus for processing PET images according to one embodiment of the present invention;
fig. 4 is a schematic view of an apparatus for processing a PET image according to another embodiment of the present invention.
Detailed Description
The present invention will be further described with reference to the following specific examples. It should be understood that the following examples are illustrative only and are not intended to limit the scope of the present invention.
It will be understood that when an element/feature is referred to as being "disposed on" another element/feature, it can be directly on the other element/feature or intervening elements/features may also be present. When a component/part is referred to as being "connected/coupled" to another component/part, it can be directly connected/coupled to the other component/part or intervening components/parts may also be present. The term "connected/coupled" as used herein may include electrical and/or mechanical physical connections/couplings. The term "comprises/comprising" as used herein refers to the presence of features, steps or components/features, but does not preclude the presence or addition of one or more other features, steps or components/features. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.
In addition, in the description of the present invention, the terms "first", "second", and the like are used for descriptive purposes only and to distinguish similar objects, and there is no precedence between them, and no indication or suggestion of relative importance is understood. In addition, in the description of the present invention, "a plurality" means two or more unless otherwise specified.
Fig. 1 is a schematic diagram illustrating steps of a method for processing a PET image according to a preferred embodiment of the present invention, and as can be seen from fig. 1, the method for processing a PET image provided by the present invention may include the following steps:
step S1: acquiring original data of a PET image;
step S2: modeling raw data of the PET image in step S1, the model including additive noise and poisson noise;
step S3: performing a first transformation on the model in step S2;
step S4: and performing inverse transformation on the data subjected to the first transformation.
Specifically, in the above step S1, the acquisition of the PET image may be performed by a reconstruction method conventional in the art, and the preferred acquired PET image is a PET image under a low count condition. The raw data of the PET image may include raw count data acquired by the PET detector and information such as voltage, time, position, energy, etc. of the pulse signal corresponding to the gamma photon acquired by the PET signal acquisition circuit, and the acquisition of these data is easy to acquire by those skilled in the art and will not be described herein again.
Regardless of the modality of the acquired PET image, the noise model commonly employed in the prior art is a gaussian noise model, which is in turn commonly referred to as additive noise or additive white noise. In the gaussian noise model, it is assumed that the noise at different spatial positions is independent and equally distributed, and therefore, the gaussian noise can be successfully suppressed through an averaging operation, which belongs to a common noise removing means for those skilled in the art and is not described herein again.
However, in step S2, since the acquired PET image is a PET image under a low count rate condition, the gaussian noise model is no longer applicable at this time, because the noise in the PET image at this time is more complex, and includes at least the following sources: true signal, random signal and scatter coincidence during gamma photon counting follow a poisson distribution, and additive noise is introduced during partial volume correction and other image reconstruction processes. Therefore, in step S2, the raw data of the PET image is modeled, and the model includes additive noise and poisson noise, so that the model becomes a hybrid poisson-gaussian model, and non-gaussian noise is converted into gaussian space, which makes data processing more convenient and closer to the real state than a single model. Utensil for cleaning buttockIn some embodiments, the modeling process may include: let P denote the Poisson distribution with a potential expectation (and variance) λ, i.e., P to P (λ), and N denote the Gaussian noise with an expectation μ and standard deviation σ, i.e., N to N (μ, σ)2) The intensity x of the observed PET image is then defined as:
x＝αp+n (1)
where x denotes the intensity of the PET image and α denotes the scale term that takes into account the relationship between the observed pixel data and the assumed poisson distribution model, in which case the model of the intensity of the PET image is also referred to as the hybrid poisson-gaussian model.
It should be noted by those skilled in the art that, in the above step S2, the hybrid poisson-gaussian model is also applicable to PET images acquired under other counting conditions, and will not be described in detail herein.
In the above step S3, for the intensity x of the observed PET image, the first transformation can be implemented as follows:
where y g (x) has an approximate unit variance, or y g (x) follows a gaussian distribution with a variance of approximately 1. These parameters can be set a priori based on the resolution and homogeneity of the PET images, e.g., clinical and preclinical PET images show different resolution features, and the homogeneity of the images can also be used to fine tune the above parameters, estimated by an inferred parameter selection method, where the parameters in a single noisy image can be estimated by fitting a global parameter model to local estimated expectation and standard deviation pairs. Resolution and homogeneity parameters of the PET image can be obtained by conventional technical means, which are not difficult for the person skilled in the art and will not be described in detail herein.
In the above step S4, the specific step of inverse-transforming the first transformed data may include inverse-transforming y ═ g (x), λ and α determined in step S3
IG(y)＝∫G(x)p(x|λ,α)dx (3)
Where p (x | λ, α) is the probability density function of the variable x in relation to the Poisson distribution family.
In the above steps, the intensity of the PET image is converted by the first transformation, then the inverse transformation of the first transformation is utilized to remove the gauss noise, and the denoised PET image can be converted back to the original image space by the inverse transformation operation, so that the intensity information of the PET image is not lost.
Fig. 2 is a schematic diagram illustrating steps of a method of processing a PET image according to another embodiment of the present invention, as can be seen from fig. 2, the method of processing a PET image may include the steps of:
step T1: acquiring original data of a PET image;
step T2: randomly searching a window image with a certain size;
step T3: sampling voxels;
step T4: and restoring the actual value of the voxel by using the region mean value.
Specifically, in the above step T1, the acquisition of the PET image may be performed by a reconstruction method conventional in the art, and the preferred acquired PET image is a PET image under a low count condition. The PET image in step T1 may also be the PET image obtained by the first transformation in the first embodiment described above. The raw data of the PET image may include raw count data acquired by the PET detector and information such as voltage, time, position, energy, etc. of the pulse signal corresponding to the gamma photon acquired by the PET signal acquisition circuit, and the acquisition of these data is easy to acquire by those skilled in the art and will not be described herein again.
In the above step T2, the searched PET image is a three-dimensional PET image having different volume elements (i.e. voxels), for example, the size of the selected window may be labeled as M × M, where M is a natural number. It should be noted by those skilled in the art that any number of window images may be searched, for example, a plurality of window images may be searched, each window image includes a different number of voxels, the intensity corresponding to the voxel u is denoted as x (u), and the intensity corresponding to the voxel v is denoted as x (v), and then the different window images are sorted and labeled, if x (u) > x (v), L (u) > L (v) is obtained, where L denotes a label number, such as 0, 1, 2, 3, … ….
In the above step T3, the following rule is followed when sampling the voxels:
first, randomly sample min { M ] in the region with class label L3D (l (u)) } voxels, D denotes the number of voxels having the same label as voxel u.
Second, if L (u)>1, randomly sampling min { M } in a first neighboring region with a class label L (u) -13D (L (u) -1) } voxel,
third, if L (u)<maxL (u), then randomly sample min { M } in the second neighboring region with the category label L (u) +13And D (L (u) +1) } voxel.
Potential artificial boundaries created by the constrained smoothing are avoided by introducing additional samples from the first and second neighboring regions in step T3. By introducing a neighboring region, this self-enhancing effect can be eliminated, more true for more separations under severe noise/artifacts.
In the above step T4, the specific steps of recovering the actual value of the voxel by using the region mean value are as follows:
where RM denotes the region mean, Ω is the sampled region of the PET image in the above step T3, w (u, v) is a weight coefficient, and w may be based on two blocks A centered on voxels u and vuAnd AvThe similarity between them is defined so that
Wherein | Au-Av‖2Representing the distance between two intensity vectors from two blocks, Z (u) being a normalization constant
The weight coefficient w satisfies the condition: 0< w (u, v) ≦ 1, Σ w (u, v) ≦ 1, and the weight coefficient w determines the degree of filtering.
By the above voxel sampling, a sufficient number of voxels can be covered, which contribute to the de-noising of voxels. By performing region mean, denoising results in a more reliable segmentation, since a smooth estimated histogram of the PET image can lead to better threshold levels, thus maximizing the distance between clusters.
Fig. 3 is a schematic diagram of an apparatus for processing a PET image according to an embodiment of the present invention, and as can be seen from fig. 3, the apparatus for processing a PET image provided by the present invention includes:
a data processing unit 11, wherein the data processing unit 11 is configured to acquire raw data of a PET image, the PET image is preferably a PET image under a low-count condition, and the raw data of the PET image may include raw count data acquired by a PET detector and information of voltage, time, position, energy, and the like of a pulse signal corresponding to a gamma photon acquired by a PET signal acquisition circuit;
the data modeling unit 12 is in communication connection with the data processing unit 11, and performs modeling processing on the acquired raw data of the PET image, wherein the model comprises additive noise and Poisson noise;
a data transformation unit 13, the data transformation unit 13 is connected with the data modeling unit 12 in a communication way and carries out first transformation on the model in the data modeling unit 12; and
an inverse transform unit 14, the inverse transform unit 14 communicatively connected with the data transform unit 13 and inversely transforms the data subjected to the first transform.
More specifically, the modeling process of the data modeling unit 12 may include: let P denote the Poisson distribution with a potential expectation (and variance) λ, i.e., P to P (λ), and N denote the Gaussian noise with an expectation μ and standard deviation σ, i.e., N to N (μ, σ)2) The intensity x of the observed PET image is then defined as:
x＝αp+n
where x represents the intensity of the PET image and α represents the scale term that accounts for the relationship between the observed pixel data and the assumed poisson distribution model.
The data transformation unit 13 may implement the first transformation by:
where y g (x) has an approximate unit variance, or y g (x) follows a gaussian distribution with a variance of approximately 1. These parameters can be set a priori based on the resolution and homogeneity of the PET images, e.g., clinical and preclinical PET images show different resolution features, and the homogeneity of the images can also be used to fine tune the above parameters, estimated by an inferred parameter selection method, where the parameters in a single noisy image can be estimated by fitting a global parameter model to local estimated expectation and standard deviation pairs.
The specific step of the inverse transformation unit 14 inverse-transforming the first transformed data may include inverse transformation into y ═ g (x), λ and α determined by the data transformation unit 13
IG(y)＝∫G(x)p(x|λ,α)dx
Where p (x | λ, α) is the probability density function of the variable x in relation to the Poisson distribution family.
Fig. 4 is a schematic diagram of an apparatus for processing a PET image according to another embodiment of the present invention, and as can be seen from fig. 4, the apparatus for processing a PET image according to the present invention may further include:
a data processing unit 21, wherein the data processing unit 21 is configured to acquire raw data of a PET image, the PET image is preferably a PET image under a low-count condition, and the PET image may also be a PET image obtained by the first transformation in the first embodiment;
a data search unit 22, the data search unit 22 being communicatively connected to the data processing unit 21 and being capable of arbitrarily searching for a window image having a certain size in the raw data of the acquired PET image;
a voxel sampling unit 23, wherein the voxel sampling unit 23 is connected with the data searching unit 22 in a communication way and performs voxel sampling on the window image at the searching position; and
a voxel recovery unit 24, the voxel recovery unit 24 being in communication with the voxel sampling unit 23 and recovering the actual value of the voxel using the region mean.
More specifically, the search process of the data search unit 22 may include: the size of the selected window may be marked as M × M, where M is a natural number, a plurality of window images are searched, each window image includes different numbers of voxels, the intensity corresponding to the voxel u is marked as x (u), and the intensity corresponding to the voxel v is marked as x (v), then different window images are sorted and marked, and if x (u) > x (v), L: > L (v) is obtained, where L denotes a serial number of a mark, such as 0, 1, 2, 3, and … ….
The voxel sampling unit 23 follows the following rule when sampling voxels:
first, randomly sample min { M ] in the region with class label L3D (l (u)) } voxels, D denotes the number of voxels having the same label as voxel u.
Second, if L (u)>1, randomly sampling min { M } in a first neighboring region with a class label L (u) -13D (L (u) -1) } voxel,
third, if L (u)<maxL (u), then randomly sample min { M } in the second neighboring region with the category label L (u) +13And D (L (u) +1) } voxel.
The specific steps of the voxel recovery unit 24 recovering the actual value of the voxel by using the region mean value are as follows:
where RM denotes the region mean, Ω is the sampling region of the voxel sampling unit 23, w (u, v) is a weight coefficient, and w may be based on two blocks a centered around voxels u and vuAnd AvThe similarity between them is defined so that
Wherein | Au-Av‖2Representing data from two blocksZ (u) is a normalization constant.
Further, the present invention also provides a computer storage medium for processing PET images, the computer storage medium storing a computer program, which when executed by a processor can implement the steps of the above method, embodiments, such as: acquiring original data of a PET image; randomly searching a window image with a certain size; sampling voxels; and restoring the actual value of the voxel by using the region mean value. Further, the computer program may be executed to model the raw data of the PET image, and perform the first transformation and the inverse transformation on the model in step S2.
For a detailed description of this embodiment, reference may be made to the detailed description of the PET image processing in the above respective embodiments, which is not described redundantly here.
The computer storage medium of the present invention may comprise any entity or device capable of carrying computer program code, such as ROM/RAM, magnetic disks, optical disks, flash memory, etc., which are not further enumerated herein.
The apparatuses, units and the like explained in the above embodiments may be specifically implemented by a computer chip and/or an entity, or implemented by a product with certain functions. For convenience of description, the above devices are described as being divided into various modules by functions, and are described separately. Of course, the functions of the modules may be integrated into one or more computer chips when implementing the embodiments of the present invention.
According to the method, the device and the computer storage medium for processing the PET image, the state of a denoising algorithm can be improved through the mixed Poisson-Gaussian model, denoising is performed through the region mean value operation, and the method, the device and the computer storage medium are not only suitable for the PET image under the condition of low counting, but also suitable for processing the common PET image.
The above embodiments are merely preferred embodiments of the present invention, which are not intended to limit the scope of the present invention, and various changes may be made in the above embodiments of the present invention. All simple and equivalent changes and modifications made according to the claims and the content of the specification of the present application fall within the scope of the claims of the present patent application. The invention has not been described in detail in order to avoid obscuring the invention.
Claims (11)
1. A method of processing a PET image, the method comprising the steps of:
step S1: acquiring original data of a PET image;
step S2: modeling raw data of the PET image, wherein the model comprises additive noise and Poisson noise;
step S3: performing a first transformation on the model, the first transformation being implemented by:
wherein G represents a function of the first transformation, x represents the intensity of the PET image, α represents a scale term for the relationship between pixel data of the PET image and the Poisson noise, μ and σ represent the expected and standard deviation, respectively, of the additive noise;
step S4: inverse transforming the data that has undergone the first transformation, the inverse transformation being:
IG(y)＝∫G(x)p(x|λ,α)dx
where IG denotes the inverse transform and p (x | λ, α) denotes the probability density function with respect to x.
2. The method of processing a PET image of claim 1, wherein the raw data of the PET image includes raw count data acquired from PET detectors and voltage, time, position, energy information of corresponding pulse signals of gamma photons.
3. The method of processing a PET image of claim 1, wherein in the step S2, the model is represented by an intensity x of the PET image:
x＝αp+n，
where p represents the distribution of poisson noise with a potentially expected value λ and n represents gaussian noise with an expected value μ and standard deviation σ.
4. The method of processing a PET image of claim 1, wherein in the step S3, the values of the parameters α are randomly selected and form an a priori database.
5. A method of processing a PET image, the method comprising the steps of:
step T1: acquiring original data of a PET image;
step T2: randomly searching a window image in the PET image;
step T3: sampling voxels within the windowed image region;
step T4: and restoring the actual value of the voxel by adopting a region mean value method, wherein the region mean value method is realized by the following formula:
wherein RM represents a region mean value, Ω is a sampling region of the PET image in the step T3, w is a weight coefficient,
wherein | Au-Av‖2Represents the distance between two intensity vectors of two blocks centered on voxels u and v, and z (u) represents a normalization constant.
6. The method of processing a PET image of claim 5, wherein the raw data of the PET image includes raw count data acquired from PET detectors and voltage, time, position, energy information of corresponding pulse signals of gamma photons.
7. The method according to claim 5, wherein in step T2, the size of the window image is denoted by M x M, M is a natural number, the intensity of the PET image corresponding to voxel u is denoted by x (u), the intensity of the PET image corresponding to voxel v is denoted by x (v), and then different window images are sorted and labeled, and if x (u) > x (v), L (u) > L (v) is obtained, where L denotes a number for labeling the window image.
8. The method of processing a PET image of claim 7, wherein in the step T3, the following rule is followed when sampling voxels:
first, randomly sample min { M ] in the region with class label L3D (l (u)) } voxels, D representing the number of voxels having the same label as voxel u;
second, when L (u)>1, randomly sampling min { M } in a first neighboring region with a class label L (u) -13D (l) (u) -1) } voxels;
third, when L (u)<maxL (u), randomly sample min { M } in the second neighboring region with the category label L (u) +13D (l (u) +1) } voxels.
9. An apparatus for processing a PET image, the apparatus comprising:
a data processing unit for acquiring raw data of a PET image;
the data modeling unit is in communication connection with the data processing unit and performs modeling processing on the acquired raw data of the PET image, and the model comprises additive noise and Poisson noise;
a data transformation unit communicatively connected with the data modeling unit and performing a first transformation on a model in the data modeling unit, the first transformation being implemented by:
wherein G represents a function of the first transformation, x represents the intensity of the PET image, α represents a scale term for the relationship between the pixel data of the PET image and the Poisson noise, μ and σ represent the expected and standard deviation, respectively, of the additive noise, and
an inverse transform unit communicatively connected to the data transform unit and inverse-transforming the first transformed data, the inverse transform to:
IG(y)＝∫G(x)p(x|λ,α)dx
where IG represents the inverse transform and p (x | λ, α) is a probability density function of the variable x associated with the poisson noise.
10. An apparatus for processing a PET image, the apparatus comprising:
a data processing unit for acquiring raw data of a PET image;
a data search unit which is connected with the data processing unit in a communication way and arbitrarily searches window images in the acquired raw data of the PET images;
a voxel sampling unit communicatively connected to the data search unit and voxel-sampling a window image at a search site; and
a voxel recovery unit, communicatively connected to the voxel sampling unit, for recovering an actual value of a voxel using a region mean, the region mean being implemented by the following formula:
wherein RM represents a region mean value, Ω is a sampling region of the PET image in the voxel sampling unit, w is a weight coefficient,
wherein | Au-Av‖2Representing the distance between two intensity vectors of two blocks centered on voxels u and v, and z (u) is a normalization constant.
11. A computer storage medium, having a computer program stored thereon, which, when executed, performs the steps of the method according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911095433.9A CN111047523A (en) | 2019-11-11 | 2019-11-11 | Method and device for processing PET image and computer storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911095433.9A CN111047523A (en) | 2019-11-11 | 2019-11-11 | Method and device for processing PET image and computer storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111047523A true CN111047523A (en) | 2020-04-21 |
Family
ID=70231907
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911095433.9A Pending CN111047523A (en) | 2019-11-11 | 2019-11-11 | Method and device for processing PET image and computer storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111047523A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111667428A (en) * | 2020-06-05 | 2020-09-15 | 北京百度网讯科技有限公司 | Noise generation method and device based on automatic search |
| CN111709897A (en) * | 2020-06-18 | 2020-09-25 | 深圳先进技术研究院 | Method for reconstructing positron emission tomography image based on domain transformation |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120281896A1 (en) * | 2010-01-28 | 2012-11-08 | Timo Aspelmeier | Tomographic imaging using poissonian detector data |
| US20150356712A1 (en) * | 2014-06-04 | 2015-12-10 | The Johns Hopkins University | Model-based tomographic reconstruction with correlated measurement noise |
| CN107169932A (en) * | 2017-03-21 | 2017-09-15 | 南昌大学 | A kind of image recovery method based on Gauss Poisson mixed noise model suitable for neutron imaging system diagram picture |
| CN108596995A (en) * | 2018-05-15 | 2018-09-28 | 南方医科大学 | A kind of PET-MRI maximum a posteriori joint method for reconstructing |
| CN109559283A (en) * | 2018-10-08 | 2019-04-02 | 浙江工业大学 | Medicine PET image denoising method based on the domain DNST bivariate shrinkage and bilateral non-local mean filtering |
-
2019
- 2019-11-11 CN CN201911095433.9A patent/CN111047523A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120281896A1 (en) * | 2010-01-28 | 2012-11-08 | Timo Aspelmeier | Tomographic imaging using poissonian detector data |
| US20150356712A1 (en) * | 2014-06-04 | 2015-12-10 | The Johns Hopkins University | Model-based tomographic reconstruction with correlated measurement noise |
| CN107169932A (en) * | 2017-03-21 | 2017-09-15 | 南昌大学 | A kind of image recovery method based on Gauss Poisson mixed noise model suitable for neutron imaging system diagram picture |
| CN108596995A (en) * | 2018-05-15 | 2018-09-28 | 南方医科大学 | A kind of PET-MRI maximum a posteriori joint method for reconstructing |
| CN109559283A (en) * | 2018-10-08 | 2019-04-02 | 浙江工业大学 | Medicine PET image denoising method based on the domain DNST bivariate shrinkage and bilateral non-local mean filtering |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111667428A (en) * | 2020-06-05 | 2020-09-15 | 北京百度网讯科技有限公司 | Noise generation method and device based on automatic search |
| CN111709897A (en) * | 2020-06-18 | 2020-09-25 | 深圳先进技术研究院 | Method for reconstructing positron emission tomography image based on domain transformation |
| CN111709897B (en) * | 2020-06-18 | 2023-10-24 | 深圳先进技术研究院 | Domain transformation-based positron emission tomography image reconstruction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Attenuation correction of PET/MR imaging | |
| US20190343477A1 (en) | Methods and systems for extracting blood vessel | |
| CN111008984B (en) | Automatic contour line drawing method for normal organ in medical image | |
| US8218848B2 (en) | System and method for the generation of attenuation correction maps from MR images | |
| WO2021030629A1 (en) | Three dimensional object segmentation of medical images localized with object detection | |
| Azhari et al. | Brain tumor detection and localization in magnetic resonance imaging | |
| US8144953B2 (en) | Multi-scale analysis of signal enhancement in breast MRI | |
| US7457447B2 (en) | Method and system for wavelet based detection of colon polyps | |
| Nasor et al. | Detection and localization of early-stage multiple brain tumors using a hybrid technique of patch-based processing, k-means clustering and object counting | |
| Koundal et al. | Challenges and future directions in neutrosophic set-based medical image analysis | |
| Zhong et al. | Automated white matter hyperintensity detection in multiple sclerosis using 3D T2 FLAIR | |
| Xu et al. | Segmentation based denoising of PET images: An iterative approach via regional means and affinity propagation | |
| US20110213243A1 (en) | Visualization and quantization of newly formed vasculature | |
| Mredhula et al. | An extensive review of significant researches on medical image denoising techniques | |
| EP2577604A1 (en) | Processing system for medical scan images | |
| CN111047523A (en) | Method and device for processing PET image and computer storage medium | |
| US20090069665A1 (en) | Automatic Lesion Correlation in Multiple MR Modalities | |
| US7711164B2 (en) | System and method for automatic segmentation of vessels in breast MR sequences | |
| Jin et al. | Mask R-CNN Models to Purify Medical Images of Training Sets | |
| Lee et al. | Tumor segmentation from small animal PET using region growing based on gradient magnitude | |
| Chan et al. | A non-local post-filtering algorithm for PET incorporating anatomical knowledge | |
| Muthu et al. | Morphological operations in medical image pre-processing | |
| Yadav et al. | Image sectionalization techniques: A review | |
| Hamid et al. | A Method for Enhancing PET Scan Images Using Nonlocal Mean Filter | |
| Gautam et al. | Implementation of NLM and PNLM for de-noising of MRI images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200421 |