CN113111874A - Rapid classification method for lateral images of lumbar vertebrae - Google Patents
Rapid classification method for lateral images of lumbar vertebrae Download PDFInfo
- Publication number
- CN113111874A CN113111874A CN202110359760.1A CN202110359760A CN113111874A CN 113111874 A CN113111874 A CN 113111874A CN 202110359760 A CN202110359760 A CN 202110359760A CN 113111874 A CN113111874 A CN 113111874A
- Authority
- CN
- China
- Prior art keywords
- image
- lateral lumbar
- lateral
- images
- lumbar
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06T2207/10116—X-ray image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20064—Wavelet transform [DWT]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30008—Bone
Abstract
The invention provides a method for rapidly classifying lateral lumbar images, which comprises the following steps of S1: collecting lateral lumbar images, and selecting an interested area on the lateral lumbar images; s2: acquiring texture features; s3: secondary extraction is carried out after feature selection; s4: and performing a classifier identification process to obtain a classification result. The invention realizes the decomposition and reconstruction of the lateral lumbar image according to the nonlinear enhancement algorithm of the controllable pyramid decomposition, can improve the spatial resolution of the image, and can effectively enhance the image of the bone microstructure and other detailed areas with low contrast on the image, so that the display on the image is clear.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a method for rapidly classifying lateral images of lumbar vertebrae.
Background
At present, the incidence rate of thoracolumbar fractures is on the rise, and particularly, severe thoracolumbar fractures with spinal cord injuries become one of the key points of orthopedic clinical research. With the development of spinal biomechanics and the development of imaging, many new theories and techniques for treating thoracolumbar fractures emerge, but there are still controversies about the research of classification systems, the grasp of surgical indications, direct or indirect decompression, the selection of fixed segments, fusion or non-fusion treatments, and the like.
Due to the factors of noise, improper X-ray exposure, over-thick human tissues and the like in the imaging data, the detailed structure of the lumbar vertebra is annihilated by the noise, the edge is fuzzy, the contrast is low, missed diagnosis and misdiagnosis of the lesion tissue are easily caused, and the development and wide use of the imaging are restricted.
The change of bone microstructure is influenced by factors such as gender, age group difference and the like, and the traditional adaptive unsharp masking algorithm adopts local variance as a parameter to enhance the proportion of high-frequency components, so that the edge and the detail of an image are effectively enhanced, but the traditional adaptive unsharp masking algorithm is sensitive to noise, cannot effectively enhance a part with low contrast in an original image and is easy to generate artifacts.
Disclosure of Invention
In view of the above, the present invention provides a method for rapidly classifying lateral lumbar images.
In order to solve the technical problems, the invention adopts the technical scheme that: a method for rapidly classifying lateral lumbar images comprises the following steps:
s1: collecting lateral lumbar images, and selecting an interested area on the lateral lumbar images;
s2: acquiring texture features;
s3: secondary extraction is carried out after feature selection;
s4: and performing a classifier identification process to obtain a classification result.
In the present invention, preferably, in the step S3, a controllable pyramid nonlinear enhancement algorithm is adopted in the secondary extraction process after the feature selection, and the controllable pyramid nonlinear enhancement algorithm specifically includes the following steps:
s31: high-frequency subband coefficients in four directions are used as directional gradients to estimate local contrast;
s32: dividing the image into a low-contrast detail region, a possible edge region and a smooth region according to a multi-mode selection strategy;
s33: carrying out edge detection on the possible edge area to obtain a target correction boundary;
s34: adopting multi-mode nonlinear mapping functions for high-frequency sub-bands in different regions;
s35: and performing inverse transformation on the adjusted controllable pyramid coefficient, and executing an image reconstruction step.
In the present invention, preferably, the step S1 is performed on the lateral lumbar image before the region of interest is selected on the lateral lumbar image in the step S1LI-MMSE AlgorithmA denoising step S5 is performed.
In the present invention, preferably, the denoising step S5 specifically includes the following steps:
s51: performing double-tree complex wavelet transformation on the lateral lumbar vertebra image to obtain a low-frequency sub-band and a high-frequency sub-band;
s52: by window size NkEstimate the noise parameter sigma by the local variance ofnAnd σk;
S53: correcting the high-frequency subband coefficient by adopting MMSE estimation;
s54: and performing inverse wavelet transform on the adjusted high-frequency wavelet coefficient, and executing an image reconstruction step.
In the present invention, preferably, the image reconstruction step is to perform contrast enhancement on the low-frequency subband and the high-frequency subband obtained by decomposition at each level, and then add the low-frequency subband and the high-frequency subband to obtain a processed image.
In the present invention, preferably, the classifier identification process uses any one or a combination of BP neural network, regression tree or SVM classifier.
In the present invention, preferably, a gray level co-occurrence matrix method is adopted to obtain texture features, where the texture features include autocorrelation coefficients, contrast, correlation, salient clusters, dark clusters, non-similarity, energy, entropy, homogeneity, maximum probability, variance, sum-average, sum-variance, sum-entropy, difference variance, difference entropy, correlation information a, correlation information B, normalized inverse difference, and normalized inverse difference.
In the present invention, it is preferable that the T value on the lateral lumbar vertebra image is acquired, and the lateral lumbar vertebra image is divided into the normal bone mass group, the reduced bone mass group, and the osteoporosis group according to the sample division rule.
In the present invention, preferably, the sample partition rule is set such that when the T > -1 condition is established, the lateral lumbar vertebra image is normal in bone mass; when the-1.0 SD > T > -2.5 condition is satisfied, the lumbar vertebra lateral image shows that the bone mass is reduced; when the condition of T < -2.5 is satisfied, the lumbar vertebra lateral image is osteoporosis.
In the present invention, preferably, the T value on the lateral lumbar image is measured by a dual-energy X-ray absorption detector.
In the present invention, it is preferable that the pixel value range of the region of interest is a difference between a maximum pixel value of the lateral lumbar image and a minimum pixel value of the lateral lumbar image.
In the present invention, preferably, the denoising step is followed by calculating using a peak signal-to-noise ratio calculation formulaAnd as a judgment index for measuring the denoising effect, wherein I (x, y) represents the pixel gray value of the image before processing, and I' (x, y) represents the pixel gray value of the image after processing.
In the present invention, preferably, the window size NkSet to 3 x 3.
The invention has the advantages and positive effects that: according to the invention, the interesting region is selected from the collected lumbar lateral image to obtain texture features, the features are extracted for the second time after being selected, the classification and identification processes are carried out through the classifier, and finally the classification result is obtained, the decomposition and reconstruction of the lumbar lateral image are realized through the nonlinear enhancement algorithm of controllable pyramid decomposition, the spatial resolution of the image can be improved, and the effective image enhancement can be carried out on the low-contrast detailed regions such as bone microstructures and the like on the image, so that the display on the image is clear, the algorithm has approximate translation invariance and more direction selectivity, and the enhancement of the lumbar overlapping edge and the bone microstructure edge is effectively improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic overall flow chart of a rapid classification method for lateral lumbar images according to the present invention;
FIG. 2 is a flowchart of a controllable pyramid non-linear enhancement algorithm of the rapid lumbar vertebrae lateral image classification method of the present invention;
FIG. 3 is a flowchart of the denoising step of the lumbar vertebrae lateral image fast classification method of the present invention;
fig. 4 is a flowchart of lumbar vertebrae lateral image division of the rapid lumbar vertebrae lateral image classification method of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It will be understood that when an element is referred to as being "secured to" another element, it can be directly on the other element or intervening elements may also be present. When a component is referred to as being "connected" to another component, it can be directly connected to the other component or intervening components may also be present. When a component is referred to as being "disposed on" another component, it can be directly on the other component or intervening components may also be present. The terms "vertical," "horizontal," "left," "right," and the like as used herein are for illustrative purposes only.
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 invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
As shown in fig. 1 to 4, the present invention provides a method for rapidly classifying lateral lumbar images, comprising the following steps:
s1: collecting lateral lumbar images, and selecting an interested area on the lateral lumbar images;
s2: acquiring texture features, wherein the texture is inherent features of an image, and can be used for describing not only pixels in a certain specified area, but also the relationship between the pixels and peripheral pixels of the pixels, reflecting the inherent attributes of the image, and realizing classification diagnosis of diseases by acquiring texture information of the image;
s3: secondary extraction is carried out after feature selection, and the process mainly screens out features with the best effect from all the features, so that the process of reducing the dimension is achieved and is used as a key step of classification and identification;
s4: and performing a classifier identification process to obtain a classification result.
The proportion of people experiencing excessive lumbago in a lifetime is high, the excessive lumbago is caused by lumbar instability to a large extent, and the factors for becoming lumbar instability are many.
The parts of the lumbar vertebra which bear the maximum gravity load of the human body have certain lordosis physiological curvature and physiological motion amplitude, and also have the capability of maintaining the physiological position balance of the parts. The lumbar vertebra stabilization is maintained by the mutual matching and linkage of all parts including inherent tension of a vertebral body, an intervertebral joint, a joint capsule, a ligament, an intervertebral disc and the ligament, a supporting effect and a stress induction effect are achieved in the spinal motion, stress change is timely fed back to a nerve control system, the nerve control system responds to the stress and achieves the purpose of stabilizing the spinal column by moving corresponding organs such as a paraspinal muscle group, and if the lumbar vertebra segment is displaced to an over limit and loses the capability of restoring the physiological state, the lumbar vertebra instability is called. At present, the stability of the lumbar vertebrae is evaluated by measuring parameters such as physiological curvature of the lumbar vertebrae, movement range between vertebral bodies, angle change and the like.
Under normal conditions, the lumbar vertebrae has a forward physiological curvature, usually about 15-20 degrees, which generally does not exceed 30 degrees, and when the physiological curvature is straightened or even reversely arched, the lumbar vertebrae is usually suggested to have degenerative changes, which are mostly caused by long-time sitting and frequent lumbar strain. The patient can judge the specific disease condition by means of the X-ray film of the lumbar vertebra.
Studies have shown that bone strength depends on bone mass (BMD measurement) and bone microarchitecture, and that changes in bone microarchitecture can reflect the extent of osteoporosis earlier and more accurately. DXA, as a standard for diagnosing osteoporosis, accurately reflects bone mass and has limited ability to assess the microstructure of bone microstructure. DXA reflects two-dimensional images, cannot independently evaluate cortical bone and cancellous bone, the T value is easily influenced by factors such as bone size and hyperosteogeny, and compared with the method for simply measuring bone density, the integrity of the cancellous bone microstructure of the vertebral body can better reflect the osteoporosis degree. The spongy bone is formed by interaction of a rod-shaped bone microstructure and a plate-shaped bone microstructure according to different proportions, and when the bone microstructure changes, such as the bone microstructure becomes thin, holes are punched, the bone microstructure is broken, gaps are increased, and the number of the bone microstructure is reduced, the absorption rate of X-rays to the spongy bone microstructure is changed, so that the spongy bone is represented as a change of gray texture in an image.
The change of bone microstructure is influenced by gender and age difference, the traditional adaptive unsharp masking algorithm adopts local variance as a parameter to enhance the proportion of high-frequency components, so as to effectively enhance the edge and the detail of an image, but the traditional adaptive unsharp masking algorithm is sensitive to noise, cannot effectively enhance the part with low contrast in the original image and is easy to generate artifacts.
Too many features will prolong the training time, increase the complexity of the classifier and even affect the actual diagnosis, so it can be seen that the more the classifier is selected, the better the classifier is. The secondary extraction is performed after the feature parameters are subjected to the feature selection in step S3, so that the accuracy of classification after the feature selection can be further improved. The method is characterized in that similar or overlapped features can be filtered out during feature selection, features with larger differences or no overlap are reserved, and feature differences among sample groups can be maximized, so that the classification accuracy is greatly improved. The classifier is adopted to classify, identify and compare every two of the normal bone mass group, the reduced bone mass group and the osteoporosis group, and the accuracy rate is higher than that of the normal bone mass group and the osteoporosis group, but the abnormal degree of the sample group with the abnormal condition can not be well distinguished because the classifier can distinguish the normal sample group from the abnormal sample group.
In this embodiment, further, in the step S3, a controllable pyramid nonlinear enhancement algorithm is adopted in the secondary extraction process after the feature selection, and the controllable pyramid nonlinear enhancement algorithm specifically includes the following steps:
s31: the high-frequency subband coefficients in four directions are used as directional gradients to estimate local contrast, and a contrast formula is utilized:where I (I, j) represents a pixel value,the average edge value of eight domains of the pixel is represented, and delta represents a Laplace operator;
estimating local contrast by using the high-frequency subband coefficients in four directions of the controllable pyramid as directional gradients, and defining the Nth-level low-frequency subband as INFour directional high frequency subbands are respectively fN1、fN2、fN3And fN4The coefficient of the high-frequency sub-band has eight decomposition directions, and the average edge value of the pixel of the Nth level low-frequency sub-band in the controllable pyramid domain is obtainedLocal contrast of the Nth order sub-band isFor contrast threshold CsmallAnd Clarge,0<Csmall<Clarge<1, when C isN(i,j)<CsmallThen, the high frequency sub-band belongs to the smooth region; when C is presentsmall<CN(i,j)<ClargeSelecting a detail region enhancement method; when C is presentN(i,j)>ClargeAnd belongs to a possible edge region.
S32: dividing the image into a low-contrast detail region, a possible edge region and a smooth region according to a multi-mode selection strategy;
s33: carrying out edge detection on the possible edge area to obtain a target correction boundary;
s34: adopting multi-mode nonlinear mapping functions for high-frequency sub-bands in different regions;
the high-frequency sub-band of the image reflects the detail information of the image, and G is obtained through image decompositioniAnd (i) 1 … L represents detail information of different levels of the image. The smooth area is not enhanced, and a mapping function is adoptedThe contrast of the low-contrast detail area is improved, so that the low-contrast detail area is enhanced to improve the image; possible edge regions using a mapping functionApplying a mapping function to the false edge regionDecomposition of the hierarchy-dependent linear attenuation, whereineRepresenting a control pseudo-edge reduction strength parameter, N representing the number of maximum decomposition levels, and k representing the decomposition level at which the control pseudo-edge reduction strength parameter is currently located.
S35: and performing inverse transformation on the adjusted controllable pyramid coefficient, and executing an image reconstruction step.
Maximum gradient is used because of false edges at edges that are not maximally suppressed along the direction of maximum gradientAnd performing edge detection in the degree direction and the direction of the sub-maximum gradient. The high-frequency sub-band amplitudes are ordered from large to small asAnddefining local gradient values ofWhen in useAndwhen the high frequency sub-band coefficients are adjacent directions, the high frequency sub-band coefficients are local maximum values to obtainAnd isThe point is an edge;
when in useAndwhen the direction of the light beams is not the adjacent direction,is larger than the Average gradient Average,gradient in vertical directionLess than the average gradient, the point is an edge, whereOtherwise the point is a false edge.
In this embodiment, further, before the step S1 selects the region of interest on the lateral lumbar image, an LI-MMSE algorithm is applied to the lateral lumbar image to perform a denoising step S5. This is because the radiography is the integral effect of the object on the X-ray absorption on the X-ray passage, and a lesion with the same size and density is presented on the X-ray image no matter in front, middle or back of the body, that is, the DR image can not reflect the three-dimensional space position of the tissue or the lesion, so that the complex structures of the chest, the pelvic cavity and the like of the DR image have the condition of mixed tissues and overlapped bones. Meanwhile, the flat noise conforming to the Gaussian noise distribution blurs the image and is difficult to remove, so that great difficulty is brought to DR image enhancement. In the prior art, a spatial domain or Fourier domain filtering method is adopted for denoising in the denoising process, but the problem that the details and the edges of an image are blurred while noise points are reduced exists. Therefore, a denoising step is performed on the lateral lumbar image before the region of interest is selected on the lateral lumbar image, specifically, the denoising step adopts an LI-MMSE algorithm, and the algorithm has a better denoising effect on Gaussian noise in the DR image.
In this embodiment, the denoising step S5 specifically includes the following steps:
s51: performing double-tree complex wavelet transformation on the lateral lumbar vertebra image to obtain a low-frequency sub-band and a high-frequency sub-band;
s52: by window size NkEstimate the noise parameter sigma by the local variance ofnAnd σk;
S53: correcting the high-frequency subband coefficient by adopting MMSE estimation;
s54: and performing inverse wavelet transform on the adjusted high-frequency wavelet coefficient, and executing an image reconstruction step.
The starting image is decomposed into a low frequency sub-band and a high frequency sub-band, then the low frequency sub-band is decomposed into a group of direction-dependent high frequency sub-bands and a low frequency sub-band, and the steps are repeated after the decomposed low frequency sub-bands are adopted, so that N levels of decomposed low frequency sub-bands and high frequency sub-bands of each decomposition level are obtained.
In this embodiment, the image reconstruction step is specifically to perform contrast enhancement on the low-frequency subband and the high-frequency subband obtained by decomposition at each level, and then add the low-frequency subband and the high-frequency subband to obtain a processed image, where the image reconstruction step is equivalent to the inverse process of the image decomposition.
In this embodiment, further, the classifier identification process uses any one or a combination of BP neural network, regression tree or SVM classifier. And comparing and analyzing through an actual experiment to obtain the highest identification accuracy obtained by adopting the SVM classifier. The regression tree classifier is relatively easy to be affected and interfered by input information, so that a binary tree structure is more complex, the condition that a data sample set is excessively matched to affect an output result is further generated, if the BP neural network is not dependent on sufficient sample data, the obtained result is possibly a local optimal solution, the generalization capability cannot be ensured, and on the other hand, the neural network is usually constructed by selecting a system structure through existing knowledge, the network structure is difficult to unify, and the stability is poor. However, SVM classifiers have significant advantages in processing small sample data.
In this embodiment, further, a gray level co-occurrence matrix method is used to obtain texture features, where the texture features include autocorrelation coefficients, contrast, correlation, salient clusters, dark clusters, non-similarity, energy, entropy, homogeneity, maximum probability, variance, sum-average, sum-variance, sum-entropy, difference variance, difference entropy, correlation information a, correlation information B, normalized inverse difference, and normalized inverse difference.
In this embodiment, further, the T value on the lateral lumbar image is acquired, and the lateral lumbar image is divided into the normal bone mass group, the reduced bone mass group, and the osteoporosis group according to the sample division rule.
In the present embodiment, further, the sample division rule is set such that when the T > -1 condition is established, the lateral lumbar vertebra image is normal in bone mass; when the-1.0 SD > T > -2.5 condition is satisfied, the lumbar vertebra lateral image shows that the bone mass is reduced; when the condition of T < -2.5 is satisfied, the lumbar vertebra lateral image is osteoporosis.
In this embodiment, further, the T value on the lateral lumbar vertebra image is measured by a dual-energy X-ray absorption detector, and we usually select lumbar vertebra or hip joint for the measured region of interest, and when the measurement of these two parts is limited, the far end of the lateral radius is not called as the candidate region.
In this embodiment, further, the pixel value range of the region of interest is a difference between the maximum pixel value of the lateral lumbar image and the minimum pixel value of the lateral lumbar image. Two more obvious peak types often exist in the graph of the lateral lumbar vertebra image, the larger pixel value corresponds to the bone region with larger density of the human body, and the smaller pixel value distribution corresponds to the background and soft tissue regions with smaller density, such as skin and muscle.
In this embodiment, further, the peak signal-to-noise ratio calculation formula is adopted to calculate after the denoising stepAnd as a judgment index for measuring the denoising effect, wherein I (x, y) represents the pixel gray value of the image before processing, and I' (x, y) represents the pixel gray value of the image after processing.
In the present embodiment, further, the window size NkSet to 3 × 3, window size NkThe larger the PSNR obtained, but with the problem that the longer the time required, and hence the window size N we choose in this embodimentkThe image is denoised for 3 x 3.
According to the invention, the interesting region is selected from the collected lumbar lateral image to obtain texture features, the features are extracted for the second time after being selected, the classification and identification processes are carried out through the classifier, and finally the classification result is obtained, the decomposition and reconstruction of the lumbar lateral image are realized through the nonlinear enhancement algorithm of controllable pyramid decomposition, the spatial resolution of the image can be improved, and the effective image enhancement can be carried out on the low-contrast detailed regions such as bone microstructures and the like on the image, so that the display on the image is clear, the algorithm has approximate translation invariance and more direction selectivity, and the enhancement of the lumbar overlapping edge and the bone microstructure edge is effectively improved.
The embodiments of the present invention have been described in detail, but the description is only for the preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. All equivalent changes and modifications made within the scope of the present invention should be covered by the present patent.
Claims (10)
1. A method for rapidly classifying lateral lumbar images is characterized by comprising the following steps:
s1: collecting lateral lumbar images, and selecting an interested area on the lateral lumbar images;
s2: acquiring texture features;
s3: secondary extraction is carried out after feature selection;
s4: and performing a classifier identification process to obtain a classification result.
2. The method for rapidly classifying lateral lumbar images as claimed in claim 1, wherein the secondary extraction process after feature selection in step S3 employs a controllable pyramid non-linear enhancement algorithm, which specifically includes the following steps:
s31: high-frequency subband coefficients in four directions are used as directional gradients to estimate local contrast;
s32: dividing the image into a low-contrast detail region, a possible edge region and a smooth region according to a multi-mode selection strategy;
s33: carrying out edge detection on the possible edge area to obtain a target correction boundary;
s34: adopting multi-mode nonlinear mapping functions for high-frequency sub-bands in different regions;
s35: and performing inverse transformation on the adjusted controllable pyramid coefficient, and executing an image reconstruction step.
3. The method for rapidly classifying the lateral lumbar image as claimed in claim 1, wherein the step S1 performs a denoising step S5 on the lateral lumbar image by LI-MMSE algorithm before selecting the region of interest on the lateral lumbar image.
4. The method for rapidly classifying lateral lumbar images as claimed in claim 3, wherein said denoising step S5 specifically comprises the following steps:
s51: performing double-tree complex wavelet transformation on the lateral lumbar vertebra image to obtain a low-frequency sub-band and a high-frequency sub-band;
s52: by window size NkEstimate the noise parameter sigma by the local variance ofnAnd σk;
S53: correcting the high-frequency subband coefficient by adopting MMSE estimation;
s54: and performing inverse wavelet transform on the adjusted high-frequency wavelet coefficient, and executing an image reconstruction step.
5. The method for rapidly classifying lateral lumbar images as claimed in claim 2 or 3, wherein the image reconstruction step is to perform contrast enhancement on the low frequency sub-band and the high frequency sub-band obtained by decomposition at each level and then add the low frequency sub-band and the high frequency sub-band to obtain the processed image.
6. The method for rapidly classifying lateral lumbar images as claimed in claim 1, wherein the classifier identification process adopts any one or a combination of BP neural network, regression tree or SVM classifier.
7. The lateral lumbar image rapid classification method according to claim 1, characterized in that a gray level co-occurrence matrix method is adopted to obtain texture features, wherein the texture features comprise autocorrelation coefficients, contrast, correlation, salient clustering, shaded clustering, non-similarity, energy, entropy, homogeneity, maximum probability, variance, and average, and variance, and entropy, difference variance, difference entropy, correlation information degree A, correlation information degree B, normalized inverse difference and normalized inverse difference.
8. The method for rapidly classifying the lateral lumbar images as claimed in claim 1, wherein T values on the lateral lumbar images are obtained, and the lateral lumbar images are divided into a normal bone mass group, a reduced bone mass group and an osteoporosis group according to a sample partition rule, wherein the sample partition rule is set such that the lateral lumbar images are normal in bone mass when T > -1 condition is established; when the-1.0 SD > T > -2.5 condition is satisfied, the lumbar vertebra lateral image shows that the bone mass is reduced; when the condition of T < -2.5 is satisfied, the lumbar vertebra lateral image is osteoporosis.
9. The method for rapidly classifying lateral lumbar images as claimed in claim 8, wherein said T value on the lateral lumbar images is measured by a dual-energy X-ray absorption detector.
10. The lateral lumbar image rapid classification method according to claim 3, characterized in that the pixel value range of the region of interest is the difference between the maximum pixel value of the lateral lumbar image and the minimum pixel value of the lateral lumbar image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110359760.1A CN113111874A (en) | 2021-04-01 | 2021-04-01 | Rapid classification method for lateral images of lumbar vertebrae |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110359760.1A CN113111874A (en) | 2021-04-01 | 2021-04-01 | Rapid classification method for lateral images of lumbar vertebrae |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113111874A true CN113111874A (en) | 2021-07-13 |
Family
ID=76713739
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110359760.1A Pending CN113111874A (en) | 2021-04-01 | 2021-04-01 | Rapid classification method for lateral images of lumbar vertebrae |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113111874A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115409834A (en) * | 2022-10-30 | 2022-11-29 | 四川大学华西医院 | Feature extraction method, system and storage medium for tendinopathy diagnosis |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020164668A1 (en) * | 2000-04-03 | 2002-11-07 | Durham L. Kathryn | Nucleic acid molecules, polypeptides and uses therefor, including diagnosis and treatment of alzheimer's disease |
CN108241865A (en) * | 2016-12-26 | 2018-07-03 | 哈尔滨工业大学 | Multiple dimensioned more subgraph liver fibrosis multi-stage quantizations based on ultrasonoscopy method by stages |
CN111599467A (en) * | 2020-05-26 | 2020-08-28 | 广州再生医学与健康广东省实验室 | Model training method and device, electronic equipment and medium |
-
2021
- 2021-04-01 CN CN202110359760.1A patent/CN113111874A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020164668A1 (en) * | 2000-04-03 | 2002-11-07 | Durham L. Kathryn | Nucleic acid molecules, polypeptides and uses therefor, including diagnosis and treatment of alzheimer's disease |
CN108241865A (en) * | 2016-12-26 | 2018-07-03 | 哈尔滨工业大学 | Multiple dimensioned more subgraph liver fibrosis multi-stage quantizations based on ultrasonoscopy method by stages |
CN111599467A (en) * | 2020-05-26 | 2020-08-28 | 广州再生医学与健康广东省实验室 | Model training method and device, electronic equipment and medium |
Non-Patent Citations (2)
Title |
---|
丰国栋: "数字化X线摄影图像增强方法研究", 《中国优秀博硕士学位论文全文数据库(博士)信息科技辑》 * |
熊春燕: "腰椎椎间盘突出的骨髓CT应用计算机辅助研究", 《中国优秀博硕士学位论文全文数据库(硕士)医药卫生科技辑》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115409834A (en) * | 2022-10-30 | 2022-11-29 | 四川大学华西医院 | Feature extraction method, system and storage medium for tendinopathy diagnosis |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Kandhway et al. | A novel reformed histogram equalization based medical image contrast enhancement using krill herd optimization | |
Borsdorf et al. | Wavelet based noise reduction in CT-images using correlation analysis | |
EP3005289B1 (en) | Image fusion scheme for differential phase contrast imaging | |
Kanade et al. | Brain tumor detection using MRI images | |
Kroon et al. | Optimized anisotropic rotational invariant diffusion scheme on cone-beam CT | |
Irrera et al. | A flexible patch based approach for combined denoising and contrast enhancement of digital X-ray images | |
Gao et al. | Deep convolutional neural network with adversarial training for denoising digital breast tomosynthesis images | |
Jeevakala | Sharpening enhancement technique for MR images to enhance the segmentation | |
Lee et al. | Noise removal in medical mammography images using fast non-local means denoising algorithm for early breast cancer detection: a phantom study | |
CN102637292B (en) | Image processing method and device | |
CN113111874A (en) | Rapid classification method for lateral images of lumbar vertebrae | |
Sahli et al. | Analytic approach for fetal head biometric measurements based on log gabor features | |
JP2008544826A (en) | Image processing | |
Zhao et al. | A multi-step blind source separation approach for the attenuation of artifacts in mobile high-density electroencephalography data | |
Das et al. | Effective image fusion method to study Alzheimer's disease using MR, PET images | |
El Hassani et al. | Texture analysis for trabecular bone X-ray images using anisotropic morlet wavelet and Rényi entropy | |
Yang et al. | Fusion of CT and MR images using an improved wavelet based method | |
Mahaboob Basha et al. | Evaluation of weighted nuclear norm minimization algorithm for ultrasound image denoising | |
Yadav et al. | A new approach of contrast enhancement for Medical Images based on entropy curve | |
Huddin et al. | Enhancement techniques for MRI human spine images | |
Lee et al. | Enhancement of blood vessels in retinal imaging using the nonsubsampled contourlet transform | |
Rodriguez-Gallo et al. | Analysis of objective quality metrics in computed tomography images affected by metal artifacts | |
Bouzeboudja et al. | Multifractal analysis for improved osteoporosis classification | |
Węgrzyn et al. | Assessment of Correlations Between Age and Textural Features of CT Images of Thoracic Vertebrae | |
Chen et al. | A region-based Retinex with data filling for the enhancement of electronic portal 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 |