CN115294125B - Tumor CT image processing method based on pattern recognition - Google Patents
Tumor CT image processing method based on pattern recognition Download PDFInfo
- Publication number
- CN115294125B CN115294125B CN202211219716.1A CN202211219716A CN115294125B CN 115294125 B CN115294125 B CN 115294125B CN 202211219716 A CN202211219716 A CN 202211219716A CN 115294125 B CN115294125 B CN 115294125B
- Authority
- CN
- China
- Prior art keywords
- slice
- tumor
- slices
- area
- semantic segmentation
- 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.)
- Active
Links
- 206010028980 Neoplasm Diseases 0.000 title claims abstract description 196
- 238000003672 processing method Methods 0.000 title claims abstract description 14
- 238000003909 pattern recognition Methods 0.000 title description 4
- 230000011218 segmentation Effects 0.000 claims abstract description 122
- 238000000034 method Methods 0.000 claims abstract description 20
- 230000000873 masking effect Effects 0.000 claims description 5
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims 1
- 238000012545 processing Methods 0.000 abstract description 8
- 238000002591 computed tomography Methods 0.000 description 104
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 6
- 230000002159 abnormal effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000009877 rendering Methods 0.000 description 2
- 101100134058 Caenorhabditis elegans nth-1 gene Proteins 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000024245 cell differentiation Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012421 spiking Methods 0.000 description 1
Images
Classifications
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- 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
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to the technical field of data processing, and particularly relates to a tumor CT image processing method based on image recognition, which comprises the steps of carrying out image recognition through related electronic equipment to obtain a tumor CT image, recognizing the tumor CT image, calculating the section neighborhood distance between any two sections according to the segmentation continuity index corresponding to each section, the area corresponding to the semantic segmentation mask of a tumor region and a tumor shape descriptor, and grading each sample according to the distance; and distributing the maximum order used for harmonic reconstruction to each slice according to the grading result. The method can correct the CT slices with wrong semantic segmentation, and improve the accuracy of the three-dimensional tumor reconstruction result. The tumor CT image processing method can also be configured into a data processing device suitable for a specific application.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a tumor CT image processing method based on image recognition.
Background
As an important link in medical image processing, three-dimensional reconstruction of Computed Tomography (CT) images of tumors can not only help doctors to diagnose disease conditions more quickly and accurately, but also provide an effective way for clinical medical actual operations and teaching research, especially when determining tumor characteristics.
The realization of accurate and efficient three-dimensional reconstruction is a problem to be solved urgently at present. Although the current tumor CT image processing method can accurately segment the tumor region, the accuracy of the boundary is still to be enhanced, and if the tumor slice image based on semantic segmentation is directly used for three-dimensional reconstruction, the false segmentation of the semantic segmentation and the error thereof may cause fracture, spiking and jagging, and the processing effect is not good.
Disclosure of Invention
The invention provides a tumor CT image processing method based on pattern recognition, aiming at solving the problem of poor processing effect of the existing tumor CT image processing method.
The invention relates to a tumor CT image processing method based on image recognition, which comprises the following steps:
acquiring semantic segmentation masks of tumor regions of the CT slices, and numbering the semantic segmentation masks of the tumor regions of the CT slices according to the shooting sequence of the CT slices;
calculating a tumor shape descriptor of a corresponding circular primitive-based Fourier descriptor for the contour of the semantic segmentation mask of the tumor region of any CT slice;
calculating the similarity of the tumor shape descriptors of the neighboring slices corresponding to each CT slice according to the tumor shape descriptors corresponding to each CT slice; calculating the area continuity index of the adjacent slice corresponding to each CT slice according to the area corresponding to the semantic segmentation mask of the tumor area of each CT slice; calculating the segmentation continuity index corresponding to each CT slice according to the tumor shape descriptor similarity of the adjacent slice and the area continuity index of the adjacent slice corresponding to each CT slice;
calculating the section neighborhood distance between any two sections according to the segmentation continuity index corresponding to each section, the area corresponding to the semantic segmentation mask of the tumor area and the tumor shape descriptor; for any slice, calculating the local reachable density corresponding to the slice according to the set number of slice neighborhood distances with the minimum slice neighborhood distance of the slice;
classifying the slices according to the corresponding local reachable density to obtain L grades, wherein L is more than or equal to 2; distributing corresponding orders for the slices according to different levels, performing harmonic reconstruction on the semantic segmentation mask contour of the tumor region of each slice according to the corresponding orders of each slice, and performing three-dimensional reconstruction on the tumor according to harmonic reconstruction results; the order of slice distribution in the level with large local reachable density is larger than that of the slice distribution in the level with small local reachable density.
Further, the calculating the similarity of the tumor shape descriptors of neighboring slices corresponding to each CT slice according to the tumor shape descriptors corresponding to each CT slice includes:
calculating the tumor shape descriptor similarity of the adjacent slices corresponding to each CT slice by using the following formula:
wherein, F i A tumor shape descriptor corresponding to the semantic segmentation mask contour of the tumor region representing the ith CT slice,
for the tumor shape descriptor corresponding to the semantic segmentation mask contour of the tumor region of the i-1 th CT slice, the corresponding->
For the tumor shape descriptor corresponding to the semantic segmentation mask contour of the tumor region of the (i + 1) th CT slice, the corresponding->
The similarity of tumor shape descriptors of adjacent slices corresponding to the ith CT slice is shown, max is the maximum value, N-1 is more than or equal to i and more than or equal to 2, and N is the number of slices.
Further, the calculating a neighboring slice area continuity index corresponding to each CT slice according to an area corresponding to the semantic segmentation mask of the tumor area of each CT slice includes:
calculating the area continuity index of the adjacent slice corresponding to each CT slice by using the following formula:
wherein,
mask the corresponding area for the semantic segmentation of the tumor region of the ith CT slice, < >>
Corresponding areas are masked for semantic segmentation of the tumor region of the i-1 th CT slice>
Corresponding area is masked for the semantic segmentation of the tumor region of the (i + 1) th CT slice>
The area continuity index of the adjacent slice corresponding to the ith CT slice is shown, min is the minimum value, N-1 is more than or equal to i and more than or equal to 2, and N is the number of slices.
Further, the calculating a segmentation continuity index corresponding to each CT slice according to the neighbor slice tumor shape descriptor similarity and the neighbor slice area continuity index corresponding to each CT slice includes:
calculating the segmentation continuity index corresponding to each CT slice by using the following formula:
wherein,
is a segmentation continuity index corresponding to the jth CT slice, N is not less than j not less than 1, N is the number of slices, and>
for the tumor shape descriptor similarity of the adjacent section corresponding to the jth CT section, the method also comprises the following steps of>
And (4) the area continuity index of the adjacent slice corresponding to the jth CT slice.
Further, the calculating a slice neighborhood distance between any two slices according to the segmentation continuity index corresponding to each slice, the area corresponding to the semantic segmentation mask of the tumor region, and the tumor shape descriptor includes:
the slice neighborhood distance between any two slices is calculated using the following formula:
wherein,
for a slice neighborhood distance between the p-th slice and the q-th slice, < >>
Masking the corresponding area for the semantic segmentation of the tumor region of the p-th slice, < >>
Masking the corresponding area for the semantic segmentation of the tumor region for the qth slice, < >>
For the segmented continuation index corresponding to the p-th slice, <' > H>
For the corresponding segmented continuation index for the qth slice, <' >>
For the tumor shape descriptor corresponding to the p-th slice, ` H `>
The tumor shape descriptor corresponding to the q slice. />
Further, for any slice, calculating the local reachable density corresponding to the slice according to the set number of slice neighborhood distances which is the minimum distance from the slice neighborhood of the slice, includes:
obtaining the K-th reachable distance corresponding to each slice
，/>
A distance for radiating outward in an assumed space with one slice sample until a K-th adjacent sample is covered;
for any slice
At the Kth reachable distance ^ of the sliced sample>
In that the Kth reachable distance of the sliced sample is->
Included slice samples are constructed to aggregate as ` Harbin `>
(ii) a Calculating the local reachable density corresponding to the slice according to the following formula:
wherein,
for the locally achievable density corresponding to this section p, q is->
Any of the above-described methods may be performed,
k is the number of neighboring samples of the slice p, which is the slice neighborhood distance between the slice p and the slice sample q.
Further, the classifying each slice according to the corresponding local reachable density to obtain L levels, where L is greater than or equal to 2, includes:
sequencing the slices according to the sequence of the corresponding local reachable density from large to small;
dividing the slices into 4 grades according to the sorting result, wherein the 1 st grade corresponds to the first 30% of the slices, and recording the 1 st grade as a high-quality slice sample grade; grade 2 corresponds to the next 30% of the slices, and grade 2 is recorded as the good quality slice sample grade; grade 3 corresponds to the next 30% of slices, and grade 3 is recorded as the medium-quality slice sample grade; grade 4 corresponds to the last 10% of the slices and grade 4 is reported as the poor quality slice sample grade.
Has the advantages that: the tumor CT image processing method based on the graph recognition can correct CT slices with wrong semantic segmentation, can improve the accuracy of a tumor three-dimensional reconstruction result, solves the problems of broken surfaces, sharp spines and sawteeth existing in the conventional semantic segmentation method for three-dimensional reconstruction of tumors, and greatly improves the data processing effect.
Drawings
Fig. 1 is a flowchart of a tumor CT image processing method based on pattern recognition according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described below with reference to the drawings in the embodiments of the present invention.
The method aims to solve the problems that the prior method directly uses a tumor slice image based on semantic segmentation to carry out three-dimensional reconstruction, and the wrong segmentation and the error of the semantic segmentation can cause broken surfaces, sharp spines and saw teeth, and ensures the rapidity and the accuracy of three-dimensional reconstruction in order to obtain smooth tumor CT three-dimensional reconstruction data, as shown in fig. 1, the method for processing the tumor CT image based on the graph recognition comprises the following steps:
(1) Acquiring semantic segmentation masks of tumor regions of the CT slices, and numbering the semantic segmentation masks of the tumor regions of the CT slices according to the shooting sequence of the CT slices;
and performing semantic segmentation operation of the tumor region based on Mask RCNN on all the slice images in the CT scanning example of the tumor to obtain the semantic segmentation result of each layer of CT slice, wherein the result of the Mask RCNN is based on the bounding box of each tumor example and the Mask segmented in the bounding box. The location of each tumor is nearly fixed in the CT slices, so methods based on target tracking or manual labeling can be used to pick the instance bounding box of the tumor that needs to be treated in each slice.
Based on the semantic segmentation result of the bounding box, a semantic segmentation mask may be obtained. The semantic segmentation mask obtained at this time has a certain degree of defects, mainly because the tumor segmentation mask obtained by MaskRCNN has a certain degree of under-segmentation, that is, a region belonging to a tumor has a certain degree of outline deletion due to network accuracy, or is over-segmented, that is, the outline overflows. If the segmentation reconstruction is directly performed, an erroneous and sharply protruded tumor three-dimensional model is generated due to low segmentation mask precision, thereby affecting the judgment of the tumor property.
To achieve optimization of the appearance of the three-dimensional model of the tumor, this example detects and segments instances of the tumor based on MaskRCNN, since the tumor is pathological tissue that continuously spans multiple CT slices. Thus, the slices of the tumor are numbered in the sequential order of the slice shots, and a mask sequence consisting of semantic segmentation masks for the tumor region of each CT slice is obtained, assuming that there are N consecutive slices in the CT data
In which>
Is the semantic segmentation mask of the tumor area of the first slice, is based on>
Is the semantic segmentation mask of the tumor area of the second slice, is based on>
Is the semantic segmentation mask for the tumor region of the nth slice.
(2) Calculating a tumor shape descriptor of a corresponding circular primitive-based Fourier descriptor for the contour of the semantic segmentation mask of the tumor region of any CT slice;
the Fourier descriptor is a characteristic parameter for describing the contour of the tumor CT image, and uses the Fourier transform of the tumor semantic segmentation boundary information as a shape characteristic so as to transform the contour characteristic from a space domain into a frequency domain. Because the boundary is continuously closed and periodic, a Fourier series can be used to approximate the boundary. When an erroneous segmentation result occurs, the high frequency classification in the frequency domain increases because a more prominent protrusion or deletion occurs in the contour boundary.
Since the growth characteristics of biological tissues are limited to cell differentiation and supply of tissue fluid, the contour curvature of tumors is low, and sharp, twisted, protruding, recessed contours are difficult to appear. Based on the characteristic that most of the tumor appearances are mellow, frequency domain information is extracted based on a one-bit equation of a circular element model to serve as a feature vector of the tumor contour, the contour is digitized, and different contours are distinguished better. For under-segmentation and over-segmentation of the tumor, the Fourier descriptor has the characteristic of scaling resistance, so that the abnormality of under-segmentation and over-segmentation of the tumor image can be better analyzed.
Each tumor semantic segmentation mask is processed as follows to obtain a Fourier descriptor of the contour: firstly, extracting the coordinates of the outline of the semantic segmentation mask of each tumor. For one image, the contour of the tumor image is represented by a one-dimensional equation of a circular primitive model:
the x-axis of the contour coordinates of the tumor image is taken as a real axis, and the y-axis is taken as an imaginary axis. Performing triangular Fourier expansion on an ellipse one-dimensional equation:
therefore, there are:
for any order k, there is a descriptor
，/>
Namely a descriptor corresponding to the k-th order; the higher the rank, the descriptor of the tumor profile->
The more the profile features of the contour, such as recesses and protrusions, which are severe compared with the circular shape, can be described. In this embodiment, when the total number of k =10 is set, there is a feature descriptor of the tumor slice outline
。
According to the method, the tumor shape descriptor F corresponding to the semantic segmentation mask outline of the tumor region of each CT slice can be obtained i ，F i Tumor region representing the ith CT sliceSemantic segmentation of the domain masks the tumor shape descriptors corresponding to the contours.
(3) Calculating the similarity of the tumor shape descriptors of the neighboring slices corresponding to each CT slice according to the tumor shape descriptors corresponding to each CT slice; calculating the area continuity index of the adjacent slice corresponding to each CT slice according to the area corresponding to the semantic segmentation mask of the tumor area of each CT slice; calculating the segmentation continuity index corresponding to each CT slice according to the tumor shape descriptor similarity of the adjacent slice and the area continuity index of the adjacent slice corresponding to each CT slice;
specifically, the method comprises the following 3 small steps:
(1) calculating the similarity of the tumor shape descriptors of the neighboring slices corresponding to each CT slice according to the tumor shape descriptors corresponding to each CT slice;
when a wrong segmentation occurs to one CT slice, the more dissimilar the tumor shape descriptor of the tumor slice based on the circle primitive is to the tumor shape descriptor corresponding to the neighboring CT slice. Therefore, in this embodiment, for slices with ct slice pictures before and after the ith slice, the tumor shape descriptor similarity of the corresponding neighboring slices is calculated:
wherein,
for the tumor shape descriptor corresponding to the semantic segmentation mask contour of the tumor region of the i-1 th CT slice, the corresponding->
For the tumor shape descriptor corresponding to the semantic segmentation mask contour of the tumor region of the (i + 1) th CT slice, the corresponding->
The tumor shape descriptor similarity of the adjacent slice corresponding to the ith CT slice is shown, max is the maximum value, and N-1 is more than or equal to i and more than or equal to 2./>
For the similarity of the tumor shape descriptors of the neighboring slices corresponding to the 1 st CT slice, the calculation is only needed to be combined with the tumor shape descriptors corresponding to the semantic segmentation mask contour of the tumor region of the 2 nd CT slice, namely
(ii) a For the similarity of the tumor shape descriptors of the adjacent slices corresponding to the Nth CT slice, the calculation is only needed to be combined with the tumor shape descriptors corresponding to the semantic segmentation mask outline of the tumor region of the (N-1) th CT slice, namely the similarity is calculated according to the result that the result of the calculation is based on the result of the comparison>
。
When the similarity of the tumor shape descriptors of the neighboring slices is high, the similarity of the characteristics of the tumor segmentation contours between the adjacent layers is proved, namely, the defects of under-segmentation and over-segmentation are low. On the contrary, the difference between the current slice and the slice of the adjacent layer is too much, so that the inconsistency of high-frequency components occurs.
(2) Calculating the area continuity index of the adjacent slice corresponding to each CT slice according to the area corresponding to the semantic segmentation mask of the tumor area of each CT slice;
the area continuity index can analyze whether the area corresponding to the current semantic segmentation mask is discontinuous due to semantic segmentation errors; in this embodiment, for slices having ct slice pictures before and after the ith slice, the area continuity index of the neighboring slice corresponding to the slice is calculated:
wherein
Mask the corresponding area for the semantic segmentation of the tumor region of the ith CT slice, < >>
Corresponding areas are masked for semantic segmentation of the tumor region of the i-1 th CT slice>
Corresponding area is masked for the semantic segmentation of the tumor region of the (i + 1) th CT slice>
And (3) the area continuity index of the adjacent slice corresponding to the ith CT slice is defined, min is the minimum value, and N-1 is more than or equal to i and more than or equal to 2.
For the area continuity index of the adjacent slice corresponding to the 1 st CT slice, the area calculation corresponding to the semantic segmentation mask of the tumor area of the 2 nd CT slice is only needed, namely
(ii) a For the area continuity index of the adjacent slice corresponding to the Nth CT slice, the area calculation corresponding to the semantic segmentation mask of the tumor region of the Nth-1 st CT slice is only needed, namely
。
Based on the characteristic that most of tumor appearances are mellow, each layer of CT slices have enough continuous curvature, when the difference between the area corresponding to the semantic segmentation mask of a certain slice tumor region and the mean value of the semantic segmentation areas of two adjacent layers of slices is large, or the difference between the mean values of the semantic segmentation areas of a first layer of CT slices and a second layer of CT slices is large, or the difference between the mean values of the semantic segmentation areas of an Nth layer of CT slices and an N-1 layer of CT slices is large, the current semantic segmentation result is judged to have a poor segmentation result.
(3) And calculating the segmentation continuity index corresponding to each CT slice according to the tumor shape descriptor similarity of the adjacent slice and the area continuity index of the adjacent slice corresponding to each CT slice.
In this embodiment, the segmentation continuity index corresponding to each CT slice is calculated by using the following formula:
wherein,
the j is a segmentation continuity index corresponding to the jth CT slice, and N is more than or equal to j and more than or equal to 1; the segmentation continuity index can calculate whether the semantic segmentation of the current CT slice is abnormal or not and whether the segmentation area is abnormal or not based on the characteristic that most of tumor appearances are mellow.
At this point, the segmentation continuity index corresponding to each CT slice can be obtained. According to the corresponding segmentation continuous index of each CT slice, a segmentation continuous index sequence of the whole tumor can be formed
Wherein is present>
Is the segmented continuation index corresponding to the first slice, <' > is>
Is the segmented continuation index for the second slice, <' > is>
Is the segmentation continuity index corresponding to the nth slice.
(4) Calculating the section neighborhood distance between any two sections according to the segmentation continuity index corresponding to each section, the area corresponding to the semantic segmentation mask of the tumor area and the tumor shape descriptor; for any slice, calculating the local reachable density corresponding to the slice according to the set number of slice neighborhood distances with the minimum slice neighborhood distance of the slice;
since the contour of the tumor CT image is approximately the same and the area continuously changes, the properties are similar, and in order to analyze erroneous CT segmentation data and thus ensure high quality of three-dimensional reconstruction, the property abnormality of the tissue slice is analyzed based on the slice data: each CT image is corresponding to a segmentation continuity index and a tumor shape descriptor, so that an assumed space for tumor segmentation is constructed, the characters of the tumor CT images are distinguished, and the abnormality of the characters is judged.
In this embodiment, the slice neighborhood distance between any two slices is calculated according to the area, the segmentation continuity index and the tumor shape descriptor corresponding to the semantic segmentation mask of the tumor region of each slice, and the calculation formula is as follows:
wherein,
for a slice neighborhood distance between the p-th slice and the q-th slice, < >>
Masking the corresponding area for the semantic segmentation of the tumor region of the p-th slice, < >>
Masking the corresponding area for the semantic segmentation of the tumor region for the qth slice, < >>
For the segmented continuation index corresponding to the p-th slice, <' > H>
For a segmented consecutive index corresponding to the qth slice>
For the tumor shape descriptor corresponding to the p-th slice, ` H `>
The tumor shape descriptor corresponding to the qth slice.
Because the tumor section is not finishedFully rounded, slowly varying contour descriptors of concavities, convexities, etc. that match physiological characteristics appear, so that when a tumor section belongs to a type of local tissue,
the features that can represent the tumor segmentation contour are similar and therefore the distances are scaled to be close in the hypothesis space. Whereas the scaling effect is lower.
Let the p-th slice be the sample point p whose local reachable density is the inverse of p's average reachable distance based on k-distance nearest neighbors. If all the reachable distances are 0, the local reachable density is likely to be ∞, and if this happens, the sample point p can be marked directly as background point.
Although the tumor slices are not completely round, when the order of the slices is not reasonably continuous, the contour is distorted to a certain extent, so that the neighborhood distance of the slices is affected
The constraint of (2) is that there is a difference from the continuity of any slice, whereas the continuity of any slice is similar, and the contour of the tumor CT slice is close to the distance in the assumed space.
The slice neighborhood distance is calculated by combining the slice sequence direction, the slice outline characteristics and the slice size, and slices with abnormal characters can be shifted to be distant in an assumed space according to the neighborhood distance of the slices. In this embodiment, based on the assumed space, the K-local reachable density of each slice is calculated as follows:
having a Kth reachable distance based on slice neighborhood distance
I.e. the distance when one sliced sample radiates outward in the assumed space until the K-th neighboring sample is covered. In this embodiment, K is 15% of the total number of slices. At the Kth reachable distance ^ of the section sample p>
In this case, more sliced samples q can be covered, so that all sliced samples q covered are constructed as a collection->
. Calculating the local reachable density corresponding to each slice according to the following formula:
when the local reachable density corresponding to a slice is higher, the higher the density of the contour feature and the continuity corresponding to the slice and the surrounding slices in the assumed space is, the more normal the slice sample is; conversely, low intensity means more abnormal sliced samples.
(5) Classifying the slices according to the corresponding local reachable density to obtain L grades, wherein L is more than or equal to 2; distributing corresponding orders for each slice according to different grades, performing harmonic reconstruction on the semantic segmentation mask contour of the tumor region of each slice according to the corresponding orders of each slice, and performing three-dimensional tumor reconstruction according to harmonic reconstruction results; the order of slice distribution in the level with larger local reachable density is larger than that of the level with smaller local reachable density.
In this embodiment, the local reachable densities corresponding to all slices are sorted, and each 30% of slice samples are classified, and the classification result is as follows: 30% of high quality sliced samples; 30% of good quality sliced samples; 30% of medium mass sliced samples; 10% bad mass sliced specimen.
Determining the downsampling level of the Fourier descriptor based on the grading, specifically, setting M downsampling levels of the descriptor, which are equivalent to a smooth level, and performing harmonic reconstruction: in this embodiment, the order of constructing the tumor CT contour feature descriptor is 10, and downsampling is performed based on high quality, good quality, medium quality, and poor quality as follows: the order of the high-quality slice sample is 10; the order of a good quality sliced sample is 8; the order of the medium-mass slice sample is 6; the order of the poor quality sliced sample was 4. Taking the medium quality as an example, namely discarding Fourier descriptors with the order greater than 6, and performing harmonic reconstruction on the contour based on a one-dimensional equation of circular elements. The harmonic reconstruction process is prior art and will not be described herein. As other embodiments, the specific content of the above division levels may be modified at the time of application.
The order corresponding to each slice can be obtained through the process, and the order is used as the maximum order when harmonic reconstruction is carried out on the mask outline of the corresponding slice; and resampling the mask outline corresponding to each slice according to the corresponding order to obtain each coordinate of each slice outline, thereby obtaining the top point of the tumor object slice.
Based on the obtained tumor object slice vertex, the tumor image is three-dimensionally reconstructed based on an isosurface extraction algorithm (MarchingCubes). Surface rendering is the mainstream algorithm for three-dimensional reconstruction of medical images at present. Marching Cubes (MC) algorithm is a classical algorithm in the surface rendering algorithm, which is a voxel-level reconstruction method proposed by w. In the MC algorithm, the tumor CT contour is of this type, assuming the raw data is a discrete three-dimensional regular data field. The basic idea of the MC algorithm is to process cubes in a data field one by one, find out a cube intersected with an isosurface, and calculate the intersection point of the isosurface and the cube edge by linear interpolation. According to the relative position of each vertex of the cube and the isosurface, connecting the isosurface with the intersection points on the edges of the cube in a certain mode to generate the isosurface which is used as an approximate representation of the isosurface in the cube. Thus, this embodiment can be based on marching cubes (e.g., vtkkmarching cubes in VTK) triangularization to obtain a tumor mesh.
The embodiment provides an analysis method for a tumor CT image based on a hypothesis space, which can correct a CT slice with wrong semantic segmentation, improve the accuracy of a tumor three-dimensional reconstruction result, and solve the problems of broken surfaces, sharp spines and sawteeth in the existing semantic segmentation method for three-dimensional reconstruction of tumors.
It should be noted that while the preferred embodiments of the present invention have been described, additional variations and modifications to those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Such variations and modifications are intended to fall within the scope of the present invention.
Claims (7)
1. A tumor CT image processing method based on image recognition is characterized by comprising the following steps:
acquiring semantic segmentation masks of tumor regions of the CT slices, and numbering the semantic segmentation masks of the tumor regions of the CT slices according to the shooting sequence of the CT slices;
calculating a tumor shape descriptor of a corresponding circular primitive-based Fourier descriptor for the contour of the semantic segmentation mask of the tumor region of any CT slice;
calculating the similarity of the tumor shape descriptors of the neighboring slices corresponding to each CT slice according to the tumor shape descriptors corresponding to each CT slice; calculating the area continuity index of the adjacent slice corresponding to each CT slice according to the area corresponding to the semantic segmentation mask of the tumor area of each CT slice; calculating the segmentation continuity index corresponding to each CT slice according to the tumor shape descriptor similarity of the adjacent slice and the area continuity index of the adjacent slice corresponding to each CT slice;
calculating the section neighborhood distance between any two sections according to the segmentation continuity index corresponding to each section, the area corresponding to the semantic segmentation mask of the tumor area and the tumor shape descriptor; for any slice, calculating the local reachable density corresponding to the slice according to the set number of slice neighborhood distances with the minimum slice neighborhood distance of the slice;
classifying the slices according to the corresponding local reachable density to obtain L grades, wherein L is more than or equal to 2; distributing corresponding orders for the slices according to different levels, performing harmonic reconstruction on the semantic segmentation mask contour of the tumor region of each slice according to the corresponding orders of each slice, and performing three-dimensional reconstruction on the tumor according to harmonic reconstruction results; the order of slice distribution in the level with large local reachable density is larger than that of the slice distribution in the level with small local reachable density.
2. The method of claim 1, wherein the calculating the similarity of the tumor shape descriptors of neighboring slices corresponding to each CT slice according to the tumor shape descriptors corresponding to each CT slice comprises:
calculating the tumor shape descriptor similarity of the adjacent slices corresponding to each CT slice by using the following formula:
wherein, F i A tumor shape descriptor corresponding to the semantic segmentation mask contour of the tumor region representing the ith CT slice,
mask the tumor shape descriptor corresponding to the outline for the semantic segmentation of the tumor region of the i-1 st CT slice,
the tumor shape descriptor corresponding to the outline of the mask is segmented for the semantics of the tumor region of the (i + 1) th CT slice,
the similarity of tumor shape descriptors of adjacent slices corresponding to the ith CT slice is shown, max is the maximum value, N-1 is more than or equal to i and more than or equal to 2, and N is the number of slices.
3. The method of claim 1, wherein the calculating the continuity index of the neighboring slice area corresponding to each CT slice according to the area corresponding to the semantic segmentation mask of the tumor region of each CT slice comprises:
calculating the area continuity index of the adjacent slice corresponding to each CT slice by using the following formula:
wherein,
the corresponding area is masked for the semantic segmentation of the tumor region of the ith CT slice,
the corresponding area is masked for the semantic segmentation of the tumor region of the i-1 st CT slice,
the corresponding area is masked for the semantic segmentation of the tumor region of the (i + 1) th CT slice,
the area continuity index of the adjacent slice corresponding to the ith CT slice is shown, min is the minimum value, N-1 is more than or equal to i and more than or equal to 2, and N is the number of slices.
4. The method as claimed in claim 1, wherein the calculating the segmentation continuity index corresponding to each CT slice according to the neighboring slice tumor shape descriptor similarity and the neighboring slice area continuity index corresponding to each CT slice comprises:
calculating the segmentation continuity index corresponding to each CT slice by using the following formula:
wherein,
is the segmentation continuity index corresponding to the jth CT slice, N is not less than j and not less than 1, N is the number of slices,
the tumor shape descriptor similarity of the adjacent slice corresponding to the jth CT slice,
and (4) the area continuity index of the adjacent slice corresponding to the jth CT slice.
5. The method of claim 1, wherein the calculating a slice neighborhood distance between any two slices according to the segmentation continuity index corresponding to each slice, the area corresponding to the semantic segmentation mask of the tumor region, and the tumor shape descriptor comprises:
the slice neighborhood distance between any two slices is calculated using the following formula:
wherein,
the slice neighborhood distance between the p-th slice and the q-th slice,
the corresponding area is masked for semantic segmentation of the tumor region of the p-th slice,
masking the corresponding area for the semantic segmentation of the tumor region of the qth slice,
for the segmentation continuity index corresponding to the p-th slice,
segmentation corresponding to the q-th sliceThe number of consecutive indices is,
for the tumor shape descriptor corresponding to the p-th slice,
the tumor shape descriptor corresponding to the qth slice.
6. The CT image processing method for tumor based on image recognition of claim 1, wherein the step of calculating the local reachable density of any slice according to the set number of slice neighborhood distances with the smallest slice neighborhood distance comprises:
obtaining the K-th reachable distance corresponding to each slice
，
The distance of one slice sample radiating outwards in an assumed space until the K-th adjacent sample is covered;
for any slice
At the Kth reachable distance of the sliced sample
The K-th reachable distance of the sliced sample
The slice sample construction set of the inner coverage is marked as
(ii) a Calculating the local reachable density corresponding to the slice according to the following formula:
wherein,
for the slice p corresponding to the local achievable density, q is
Any of the above-described methods may be performed,
k is the number of neighboring samples of the slice p, which is the slice neighborhood distance between the slice p and the slice sample q.
7. The tumor CT image processing method based on image recognition of claim 1, wherein the step of classifying each slice according to the corresponding local reachable density size to obtain L levels, wherein L is greater than or equal to 2, comprises:
sequencing the slices according to the sequence of the corresponding local reachable density from large to small;
dividing the slices into 4 grades according to the sorting result, wherein the 1 st grade corresponds to the first 30% of the slices, and recording the 1 st grade as a high-quality slice sample grade; grade 2 corresponds to the next 30% of the slices, and grade 2 is recorded as a good quality slice sample grade; grade 3 corresponds to the next 30% of slices, and grade 3 is recorded as the medium-quality slice sample grade; grade 4 corresponds to the last 10% of the slices and grade 4 is reported as the poor quality slice sample grade.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211219716.1A CN115294125B (en) | 2022-10-08 | 2022-10-08 | Tumor CT image processing method based on pattern recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211219716.1A CN115294125B (en) | 2022-10-08 | 2022-10-08 | Tumor CT image processing method based on pattern recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115294125A CN115294125A (en) | 2022-11-04 |
| CN115294125B true CN115294125B (en) | 2023-03-24 |
Family
ID=83833761
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211219716.1A Active CN115294125B (en) | 2022-10-08 | 2022-10-08 | Tumor CT image processing method based on pattern recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115294125B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116153472B (en) * | 2023-02-24 | 2023-10-24 | 萱闱(北京)生物科技有限公司 | Image multidimensional visualization method, device, medium and computing equipment |
| CN116844697B (en) * | 2023-02-24 | 2024-01-09 | 萱闱(北京)生物科技有限公司 | Image multidimensional visualization method, device, medium and computing equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488781A (en) * | 2015-06-01 | 2016-04-13 | 深圳市第二人民医院 | Dividing method based on CT image liver tumor focus |
| CN107507195A (en) * | 2017-08-14 | 2017-12-22 | 四川大学 | The multi-modal nasopharyngeal carcinoma image partition methods of PET CT based on hypergraph model |
| CN110610491A (en) * | 2019-09-17 | 2019-12-24 | 湖南科技大学 | Liver tumor region segmentation method of abdominal CT image |
| CN110969619A (en) * | 2019-12-19 | 2020-04-07 | 广州柏视医疗科技有限公司 | Method and device for automatically identifying primary tumor of nasopharyngeal carcinoma |
-
2022
- 2022-10-08 CN CN202211219716.1A patent/CN115294125B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488781A (en) * | 2015-06-01 | 2016-04-13 | 深圳市第二人民医院 | Dividing method based on CT image liver tumor focus |
| CN107507195A (en) * | 2017-08-14 | 2017-12-22 | 四川大学 | The multi-modal nasopharyngeal carcinoma image partition methods of PET CT based on hypergraph model |
| CN110610491A (en) * | 2019-09-17 | 2019-12-24 | 湖南科技大学 | Liver tumor region segmentation method of abdominal CT image |
| CN110969619A (en) * | 2019-12-19 | 2020-04-07 | 广州柏视医疗科技有限公司 | Method and device for automatically identifying primary tumor of nasopharyngeal carcinoma |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115294125A (en) | 2022-11-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115294125B (en) | Tumor CT image processing method based on pattern recognition | |
| CN109146854B (en) | Analysis method for association relationship between pulmonary nodule and pulmonary blood vessel | |
| CN102663416B (en) | Segmentation method of viscera and internal blood vessels thereof in surgical planning system | |
| CN105931226A (en) | Automatic cell detection and segmentation method based on deep learning and using adaptive ellipse fitting | |
| Pandiselvi et al. | Efficient framework for identifying, locating, detecting and classifying MRI brain tumor in MRI images | |
| CN112819747A (en) | Method for automatically diagnosing benign and malignant nodules based on lung tomography image | |
| WO2009058315A1 (en) | Structure segmentation via mar-cut | |
| CN115546605A (en) | Training method and device based on image labeling and segmentation model | |
| CN113963138A (en) | Complete and accurate extraction method of three-dimensional laser point cloud characteristic point line | |
| CN102081740B (en) | 3D image classification method based on scale invariant features | |
| CN113780421B (en) | Brain PET image identification method based on artificial intelligence | |
| Zhang et al. | Difference-deformable convolution with pseudo scale instance map for cell localization | |
| CN110428437B (en) | GGO segmentation method based on edge-sensitive SLIC and quadratic density clustering | |
| Dickson et al. | A Dual Channel Multiscale Convolution U-Net Methodfor Liver Tumor Segmentation from Abdomen CT Images | |
| CN116468103A (en) | Training method, application method and system for lung nodule benign and malignant recognition model | |
| JP4544891B2 (en) | Image processing method and program for line extraction, line concentration image filter | |
| CN111415350B (en) | Colposcope image identification method for detecting cervical lesions | |
| CN100538741C (en) | The candidate of lung nodule detection produces | |
| Agnes | Boundary Aware Semantic Segmentation using Pyramid-dilated Dense U-Net for Lung Segmentation in Computed Tomography Images | |
| CN114764810A (en) | Medical image segmentation method | |
| CN111311586A (en) | Nonlinear health analysis system data-based multi-index dynamic integration algorithm and system | |
| Liu et al. | Novel Approach for Automatic Region of Interest and Seed Point Detection in CT Images Based on Temporal and Spatial Data. | |
| CN117542527B (en) | Lung nodule tracking and change trend prediction method, device, equipment and storage medium | |
| Song et al. | A new separation algorithm for overlapping blood cells using shape analysis | |
| Li et al. | A Framework for Cervical Cell EDF Image Segmentation Using Similarity Measure |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |