Fine marking method for colon cancer lesion in CT (Computed Tomography) image

Abstract

The invention relates to the technical field of CT (Computed Tomography) image analysis and labeling, in particular to a method for finely labeling colon cancer lesions in CT images. The method comprises the following steps: firstly, acquiring an intestinal wall section corresponding to each non-intestinal wall section area in an intestinal wall section area in a CT image; analyzing and acquiring an abnormal index of each pixel point in each non-intestinal-wall section area; based on the abnormal index and the distribution characteristic of each pixel point, obtaining all focus sub-regions; analyzing and acquiring a focus score of each focus sub-region, and further combining different focus sub-regions to acquire all focus regions; and finally, screening out a to-be-marked focus area from all focus areas and marking the to-be-marked focus area. According to the method, by analyzing the characteristics that the texture of the intestinal wall surface and the shape of the intestinal wall cross section are changed due to colon cancer, and further combining the area characteristics and the distribution characteristics of the lesions in different lesion stages, the lesion region is accurately obtained and marked, and the marking effect of the colon cancer lesion is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of CT image analysis and annotation, and in particular to a fine annotation method for colon cancer lesions in CT images. Background Art

[0002] Computed tomography (CT) is a medical imaging technology that combines X-rays and computer technology to generate cross-sectional images of tissue structures in the body. Colon cancer is one of the most common malignant tumors of the digestive system. Its symptoms in CT images are usually characterized by structural changes in the intestine, such as intestinal wall thickening or intestinal obstruction. By evaluating and annotating cancer lesions in colon CT images, relevant medical personnel can be assisted in accurately locating the location, size and shape of the lesions, providing a reference basis for subsequent diagnosis and treatment. Therefore, the accuracy of lesion annotation is crucial.

[0003] In the prior art, lesion areas are usually automatically labeled based on deep learning algorithms. However, deep learning algorithms rely on a large number of labeled training data sets to optimize model parameters, and the effect of automatic labeling is affected by the quality of the training data sets and the model training effect. The acquisition and training of training data sets occupy a lot of resources, and there may be deviations in the lesion types and image specifications in the training data sets, which limits the generalization ability of the algorithm, resulting in poor model training results, which in turn affects the labeling effect of colon cancer lesions. Summary of the invention

[0004] In order to solve the technical problem that the existing technology has poor effect on the annotation of colon cancer lesions, the purpose of the present invention is to provide a fine annotation method for colon cancer lesions in CT images. The technical solution adopted is as follows: Acquire all intestinal wall cross-sectional areas and all non-intestinal wall cross-sectional areas within the colon tissue area in the CT image; According to the shape features of each non-intestinal wall cross-sectional area, the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area in the intestinal wall cross-sectional area is obtained; in each non-intestinal wall cross-sectional area, the abnormality index of each pixel is obtained according to the local grayscale change of each pixel and the shape features of the corresponding intestinal wall segment; Based on the abnormal index and distribution characteristics of each pixel point, all pixel points in all non-intestinal wall cross-section areas are clustered to obtain all lesion sub-areas; according to the abnormal index of each pixel point in each lesion sub-area and the boundary rule of the lesion sub-area, the lesion score of each lesion sub-area is obtained; according to the spatial distance and area difference between different lesion sub-areas, combined with the difference between the lesion scores, different lesion sub-areas are merged to obtain all lesion areas; According to the lesion scores of all the lesion sub-regions in each lesion region, lesion regions to be marked are screened out from all the lesion regions and marked.

[0005] Furthermore, the method for obtaining the intestinal wall segment includes: On the regional boundary of each non-intestinal wall cross-sectional area, the two boundary pixel points with the farthest distance are taken as target pixel points; with the center of the CT image as the starting point of the ray, a ray is made through each of the target pixel points respectively, and the intestinal wall cross-sectional sub-area between the two line segments obtained by the intersection of the two rays with the intestinal wall cross-sectional area is taken as the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area in the intestinal wall cross-sectional area.

[0006] Furthermore, the method for obtaining the abnormality index includes: In each of the non-intestinal wall cross-sectional areas, according to the deviation of the gray value of each pixel point relative to the average level, combined with the gray values ​​of all neighboring pixels of each pixel point, the gray abnormality coefficient of each pixel point is obtained; According to the width and boundary regularity of the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area, a morphological abnormality coefficient of the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area is obtained; The abnormality index of each pixel is obtained according to the grayscale abnormality coefficient of each pixel and the morphological abnormality coefficient of the intestinal wall segment where each pixel is located.

[0007] Furthermore, the method for obtaining the grayscale anomaly coefficient includes: The ratio of the gray value of each pixel to the average gray value of all pixels in the non-intestinal wall cross-section area is used as the first gray abnormality parameter; The variance of the grayscale values ​​of all pixels in the eight neighborhoods of each pixel is used as the second grayscale anomaly parameter; The first grayscale anomaly parameter and the second grayscale anomaly parameter are fused to obtain a grayscale anomaly coefficient of a corresponding pixel.

[0008] Furthermore, the method for obtaining the width includes: In each of the intestinal wall segments, two boundary pixel points where the two rays intersect the intestinal wall cross-sectional area for the first time are used as inner wall boundary corner points in the intestinal wall segment; a boundary formed by all boundary pixel points between the two inner wall boundary corner points is used as the inner wall boundary of the intestinal wall segment; Taking the center of the CT image as the starting point of the ray, a ray is drawn through each boundary pixel point on the inner wall boundary, and each ray is used as a reference ray; the length of the line segment where each reference ray intersects with the intestinal wall segment is used as all the reference widths corresponding to the intestinal wall segment, and the average of all the reference widths is used as the width of the intestinal wall segment.

[0009] Furthermore, the method for obtaining the morphological abnormality coefficient includes: The ratio of the width of each intestinal wall segment to the mean of the widths of all the intestinal wall segments is used as a width deviation parameter; the variance of all the reference widths of each intestinal wall segment is used as a width fluctuation parameter; the width deviation parameter and the width fluctuation parameter are integrated to obtain a first morphological abnormality parameter; In each of the intestinal wall segments, taking any of the inner wall boundary corner points as a starting point, obtaining a chain code of the inner wall boundary; taking the sum of the absolute values ​​of the differences between all adjacent chain code values ​​in the chain code as an inner wall irregularity index; taking the negative correlation mapping result of the length of the inner wall boundary as a weighted weight of the inner wall irregularity index, and taking the weighted result as a second morphological abnormality parameter; The first morphological abnormality parameter and the second morphological abnormality parameter are fused to obtain the morphological abnormality coefficient of the intestinal wall segment.

[0010] Furthermore, the method for acquiring the lesion sub-region includes: All the pixel points in the non-intestinal wall cross-sectional area are taken as pixel points to be clustered, and the metric distance between different pixel points to be clustered is obtained according to the difference of the abnormal index and the spatial distance between different pixel points to be clustered; based on the metric distance between different pixel points to be clustered and the preset K value, K-means clustering is performed on all pixel points to be clustered to obtain all clusters; regional connectivity detection is performed on all pixel points in each cluster, and all connected domains are regarded as a lesion sub-region.

[0011] Furthermore, the method for obtaining the lesion score includes: Acquire a first lesion score for each lesion sub-region according to a concentrated feature of the abnormality index of all pixels in each lesion sub-region; Among all the boundary pixel points of each lesion sub-region, taking any boundary pixel point as a starting point, obtaining a boundary chain code of the lesion sub-region; taking the sum of the absolute values ​​of the differences between all adjacent chain code values ​​in the boundary chain code as a second lesion score corresponding to the lesion sub-region; The first lesion score and the second lesion score are fused to obtain a lesion score corresponding to the lesion sub-region.

[0012] Furthermore, the method for obtaining the lesion area includes: According to the spatial distances between different lesion sub-regions and the score differences between the corresponding lesion scores, obtaining first combined difference parameters corresponding to different lesion sub-regions; the spatial distances and the score differences are both positively correlated with the first combined difference parameters; using the negative correlation mapping result of the area difference between different lesion sub-regions as the second combined difference parameter corresponding to the different lesion sub-regions; The first merged difference parameter and the second merged difference parameter are integrated to obtain the merged distance between the corresponding different lesion sub-regions; based on the merged distance and the distance clustering algorithm, all the lesion sub-regions are clustered to obtain all clusters, and each cluster is regarded as a lesion region.

[0013] Furthermore, the method for obtaining the lesion area to be marked includes: In each lesion area, the average of the lesion scores of all the lesion sub-areas is used as the index to be screened of the lesion area; and the lesion area whose index to be screened is greater than a preset threshold is used as the lesion area to be marked.

[0014] The present invention has the following beneficial effects: The present invention obtains the intestinal wall segment corresponding to each non-intestinal wall cross-section area in the intestinal wall cross-section area in the colon tissue area in the CT image, so as to prepare for the subsequent analysis of the morphological characteristics of the intestinal wall cross-section and the texture characteristics of the non-intestinal wall cross-section, and the evaluation and annotation of the lesions; then, in each non-intestinal wall cross-section area, according to the local grayscale change of each pixel point and the shape characteristics of the corresponding intestinal wall segment, an abnormality index reflecting the possibility of the lesion of each pixel point is obtained; further, based on the abnormality index and distribution characteristics of each pixel point, all the pixel points in all non-intestinal wall cross-section areas are clustered to preliminarily obtain all the lesion sub-areas; then, according to each lesion sub-area, an abnormality index is obtained. The abnormal index of each pixel in the region and the boundary rule of the lesion sub-region are used to obtain the lesion score of each lesion sub-region; the larger the lesion score, the greater the possibility and degree of the lesion in the lesion sub-region; then based on the gradual diffusion characteristics of the lesion, the spatial distance and area difference between different lesion sub-regions are analyzed, and the difference between the lesion scores is combined to evaluate the lesion performance of different lesion sub-regions, so as to merge different lesion sub-regions and obtain all lesion regions; finally, according to the lesion scores of all lesion sub-regions in each lesion region, the lesion regions to be marked are screened out from all lesion regions and marked. The present invention analyzes the characteristics of the changes in the texture of the intestinal wall surface and the morphology of the intestinal wall cross section caused by colon canceration, further combines the area characteristics and distribution characteristics of the lesions in different lesion stages, accurately obtains and marks the lesion region, and improves the marking effect of colon cancer lesions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flow chart of a method for finely marking colon cancer lesions in CT images provided by one embodiment of the present invention; Figure 2 A partial schematic diagram of an intestinal wall segment provided by one embodiment of the present invention; Figure 3 A flow chart of a method for obtaining an abnormality index provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the method for finely marking colon cancer lesions in CT images proposed by the present invention, in combination with the accompanying drawings and preferred embodiments, and its specific implementation, structure, characteristics and effects are described as follows. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0018] 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.

[0019] The specific scheme of the method for finely marking colon cancer lesions in CT images provided by the present invention is described in detail below with reference to the accompanying drawings.

[0020] See also Figure 1 , which shows a flow chart of a method for finely marking colon cancer lesions in CT images provided by an embodiment of the present invention, specifically comprising: Step S1, acquiring all intestinal wall cross-sectional areas and all non-intestinal wall cross-sectional areas in the colon tissue area in the CT image.

[0021] Considering that CT scans generate cross-sectional images of body structures, there may be some intestinal wall cross-sectional areas in the colon tissue area in the CT image; and colon cancer lesions are mainly lesions on the inner wall of the colon, which may cause lesions in the entire layer of the colon wall structure or adjacent organs in severe cases, thereby causing changes in the texture of the intestinal wall surface and the morphology of the intestinal wall cross-section, specifically manifested as intestinal wall thickening or fibrosis; therefore, by analyzing the morphological characteristics of the intestinal wall cross-section in the colon tissue area and the texture characteristics of the non-intestinal wall cross-section, the possibility of each pixel point in the non-intestinal wall cross-sectional area being a lesion can be evaluated, and the lesions can be accurately marked; Based on this, an embodiment of the present invention will first obtain the colon tissue area in the CT image of the patient, and further obtain all intestinal wall cross-sectional areas and all non-intestinal wall cross-sectional areas within the colon tissue area for subsequent analysis and marking of lesions; wherein the intestinal wall cross-sectional area corresponds to the cross-section of the intestinal wall, and the non-intestinal wall cross-sectional area corresponds to the inner or outer wall of the colon.

[0022] As an example, first, preparations before CT scanning are done, such as requiring patients to clean their intestines and take contrast agents, etc., to enhance image clarity and contrast; then a CT scan is performed to obtain the CT scan results of the patient's abdominal cavity, and the CT scan results are preprocessed by denoising or enhancement to reduce noise interference, thereby obtaining a CT image to be labeled; considering that the contrast agent will cause the grayscale value of the colon tissue area to have an obvious grayscale difference with the surrounding tissue, the CT image is segmented based on the Otsu threshold segmentation algorithm, and the segmented foreground area is used as the colon tissue area; then all intestinal wall cross-sectional areas in the colon tissue area are obtained based on the convolutional neural network; all pixel points in the colon tissue area except the intestinal wall cross-sectional area are subjected to regional connectivity detection, and each connected domain is used as a non-intestinal wall cross-sectional area.

[0023] It should be noted that there are usually multiple CT images of different sections in the CT scan results. Since the analysis and annotation method of each CT image is consistent, only any CT image is used as the object to be annotated for analysis and description; CT images are already grayscale images, so there is no need to perform grayscale processing.

[0024] It should be noted that obtaining CT images, Otsu threshold segmentation, training and application of convolutional neural networks, and regional connectivity detection are all existing technologies well known to those skilled in the art and will not be described in detail here; in other embodiments, the implementer can also determine the segmentation threshold based on the image texture features to segment and obtain the colon tissue area, which will not be described in detail.

[0025] Step S2, according to the shape characteristics of each non-intestinal wall cross-sectional area, obtain the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area in the intestinal wall cross-sectional area; in each non-intestinal wall cross-sectional area, according to the local grayscale change of each pixel point and the shape characteristics of the corresponding intestinal wall segment, obtain the abnormality index of each pixel point.

[0026] Considering that intestinal wall thickening is a further manifestation of lesions in the non-intestinal wall cross-sectional area, the occurrence of lesions in the non-intestinal wall cross-sectional area and intestinal wall thickening are usually in a corresponding relationship, that is, lesions of different shapes and sizes in the non-intestinal wall cross-sectional area usually lead to thickening of intestinal wall segments of different lengths; and analyzing the cross-section of the intestinal wall, that is, the intestinal wall cross-sectional area, can help determine whether the intestinal wall is thickened; therefore, the embodiment of the present invention will first obtain the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area in the intestinal wall cross-sectional area based on the shape characteristics of each non-intestinal wall cross-sectional area, so as to subsequently accurately evaluate the possibility of lesions in each non-intestinal wall cross-sectional area.

[0027] Preferably, in one embodiment of the present invention, considering that the two boundary pixel points with the farthest distance on the boundary of each non-intestinal wall cross-sectional area can roughly reflect the tissue size or contour span of the non-intestinal wall cross-sectional area, the intestinal wall segment of the corresponding length in the intestinal wall cross-sectional area can be obtained; therefore, the method for obtaining the intestinal wall segment includes: On the boundary of each non-intestinal wall cross-sectional area, the two boundary pixel points with the farthest distance are taken as target pixel points; with the center of the CT image as the starting point of the ray, a ray is made through each target pixel point respectively, and the intestinal wall cross-sectional sub-area between the two line segments obtained by the intersection of the two rays with the intestinal wall cross-sectional area is taken as the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area in the intestinal wall cross-sectional area.

[0028] As an example, see Figure 2 , which shows a partial schematic diagram of an intestinal wall segment provided by an embodiment of the present invention; Figure 2 In the method, two boundary pixel points A and B with the farthest Euclidean distance on each non-intestinal wall cross-sectional area are selected, and two rays are drawn through A and B respectively with the center of the CT image as the starting point of the ray. Combined with the intersection of the ray and the intestinal wall cross-sectional area, the geometric span and corresponding position of the intestinal wall segment are accurately obtained, which is convenient for the subsequent imaging analysis of the intestinal wall segment thickness.

[0029] Considering that lesions in non-intestinal wall cross-sectional areas will not only lead to intestinal wall thickening, but also intestinal wall fibrosis, which is specifically manifested in that the texture characteristics of the lesion area are greatly different from those of other normal tissue areas, and the local grayscale of the lesion area will change; therefore, the embodiment of the present invention will, in each non-intestinal wall cross-sectional area, obtain the abnormality index of each pixel point according to the local grayscale change of each pixel point and the shape characteristics of the corresponding intestinal wall segment; the abnormality index reflects the possibility of lesions at each pixel point, providing a basis for subsequent evaluation and annotation of lesion areas.

[0030] Preferably, in one embodiment of the present invention, the method for obtaining the abnormality index includes: See also Figure 3 , which shows a flow chart of a method for obtaining an abnormality index provided by an embodiment of the present invention, specifically comprising: Step S201, in each non-intestinal wall cross-sectional area, according to the deviation of the gray value of each pixel relative to the average level, combined with the gray values ​​of all neighboring pixels of each pixel, obtain the gray abnormality coefficient of each pixel.

[0031] Considering that the grayscale of normal tissue areas of the colon is usually relatively uniform and similar, while the lesion area of ​​colon cancer usually has a high grayscale value and uneven local grayscale distribution due to changes or uneven distribution of local tissue density; based on this disease characteristic, the grayscale anomaly coefficient of each pixel point in each non-intestinal wall cross-section area can be analyzed and evaluated; the grayscale anomaly coefficient analyzes and evaluates the possibility of the pixel point being a pixel point in the lesion area from the perspective of the grayscale texture of the pixel point. The larger the grayscale anomaly coefficient, the more likely the pixel point is a pixel point in the lesion area.

[0032] In a preferred embodiment of the present invention, the method for obtaining the grayscale anomaly coefficient includes: The ratio of the grayscale value of each pixel to the mean grayscale value of all pixels in the non-intestinal wall cross-section area is used as the first grayscale abnormality parameter; the variance of the grayscale values ​​of all pixels in the eight-neighborhood of each pixel is used as the second grayscale abnormality parameter; the first grayscale abnormality parameter and the second grayscale abnormality parameter are fused to obtain the grayscale abnormality coefficient of the corresponding pixel.

[0033] As an example, taking any non-intestinal wall cross-section area as an example, the calculation formula of the grayscale abnormality coefficient is: ; Where a is the serial number of the pixel point in the non-intestinal wall cross-section area; is the grayscale abnormality coefficient of the ath pixel in the non-intestinal wall cross-section area; is the gray value of the ath pixel in the non-intestinal wall cross-section area; It is the mean gray value of all pixels in the non-intestinal wall cross-section area in the CT image; is the first grayscale abnormal parameter of the ath pixel in the non-intestinal wall cross-section area; is the variance of the grayscale values ​​of all pixels in the eight-neighborhood of the a-th pixel in the non-intestinal wall cross-section area, that is, the second grayscale abnormal parameter of the a-th pixel in the non-intestinal wall cross-section area; norm() is the standard normalization function.

[0034] It should be noted that the mean grayscale value of all pixels in the non-intestinal wall cross-section area cannot be 0, so the ratio is meaningful.

[0035] In this example, when the ratio of the grayscale value of a pixel to the average grayscale value of all pixels in the non-intestinal wall cross-sectional area is larger, it means that the grayscale value of the pixel is larger and the deviation from the average level is larger, then it is more likely to be a pixel in the lesion area, and the first grayscale abnormality parameter is larger; the larger the variance of the grayscale values ​​of all pixels in the eight neighborhoods of the pixel, the more uneven its local grayscale distribution is, and the more likely it is caused by local density changes, the greater the possibility that it is a pixel in the lesion area, and the larger the second grayscale abnormality parameter; then the first grayscale abnormality parameter and the second grayscale abnormality parameter are multiplied and fused, and then linearly normalized to obtain the grayscale abnormality coefficient.

[0036] In other embodiments, the implementer may also evaluate the grayscale abnormality of the pixel by the difference between the grayscale value of the pixel and the grayscale mean. The larger the linear normalization result of the difference, the larger the first grayscale abnormality parameter. The implementer may also obtain the sum of the absolute values ​​of the difference between the grayscale value of the pixel and each pixel in its eight neighborhoods instead of the variance to evaluate the local grayscale unevenness and obtain the second grayscale abnormality parameter. Then, other basic mathematical operations such as addition or weighted summation or related mapping methods are used to combine the two to obtain a grayscale abnormality coefficient.

[0037] Step S202, obtaining the morphological abnormality coefficient of the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area according to the width and boundary regularity of the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area.

[0038] Taking into account that the lesions in each non-intestinal wall cross-sectional area will lead to the thickening of the intestinal wall and the changes in the width and boundary of the intestinal wall segment, the morphological abnormality coefficient of the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area can be analyzed and evaluated based on this; the morphological abnormality coefficient indirectly reflects the possibility of lesions in the intestinal wall segment corresponding to the non-intestinal wall cross-sectional area, and prepares for the subsequent evaluation and annotation of the lesion area in combination with the grayscale abnormality coefficient of the pixel points in the non-intestinal wall cross-sectional area.

[0039] It should be noted that the regularity of the intestinal wall boundary refers to whether the intestinal wall boundary is relatively smooth. If the intestinal wall boundary appears to be rugged or tortuous, it indicates that there may be lesions in the rugged or tortuous morphologically abnormal areas.

[0040] To evaluate the morphological abnormality coefficient, the width of the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area is first evaluated; considering that the intestinal wall segment corresponding to the non-intestinal wall cross-sectional area is usually divided into an inner wall and an outer wall, the distance between the inner wall and the outer wall roughly reflects the width of the intestinal wall segment; based on this, in a preferred embodiment of the present invention, the method for obtaining the width includes: In each intestinal wall segment, the two boundary pixel points where the two rays intersect the intestinal wall cross-section area for the first time are taken as the inner wall boundary corner points in the intestinal wall segment; the boundary formed by all boundary pixel points between the two inner wall boundary corner points is taken as the inner wall boundary of the intestinal wall segment; Taking the center of the CT image as the starting point of the ray, a ray is drawn through each boundary pixel point on the inner wall boundary, and each ray is used as a reference ray; the length of the line segment where each reference ray intersects with the intestinal wall segment is used as all the reference widths of the corresponding intestinal wall segment, and the average of all the reference widths is used as the width of the intestinal wall segment.

[0041] See also Figure 2 , Figure 2 In the figure, the two rays passing through the two target pixels in the non-intestinal wall cross-section area and the two boundary pixels that first intersect with the intestinal wall cross-section area are the two inner wall boundary corner points C and D, thereby obtaining the inner wall boundary CD; then the reference ray passing through each boundary pixel on the inner wall boundary CD and the length of the line segment intersecting with the intestinal wall segment are used as the reference width, as shown in Figure 2 The length of the line segment EF where the reference ray intersects the intestinal wall segment corresponding to the boundary pixel point E on the inner wall boundary CD is the reference width, and then the reference widths of all boundary pixel points on the inner wall boundary CD are combined to obtain the width of the intestinal wall segment.

[0042] After obtaining the width of the intestinal wall segment, the morphological abnormality coefficient can be further evaluated in combination with the boundary regularity.

[0043] In a preferred embodiment of the present invention, it is considered that the wider a certain intestinal wall segment is and the greater the deviation is from the average width of all other intestinal wall segments, the more likely it is that the morphology of the segment will change due to the influence of the lesion; and considering that the lesion usually develops from the inner wall, the regularity of the inner wall boundary can be analyzed and evaluated to determine whether the intestinal wall morphology is abnormal; and the chain code can be used to evaluate whether the boundary change is regular. If the inner wall boundary is relatively regular, the chain code values ​​of adjacent boundary pixels should be similar; Based on this, the methods for obtaining the morphological anomaly coefficient include: The ratio of the width of each intestinal wall segment to the mean width of all intestinal wall segments is used as a width deviation parameter; the variance of all reference widths of each intestinal wall segment is used as a width fluctuation parameter; the width deviation parameter and the width fluctuation parameter are fused to obtain the first morphological abnormality parameter; In each intestinal wall segment, any inner wall boundary corner point is used as the starting point to obtain the chain code of the inner wall boundary; the sum of the absolute values ​​of the differences between all adjacent chain code values ​​in the chain code is used as the inner wall irregularity index; the negative correlation mapping result of the length of the inner wall boundary is used as the weighted weight of the inner wall irregularity index, and the weighted result is used as the second morphological abnormality parameter; The first morphological abnormality parameter and the second morphological abnormality parameter are integrated to obtain the morphological abnormality coefficient of the intestinal wall segment.

[0044] As an example, the inner wall boundary corner point C is taken as the starting point to obtain the chain code of the inner wall boundary; it should be noted that the chain code is a well-known technology and will not be described in detail here; The calculation formula of the morphological abnormality coefficient is: ; Where i is the serial number of the non-intestinal wall cross-section area; is the morphological abnormality coefficient of the i-th non-intestinal wall cross-sectional area; is the width of the intestinal wall segment corresponding to the i-th non-intestinal wall cross-sectional area; is the mean width of all intestinal wall segments; is the width deviation parameter of the intestinal wall segment corresponding to the i-th non-intestinal wall cross-section area; are all reference widths of the intestinal wall segments corresponding to the i-th non-intestinal wall cross-section area; is the first morphological abnormality parameter of the intestinal wall segment corresponding to the i-th non-intestinal wall cross-section area; m is the sequence number of the absolute value of the difference between adjacent chain code values ​​in the chain code of the inner wall boundary of the intestinal wall segment corresponding to the i-th non-intestinal wall cross-section area; is the mth absolute value of the difference among the absolute values ​​of the differences between all adjacent chain code values ​​in the chain code of the inner wall boundary of the intestinal wall segment corresponding to the i-th non-intestinal wall cross-section area; is the inner wall irregularity index of the intestinal wall segment corresponding to the i-th non-intestinal wall cross-section area; is the length of the inner wall boundary of the intestinal wall segment corresponding to the i-th non-intestinal wall cross-section area; is the second morphological abnormality parameter of the intestinal wall segment corresponding to the i-th non-intestinal wall cross-sectional area; norm() is the standard normalization function.

[0045] In this example, when the ratio of the width of the intestinal wall segment to the average width of all intestinal wall segments is larger, it means that the width of the intestinal wall segment is wider and the deviation from the average level is larger, and the width deviation parameter is larger; when the variance of all reference widths of the intestinal wall segment is larger, it means that the width of the intestinal wall segment is more uneven and the width fluctuation parameter is larger; then the width fluctuation parameter is multiplied and fused with the width deviation parameter to obtain the first morphological abnormality parameter. The wider the width and the uneven width change, the more likely the intestinal wall segment is to change its morphology due to the lesion; the smaller the absolute value of the difference between the chain code values ​​corresponding to adjacent boundary pixels on the inner wall boundary, the more regular the inner wall boundary is and the smaller the inner wall irregularity index is; considering that the inner wall irregularity index is also affected by the length of the inner wall boundary, the length of the inner wall boundary is reciprocally calculated and used as the weighted weight of the inner wall irregularity index, and then multiplied to obtain the second morphological abnormality parameter; finally, the first morphological abnormality parameter and the second morphological abnormality parameter are multiplied and fused, and then linearly normalized to obtain the morphological abnormality coefficient.

[0046] In other embodiments, the implementer may also evaluate the width deviation parameter by the difference between the width of the intestinal wall segment and the mean width. The larger the linear normalization result of the difference is, the larger the width deviation parameter is. Other basic mathematical operations such as addition or weighted summation may be used to combine the width fluctuation parameter and the width deviation parameter to obtain a first morphological abnormality parameter. The implementer may also use other negative correlation mapping methods to map the length of the inner wall boundary, such as using it as the x in the exponential function exp(-x) with the natural constant e as the base to obtain a weighted weight. Other basic mathematical operations such as addition or weighted summation may be used to combine the first morphological abnormality parameter and the second morphological abnormality parameter to obtain a morphological abnormality coefficient.

[0047] Step S203, obtaining an abnormality index of each pixel point according to the grayscale abnormality coefficient of each pixel point and the morphological abnormality coefficient of the intestinal wall segment where each pixel point is located.

[0048] As an example, the grayscale abnormality coefficient of each pixel and the morphological abnormality coefficient of the intestinal wall segment where each pixel is located are added together, the sum is linearly normalized, and the normalized value is used as the abnormality index of each pixel; in other examples, the implementer may also use other basic mathematical operations such as weighted summation or multiplication or positive correlation mapping methods, which are not described here.

[0049] Step S3, based on the abnormal index and distribution characteristics of each pixel point, all pixels in all non-intestinal wall cross-sectional areas are clustered to obtain all lesion sub-areas; according to the abnormal index of each pixel point in each lesion sub-area and the boundary rules of the lesion sub-area, the lesion score of each lesion sub-area is obtained; according to the spatial distance and area difference between different lesion sub-areas, combined with the difference between lesion scores, different lesion sub-areas are merged to obtain all lesion areas.

[0050] Considering that the closer the abnormality index of the pixel points in the non-intestinal wall cross-sectional area is and the closer the spatial position distribution is, the more similar the lesion degree or lesion possibility of the corresponding pixel points is, and the more likely it is that the pixel points are in the same lesion area; and the clustering algorithm can cluster pixel points with similar characteristics into a cluster; therefore, the embodiment of the present invention clusters all pixel points in all non-intestinal wall cross-sectional areas based on the abnormality index and distribution characteristics of each pixel point to obtain all lesion sub-areas; after clustering, all pixel points in each cluster will be more likely to be pixel points in the same lesion area, and the lesion sub-area initially clusters pixel points with similar lesion degrees into a local sub-area for subsequent analysis and merging to obtain a complete lesion area.

[0051] Preferably, in one embodiment of the present invention, the method for acquiring the lesion sub-region includes: All pixels in the non-intestinal wall cross-sectional area are taken as pixels to be clustered. According to the difference of abnormal index and spatial distance between different pixels to be clustered, the metric distance between different pixels to be clustered is obtained. Based on the metric distance between different pixels to be clustered and the preset K value, all pixels to be clustered are clustered by K-means to obtain all clusters. Regional connectivity detection is performed on all pixels in each cluster, and all connected domains are regarded as a lesion sub-region.

[0052] As an example, the difference in anomaly index and spatial distance between different pixels to be clustered are first obtained, and the difference in anomaly index and the Euclidean norm of the spatial distance are used as the metric distance between different pixels to be clustered; then the elbow method is used to obtain the preset K value, and K-means clustering is performed on all pixels to be clustered, and then regional connectivity detection is performed within each cluster to obtain all lesion sub-regions within each cluster.

[0053] In other examples, the implementer may also perform weighted averaging of the difference in anomaly index and spatial distance to obtain the metric distance; other distance clustering algorithms may also be used instead of the K-means clustering algorithm, which, together with the K-means clustering algorithm, the elbow method, and the regional connectivity detection, are all well-known technologies and will not be elaborated here.

[0054] Considering that a lesion sub-region may contain different pixel points in the adjacent non-intestinal wall cross-sectional area, there may be certain differences in the abnormal coefficients of different pixel points, which is not convenient for accurately evaluating the lesion possibility of each lesion sub-region; considering that the more irregular the boundary of the lesion sub-region is, the greater the possibility of its lesion is; therefore, the embodiment of the present invention obtains the lesion score of each lesion sub-region based on the abnormal index of each pixel point in each lesion sub-region and the boundary regularity of the lesion sub-region; the larger the lesion score is, the greater the possibility and degree of lesion of the lesion sub-region is.

[0055] Preferably, in one embodiment of the present invention, the method for obtaining the lesion score includes: Obtaining the first lesion score of each lesion sub-region according to the concentrated features of the abnormal indexes of all pixels in each lesion sub-region; Among all the boundary pixel points of each lesion sub-region, take any boundary pixel point as the starting point to obtain the boundary chain code of the lesion sub-region; take the average of the absolute values ​​of the differences between all adjacent chain code values ​​in the boundary chain code as the second lesion score of the corresponding lesion sub-region; The first lesion score and the second lesion score are fused to obtain the lesion score of the corresponding lesion sub-region.

[0056] As an example, the lesion score is calculated as: ; Where r is the serial number of the lesion sub-region; is the lesion score for the rth lesion sub-region; is the mean of the abnormality indexes of all pixels in the r-th lesion sub-region, and is also the first lesion score of the r-th lesion sub-region; is the sequence number of the chain code value in the boundary chain code of the rth lesion sub-region; is the total length of the boundary chain code of the rth lesion sub-region; is the sth chain code value in the boundary chain code of the rth lesion sub-region; is the s+1th chain code value in the boundary chain code of the rth lesion sub-region; is the absolute value symbol; is the second lesion score of the rth lesion sub-region; norm() is the standard normalization function.

[0057] In this example, the mean is used to evaluate the first lesion score. The larger the mean, the larger the first lesion score. The larger the absolute value of the difference between adjacent chain code values ​​in the boundary chain code, the more obvious the boundary change, the more irregular the boundary, and the larger the second lesion score. The first lesion score and the second lesion score are then multiplied and fused, and then linearly normalized to obtain the lesion score of the corresponding lesion sub-area.

[0058] In other examples, implementers may also use the mode or median instead of the mean to evaluate the first lesion score, or use other basic mathematical operations such as addition or weighted summation to combine the first lesion score and the second lesion score, or use other normalization methods, which are not described here.

[0059] Considering that in the course of colon cancer pathology, there may be a trend of gradual diffusion, that is, there may be lesion manifestations of different degrees or stages of pathology at a certain intestinal wall; if the areas of different lesion sub-regions differ greatly, while the lesion scores are similar and the distances are close, it means that the different lesion sub-regions are more likely to correspond to lesion manifestations of different degrees or stages of pathology, and they are more likely to belong to the same lesion area. By merging and labeling, it will be more helpful for relevant medical personnel to evaluate the overall distribution, development trend and potential impact of the lesions from a macro perspective; therefore, the embodiment of the present invention will merge different lesion sub-regions according to the spatial distance and area differences between different lesion sub-regions, combined with the differences between the lesion scores, to obtain all lesion areas.

[0060] Preferably, in one embodiment of the present invention, considering that if the areas of different lesion sub-regions differ greatly, but the lesion scores are similar and the distances are close, the different lesion sub-regions should be merged, based on which the merged distances of the different lesion sub-regions can be obtained; the larger the merged distance, the less suitable it is to merge; considering that clustering can be based on the merged distances between different lesion sub-regions, the lesion sub-regions can be clustered; therefore, the method for obtaining the lesion region includes: According to the spatial distances between different lesion sub-regions and the score differences between the corresponding lesion scores, first merged difference parameters corresponding to the different lesion sub-regions are obtained; the spatial distances and the score differences are both positively correlated with the first merged difference parameters; The negative correlation mapping result of the area difference between different lesion sub-regions is used as the second merged difference parameter corresponding to the different lesion sub-regions; The first merged difference parameter and the second merged difference parameter are integrated to obtain the merged distance between the corresponding different lesion sub-regions; based on the merged distance and the distance clustering algorithm, all lesion sub-regions are clustered to obtain all clusters, and each cluster is regarded as a lesion region.

[0061] As an example, firstly, the merge distance between different lesion sub-regions is calculated. The merge distance reflects the necessity of merging different lesion sub-regions. The larger the merge distance, the less likely they should be merged. Take any two different lesion sub-regions as an example; the spatial distance between different lesion sub-regions is the Euclidean distance between their centroids, the area can be evaluated by the number of pixels in the lesion sub-region, the area difference is evaluated by the absolute value of the difference in the number of pixels, and the score difference is evaluated by the absolute value of the difference in the lesion scores; the calculation formula for the combined distance is: ; Where F is the merged distance between different lesion sub-regions; is the area difference between different lesion sub-regions; is the score difference between lesion scores in different lesion sub-areas; is the spatial distance between lesion scores of different lesion sub-regions; is the first merge difference parameter; is the second merge difference parameter.

[0062] In this example, the specific score difference and the area difference are multiplied and combined to obtain the first combined difference parameter. The larger the product, the less likely it is to be merged and the larger the merge distance. The area difference is added with the minimum positive parameter 0.1 and then the inverse operation is performed to adjust the logic of negative correlation mapping, so that the smaller the area difference, the less necessary the merge and the larger the second combined difference parameter. Since the areas of different lesion sub-regions may be equal, the minimum positive parameter 0.1 does not affect the calculation results, so as to avoid the area difference being 0 and causing the inverse to be meaningless, and then the first combined difference parameter and the second combined difference parameter are multiplied and combined to obtain the merge distance. The larger the merge distance, the less likely the lesion sub-regions should be merged.

[0063] It should be noted that, in other embodiments, the implementer may also use other basic mathematical operations such as weighted sum to combine two combined difference parameters; or may use other negative correlation mapping means to perform negative correlation mapping on the area difference, which will not be elaborated here.

[0064] After obtaining the merged distances between all different lesion sub-regions, all different lesion sub-regions can be clustered and merged to obtain the merged distances between all different lesion sub-regions; then the elbow method is used to obtain the preset K value, K-means clustering is performed on all lesion sub-regions, and then all lesion sub-regions in the same cluster are merged into one lesion region.

[0065] It should be noted that the K-means clustering algorithm and the elbow method are both existing technologies and will not be described in detail here.

[0066] Step S4, based on the lesion scores of all lesion sub-regions in each lesion region, select the lesion regions to be marked from all lesion regions and mark them.

[0067] Taking into account that after the lesion areas are merged, each lesion area may contain multiple lesion sub-areas. The larger the lesion score of the lesion sub-area, the more it should be marked, so as to facilitate the attention and analysis of relevant medical personnel; therefore, the embodiment of the present invention will further screen out the lesion areas to be marked from all lesion areas based on the lesion scores of all lesion sub-areas in each lesion area and mark them.

[0068] Preferably, in one embodiment of the present invention, the method for acquiring the lesion area to be marked includes: In each lesion area, the average of the lesion scores of all lesion sub-areas is used as the index to be screened for the lesion area; the lesion area whose index to be screened is greater than the preset threshold is used as the lesion area to be marked; considering that the value range of the lesion scores of all lesion sub-areas is 0-1, the value range of the index to be screened is also 0-1, so in this example, the preset threshold is set to 0.6, and the implementer can also set it by himself, which will not be repeated here.

[0069] In one embodiment of the present invention, after screening out all lesion areas to be marked, the boundaries of each representative and lesion area can be marked, and then the marked information can be mapped to the CT image through masking technology, so as to obtain a CT image with finely marked colon lesions.

[0070] In summary, the present invention first obtains the intestinal wall cross-sectional area and the non-intestinal wall cross-sectional area in the CT image, and obtains the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area in the intestinal wall cross-sectional area; then analyzes and obtains the abnormal index of each pixel point in each non-intestinal wall cross-sectional area; based on the abnormal index and distribution characteristics of each pixel point, obtains all lesion sub-areas; then analyzes and obtains the lesion score of each lesion sub-area, further merges different lesion sub-areas, and obtains all lesion areas; finally, the lesion areas to be marked are screened out from all lesion areas and annotated. The present invention analyzes the characteristics of the changes in the texture of the intestinal wall surface and the morphology of the intestinal wall cross-section caused by colon cancer, and further combines the area characteristics and distribution characteristics of the lesions at different lesion stages, accurately obtains and annotates the lesion area, and improves the annotation effect of colon cancer lesions.

[0071] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for finely marking colon cancer lesions in CT images, characterized in that: The method comprises: Acquire all intestinal wall cross-sectional areas and all non-intestinal wall cross-sectional areas within the colon tissue area in the CT image; According to the shape features of each non-intestinal wall cross-sectional area, the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area in the intestinal wall cross-sectional area is obtained; in each non-intestinal wall cross-sectional area, the abnormality index of each pixel is obtained according to the local grayscale change of each pixel and the shape features of the corresponding intestinal wall segment; Based on the abnormal index and distribution characteristics of each pixel point, all pixel points in all non-intestinal wall cross-section areas are clustered to obtain all lesion sub-areas; according to the abnormal index of each pixel point in each lesion sub-area and the boundary rule of the lesion sub-area, the lesion score of each lesion sub-area is obtained; according to the spatial distance and area difference between different lesion sub-areas, combined with the difference between the lesion scores, different lesion sub-areas are merged to obtain all lesion areas; According to the lesion scores of all the lesion sub-regions in each lesion region, lesion regions to be marked are screened out from all the lesion regions and marked.

2. The method for finely marking colon cancer lesions in CT images according to claim 1, characterized in that: The method for obtaining the intestinal wall segment comprises: On the regional boundary of each non-intestinal wall cross-sectional area, the two boundary pixel points with the farthest distance are taken as target pixel points; with the center of the CT image as the starting point of the ray, a ray is made through each of the target pixel points respectively, and the intestinal wall cross-sectional sub-area between the two line segments obtained by the intersection of the two rays with the intestinal wall cross-sectional area is taken as the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area in the intestinal wall cross-sectional area.

3. The method for finely marking colon cancer lesions in CT images according to claim 2, characterized in that: The method for obtaining the abnormality index includes: In each of the non-intestinal wall cross-sectional areas, according to the deviation of the gray value of each pixel point relative to the average level, combined with the gray values ​​of all neighboring pixels of each pixel point, the gray abnormality coefficient of each pixel point is obtained; According to the width and boundary regularity of the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area, a morphological abnormality coefficient of the intestinal wall segment corresponding to each non-intestinal wall cross-sectional area is obtained; The abnormality index of each pixel is obtained according to the grayscale abnormality coefficient of each pixel and the morphological abnormality coefficient of the intestinal wall segment where each pixel is located.

4. The method for finely marking colon cancer lesions in CT images according to claim 3, characterized in that: The method for obtaining the grayscale anomaly coefficient includes: The ratio of the gray value of each pixel to the average gray value of all pixels in the non-intestinal wall cross-section area is used as the first gray abnormality parameter; The variance of the grayscale values ​​of all pixels in the eight neighborhoods of each pixel is used as the second grayscale anomaly parameter; The first grayscale anomaly parameter and the second grayscale anomaly parameter are fused to obtain a grayscale anomaly coefficient of a corresponding pixel.

5. The method for finely marking colon cancer lesions in CT images according to claim 3, characterized in that: The method for obtaining the width includes: In each of the intestinal wall segments, two boundary pixel points where the two rays intersect the intestinal wall cross-sectional area for the first time are used as inner wall boundary corner points in the intestinal wall segment; a boundary formed by all boundary pixel points between the two inner wall boundary corner points is used as the inner wall boundary of the intestinal wall segment; Taking the center of the CT image as the starting point of the ray, a ray is drawn through each boundary pixel point on the inner wall boundary, and each ray is used as a reference ray; the length of the line segment where each reference ray intersects with the intestinal wall segment is used as all the reference widths corresponding to the intestinal wall segment, and the average of all the reference widths is used as the width of the intestinal wall segment.

6. The method for finely marking colon cancer lesions in CT images according to claim 5, characterized in that: The method for obtaining the morphological abnormality coefficient comprises: The ratio of the width of each intestinal wall segment to the mean of the widths of all the intestinal wall segments is used as a width deviation parameter; the variance of all the reference widths of each intestinal wall segment is used as a width fluctuation parameter; the width deviation parameter and the width fluctuation parameter are integrated to obtain a first morphological abnormality parameter; In each of the intestinal wall segments, taking any of the inner wall boundary corner points as a starting point, obtaining a chain code of the inner wall boundary; taking the sum of the absolute values ​​of the differences between all adjacent chain code values ​​in the chain code as an inner wall irregularity index; taking the negative correlation mapping result of the length of the inner wall boundary as a weighted weight of the inner wall irregularity index, and taking the weighted result as a second morphological abnormality parameter; The first morphological abnormality parameter and the second morphological abnormality parameter are fused to obtain the morphological abnormality coefficient of the intestinal wall segment.

7. The method for finely marking colon cancer lesions in CT images according to claim 1, characterized in that: The method for acquiring the lesion sub-region comprises: All the pixel points in the non-intestinal wall cross-sectional area are taken as pixel points to be clustered, and the metric distance between different pixel points to be clustered is obtained according to the difference of the abnormal index and the spatial distance between different pixel points to be clustered; based on the metric distance between different pixel points to be clustered and the preset K value, K-means clustering is performed on all pixel points to be clustered to obtain all clusters; regional connectivity detection is performed on all pixel points in each cluster, and all connected domains are regarded as a lesion sub-region.

8. The method for finely marking colon cancer lesions in CT images according to claim 1, characterized in that: The method for obtaining the lesion score includes: Acquire a first lesion score for each lesion sub-region according to a concentrated feature of the abnormality index of all pixels in each lesion sub-region; Among all the boundary pixel points of each lesion sub-region, taking any boundary pixel point as a starting point, obtaining a boundary chain code of the lesion sub-region; taking the sum of the absolute values ​​of the differences between all adjacent chain code values ​​in the boundary chain code as a second lesion score corresponding to the lesion sub-region; The first lesion score and the second lesion score are fused to obtain a lesion score corresponding to the lesion sub-region.

9. The method for finely marking colon cancer lesions in CT images according to claim 1, characterized in that: The method for obtaining the lesion area comprises: According to the spatial distances between different lesion sub-regions and the score differences between the corresponding lesion scores, obtaining first combined difference parameters corresponding to different lesion sub-regions; the spatial distances and the score differences are both positively correlated with the first combined difference parameters; using the negative correlation mapping result of the area difference between different lesion sub-regions as the second combined difference parameter corresponding to the different lesion sub-regions; The first merged difference parameter and the second merged difference parameter are integrated to obtain the merged distance between the corresponding different lesion sub-regions; based on the merged distance and the distance clustering algorithm, all the lesion sub-regions are clustered to obtain all clusters, and each cluster is regarded as a lesion region.

10. The method for finely marking colon cancer lesions in CT images according to claim 1, characterized in that: The method for acquiring the lesion area to be marked comprises: In each lesion area, the average of the lesion scores of all the lesion sub-areas is used as the index to be screened of the lesion area; and the lesion area whose index to be screened is greater than a preset threshold is used as the lesion area to be marked.

Images