Intelligent detection method for minimal lesion of central nervous system image

Abstract

The invention relates to the technical field of brain MRI image analysis, in particular to an intelligent detection method for central nervous system image minimal lesions. Obtaining a fuzzy index and a gray scale significant feature according to the gray scale difference of the pixel point neighborhood; screening the pixel points to obtain a suspected disease connected domain; obtaining a lesion expression factor according to the gray difference between the connected domain and the symmetric connected domain, the connected domain difference feature and the distance; obtaining a result saliency factor according to the pixel point gradient features; screening to obtain special pixel points; obtaining a descriptor of the connected domain according to the special pixel points; according to the descriptor difference and the lesion expression factor difference between different images, the lesion possibility degree is obtained; and intelligent detection of minimal lesions is carried out according to the lesion possible degree. According to the invention, the minimal lesion area in the brain MRI image can be accurately identified.

Description

Technical Field

[0001] The present invention relates to the technical field of brain MRI image analysis, and particularly relates to an intelligent detection method for minute lesions in central nervous system imaging. Background Art

[0002] The central nervous system is the most main part of the human nervous system composed of the brain and spinal cord, which dominates and controls all human behaviors. Among them, the brain, as the main component of the central nervous system, is very likely to have a great impact on the human body once a disease occurs. Therefore, early detection of the brain and early discovery of potential lesion risks are of crucial significance and are vital for improving the survival rate and quality of life of patients.

[0003] MRI (Magnetic Resonance Imaging) is a medical imaging technology commonly used for detecting brain lesions. In the prior art, regional segmentation of the patient's brain MRI images is usually performed to detect brain lesion areas. However, early brain lesions are small in area in MRI images, and the image manifestations are not obvious, which is easy to be confused with other brain tissue structures. In addition, the imaging quality of MRI technology is limited, resulting in poor effects of conventional intelligent detection methods in dealing with such minute lesions, and the accuracy of the segmentation results is limited, making it difficult to achieve the purpose of early diagnosis. Summary of the Invention

[0004] In order to solve the technical problem that the gray-scale manifestation of tiny lesions in brain MRI images is not obvious enough and is easily confused with other brain tissue structures, resulting in the inability of conventional detection methods to accurately diagnose tiny lesion areas, the purpose of the present invention is to provide an intelligent detection method for tiny lesions in central nervous system images. The specific technical solution adopted is as follows: An intelligent detection method for tiny lesions in central nervous system images, the method comprising: acquiring a brain MRI image of a patient to be tested; arbitrarily selecting a pixel point in the brain MRI image as a reference pixel point; obtaining a fuzziness index of the reference pixel point according to the gray-scale difference between adjacent pixel points within a preset first neighborhood of the reference pixel point; obtaining a gray-scale significance index of the reference pixel point according to the gray-scale significant features of each pixel point within a preset second neighborhood of the reference pixel point; screening and merging each pixel point in the brain MRI image according to the fuzziness index and the gray-scale significance index to obtain all suspected diseased connected regions of the brain MRI image; arbitrarily selecting a suspected diseased connected region as a reference connected region; obtaining a symmetric connected region of the reference connected region according to the symmetry of the brain MRI image; obtaining a lesion manifestation factor of the reference connected region according to the gray-scale difference, connected region difference feature and distance difference between the reference connected region and the symmetric connected region; obtaining a structural significance factor of each pixel point according to the gradient feature of each pixel point in the brain MRI image; screening all pixel points according to the structural significance factor to obtain special pixel points of the centroid pixel point within the reference connected region; obtaining a descriptor of the reference connected region by using the distance between the centroid pixel point and the special pixel point within the reference connected region, and the structural significance factor of the special pixel point; obtaining brain MRI images of other historical patients as other brain MRI images; obtaining the possible degree of lesions of the reference connected region according to the descriptor difference and the lesion manifestation factor difference between the reference connected region and the suspected diseased connected region at the closest position in the other brain MRI images; performing intelligent detection of tiny lesions according to the possible degree of lesions.

[0005] Further, the method for obtaining the fuzziness index includes: starting from the upper left corner, traversing the pixel points within the preset first neighborhood in a row-by-row alternating direction to obtain the serial number of each pixel point within the preset first neighborhood; obtaining the fuzziness index according to the fuzziness index calculation formula, and the fuzziness index calculation formula is as follows: In the formula, represents the fuzziness index of the reference pixel point; represents the information entropy of the gray-scale values of the pixel points within the preset first neighborhood; represents the number of pixel points within the preset first neighborhood; represents the th pixel point within the preset first neighborhood; represents the gray-scale value of the th pixel point within the preset first neighborhood; Represents the absolute value function.

[0006] Furthermore, the method for obtaining the grayscale significance index includes: starting from the upper left corner, traversing the pixel points in the preset second neighborhood in a row-by-row alternating direction to obtain the serial numbers of each pixel point in the preset second neighborhood; obtaining the grayscale significance index according to the grayscale significance index calculation formula, and the grayscale significance index calculation formula is as follows: In the formula, Represents the grayscale significance index of the reference pixel point; Represents the number of pixel points in the preset second neighborhood; Represents the th pixel point in the preset second neighborhood; Represents the maximum grayscale value of the pixel points in the preset second neighborhood; Represents the minimum grayscale value of the pixel points in the preset second neighborhood; Represents the maximum value function.

[0007] Furthermore, the method for obtaining the suspected diseased connected region includes: taking the pixel points where both the fuzzy index and the grayscale significance index are greater than a preset first threshold as suspected diseased pixel points; taking the region composed of all suspected diseased pixel points as the suspected diseased connected region.

[0008] Furthermore, the method for obtaining the symmetric connected region includes: obtaining the vertical midline of the brain MRI image, taking the vertical midline as the axis of symmetry, and obtaining the symmetric point of the centroid pixel point of the reference connected region; taking the suspected diseased connected region closest to the symmetric point as the symmetric connected region.

[0009] Furthermore, the method for obtaining the lesion manifestation factor includes: obtaining the lesion manifestation factor according to the lesion manifestation factor calculation formula, and the lesion manifestation factor calculation formula is as follows: In the formula, Represents the serial number of the reference connected region, Represents the serial number of the symmetric connected region; Represents the connected region difference feature between the reference connected region and the symmetric connected region; Represents the Hausdorff distance between the reference connected region and the symmetric connected region; Represents the grayscale mean value of the reference connected region; Represents the grayscale mean value of the symmetric connected region; Represents the lesion manifestation factor of the reference connected region; Represents the Euclidean distance between the centroid pixel point of the reference connected region and the centroid pixel point of the symmetric connected region; Represents the number of all other suspected diseased connected regions; Represents the reference connected region and the The connectivity domain difference features between other suspected diseased connectivity domains.

[0010] Further, the method for obtaining the structure significance factor includes: calculating the gradient value and gradient direction angle of each pixel point; calculating the variance of the gradient direction angles of all pixel points as the first variance; calculating the ratio between the gradient value of each pixel point and the first variance as the structure significance factor of the pixel point.

[0011] Further, the method for obtaining the special pixel points of the centroid pixel point in the reference connectivity domain includes: drawing four rays in the horizontal and vertical directions respectively starting from the centroid pixel point in the reference connectivity domain; taking the pixel points that first reach the preset third threshold among the four rays as the special pixel points of the centroid pixel point in the reference connectivity domain.

[0012] Further, the method for obtaining the descriptor of the reference connectivity domain includes: obtaining the descriptor according to the descriptor calculation formula, and the method for obtaining the descriptor calculation formula is as follows: In the formula, represents the descriptor of the reference connectivity domain; represents the structure significance factor of the special pixel point of the first ray; represents the structure significance factor of the special pixel point of the second ray; represents the structure significance factor of the special pixel point of the third ray; represents the structure significance factor of the special pixel point of the fourth ray; represents the distance between the special pixel point of the first ray and the centroid pixel point in the reference connectivity domain; represents the distance between the special pixel point of the second ray and the centroid pixel point in the reference connectivity domain; represents the distance between the special pixel point of the third ray and the centroid pixel point in the reference connectivity domain; represents the distance between the special pixel point of the fourth ray and the centroid pixel point in the reference connectivity domain.

[0013] Further, the method for obtaining the possible degree of lesion includes: obtaining the possible degree of lesion according to the possible degree of lesion calculation formula, and the possible degree of lesion calculation formula is as follows: In the formula, represents the possible degree of lesion of the reference connectivity domain; represents the number of other brain MRI images; represents the descriptor of the reference connectivity domain; represents the th descriptor of the connectivity domain at the same position as the reference connectivity domain in the represents the lesion manifestation factor of the reference connectivity domain; Indicates the lesion manifestation factor of the connected domain at the same position as the reference connected domain in the nth other brain MRI image; Indicates the cosine similarity function; Indicates the normalization function.

[0014] The present invention has the following beneficial effects: The present invention obtains the brain MRI image of a patient; Since the lesion is in the early stage, the boundary is usually relatively blurred, which is specifically manifested as a gray-scale prominent connected domain with a blurred edge in the MRI image. Therefore, first, the suspected diseased pixel points are searched, and according to the gray-scale difference between adjacent pixel points within the preset first neighborhood of the reference pixel point, the blur index of the reference pixel point is obtained; According to the gray-scale significant feature of each pixel point within the preset second neighborhood of the reference pixel point, the gray-scale significant index of the reference pixel point is obtained; Each pixel point in the brain MRI image is screened and merged through the blur index and the gray-scale significant index to obtain all suspected diseased connected domains in the brain MRI image; After obtaining the suspected diseased connected domains in the patient's brain MRI image, it is necessary to determine the lesion manifestation of the suspected diseased connected domains. Since there are many normal brain tissues in the brain that are similar to the manifestation of micro-lesions, it is also necessary to calculate the corresponding lesion manifestation factors according to the basic feature manifestations of the suspected connected domains; Since it is easy to miss detections by only using the single-image analysis method, the brain MRI images of historical patients are obtained; Since there are certain individual differences in the shape and size of the brains of different patients, but the relative positions conform to the physiological characteristics of the human brain, the structural significant factors of each pixel point can be analyzed to represent the importance of the position where the pixel point is located, and then the descriptor of the reference connected domain is obtained, and the suspected diseased connected domains are located with the descriptor as a reference, and finally the final micro-lesion area is determined by combining the located lesion manifestation factors. The present invention can accurately identify the micro-lesion area in the brain MRI image, and thus relevant personnel can accurately identify the micro-lesions in the brain MRI image. 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 following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a flowchart of an intelligent detection method for micro-lesions in central nervous system imaging provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following specifically describes, with reference to the accompanying drawings and preferred embodiments, a method for intelligent detection of minute lesions in central nervous system images according to the present invention, including its specific implementation manner, structure, features, and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0019] The following specifically describes the specific solution of a method for intelligent detection of minute lesions in central nervous system images provided by the present invention with reference to the accompanying drawings.

[0020] Please refer to Figure 1 , which shows a method for intelligent detection of minute lesions in central nervous system images provided by an embodiment of the present invention. The method includes: Step S1: Obtain the brain MRI image of the patient to be tested.

[0021] The embodiment of the present invention is mainly applied to the scenario of brain lesion detection based on MRI images. Therefore, the brain MRI image of the patient is first obtained. In one embodiment of the present invention, the brain MRI image of the patient is obtained by using an MRI device. Since the MRI image itself is a grayscale image, no additional grayscale processing is required. Then, the part belonging to the brain tissue is extracted by using semantic segmentation technology to avoid the influence of other irrelevant regions on the processing effect. It should be noted that semantic segmentation technology is a well-known technical means for those skilled in the art and will not be elaborated herein.

[0022] Step S2: Arbitrarily select a pixel point in the brain MRI image as a reference pixel point; obtain the fuzzy index of the reference pixel point according to the gray-scale difference between adjacent pixel points within the preset first neighborhood of the reference pixel point; obtain the gray-scale significant index of the reference pixel point according to the gray-scale significant feature of each pixel point within the preset second neighborhood of the reference pixel point; screen and merge each pixel point in the brain MRI image according to the fuzzy index and the gray-scale significant index to obtain all suspected diseased connected regions in the brain MRI image; arbitrarily select a suspected diseased connected region as a reference connected region; obtain the symmetric connected region of the reference connected region according to the symmetry of the brain MRI image; obtain the lesion manifestation factor of the reference connected region according to the gray-scale difference, connected region difference feature, and distance difference between the reference connected region and the symmetric connected region; obtain the structural significant factor of each pixel point according to the gradient feature of each pixel point in the brain MRI image.

[0023] In actual situations, there are obvious high and low signal changes in the tiny lesions in the brain during MRI scans, and they are relatively concentrated in terms of morphology. However, since the lesions are in the early stage, their boundaries are usually relatively blurred, and they are specifically manifested as gray-scale prominent connected regions with blurred edges in the MRI images. Therefore, in the embodiments of the present invention, first, the suspected diseased pixel points are searched, and according to the gray-scale differences between adjacent pixel points within the preset first neighborhood of the reference pixel point, the blur index of the reference pixel point is obtained; according to the gray-scale significant features of each pixel point within the preset second neighborhood of the reference pixel point, the gray-scale significant index of the reference pixel point is obtained.

[0024] Preferably, in one embodiment of the present invention, the method for obtaining the blur index includes: starting from the upper left corner, traversing the pixel points within the preset first neighborhood in a row-by-row alternating direction to obtain the serial numbers of each pixel point within the preset first neighborhood. In one embodiment of the present invention, taking the reference pixel point as the center, the rectangular area of is used as the preset first neighborhood. It should be noted that the preset first neighborhood can be set by itself and is not limited here.

[0025] The blur index is obtained according to the blur index calculation formula, and the blur index calculation formula is as follows: In the formula, represents the blur index of the reference pixel point; represents the information entropy of the gray-scale values of the pixel points within the preset first neighborhood; represents the number of pixel points within the preset first neighborhood; represents the th pixel point within the preset first neighborhood; represents the th pixel point within the preset first neighborhood; represents the absolute value function.

[0026] In the blur index calculation formula, the larger the information entropy of the gray-scale values of the pixel points within the preset first neighborhood, the richer the gray-scale distribution of the pixel points within the preset first neighborhood, and at this time, the gray-scale value of the reference pixel point is less prominent, and the blur index of the reference pixel point is higher; the greater the gray-scale difference between adjacent pixel points within the preset first neighborhood, and at this time is larger, and at this time, the gray-scale change speed between adjacent pixel points is faster, the gray-scale value of the reference pixel point is less prominent, and the blur index of the reference pixel point is higher.

[0027] Preferably, in one embodiment of the present invention, the method for obtaining the gray-scale significant index includes: starting from the upper left corner, traversing the pixel points within the preset second neighborhood in a row-by-row alternating direction to obtain the serial numbers of each pixel point within the preset second neighborhood. In one embodiment of the present invention, taking the reference pixel point as the center, The rectangular area is used as the preset second neighborhood. It should be noted that the preset second neighborhood can be set by oneself and is not limited here.

[0028] The gray-scale saliency index is obtained according to the gray-scale saliency index calculation formula, and the gray-scale saliency index calculation formula is as follows: In the formula, represents the gray-scale saliency index of the reference pixel point; represents the number of pixel points in the preset second neighborhood; represents the th pixel point in the preset second neighborhood; represents the maximum gray-scale value of the pixel points in the preset second neighborhood; represents the minimum gray-scale value of the pixel points in the preset second neighborhood; represents the maximum value function.

[0029] In the gray-scale saliency index calculation formula, by judging the degree to which each pixel point in the preset second neighborhood approaches the gray-scale extreme value, the larger, the more prominent the gray-scale value of the th pixel point in the preset second neighborhood. Taking the average of them, is obtained, which represents whether the gray-scale of the pixel points in the preset second neighborhood is prominent. Among them, the larger, the more prominent the gray-scale of the pixel points in the preset second neighborhood, and at this time, the gray-scale saliency index of the reference pixel point is larger.

[0030] Each pixel point in the brain MRI image is screened and merged through the fuzzy index and the gray-scale saliency index to obtain all suspected diseased connected regions of the brain MRI image. Preferably, in an embodiment of the present invention, the method for obtaining the suspected diseased connected region includes: taking the pixel points with both the fuzzy index and the gray-scale saliency index greater than the preset first threshold as the suspected diseased pixel points; taking the region composed of all the suspected diseased pixel points as the suspected diseased connected region. In an embodiment of the present invention, the preset first threshold is set to 0.8. It should be noted that the preset first threshold can be set by oneself and is not limited here.

[0031] After obtaining the suspected diseased connected regions in the patient's brain MRI image, it is necessary to determine the lesion manifestations of the suspected diseased connected regions. Since there are many normal brain tissues in the brain that are similar to the manifestations of minute lesions, it is also necessary to calculate the corresponding lesion manifestation factors based on the basic characteristic manifestations of the suspected connected regions. It is known that in MRI images, some normal brain tissues such as the basal ganglia, gray matter, and white matter are also prone to showing concentrated gray-scale differences. Therefore, the suspected connected regions may also contain non-lesioned normal brain tissue regions. However, due to its physiological structure, the human brain usually shows strong symmetry in MRI images, that is, most normal brain tissues are symmetrically distributed in the left and right brains, and there may be multiple similar brain tissues, while minute lesions usually appear randomly alone and do not have this characteristic. Therefore, by calculating the lesion manifestation factors of the suspected diseased connected regions using their basic characteristics, since minute lesion regions usually appear randomly alone, the worse the symmetry of the suspected diseased connected regions and the fewer the connected regions similar to them, the greater the lesion manifestation factors. Therefore, in the embodiments of the present invention, first, the symmetric connected region of the reference connected region is obtained, and then, based on the gray-scale difference, connected region difference characteristics, and distance difference between the reference connected region and the symmetric connected region, the lesion manifestation factor of the reference connected region is obtained.

[0032] Preferably, in one embodiment of the present invention, the method for obtaining the symmetric connected region includes: obtaining the vertical midline of the brain MRI image, using the vertical midline as the axis of symmetry to obtain the symmetric point of the centroid pixel point of the reference connected region; taking the suspected diseased connected region closest to the symmetric point as the symmetric connected region.

[0033] Preferably, in one embodiment of the present invention, the method for obtaining the lesion manifestation factor includes: obtaining the lesion manifestation factor according to the lesion manifestation factor calculation formula, and the lesion manifestation factor calculation formula is as follows: In the formula, represents the serial number of the reference connected region, represents the serial number of the symmetric connected region; represents the connected region difference characteristics between the reference connected region and the symmetric connected region; represents the Hausdorff distance between the reference connected region and the symmetric connected region; represents the average gray scale of the reference connected region; represents the average gray scale of the symmetric connected region; represents the lesion manifestation factor of the reference connected region; represents the Euclidean distance between the centroid pixel point of the reference connected region and the centroid pixel point of the symmetric connected region; represents the number of all other suspected diseased connected regions; represents the connected region difference characteristics between the reference connected region and the th other suspected diseased connected region.

[0034] In the calculation formula of the lesion manifestation factor, the contour similarity degree between the reference connected domain and the symmetric connected domain is compared through the connected domain difference feature The larger the Hausdorff distance between the reference connected domain and the symmetric connected domain the greater the contour difference between the reference connected domain and the symmetric connected domain, that is, the less similar the contours are, and the greater the connected domain difference feature between the reference connected domain and the symmetric connected domain; the greater the difference between the average gray values of the reference connected domain and the symmetric connected domain, the greater the gray difference between the reference connected domain and the symmetric connected domain. At this time, the signal difference between the reference connected domain and the symmetric connected domain is greater, and the connected domain difference feature between the reference connected domain and the symmetric connected domain is greater; the larger the Euclidean distance between the centroid pixel points of the reference connected domain and the symmetric connected domain the greater the asymmetry between the reference connected domain and the symmetric connected domain, and the greater the connected domain difference feature between the reference connected domain and the symmetric connected domain, the more asymmetric the connected domain difference feature between the reference connected domain and the symmetric connected domain is. At this time, the lesion manifestation factor of the reference connected domain is greater; the greater the average value of the connected domain difference features between the reference connected domain and other suspected diseased connected domains the more special the contour and gray level performance of the reference connected domain are. At this time, the lesion manifestation factor of the reference connected domain is greater.

[0035] Step S3: Obtain the structural significance factor of each pixel point according to the gradient feature of each pixel point in the brain MRI image; screen all pixel points according to the structural significance factor to obtain the special pixel points of the centroid pixel points in the reference connected domain; use the distance between the centroid pixel points and the special pixel points in the reference connected domain, and the structural significance factor of the special pixel points to obtain the descriptor of the reference connected domain; obtain the brain MRI images of other historical patients as other brain MRI images; obtain the possible degree of lesions of the reference connected domain according to the descriptor difference and the lesion manifestation factor difference between the reference connected domain and the suspected diseased connected domain at the closest position in the other brain MRI images; perform intelligent detection of minute lesions in the central nervous system imaging according to the possible degree of lesions.

[0036] Due to the limited quality of brain MRI imaging and the mostly extremely vague manifestations of lesions in the early stage, it is easy to miss detections through the method of single-image analysis alone. However, many brain lesions have certain location distribution characteristics, such as being located in specific brain regions such as periventricular, prefrontal, or basal ganglia, and multiple lesions. It is of relatively important significance to mine the location information of these tiny lesions through the MRI image dataset of historical patients. It is known that there are certain individual differences in the morphology and size of different patients' brains, but the relative positions conform to the physiological characteristics of the human brain. Therefore, the positions of important anatomical structures in the MRI image can be analyzed and marked, and these positions can be used as references to locate the suspected diseased connected regions. Finally, the final tiny lesion area is determined by combining the located lesion manifestation factors. Therefore, in the embodiments of the present invention, first, according to the gradient characteristics of each pixel point in the brain MRI image, the structural significance factor of each pixel point is obtained to represent the importance of the position where the pixel point is located.

[0037] Preferably, in one embodiment of the present invention, the method for obtaining the structural significance factor includes: calculating the gradient value and gradient direction angle of each pixel point; calculating the variance of the gradient direction angles of all pixel points within the preset third neighborhood of each pixel point as the first variance. The smaller the first variance, the stronger the gradient direction consistency in the local area of each pixel point, and at this time, the structure of the pixel point is more prominent; calculating the ratio between the gradient value of each pixel point and the first variance as the structural significance factor of the pixel point. The larger the gradient value of the pixel point, the more prominent the pixel point, and at this time, the structural significance factor of the pixel point is larger. In one embodiment of the present invention, a rectangular area centered on the reference pixel point is used as the preset third neighborhood. It should be noted that the preset third neighborhood can be set by itself and is not limited here.

[0038] All pixel points are screened using the structural significance factor to obtain special pixel points of the centroid pixel point within the reference connected region. Preferably, in one embodiment of the present invention, the method for obtaining special pixel points of the centroid pixel point within the reference connected region includes: drawing four rays in the horizontal and vertical directions respectively starting from the centroid pixel point within the reference connected region; taking the pixel points that first reach the preset third threshold among the four rays as special pixel points of the centroid pixel point within the reference connected region, that is, 4 special pixel points are obtained in the horizontal and vertical directions of the reference connected region. In one embodiment of the present invention, first, the maximum structural significance factor among all pixel points is obtained, denoted as , and at this time, the preset third threshold is set to . It should be noted that the preset third threshold can be set by itself and is not limited here.

[0039] The relative position of the reference connected region is described using the above 4 special pixel points, that is, the descriptor of the reference connected region is analyzed and obtained.

[0040] Preferably, in an embodiment of the present invention, the method for obtaining the descriptor of the reference connected component includes: obtaining the descriptor according to the descriptor calculation formula, and the method for obtaining the descriptor calculation formula is as follows: In the formula, represents the descriptor of the reference connected component; represents the structural significance factor of the special pixel point of the first ray; represents the structural significance factor of the special pixel point of the second ray; represents the structural significance factor of the special pixel point of the third ray; represents the structural significance factor of the special pixel point of the fourth ray; represents the distance between the special pixel point of the first ray and the centroid pixel point within the reference connected component; represents the distance between the special pixel point of the second ray and the centroid pixel point within the reference connected component; represents the distance between the special pixel point of the third ray and the centroid pixel point within the reference connected component; represents the distance between the special pixel point of the fourth ray and the centroid pixel point within the reference connected component.

[0041] Since the above operations mention that it is necessary to compare with the brain MRI images of other historical patients, in the embodiment of the present invention, the brain MRI images of other historical patients are obtained through channels such as medical research institutions or public databases.

[0042] Analyze the difference between the relative position of the reference connected component and the nearest suspected diseased connected component in other images, and obtain the possible degree of lesion of the reference connected component.

[0043] Preferably, in an embodiment of the present invention, the method for obtaining the possible degree of lesion includes: obtaining the possible degree of lesion according to the possible degree of lesion calculation formula, and the possible degree of lesion calculation formula is as follows: In the formula, represents the possible degree of lesion of the reference connected component; represents the number of other brain MRI images; represents the descriptor of the reference connected component; represents the descriptor of the connected component at the same position as the reference connected component in the th other brain MRI image; represents the lesion manifestation factor of the reference connected component; represents the lesion manifestation factor of the connected component at the same position as the reference connected component in the th other brain MRI image; represents the cosine similarity function;

[0044] In the formula for calculating the likelihood of a lesion, the brain MRI images of each other historical patient are analyzed. If the cosine similarity of the descriptors between the reference connected region and the connected region at the same position in the th other brain MRI image is greater, it indicates that the relative position is closer. At this time, if the difference in the lesion manifestation factors between the reference connected region and the connected region at the same position in the th other brain MRI image is smaller, it indicates that the lesion manifestations of the patients to which the reference connected region and the th other brain MRI image belong are more similar. Analyze the similar lesion manifestations at the same positions of all other brain MRI images. The greater it is, the higher the likelihood of a lesion in the reference connected region, and the more likely it is to be a micro-lesion area.

[0045] Calculate the likelihood of a lesion for all suspected diseased connected regions of the patient to be tested. Set the threshold to 0.75, and regard the suspected diseased connected regions with a lesion likelihood greater than 0.75 as micro-lesion areas. It should be noted that this threshold can be set by oneself and is not limited here.

[0046] Perform micro-lesion detection on the brain MRI image of the patient to be tested through the above steps to determine whether there is a micro-lesion area. If a micro-lesion is detected, further analysis should be carried out on the lesion area to detect the problems existing in the patient's brain as early as possible, formulate corresponding treatment plans at the early stage of the disease, avoid more serious impacts in the future, and finally complete the intelligent detection of micro-lesions in central nervous system imaging.

[0047] Thus, the intelligent detection of micro-lesions in brain MRI images is completed.

[0048] In summary, obtain the brain MRI image of the patient; select an arbitrary pixel point in the brain MRI image as the reference pixel point; obtain the fuzziness index of the reference pixel point according to the gray - level difference between adjacent pixel points within the preset first neighborhood of the reference pixel point; obtain the gray - level significance index of the reference pixel point according to the gray - level significant features of each pixel point within the preset second neighborhood of the reference pixel point; screen and merge each pixel point in the brain MRI image according to the fuzziness index and the gray - level significance index to obtain all suspected diseased connected regions of the brain MRI image; select an arbitrary suspected diseased connected region as the reference connected region; obtain the symmetric connected region of the reference connected region according to the symmetry of the brain MRI image; obtain the lesion manifestation factor of the reference connected region according to the gray - level difference, the connected - region difference feature, and the distance difference between the reference connected region and the symmetric connected region; obtain the structural significance factor of each pixel point according to the gradient feature of each pixel point in the brain MRI image; screen all pixel points according to the structural significance factor to obtain the special pixel points of the centroid pixel point within the reference connected region; use the distance between the centroid pixel point and the special pixel point within the reference connected region, and the structural significance factor of the special pixel point to obtain the descriptor of the reference connected region; obtain the brain MRI images of other historical patients as other brain MRI images; obtain the possible degree of lesion of the reference connected region according to the descriptor difference and the lesion manifestation factor difference between the reference connected region and the suspected diseased connected region at the closest position in the other brain MRI images; perform intelligent detection of micro - lesions according to the possible degree of lesion.

[0049] It should be noted that the above - mentioned order of the embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or the sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0050] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. An intelligent detection method for microlesions in central nervous system images, characterized in that: The method comprises: obtaining a brain MRI image of a patient to be tested; selecting any pixel point in the brain MRI image as a reference pixel point; obtaining a fuzzy index of the reference pixel point according to the grayscale difference between adjacent pixels in a preset first neighborhood of the reference pixel point; obtaining a grayscale significant index of the reference pixel point according to the grayscale significant features of each pixel point in a preset second neighborhood of the reference pixel point; screening and merging each pixel point in the brain MRI image according to the fuzzy index and the grayscale significant index to obtain all suspected diseased connected domains of the brain MRI image; selecting any suspected diseased connected domain as a reference connected domain; obtaining a symmetric connected domain of the reference connected domain according to the symmetry of the brain MRI image; and obtaining a symmetric connected domain of the reference connected domain according to the grayscale difference and connected domain between the reference connected domain and the symmetric connected domain. The difference feature and the distance difference are used to obtain the lesion expression factor of the reference connected domain; the structural significance factor of each pixel point is obtained according to the gradient feature of each pixel point in the brain MRI image; all pixels are screened according to the structural significance factor to obtain the special pixel point of the centroid pixel point in the reference connected domain; the descriptor of the reference connected domain is obtained by using the distance between the centroid pixel point and the special pixel point in the reference connected domain, and the structural significance factor of the special pixel point; the brain MRI images of other historical patients are obtained as other brain MRI images; the lesion possibility degree of the reference connected domain is obtained according to the descriptor difference between the suspected diseased connected domain in the closest position in the reference connected domain and other brain MRI images, and the difference in lesion expression factors; and the intelligent detection of small lesions is performed according to the lesion possibility degree.

2. The intelligent detection method for microlesions of central nervous system images according to claim 1, characterized in that: The method for obtaining the fuzzy index includes: starting from the upper left corner, traversing the pixel points in the preset first neighborhood in an alternating direction row by row, and obtaining the sequence number of each pixel point in the preset first neighborhood; obtaining the fuzzy index according to the fuzzy index calculation formula, and the fuzzy index calculation formula is as follows: In the formula, Indicates the blur index of the reference pixel; Represents the information entropy of the grayscale value of the pixel in the preset first neighborhood; Indicates the number of pixels in the preset first neighborhood; Indicates the first neighborhood in the preset The gray value of each pixel; Indicates the first neighborhood in the preset The gray value of each pixel; represents the absolute value function.

3. The intelligent detection method for microlesions of central nervous system images according to claim 1, characterized in that: The grayscale significant index acquisition method includes: starting from the upper left corner, traversing the pixel points in the preset second neighborhood in an alternating direction row by row, and obtaining the sequence number of each pixel point in the preset second neighborhood; obtaining the grayscale significant index according to the grayscale significant index calculation formula, and the grayscale significant index calculation formula is as follows: In the formula, Represents the grayscale significance index of the reference pixel; Indicates the number of pixels in the preset second neighborhood; Indicates the preset second neighborhood The gray value of each pixel; Indicates the maximum grayscale value of the pixels in the preset second neighborhood; Indicates the minimum grayscale value of the pixels in the preset second neighborhood; Represents the maximum value function.

4. The intelligent detection method for central nervous system imaging microlesions according to claim 1, characterized in that: The method for obtaining the suspected disease connected domain includes: taking the pixel points whose fuzzy index and the grayscale significance index are both greater than a preset first threshold as suspected diseased pixel points; and taking the area composed of all suspected diseased pixel points as the suspected diseased connected domain.

5. The intelligent detection method for central nervous system imaging microlesions according to claim 1, characterized in that: The method for obtaining the symmetrical connected domain includes: obtaining the vertical midline of the brain MRI image, taking the vertical midline as the symmetry axis, obtaining the symmetrical points of the centroid pixel point of the reference connected domain; and taking the suspected diseased connected domain closest to the symmetrical point as the symmetrical connected domain.

6. The intelligent detection method for microlesions in central nervous system images according to claim 1, characterized in that: The method for obtaining the lesion manifestation factor includes: obtaining the lesion manifestation factor according to a lesion manifestation factor calculation formula, and the lesion manifestation factor calculation formula is as follows: In the formula, represents the serial number of the reference connected domain, Indicates the serial number of the symmetric connected domain; Represents the difference characteristics of the connected domain between the reference connected domain and the symmetric connected domain; represents the Hausdorff distance between the reference connected domain and the symmetric connected domain; Represents the grayscale mean of the reference connected domain; Represents the grayscale mean of the symmetrical connected domain; represents the lesion manifestation factor of the reference connectivity domain; Represents the Euclidean distance between the centroid pixel of the reference connected domain and the centroid pixel of the symmetric connected domain; represents the number of all other suspected diseased connected domains; Represents the reference connected domain and the The connected domain difference characteristics between other suspected disease connected domains.

7. The intelligent detection method for central nervous system imaging microlesions according to claim 1, characterized in that: The method for obtaining the structural significance factor includes: calculating the gradient value and gradient direction angle of each pixel point; calculating the variance of the gradient direction angles of all pixel points as a first variance; and calculating the ratio between the gradient value of each pixel point and the first variance as the structural significance factor of the pixel point.

8. The intelligent detection method for microlesions in central nervous system images according to claim 1, characterized in that: The method for obtaining a special pixel point of the centroid pixel point in a reference connected domain includes: taking the centroid pixel point in the reference connected domain as the starting point, making four rays in four horizontal and vertical directions respectively; and taking the pixel point that first reaches a preset third threshold in the four rays as the special pixel point of the centroid pixel point in the reference connected domain.

9. The intelligent detection method for central nervous system image microlesions according to claim 1, characterized in that: The method for obtaining the descriptor of the reference connected domain includes: obtaining the descriptor according to a descriptor calculation formula, and the method for obtaining the descriptor calculation formula is as follows: In the formula, Descriptor representing the reference connected domain; The structural significance factor of the special pixel point representing the first ray; The structural significance factor of the special pixel point representing the second ray; The structural significance factor of the special pixel point representing the third ray; The structural significance factor of the special pixel point representing the fourth ray; Represents the distance between the special pixel point of the first ray and the centroid pixel point in the reference connected domain; Represents the distance between the special pixel point of the second ray and the centroid pixel point in the reference connected domain; Represents the distance between the special pixel point of the third ray and the centroid pixel point in the reference connected domain; Represents the distance between the special pixel point of the fourth ray and the centroid pixel point in the reference connected domain.

10. The intelligent detection method for central nervous system image microlesions according to claim 1, characterized in that: The method for obtaining the possible degree of lesions includes: obtaining the possible degree of lesions according to a calculation formula for the possible degree of lesions, and the calculation formula for the possible degree of lesions is as follows: In the formula, Indicates the possible degree of lesion in the reference connected domain; represents the number of other brain MRI images; Descriptor representing the reference connected domain; Indicates Descriptors of connected domains in the same positions as the reference connected domains in other brain MRI images; represents the lesion manifestation factor of the reference connectivity domain; Indicates The lesion expression factor of the connected domain in the same position as the reference connected domain in other brain MRI images; represents the cosine similarity function; Represents the normalization function.

Images