Cerebral small vessel lesion recognition and detection system based on MRI image analysis

Abstract

The application relates to the technical field of image processing, in particular to a cerebral small vessel lesion recognition and detection system based on MRI image analysis, which comprises a processor and a memory, and the processor executes a computer program stored in the memory to realize the following steps: obtaining a normal possibility representation value of a pixel point; obtaining a suspected slice image and a suspected lesion area on the suspected slice image according to the normal possibility representation value; obtaining a same-layer area corresponding to the suspected lesion area on the suspected slice image according to all slice images with the same layer as the suspected slice image in other brain MRI images except the brain MRI image where the suspected slice image is located, and obtaining a real lesion area according to the same-layer area corresponding to the suspected lesion area. The application can improve the accuracy and integrity of the recognition and detection of the real lesion area or the real cerebral small vessel lesion area.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to a system for identifying and detecting small blood vessel lesions in the brain based on MRI image analysis. Background Technology

[0002] Due to the advantages of high resolution and radiation-free imaging, MRI plays a crucial role in the early identification and localization of cerebral small vessel lesions, as well as in assessing disease progression. Currently, MRI images of patients are generally used to identify and locate cerebral small vessel lesions. However, when identifying cerebral small vessel lesions based on patient MRI images, existing image processing methods are typically used to segment the lesion area, such as edge detection and grayscale thresholding. However, issues such as small lesion size, unclear transition between the lesion and surrounding tissue, and noise can prevent accurate or complete segmentation of the lesion area using existing image processing methods alone. This can affect subsequent analysis of the morphology and development of cerebral small vessel lesions. Therefore, improving the accuracy and completeness of lesion area identification is an urgent problem to be solved. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides a system for identifying and detecting small cerebral vessel lesions based on MRI image analysis. The specific technical solution adopted is as follows:

[0004] One embodiment of the present invention provides a system for identifying and detecting small cerebral vessel lesions based on MRI image analysis, comprising a processor and a memory, wherein the processor executes a computer program stored in the memory to perform the following steps:

[0005] Acquire brain MRI images of the patient in different sequences and the local windows corresponding to each pixel point in each slice image of the brain MRI images;

[0006] For any slice image a in a brain MRI image of any sequence, all other slice images in the brain MRI image of the sequence other than slice image a are recorded as the neighborhood images of slice image a. Based on the pixel gradient value in the local window corresponding to each pixel point on slice image a and the local window corresponding to the pixel point in the neighborhood image that has the same target coordinate value as each pixel point on slice image a, the normal probability characterization value of each pixel point on slice image a is obtained.

[0007] Based on the normal probability characterization value, a suspected slice image and a suspected lesion region on the suspected slice image are obtained; based on all slice images with the same layer number as the suspected slice image in other brain MRI images other than the brain MRI image where the suspected slice image is located, the same layer region corresponding to the suspected lesion region on the suspected slice image is obtained, and based on the same layer region corresponding to the suspected lesion region, the actual lesion region is obtained.

[0008] Beneficial Effects: This invention first acquires brain MRI images of a patient under different sequences and the local windows corresponding to each pixel in each slice image of the brain MRI images; then, for any slice image 'a' in any sequence of brain MRI images, all other slice images in the sequence of brain MRI images except slice image 'a' are recorded as neighborhood images of slice image 'a'; based on the pixel gradient values ​​in the local windows corresponding to each pixel in slice image 'a' and the local windows corresponding to pixels in the neighborhood images that have the same target coordinate values ​​as each pixel in slice image 'a', the normal probability characterization value of each pixel in slice image 'a' is obtained; then, based on the normal probability characterization value, a suspected slice image and a suspected lesion region on the suspected slice image are obtained; finally, based on all slice images with the same layer number as the suspected slice image in other brain MRI images besides the brain MRI image containing the suspected slice image, the same layer region corresponding to the suspected lesion region on the suspected slice image is obtained, and based on the same layer region corresponding to the suspected lesion region, the actual lesion region is obtained. Furthermore, the present invention, based on the normal probability representation value of pixels, obtains the suspected lesion area and the corresponding same layer area, which can improve the accuracy and completeness of the identification and detection of real lesion areas or real cerebral small blood vessel lesion areas. Attached Figure Description

[0009] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of a method for identifying and detecting small cerebral vessel lesions based on MRI image analysis according to the present invention. Detailed Implementation

[0011] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the protection scope of the embodiments of the present invention.

[0012] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.

[0013] This embodiment provides a system for identifying and detecting small cerebral vessel lesions based on MRI image analysis, including a processor and a memory. The processor executes a computer program stored in the memory to implement an artificial intelligence-based method for monitoring the processing of intestinal-friendly pet food. Figure 1 As shown, this method for identifying and detecting small cerebral vessel lesions based on MRI image analysis includes the following steps:

[0014] Step S001: Obtain the patient's brain MRI images in different sequences and the local windows corresponding to each pixel point on each slice image in the brain MRI images.

[0015] The main purpose of this embodiment is to improve the accuracy and completeness of identifying cerebral small vessel lesion areas, and to provide reference or data support for relevant personnel to analyze and judge the morphology, development and other conditions of cerebral small vessel lesions in patients. In addition, for ease of understanding and subsequent analysis, this embodiment will describe the process of identifying cerebral small vessel lesion areas of any patient as an example, that is, the patients that appear in the following examples are the same patients.

[0016] Because different small vessel lesions in the brain exhibit varying degrees of appearance in MRI images across different sequences, to ensure the accuracy and completeness of small vessel lesion identification, it is generally necessary to acquire brain MRI images of the patient in different sequences. This embodiment first acquires brain MRI images of the patient in different sequences. In specific applications, relevant personnel will set the number and type of sequences according to the actual situation. Sequence types include, but are not limited to, T1, T2, FLAIR, DWI, and SWI. For example, if the sequence type selected when acquiring MRI images of the patient is T1, T2, or FLAIR, then... The system obtains brain MRI images of the patient under sequence T1, sequence T2, and sequence FLAIR. In practical applications, the slice thickness is set according to the sequence accuracy and other practical considerations. However, this embodiment requires that the slice thickness be consistent for brain MRI images acquired under different sequences. For example, if the slice thickness is set to 3mm, then each sequence's brain MRI image will consist of 50 to 60 slice images, with one slice corresponding to one image. If a slice image is the 19th slice, then the slice number is 19. Furthermore, before acquiring the patient's brain MRI images, the patient must lie on the MRI examination table with their head fixed. The examination table is then moved to the center of the machine, and the head coil is used for signal reception. Since small vessel lesions in the brain are usually clearer in axial images, this embodiment selects to capture axial images of the patient's brain.

[0017] Therefore, this embodiment can obtain the patient's brain MRI images in different sequences and the slice images in the brain MRI images through the above process. After obtaining the slice images, a local window centered on each pixel is obtained. Obtaining the local window corresponding to each pixel is for subsequent analysis of the possibility that the corresponding pixel is a normal tissue area. The specific process of obtaining the local window corresponding to each pixel in each slice image of the brain MRI image is as follows: For any slice image, a window with a length and width of A×A is constructed with the pixel on the slice image as the center, and it is recorded as the local window corresponding to the corresponding pixel. The center pixel in the local window corresponding to any pixel is the pixel. In specific applications, the implementer can set the value of A according to the actual situation such as lesion size, image resolution or experience. For example, in this embodiment, A can be set to 7, then the size of the local window corresponding to the obtained pixel is 7×7.

[0018] Normal brain tissue has a strong structural character and exhibits a gradual, continuous change across multiple slices within a sequence. Small vessel lesions, however, are typically small and highly random, appearing only in a few slices and lacking a clear pattern with other tissues. Therefore, to improve the accuracy and completeness of small vessel lesion identification, this embodiment will subsequently analyze the probability that each pixel in the slice image belongs to a normal tissue region by combining pixels from multiple slices within the same sequence that belong to the same brain location. During MRI image acquisition, the patient's head may move slightly. Movement can cause the position of the same brain region to change on different slice images, thus affecting the accuracy of subsequent analysis of the possibility that a pixel is a normal tissue region. In order to avoid this situation, this embodiment will construct a midline coordinate system for each slice image based on the symmetry between the left and right hemispheres of the brain. The coordinate values ​​of each pixel on the slice image in the midline coordinate system are called target coordinate values. That is, the target coordinate values ​​of a pixel on any slice image are the coordinate values ​​of the corresponding slice image in the midline coordinate system of the corresponding slice image. The specific process of obtaining the midline coordinate system of each slice image in this embodiment is as follows:

[0019] For any slice image a0: First, construct a coordinate system with the horizontal direction of pixels in slice image a0 as the x-axis, the vertical direction of pixels as the y-axis, and the top-left corner of slice image a0 as the origin, and denote it as the original coordinate system of slice image a0. The coordinates of each pixel in slice image a0 under the original coordinate system of slice image a0 are then used as the original coordinates of the corresponding pixel. Next, use the Canny edge detection algorithm to extract edges from slice image a0, obtaining all edges on slice image a0. Identify the closed edges among all edges on slice image a0 and calculate the area of ​​each closed edge. Select the closed edge with the largest area on slice image a0 as the edge to be analyzed, and the edge to be analyzed is the outermost edge of the head of slice image a0. The area of ​​a closed edge refers to the total number of pixels occupied by the closed region enclosed by the closed edge curve in the image plane. Then, perform pairwise permutations of the edge pixels on the edge to be analyzed without repetition to obtain all edge pixel combinations, and calculate the area of ​​each edge pixel. The Euclidean distance between two pixels in an edge pixel combination is calculated using the original coordinates of the pixels. The magnitude of the Euclidean distance between any two pixels remains unchanged and is recorded as the feature distance of the corresponding edge pixel combination. All edge pixel combinations are sorted in descending order of feature distance, and the sorting result is recorded as the first sequence. Then, a sequence of a predetermined number of edge pixel combinations at the beginning of the first sequence is recorded as the second sequence. The optimization degree of each edge pixel combination in the second sequence is calculated, and the edge pixel combination with the highest optimization degree is recorded as the target combination. The two edge pixels in the target combination are connected, and the resulting line segment is recorded as the target line segment. A coordinate system is established with the vertical axis coinciding with the target line segment, the horizontal axis perpendicular to the midpoint of the target line segment, and the origin at the midpoint of the target line segment. This coordinate system is used as the midline coordinate system of the slice image a0, that is, the target line segment is the midline between the left and right hemispheres of the slice image.

[0020] In the midline coordinate system of slice image a0, the positive direction of the vertical axis is upward, and the positive direction of the horizontal axis is to the right. Since the head appears elliptical in axial MRI images, the midline between the left and right hemispheres should be close to the longitudinal direction of the head. Therefore, the longitudinal direction of the head region in the slice image is longer than the lateral direction. The edge to be analyzed is the outermost edge of the head in the image. Therefore, the line connecting two edge pixels in the target combination is not only the midline between the left and right hemispheres, but also their positions in slice image a0 are one above and one below. That is, in this embodiment, the positive direction of the vertical axis of the midline coordinate system of slice image a0 is the direction from the lower pixel to the upper pixel in the target combination. Furthermore, since the patient's head is not necessarily perfectly symmetrical, the line connecting the farthest edge pixels is not necessarily the midline between the left and right hemispheres. However, the pixel coordinates of similar structures in the left and right hemispheres are symmetrically distributed. Therefore, in this embodiment, the coordinates of pixels based on similar structures in the left and right hemispheres are... The symmetrical distribution of the markers is used to calculate the preference of multiple edge pixel combinations. The higher the preference, the greater the probability that the straight line connecting the two edge pixels in the corresponding edge pixel combination is the midline between the left and right hemispheres on slice image a0. The preference of all edge pixel combinations is not calculated because although the midline between the left and right hemispheres does not completely coincide with the line connecting the edge pixels with the largest distance, their overall directions are relatively close. Therefore, in order to reduce the amount of calculation, this embodiment selects only the combinations ranked first in the first sequence for screening. In addition, since the normal brain structure in axial MRI images usually has good symmetry, this embodiment selects the midline position between the left and right hemispheres based on the symmetry of the left and right hemispheres in the slice image to establish the midline coordinate system of each slice image. That is, the vertical axis of the midline coordinate system of any slice image is symmetrical about the left and right hemispheres. Therefore, pixels with the same coordinate value on the slice image in the midline coordinate system of the slice image belong to the same brain location.

[0021] In practical applications, implementers need to set a preset quantity value based on experimental statistics and actual conditions. For example, in this embodiment, the preset quantity value can be the rounded-up value of 10% of the total number of combinations in the first sequence.

[0022] In this embodiment, the specific process for obtaining the optimality of each edge pixel combination is as follows: For any edge pixel combination: First, obtain the straight line connecting the two edge pixels in the edge pixel combination and record it as the feature line. Then, on the slice image a0, obtain all pixel pairs symmetrical about the feature line and record them as symmetrical pixel pairs. Next, calculate the reciprocal of the sum of the absolute value of the grayscale difference between the two pixels in each symmetrical pixel pair and a preset constant, and use this as the symmetry representation value of the corresponding symmetrical pixel pair. That is, the expression for the symmetry representation value of any symmetrical pixel pair is: Where g1 is the grayscale value of the first pixel in the symmetrical pixel pair, and g2 is the grayscale value of the second pixel in the symmetrical pixel pair. The first pixel and the second pixel in the symmetrical pixel pair are not the same pixel. This is a preset constant, intended to prevent the denominator from being zero. It is generally set as... Set to a very small constant, such as if it can be set. The value is 0.01. Then, the symmetry representation values ​​of all symmetrical pixel pairs are accumulated and normalized. This result is used as the degree of preference for the edge pixel combination. Here, the normalization function Norm() is used for normalization. The larger the symmetry representation value of the symmetrical pixel pair, that is, the smaller the absolute value of the gray level difference between the two pixels in the symmetrical pixel pair, the greater the probability that the feature line is the midline between the left and right hemispheres of the slice image a0. Conversely, the smaller the symmetry representation value of the symmetrical pixel pair, that is, the larger the absolute value of the gray level difference between the two pixels in the symmetrical pixel pair, the smaller the probability that the feature line is the midline between the left and right hemispheres of the slice image a0.

[0023] Therefore, this embodiment obtains the patient's brain MRI images in different sequences, the local window corresponding to each pixel in each slice of the brain MRI image, the number of layers in each slice of the brain MRI image, the midline coordinate system of each slice of the brain MRI image, and the target coordinate value of each pixel in the midline coordinate system of the corresponding slice image through the above process.

[0024] Step S002: For any slice image a in any sequence of brain MRI images, all other slice images in the sequence of brain MRI images except slice image a are recorded as neighborhood images of slice image a. Based on the pixel gradient value in the local window corresponding to each pixel point on slice image a and the local window corresponding to the pixel point in the neighborhood image that has the same target coordinate value as each pixel point on slice image a, the normal probability characterization value of each pixel point on slice image a is obtained.

[0025] Because normal brain tissue has a strong structural character and exhibits a gradually changing, continuous feature in multi-slice images within a sequence, while small cerebral vessel lesions are typically small and highly random, they only appear in a few slice images and show no obvious regularity with other tissues. Therefore, this embodiment will combine the described characteristics of small cerebral vessel lesions with the pixel gradient values ​​in the windows corresponding to pixels in various slice images of the same sequence and the pixel gradient values ​​in the windows corresponding to pixels in other slice images with the same target coordinate values ​​to calculate the normal probability representation value of each pixel in the slice image. The normal probability representation value of the pixel is an important parameter for subsequently obtaining the actual lesion area. In addition, for ease of understanding, this embodiment will describe the specific process of obtaining the normal probability representation value of the i-th pixel in any slice image a of any sequence of brain MRI images as an example, that is, the specific process of obtaining the normal probability representation value of the i-th pixel in any slice image a of the brain MRI images of this sequence is as follows:

[0026] First, based on the pixel gradient values ​​in the local windows corresponding to each pixel in each slice of the brain MRI images in this sequence, the tissue edge probability representation value of the local window corresponding to each pixel in each slice image is obtained. Then, in the brain MRI images in this sequence, other slice images besides slice image a are obtained and are all used as neighborhood images of slice image a. Next, in each neighborhood image of slice image a, pixels with the same target coordinate value as the i-th pixel in slice image a are obtained and recorded as neighborhood co-position pixels of the i-th pixel in slice image a. That is, the target coordinate value of the neighborhood co-position pixels of the i-th pixel in slice image a is the same as the target coordinate value of the i-th pixel. Furthermore, the neighboring pixels of the i-th pixel in slice image a belong to different slice images but are brain MRI images under the same sequence as the i-th pixel in slice image a. Then, based on the tissue edge probability representation value of the local window corresponding to the i-th pixel in slice image a and the local window corresponding to the neighboring pixels of the i-th pixel, the similarity representation value between the i-th pixel and the neighboring pixels of the i-th pixel is obtained. Then, based on the similarity representation value between the i-th pixel and the neighboring pixels of the i-th pixel and the layer difference between slice image a and the images to which the neighboring pixels belong, the normal probability representation value of the i-th pixel in slice image a is obtained.

[0027] In this embodiment, the specific process of obtaining the tissue edge probability representation value of the local window corresponding to each pixel in each slice image of the brain MRI image in this sequence is as follows: For any local window corresponding to any pixel in slice image a: obtain the gradient values ​​of all pixels in the local window corresponding to that pixel, and if there is a clear edge in the window, the gradient values ​​of the pixels in the window will differ greatly, that is, the gradients of pixels at the edge position and pixels at the non-edge position will differ greatly; then calculate the standard deviation of the gradient values ​​of all pixels in the local window corresponding to the pixel, and record it as the gradient standard deviation of the local window corresponding to that pixel, and then normalize the gradient standard deviation of the local window corresponding to that pixel. The normalized result is used as the tissue edge probability representation value of the local window corresponding to the pixel. The larger the tissue edge probability representation value of the local window corresponding to the pixel, the greater the probability that the local window corresponding to the pixel is located at the tissue edge. The smaller the tissue edge probability representation value of the local window corresponding to the pixel, the greater the probability that the local window corresponding to the pixel is located inside the tissue. In addition, the purpose of calculating the tissue edge probability representation value of the local window is to analyze the possibility that the window has a tissue edge. In order to more accurately represent whether the pixel belongs to the normal tissue area or the lesion area, this embodiment uses two methods to calculate the similarity representation value. That is, the calculation method of the similarity representation value is different for the presence of tissue edge in the window and the absence of tissue edge.

[0028] The process of normalizing the gradient standard deviation of the local window corresponding to the pixel is as follows: the gradient standard deviation of the local window corresponding to each pixel in the slice image a is obtained in the manner described above, and the maximum and minimum gradient standard deviations among the gradient standard deviations of the local windows corresponding to all pixels in the slice image a are recorded as the first gradient standard deviation and the second gradient standard deviation, respectively. The result of subtracting the second gradient standard deviation from the first gradient standard deviation is recorded as the maximum gradient standard deviation difference. The ratio of the result of subtracting the second gradient standard deviation from the gradient standard deviation of the local window corresponding to the pixel to the maximum gradient standard deviation difference is used as the result of normalizing the gradient standard deviation of the local window corresponding to the pixel.

[0029] In this embodiment, the specific process of obtaining the similarity representation value between the i-th pixel and its neighboring pixels based on the tissue edge probability representation value of the local window corresponding to the i-th pixel in the slice image a and the local window corresponding to the i-th pixel's neighboring pixels at the same position is as follows: For the i-th pixel and any neighboring pixel c of the i-th pixel:

[0030] First, determine if the tissue edge probability representation value of the i-th pixel is greater than a preset first threshold. If so, determine that the local window corresponding to the i-th pixel is at a tissue edge position. Then, measure the similarity between the i-th pixel and its neighboring pixel c by the difference or similarity between the edges in the local windows corresponding to the i-th pixel and its neighboring pixel c. That is, first use the Canny edge detection algorithm to perform edge detection on the region corresponding to the local window of the i-th pixel to obtain all edges in the local window of the i-th pixel, then use the Canny edge detection algorithm to perform edge detection on the region corresponding to the local window of its neighboring pixel c to obtain all edges in the local window of its neighboring pixel c, and finally obtain the local window of the i-th pixel. The similarity representation value between the i-th pixel and the neighboring pixel c is calculated by calculating the DTW distance between each edge in the local window corresponding to the i-th pixel and each edge in the local window corresponding to the neighboring pixel c. Furthermore, there may be cases where no edge exists in the local window corresponding to the neighboring pixel c. When no edge exists in the local window corresponding to the neighboring pixel c, it indicates that the probability of the local window corresponding to the neighboring pixel c being located at a tissue edge is extremely low, and the similarity between the i-th pixel and the neighboring pixel c is extremely low. Therefore, to reduce computational cost, 0 can be directly used as the similarity representation value between the i-th pixel and the neighboring pixel c.If the probability representation value of the tissue edge of the i-th pixel is not greater than a preset first threshold, then the local window corresponding to the i-th pixel is determined to be inside the tissue. In this case, the similarity between the i-th pixel and its neighboring pixel c is measured by the gray-level similarity. That is, firstly, based on the gray-level values ​​of the pixels in the local window corresponding to the i-th pixel, the gray-level values ​​of the pixels in the local window corresponding to the neighboring pixel c, the gray-level value of the i-th pixel, and the gray-level value of the neighboring pixel c, the gray-level similarity between the i-th pixel and its neighboring pixel c is obtained. Then, the gray-level similarity between the i-th pixel and its neighboring pixel c is calculated. The gray-level similarity between pixels is used as the similarity representation value between the i-th pixel and its neighboring pixel c at the same position. That is, when there is no obvious edge within the window, it indicates that the window is inside an organization, because the gray-level values ​​within the same organization are basically consistent. Furthermore, since the size of normal organization does not change suddenly, if a pixel in the window is in a certain structure in the image, then most of them will still be in the same structure in the next consecutive image. Therefore, the similarity of the gray-level values ​​of pixels within the window in multiple consecutive images can reflect the continuity of the structural changes of the window, and thus reflect the continuity of the structural changes of the pixels. In addition, the larger the similarity representation value, the more similar the i-th pixel is to its neighboring pixel c at the same position.

[0031] In practical applications, implementers need to set a preset first threshold based on the value range of the similarity representation value, experimental statistics, and other actual conditions. For example, if the value range of the similarity representation value in this embodiment is 0 to 1, then the preset first threshold can be set to 0.5.

[0032] In this embodiment, the specific process of obtaining the similarity representation value between the i-th pixel and the neighboring pixel c based on the DTW distance between each edge in the local window corresponding to the i-th pixel and each edge in the local window corresponding to the neighboring pixel c at the same position is as follows: First, obtain the similarity feature values ​​of all edges in the local window corresponding to the i-th pixel and the edges in the local window corresponding to the neighboring pixel c at the same position, and calculate the mean of the similarity feature values ​​of all edges in the local window corresponding to the i-th pixel and the edges in the local window corresponding to the neighboring pixel c at the same position. Then, normalize the calculated mean, and use the result of the normalization as the similarity representation value between the i-th pixel and the neighboring pixel c at the same position. That is, the calculation expression for the similarity representation value between the i-th pixel and the neighboring pixel c at the same position is as follows: Where Norm() is the normalization function, and J is the total number of edges in the local window corresponding to the i-th pixel. Let be the similarity feature value between the j-th edge in the local window corresponding to the i-th pixel and the edge in the local window corresponding to the neighboring pixel c at the same position. The larger the value, the more similar the attributes of the i-th pixel are to those of its neighboring pixel c at the same position.

[0033] The process of obtaining similarity feature values ​​is as follows: For the j-th edge in the local window corresponding to the i-th pixel: First, the set of all edges in the local window corresponding to the neighboring pixel c is denoted as the edge set, and the DTW distance between the j-th edge and each edge in the edge set is calculated; then, the minimum DTW distance between the j-th edge and each edge in the edge set is selected, and the minimum DTW distance is negatively correlated and mapped, and the mapping result is used as the matching degree between the j-th edge and the edge in the local window corresponding to the neighboring pixel c; then, the minimum distance between the j-th edge and the center pixel in the local window corresponding to the i-th pixel is calculated and denoted as the feature difference, and the reciprocal of the feature difference plus a preset constant is used as the distance weight factor of the j-th edge; the product of the distance weight factor of the j-th edge and the matching degree between the j-th edge and the edge in the local window corresponding to the neighboring pixel c is used as the similarity feature value between the j-th edge and the edge in the local window corresponding to the neighboring pixel c. The matching degree of the edge in the local window corresponding to the j-th edge and the neighboring pixel c is exp(-D), where D is the minimum DTW distance among the DTW distances between the j-th edge and all edges in the edge set, and exp() is an exponential function with a constant e as the base. Alternatively, the reciprocal of D can be used as the matching degree of the edge in the local window corresponding to the j-th edge and the neighboring pixel c. Furthermore, as another implementation, when calculating the DTW distance between the j-th edge and the z-th edge in the edge set, a predetermined number of pixels uniformly sampled from the j-th edge can be used. The DTW distance between the sequence formed and the sequence formed by uniformly sampling a preset number of pixels on the z-th edge is taken as the DTW distance between the j-th edge and the z-th edge in the edge set. In other implementations, the implementer can set the preset number of values ​​according to the actual situation, such as setting the preset number to 10. If the number of points on a certain edge is less than the preset number, then all points on the corresponding edge are selected to form the sequence. The smaller the DTW distance between the j-th edge and the z-th edge in the edge set, the more similar the shapes of the j-th edge and the z-th edge in the edge set are.

[0034] The minimum distance between the j-th edge and the center pixel in the local window corresponding to the i-th pixel is the minimum of the Euclidean distances between each pixel on the j-th edge and the center pixel in the local window corresponding to the i-th pixel. The reason for using the minimum distance between the j-th edge and the center pixel in the local window corresponding to the i-th pixel is that the center pixel in the local window corresponding to the i-th pixel is the i-th pixel itself. Edges closer to the i-th pixel are more representative of its features. Therefore, to minimize the impact of blurred lesion edges on the completeness or accuracy of subsequent lesion region identification, and to more accurately distinguish whether the i-th pixel belongs to a lesion region or normal tissue region, the matching degree between the j-th edge closer to the i-th pixel and the edges in the local window corresponding to the neighboring pixel c contributes more to the subsequent calculation of the similarity representation value between the i-th pixel and the neighboring pixel c. The specific calculation expression for the similarity feature value between the j-th edge and the edges in the local window corresponding to the neighboring pixel c is as follows: ,in, Let be the matching degree of the j-th edge with the edge in the local window corresponding to the pixel c at the same position in its neighborhood. Let be the minimum distance between the j-th edge and the center pixel of the local window corresponding to the i-th pixel. This is a preset constant, the purpose of which is to prevent the denominator from being 0. Additionally, when... The larger and The smaller the value, the greater the similarity feature value between the j-th edge and the edge in the local window corresponding to the pixel c at the same position in the neighborhood.

[0035] In this embodiment, based on the grayscale value of the pixel in the local window corresponding to the i-th pixel, the grayscale value of the pixel in the local window corresponding to the neighboring pixel c at the same position, the grayscale value of the i-th pixel, and the grayscale value of the neighboring pixel c at the same position, the specific calculation expression for the grayscale similarity between the i-th pixel and the neighboring pixel c at the same position is as follows:

[0036] in, Let H be the grayscale similarity between the i-th pixel and its neighboring pixel c at the same position, and let H be the total number of pixels in the local window corresponding to the i-th pixel. Let be the grayscale value of the h-th pixel in the local window corresponding to the i-th pixel. Let be the grayscale value of the h-th pixel in the local window corresponding to the pixel c at the same location in the neighborhood. As a preset constant, here This is also to prevent the denominator from being 0. Let be the grayscale value of the i-th pixel. Let c be the grayscale value of the neighboring pixel at the same location. Let be the Euclidean distance between the h-th pixel in the local window corresponding to the i-th pixel and the center pixel in the local window corresponding to the i-th pixel; the h-th pixel in the local window corresponding to the i-th pixel is not the center pixel in the local window corresponding to the i-th pixel, the h-th pixel in the local window corresponding to a neighboring pixel c is not the center pixel in the local window corresponding to a neighboring pixel c, and the h-th pixel in the local window corresponding to the i-th pixel has the same target coordinate value or the same position in the window as the h-th pixel in the local window corresponding to a neighboring pixel c; and when calculating grayscale similarity, consider... The reason is as follows: Since the center pixel of the local window corresponding to the i-th pixel is the i-th pixel itself, and there may be situations where the window simultaneously contains lesion pixels and normal tissue pixels, and generally, pixels closer to the i-th pixel can better reflect the true features of the i-th pixel or the features of the area it is located in, in order to minimize the impact of blurred lesion edges on the complete or accurate identification of the lesion area, and in order to more accurately distinguish whether the i-th pixel belongs to the lesion area or the normal tissue area, the grayscale difference between the pixels closer to the i-th pixel and the pixels in the local window corresponding to the neighboring pixel c at the same position contributes more to the subsequent calculation of the similarity representation value between the i-th pixel and the neighboring pixel c at the same position; in addition, when The larger and The smaller the value, the more similar the gray levels of the i-th pixel are to those of its neighboring pixel c at the same position; conversely, the larger the value, the less similar they are. Furthermore, the calculation of gray level similarity takes into account... as well as In two aspects, it can minimize the impact of lesion pixels at the junction of lesions and normal tissue or lesion pixels being close to normal tissue pixels on lesion area identification, thereby improving the accuracy and completeness of lesion area identification. Lesion pixels refer to pixels belonging to the lesion area.

[0037] In this embodiment, the specific process of obtaining the normal probability representation value of the i-th pixel on slice image a based on the similarity representation value between the i-th pixel and its neighboring pixels at the same position, and the layer difference between slice image a and the image to which the neighboring pixels at the same position belong, is as follows:

[0038] First, obtain the set of all neighboring pixels at the same position of the i-th pixel in the slice image a, and denote it as the set of neighboring pixels at the same position of the i-th pixel. Then, obtain the set of weighted similarity representation values ​​corresponding to the i-th pixel. Next, accumulate all weighted similarity representation values ​​in the set, and use the accumulated result as the normal probability representation value of the i-th pixel in the slice image a; and the f-th weighted similarity representation value in the set of weighted similarity representation values ​​corresponding to the i-th pixel is... , Let f be the similarity representation value between the i-th pixel and the f-th neighboring pixel in the set of neighboring pixels at the same position. The result is the result of performing a negative correlation mapping and then normalization on the absolute value of the difference between the layer number of the slice image to which the i-th pixel belongs and the layer number of the slice image to which the f-th neighboring pixel at the same position belongs. It can also be called the layer difference weighting factor between the slice image to which the i-th pixel belongs and the slice image to which the f-th neighboring pixel at the same position belongs, i.e. Where exp() is an exponential function with base e. Let F be the absolute value of the difference between the layer number of the slice image to which the i-th pixel belongs and the layer number of the slice image to which the f-th neighboring pixel at the same position belongs, and let F be the total number of neighboring pixels at the same position in the set of neighboring pixels at the i-th pixel. This is the accumulated result of the layer difference weighting factors between the slice image to which the i-th pixel belongs and the slice images to which each neighboring pixel at the same position belongs in the set of neighboring pixels at the same position belongs. The purpose of this accumulation is to... Normalization is performed. In this embodiment, the layer difference weighting factor is considered because the slice image with a layer number closer to slice image a is more similar to slice image a in terms of normal tissue shape and other features. Conversely, the slice image with a larger layer number difference from slice image a is less similar to slice image a in terms of normal tissue shape and other features. Therefore, slice images with a layer number closer to slice image a should have a greater weight in analyzing whether the pixels on slice image a are lesion area pixels or normal tissue area pixels. Furthermore, because normal brain tissue has a strong structural integrity, it exhibits a gradual and continuous change in multi-slice images within a sequence. In contrast, small cerebral vessel lesions are typically small and highly random, appearing only in a few slice images and lacking a clear regularity with other tissues. Therefore, the larger the sum of all weighted similarity values ​​in the weighted similarity representation set corresponding to the i-th pixel, the greater the normal probability representation value of the i-th pixel in slice image a. This indicates a stronger regularity between the i-th pixel and its neighboring pixels in the brain MRI image where slice image a is located, or a more obvious continuous change in the tissue structure of the local window corresponding to the i-th pixel. It also indicates that the i-th pixel is less likely to have the characteristics of a lesion pixel or more likely to have the characteristics of a normal tissue pixel. Conversely, the smaller the normal probability representation value of the i-th pixel in slice image a, the more likely the i-th pixel is to have the characteristics of a lesion pixel or less likely to have the characteristics of a normal tissue pixel.

[0039] Therefore, based on the process of obtaining the normal probability representation value of the i-th pixel on the slice image a, this embodiment can obtain the normal probability representation value of any pixel on any slice image.

[0040] Step S003: Based on the normal probability characterization value, obtain the suspected slice image and the suspected lesion area on the suspected slice image; based on all slice images with the same layer number as the suspected slice image in other brain MRI images other than the brain MRI image where the suspected slice image is located, obtain the same layer area corresponding to the suspected lesion area on the suspected slice image, and obtain the actual lesion area based on the same layer area corresponding to the suspected lesion area.

[0041] After obtaining the normal probability representation values ​​of each pixel on the slice image, the suspected slice image and the suspected lesion region on the suspected slice image are obtained based on the normal probability representation values ​​of each pixel on the slice image. Since the acquisition of brain MRI images may contain noise points due to other interference factors, and noise and lesions are quite similar, both having the characteristic of abrupt changes in multi-layer images of the same sequence, the obtained suspected lesion region may also contain noise points. Therefore, it is necessary to further analyze the suspected lesion region to further distinguish noise from real lesions. Since the location and time of noise are often very random, they will basically not appear in the same position in different sequence images at the same time. However, lesions are real tissue structures. Although the degree of manifestation in different sequence images is different, they will still appear in the same position in a few images. Therefore, after obtaining the suspected slice image and the suspected lesion region on the suspected slice image, the corresponding layer region on the suspected slice image is obtained based on all slice images with the same layer number as the corresponding suspected slice image in other brain MRI images other than the brain MRI image where the suspected slice image is located. Then, the real lesion region is obtained based on the corresponding layer region on the suspected lesion region.

[0042] In this embodiment, the specific process of obtaining the actual lesion region based on all slice images with the same layer number as the corresponding suspected slice image from other brain MRI images (excluding the brain MRI image containing the suspected slice image) is as follows:

[0043] For any suspected lesion region m on any suspected slice image M: First, in other brain MRI images besides the one containing the suspected slice image M, acquire all slice images with the same layer number as the suspected slice image M, and record all slice images with the same layer number as the suspected slice image M as slice images of the same layer as the suspected slice image M; then, on each slice image of the same layer, acquire regions with the same target coordinates as the suspected lesion region m, and record them as regions of the same layer corresponding to the suspected lesion region m, that is, the area of ​​any region of the same layer is the same as that of the suspected lesion region m, and there are pixels in the region of the same layer with the same target coordinate values ​​as each pixel in the suspected lesion region m. On a slice image of the same layer There exists a corresponding region in the same layer as a suspected lesion region m. This means that within this region, pixels with the same target coordinate values ​​as those in the suspected lesion region m can be found. In other words, each region in the same layer is constructed from all pixels in the slice image of the corresponding region that have the same target coordinate values ​​as those in the suspected lesion region m. Then, the lesion probability characterization value of each pixel in the same region is obtained. The lesion probability characterization value of any pixel refers to the result of a negative correlation mapping with its normal probability characterization value. Alternatively, the lesion probability characterization value of the pixel can be obtained by subtracting its normal probability characterization value from a constant 1, provided that the pixel's... The normal probability representation value ranges from 0 to 1, and in this embodiment, the normal probability representation value of a pixel ranges from 0 to 1. Then, the mean value of the lesion probability representation values ​​of all pixels in each same-layer region corresponding to the suspected lesion region m is obtained and recorded as the same-layer lesion probability representation value for that same-layer region. That is, the mean value of the lesion probability representation values ​​of all pixels in any same-layer region corresponding to the suspected lesion region m is the same-layer lesion probability representation value for that same-layer region. Next, the mean value of the same-layer lesion probability representation values ​​of all same-layer regions corresponding to the suspected lesion region m is obtained and recorded as the same-layer lesion probability mean of the suspected lesion region m. The same-layer lesion probability mean of the suspected lesion region m is then further processed... Normalization is performed, and the result of normalization is used as the judgment index value of the suspected lesion region m. The larger the judgment index value of the suspected lesion region m, the greater the probability that the suspected lesion region m is a real lesion region or a real cerebral small vessel lesion region. Therefore, after obtaining the judgment index value of the suspected lesion region m, it is judged whether the judgment index value of the suspected lesion region m is greater than the judgment threshold. If it is, it indicates that the suspected lesion region m is a real lesion region or a real cerebral small vessel lesion region, and the suspected lesion region m is recorded as a real lesion region. Otherwise, it indicates that the suspected lesion region m is not a real lesion region or a real cerebral small vessel lesion region, and the suspected lesion region m is not judged as a real lesion region.Furthermore, in practical applications, implementers need to set the judgment threshold according to the value range of the judgment indicator, experimental statistics, and other actual situations. For example, since the value range of the judgment threshold is 0 to 1, this embodiment can set the judgment threshold to 0.5.

[0044] Furthermore, the specific calculation expression for the judgment index value of the suspected lesion area m is as follows:

[0045]

[0046] in, Let be the judgment index value for suspected lesion region m, and U be the number of regions in the same layer corresponding to suspected lesion region m, which is also the number of slice images in the same layer as suspected slice image M. This represents the probability characterization value of a lesion in the same layer of the u-th same-layer region corresponding to the suspected lesion region m; and When it is larger, The larger, and The larger the value, the greater the probability that the suspected lesion area m is the actual lesion area or the actual cerebral small vessel lesion; conversely, the smaller the value, the lower the probability that the suspected lesion area m is the actual lesion area or the actual cerebral small vessel lesion.

[0047] In this embodiment, the specific process of obtaining a suspected slice image and a suspected lesion region on the suspected slice image based on the normal probability representation value of each pixel on the slice image is as follows: For any slice image, firstly, it is determined whether there are pixels on the slice image whose normal probability representation value is less than a preset second threshold. If so, it indicates that there may be a lesion region on the slice image, and the slice image is recorded as a suspected slice image. After obtaining the suspected slice image, all pixels on the suspected slice image whose normal probability representation value is less than the preset second threshold are obtained and recorded as suspected lesion pixels on the corresponding suspected slice image. Then, connected component analysis is performed on all suspected lesion pixels on each suspected slice image, and the resulting connected component is recorded as the suspected lesion region on the corresponding suspected slice image. The process of obtaining connected components by performing connected component analysis on all suspected lesion pixels on the suspected slice image is well known. In specific applications, the implementer needs to set the preset second threshold according to the value range of the normal probability representation value, experimental statistics, and other actual conditions. For example, since the value range of the normal probability representation value is 0 to 1, in this embodiment, the preset second threshold can be set to 0.5.

[0048] In addition, this embodiment can also identify the cerebral small vessel lesion region of the patient based on the obtained real lesion region and the cerebral small vessel lesion classification model. For example, a cerebral small vessel lesion classification model can be constructed first, which can be a random forest, and the cerebral small vessel lesion classification model can be trained. The specific training process is well known and will not be described in detail here. Then, the features of the obtained real lesion region are input into the cerebral small vessel lesion classification model, and the cerebral small vessel lesion type of the real lesion region is output. The features of the real lesion region can be extracted by a morphological recognition model, which can be a CNN convolutional neural network. That is, the real lesion region is input into the trained morphological recognition model to extract the size, shape, grayscale and other features of the real lesion region. The training process of the morphological recognition model is also well known.

[0049] Thus, this embodiment completes the identification and detection of cerebral small vessel lesions.

[0050] In summary, this embodiment first acquires brain MRI images of the patient under different sequences and the local windows corresponding to each pixel in each slice image of the brain MRI images; then, for any slice image 'a' in any sequence of brain MRI images, all other slice images in the sequence of brain MRI images except slice image 'a' are recorded as neighborhood images of slice image 'a'; based on the pixel gradient values ​​in the local windows corresponding to each pixel in slice image 'a' and the local windows corresponding to pixels in the neighborhood images that have the same target coordinate values ​​as each pixel in slice image 'a', the normal probability characterization value of each pixel in slice image 'a' is obtained; then, based on the normal probability characterization value, a suspected slice image and a suspected lesion region on the suspected slice image are obtained; finally, based on all slice images with the same layer number as the suspected slice image in other brain MRI images except the brain MRI image where the suspected slice image is located, the same layer region corresponding to the suspected lesion region on the suspected slice image is obtained, and based on the same layer region corresponding to the suspected lesion region, the actual lesion region is obtained. Furthermore, the suspected lesion region and the corresponding same-layer region obtained by this embodiment based on the normal probability representation value of each pixel can improve the accuracy and completeness of the identification and detection of real lesion regions or real cerebral small vessel lesion regions.

[0051] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A system for identifying and detecting small cerebral vessel lesions based on MRI image analysis, comprising a processor and a memory, characterized in that, The processor executes the computer program stored in the memory to perform the following steps: Acquire brain MRI images of the patient in different sequences and the local windows corresponding to each pixel point in each slice image of the brain MRI images; For any slice image a in a brain MRI image of any sequence, all other slice images in the brain MRI image of the sequence other than slice image a are recorded as the neighborhood images of slice image a. Based on the pixel gradient value in the local window corresponding to each pixel point on slice image a and the local window corresponding to the pixel point in the neighborhood image that has the same target coordinate value as each pixel point on slice image a, the normal probability characterization value of each pixel point on slice image a is obtained. Based on the normal probability characterization value, a suspected slice image and a suspected lesion region on the suspected slice image are obtained; based on all slice images with the same layer number as the suspected slice image in other brain MRI images other than the brain MRI image where the suspected slice image is located, the same layer region corresponding to the suspected lesion region on the suspected slice image is obtained, and based on the same layer region corresponding to the suspected lesion region, the actual lesion region is obtained.

2. The cerebral small vessel lesion identification and detection system based on MRI image analysis as described in claim 1, characterized in that, The method for obtaining the normal probability representation value of each pixel on the slice image a includes: Based on the gradient value of each pixel in the local window corresponding to each pixel in each slice image, the tissue edge probability representation value of the local window corresponding to the pixel is obtained. For the i-th pixel in the slice image a: all pixels in the neighboring images of the slice image a that have the same target coordinate value as the i-th pixel in the slice image a are recorded as neighboring pixels in the slice image a. Based on the tissue edge probability representation value of the local window corresponding to the i-th pixel and the local window corresponding to the neighboring pixels in the same position, the similarity representation value between the i-th pixel and the neighboring pixels in the same position is obtained. Based on the similarity representation value between the i-th pixel and the neighboring pixels in the same position and the layer difference between the slice image a and the image to which the neighboring pixels in the same position belong, the normal probability representation value of the i-th pixel in the slice image a is obtained. The target coordinate value of a pixel in any slice image is the coordinate value of the corresponding slice image in the midline coordinate system of the corresponding slice image.

3. The cerebral small vessel lesion identification and detection system based on MRI image analysis as described in claim 2, characterized in that, Methods for obtaining organizational marginal probability characterization values ​​include: For any pixel, obtain the gradient values ​​of all pixels in the local window corresponding to the pixel, calculate the standard deviation of the gradient values ​​of all pixels in the local window corresponding to the pixel, and record it as the gradient standard deviation of the local window corresponding to the pixel. The normalization result of the gradient standard deviation is used as the organization edge probability characterization value of the local window corresponding to the pixel.

4. The cerebral small vessel lesion identification and detection system based on MRI image analysis as described in claim 2, characterized in that, The method for obtaining the similarity representation value between the i-th pixel and the neighboring pixels at the same position includes: For the i-th pixel and any neighboring pixel c at the same position: if the tissue edge probability representation value of the i-th pixel is greater than a preset first threshold, then the similarity representation value between the i-th pixel and the neighboring pixel c is obtained based on the DTW distance between the edge in the local window corresponding to the i-th pixel and the edge in the local window corresponding to the neighboring pixel c; if the tissue edge probability representation value of the i-th pixel is not greater than the preset first threshold, then the grayscale similarity between the i-th pixel and the neighboring pixel c is used as the similarity representation value between the i-th pixel and the neighboring pixel c.

5. The cerebral small vessel lesion identification and detection system based on MRI image analysis as described in claim 4, characterized in that, A method for obtaining the similarity representation value between the i-th pixel and its neighboring pixel c based on the DTW distance between the edge of the local window corresponding to the i-th pixel and the edge of the local window corresponding to the neighboring pixel c at the same position includes: For the j-th edge in the local window corresponding to the i-th pixel, the minimum DTW distance among the DTW distances between the j-th edge and each edge in the local window corresponding to the neighboring pixel c is negatively correlated and mapped as the matching degree between the j-th edge and the edge in the local window corresponding to the neighboring pixel c. The minimum distance between the j-th edge and the center pixel in the local window corresponding to the i-th pixel, plus the reciprocal of a preset constant, is used as the distance weight factor of the j-th edge. The product of the distance weight factor and the matching degree is used as the similarity feature value between the j-th edge and the edge in the local window corresponding to the neighboring pixel c. The result of normalizing the mean of the similarity feature values ​​of all edges in the local window corresponding to the i-th pixel and the edges in the local window corresponding to the neighboring pixel c is denoted as the similarity representation value between the i-th pixel and the neighboring pixel c.

6. The cerebral small vessel lesion identification and detection system based on MRI image analysis as described in claim 2, characterized in that, The method for obtaining the normal probability representation value of the i-th pixel on the slice image a includes: The set of all neighboring pixels at the same position of the i-th pixel in the slice image a is denoted as the set of neighboring pixels at the same position of the i-th pixel. The set of weighted similarity representation values ​​corresponding to the i-th pixel is obtained, and the f-th weighted similarity representation value in the set of weighted similarity representation values ​​is... , Let f be the similarity representation value between the i-th pixel and the f-th neighboring pixel in the set of neighboring pixels at the same position. The result is obtained by performing negative correlation mapping and normalization on the absolute value of the difference between the layer number of the slice image to which the i-th pixel belongs and the layer number of the image to which the f-th neighboring pixel at the same position belongs; the accumulated result of the weighted similarity characterization value set is calculated and used as the normal probability characterization value of the i-th pixel on the slice image a.

7. The cerebral small vessel lesion identification and detection system based on MRI image analysis as described in claim 2, characterized in that, Methods for obtaining the midline coordinate system of a sliced ​​image include: For any slice image, an edge detection algorithm is used to extract edges from the slice image to obtain all edges on the slice image. The closed edge with the largest area on the slice image is taken as the edge to be analyzed. The edge pixels on the edge to be analyzed are combined in pairs without repetition to obtain all edge pixel combinations. The Euclidean distance between the two pixels in each edge pixel combination is calculated and recorded as the feature distance of the corresponding edge pixel combination. All edge pixel combinations are sorted in descending order of feature distance to obtain a first sequence. The sequence constructed by the first preset number of edge pixel combinations in the first sequence is recorded as the second sequence. The optimization degree of each edge pixel combination in the second sequence is calculated. The edge pixel combination corresponding to the maximum optimization degree is recorded as the target combination. The line segment connecting the two edge pixels in the target combination is recorded as the target line segment. The coordinate system with the vertical axis coinciding with the target line segment and the horizontal axis perpendicular to the midpoint of the target line segment is used as the midline coordinate system of the slice image. The method for obtaining the preference of edge pixel combination includes: for any edge pixel combination, the straight line connecting the two edge pixels in the edge pixel combination is recorded as a feature line; on the slice image, all pixel pairs symmetrical about the feature line are obtained and recorded as symmetrical pixel pairs; the reciprocal of the sum of the absolute value of the gray level difference between the two pixels in the symmetrical pixel pair and a preset constant is used as the symmetry representation value of the corresponding symmetrical pixel pair; the result of summing the symmetry representation values ​​of all symmetrical pixel pairs and then normalizing them is used as the preference of the edge pixel combination.

8. The cerebral small vessel lesion identification and detection system based on MRI image analysis as described in claim 1, characterized in that, Methods for obtaining suspected slice images and suspected lesion areas on suspected slice images include: For any slice image, determine whether there are any pixels on the slice image whose normal probability representation value is less than a preset second threshold. If so, the slice image is recorded as a suspected slice image, and all pixels on the suspected slice image whose normal probability representation value is less than the preset second threshold are recorded as suspected lesion pixels on the corresponding suspected slice image. Connectivity analysis is performed on all suspected lesion pixels on the suspected slice image to obtain the suspected lesion region on the corresponding suspected slice image.

9. The cerebral small vessel lesion identification and detection system based on MRI image analysis as described in claim 1, characterized in that, A method for obtaining the corresponding layer region of the suspected lesion area on the suspected lesion image from all slice images with the same layer number as the suspected slice image in other brain MRI images besides the brain MRI image containing the suspected slice image, and obtaining the actual lesion area based on the corresponding layer region of the suspected lesion area, includes: For any suspected lesion region on any suspected slice image: Obtain all slice images with the same layer number as the suspected slice image from other brain MRI images besides the brain MRI image containing the suspected slice image, and record them as slice images of the same layer as the suspected slice image. Construct regions from all pixels in each slice image that have the same target coordinate values ​​as pixels in the suspected lesion region, and record these regions as the corresponding layer regions of the suspected lesion region. Obtain the average lesion probability representation value of all pixels in each layer region corresponding to the suspected lesion region, and record this as the corresponding layer lesion probability representation value. The lesion probability representation value of any pixel is the result of negatively correlated mapping with the normal probability representation value of the corresponding pixel. The average of the layer lesion probability representation values ​​of all layer regions is then normalized and used as the judgment index value for the suspected lesion region. If the judgment index value is greater than the judgment threshold, the suspected lesion region is recorded as a true lesion region.

Images