A prostate ultrasound image segmentation method and device based on outlier detection

Abstract

The application discloses a prostate ultrasonic image segmentation method and device based on abnormal point detection, and the method comprises the following steps: acquiring an average contour representing a prostate shape; performing deformation processing on the average contour, so that the deformed contour is preliminarily consistent with a real boundary of the prostate on the prostate ultrasonic image; adjusting each point on the deformed contour to a gray value mutation position on the prostate ultrasonic image by using a normal vector contrast boundary algorithm, so as to obtain a rough segmentation contour; detecting and removing position abnormal points on the rough segmentation contour; and sequentially connecting each point on the rough segmentation contour after the position abnormal points are removed, so as to obtain a final prostate ultrasonic image segmentation contour. The method has the beneficial effects that the average contour is used to preliminarily frame the prostate contour, so that the calculation amount required for subsequent contour correction is greatly reduced, and meanwhile, the abnormal point detection algorithm is used to identify the position abnormal points, so that the calculation amount can be further reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of prostate ultrasound image processing, in particular to a prostate ultrasound image segmentation method and device based on abnormal point detection. BACKGROUND

[0002] The prostate is the largest accessory gland unique to men, which is shaped like a chestnut and is slightly flat from front to back, located in the deep pelvic cavity of the human body. The prostate belongs to part of the urethra, and the prostate fluid secreted by the prostate is an important part of semen, so the prostate plays an important role in urinary control, reproduction and sexual function. The symptoms of prostate disease are usually mild, such as frequent urination, urgency, and discomfort of scrotal swelling and pain, and severe, affecting male reproduction and even life and health. The annual incidence of prostate disease in China is also not optimistic, so the treatment of prostate disease has a special significance for improving the health level of Chinese men. In clinical practice, in order to facilitate the observation of the lesion part and quantitative analysis, the physician often needs to segment the prostate, and the current prostate ultrasound image segmentation is mainly completed by experienced physicians, but this method has the disadvantages of low efficiency and different segmentation effects for different people, therefore, an automatic, fast and accurate segmentation method is very necessary in clinical application.

[0003] At present, the prostate ultrasound image segmentation method is divided into two categories, one is the traditional boundary operator extraction image edge information according to specific problems such as Sobel, Robert, etc., and the other is the end-to-end deep learning method popular in recent years, the network model represented by U-net structure and its derivative method. However, the detection effect of the former is easily affected by the image itself artifact factors, and the latter often needs a large amount of data set to support iterative training, and the calculation efficiency is low, which is not cost-effective in terms of calculation time and computing power. SUMMARY

[0004] Therefore, it is necessary to provide a prostate ultrasound image segmentation method and device based on abnormal point detection, to solve the technical problems that the existing prostate ultrasound image segmentation method is greatly affected by artifact factors and / or has low calculation efficiency.

[0005] In order to achieve the above purpose, the present application provides a prostate ultrasound image segmentation method based on abnormal point detection, comprising the following steps:

[0006] S1, obtaining an average contour representing the shape of the prostate;

[0007] S2, obtaining a prostate ultrasound image to be segmented, displaying the prostate ultrasound image and the average contour on the same plane, and deforming the average contour to preliminarily match the real boundary of the prostate on the prostate ultrasound image after deformation;

[0008] S3. Using the normal vector contrast boundary algorithm, each point on the deformed contour is adjusted to the gray value abrupt change point on the prostate ultrasound image to obtain the coarse segmentation contour.

[0009] S4. Detect and remove abnormal points on the coarse segmentation contour;

[0010] S5. Connect the points on the coarse segmentation contour after removing the abnormal points in sequence to obtain the final prostate ultrasound image segmentation contour.

[0011] In some embodiments, step S1, obtaining the average contour characterizing the shape of the prostate, specifically includes the following steps:

[0012] S11. Select a certain number of typical prostate boundary contours that are manually outlined;

[0013] S12. Sample each of the boundary contours at equal intervals in a certain order to obtain several boundary contour point sets. Based on each boundary contour point set, obtain the average contour representing the shape of the prostate.

[0014] In some embodiments, step S12 involves sampling each of the boundary contours at equal intervals in a certain order to obtain several boundary contour point sets. Based on each boundary contour point set, an average contour representing the shape of the prostate is obtained, specifically including:

[0015] S121. For each boundary contour, taking the contour point directly above the midpoint of the boundary contour as the starting point, select a predetermined number of boundary contour points at equal intervals in a counterclockwise direction to obtain several boundary contour point sets. Let the coordinates of the j-th point on the i-th boundary contour be (x...). ij y ij );

[0016] S122. Calculate the coordinates of each point on the average profile using the following formula.

[0017]

[0018]

[0019] Where i is the index of the boundary contour, n is the number of boundary contours, and j is the index of the contour point.

[0020] In some embodiments, in step S2, the deformation process of the average contour includes at least one of scaling, translation, and rotation.

[0021] In some embodiments, in the step S3, the normal vector contrast boundary algorithm is used to adjust each point on the deformed contour to a gray value mutation on the prostate ultrasound image, to obtain a rough segmentation contour, specifically including the following steps:

[0022] S31, a normal vector is made for each point on the deformed contour, the gray value of each pixel point passed by the normal vector corresponding to each point on the prostate ultrasound image is extracted, and the gray value list is combined in order;

[0023] S32, a point is selected, the first gray value sum within a preset length range on one side of the selected point is calculated, and the second gray value sum within a preset length range on the other side of the selected point is calculated, to obtain the gray value difference on both sides of the selected point;

[0024] S33, the selected point is moved along the direction of the normal vector, and the gray value difference on both sides of the selected point is calculated each time the selected point is moved to the position, and the selected point is adjusted to the point position with the maximum gray value difference;

[0025] S34, other points on the deformed contour are processed in turn according to the steps S32 and S33, to obtain a rough segmentation contour.

[0026] In some embodiments, in the step S4, the position abnormal point on the rough segmentation contour is detected and removed, specifically including the following steps:

[0027] S41, the coordinates of the center point of the rough segmentation contour are obtained according to the rough segmentation contour;

[0028] S42, the Euclidean distance between each point on the rough segmentation contour and the center point is calculated in turn in a certain order, to obtain a function relationship about the Euclidean distance and the number of contour points;

[0029] S43, the gradient of the function relationship about the Euclidean distance and the number of contour points at each number is calculated, and when the gradient satisfies a preset relationship, the contour point at the number is determined as a position abnormal point;

[0030] S44, the position abnormal point is removed on the rough segmentation contour.

[0031] In some embodiments, in the step S41, the coordinates of the center point of the rough segmentation contour are:

[0032] center_x=(x min +x max ) / 2

[0033] center_y=(y min +y max ) / 2

[0034] wherein, center_x is the horizontal coordinate of the center point of the rough segmentation contour, center_y is the vertical coordinate of the center point of the rough segmentation contour, x max is the maximum horizontal coordinate value corresponding to the rough segmentation contour, x min is the minimum horizontal coordinate value corresponding to the rough segmentation contour, y max is the maximum vertical coordinate value corresponding to the rough segmentation contour, y min is the minimum vertical coordinate value corresponding to the rough segmentation contour.

[0035] In some embodiments, in the step S42, the function relationship between the Euclidean distance and the number of the contour point is:

[0036]

[0037] wherein, l i is the Euclidean distance between the contour point numbered i and the center point, point_x[i] is the horizontal coordinate of the contour point numbered i, and point_y[i] is the vertical coordinate of the contour point numbered i.

[0038] In some embodiments, in the step S43, the gradient of the function between the Euclidean distance and the number of the contour point at each number is calculated, and when the gradient satisfies a preset relationship, the contour point at the number is determined as a position abnormal point, and specifically includes:

[0039] S431, the gradient of the function between the Euclidean distance and the number of the contour point at each number is calculated according to the following formula:

[0040] f i = (l i -l i-1 ) / dis

[0041] wherein, l i is the Euclidean distance between the contour point numbered i and the center point, f i is the gradient at the contour point numbered i, and dis is the distance between two adjacent contour points.

[0042] S432, when the following relationship is satisfied, the contour point numbered i is a position abnormal point:

[0043]

[0044] wherein, f i is the gradient at the contour point numbered i, and M and m are both preset values.

[0045] The application further provides a prostate ultrasound image segmentation device, comprising a processor, a memory and a communication bus; the memory stores a computer readable program which can be executed by the processor; the communication bus realizes the connection communication between the processor and the memory; the processor realizes the prostate ultrasound image segmentation method based on the abnormal point detection when executing the computer readable program.

[0046] Compared with the prior art, the technical scheme has the beneficial effects that: the average profile representing the shape of the prostate is used to preliminarily frame the prostate profile on the prostate ultrasound image, the preliminarily framed profile is corrected by the normal vector contrast boundary algorithm, so that the corrected profile conforms to the gray scale information on the prostate ultrasound image, then the abnormal point detection is performed on the corrected profile to eliminate the noise points thereon, and then the remaining normal points are connected to obtain the final prostate ultrasound image segmentation profile, compared with the prior art, since the average profile is used to preliminarily frame the prostate profile on the prostate ultrasound image, the calculation amount required for subsequent profile correction is greatly reduced, and the abnormal point detection algorithm is used to identify the position abnormal point, so that the calculation amount can be further reduced, and the prostate ultrasound image segmentation efficiency is greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a process of performing equal-interval sampling on a typical prostate boundary profile manually outlined to obtain a plurality of boundary profile point sets;

[0048] Figure 2 is an average profile representing the shape of the prostate extracted according to Figure 1

[0049] Figure 3 is a schematic diagram of a process of performing deformation processing on the average profile to preliminarily match the deformed profile with the real boundary of the prostate on the prostate ultrasound image;

[0050] Figure 4 is a schematic diagram of the normal vector contrast boundary algorithm;

[0051] Figure 5 is a schematic diagram of performing normal vectors on each point of the deformed profile in Figure 3

[0052] Figure 6 is a prostate ultrasound image rough segmentation profile obtained by performing the normal vector contrast boundary algorithm on the deformed profile in Figure 5

[0053] is a schematic diagram of the Euclidean distance between a point on the rough segmentation profile in Figure 7 Figure 6 ​​​​

[0054] Figure 8 is Figure 7 a set of distances from each point on the coarse segmentation contour in

[0055] Figure 9 is Figure 8 a gradient of the Euclidean distance in

[0056] Figure 10 is Figure 7 a schematic diagram of the connection of normal points on the coarse segmentation contour in

[0057] Figure 11 is the final prostate ultrasound image segmentation contour. DETAILED DESCRIPTION

[0058] The preferred embodiments of the present application will be described in detail below with reference to the drawings, which form a part of this application. The drawings illustrate embodiments of the application and serve to explain the principles of the application, but are not intended to limit the scope of the present application.

[0059] Referring to Figure 1 , the present application provides a prostate ultrasound image segmentation method based on abnormal point detection, comprising the following steps:

[0060] S1, obtaining an average contour representing the shape of the prostate;

[0061] Step S1 specifically comprises the following steps:

[0062] S11, selecting a certain number of typical prostate boundary contours manually outlined (such as Figure 1 );

[0063] S12, sampling each of the boundary contours at equal intervals in a certain order to obtain a plurality of boundary contour point sets (such as Figure 1 ), and obtaining an average contour representing the shape of the prostate (such as Figure 2 ) according to each boundary contour point set.

[0064] Step S12 specifically comprises:

[0065] S121, for each boundary contour, taking the contour point directly above the midpoint of the boundary contour as the starting point, and selecting a preset number of boundary contour points at equal intervals in the counterclockwise direction to obtain a plurality of boundary contour point sets, and setting the coordinates of the jth point on the ith boundary contour as (x ij , y ij );

[0066] S122, calculating the coordinates of each point on the average contour according to the following formula

[0067]

[0068]

[0069] Wherein, the serial number of i boundary contours, n is the number of boundary contours, j is the serial number of contour points.

[0070] S2, obtaining the prostate ultrasound image to be segmented, displaying the prostate ultrasound image and the average contour on the same plane, and deforming the average contour to preliminarily match the deformed contour with the real boundary of the prostate on the prostate ultrasound image;

[0071] Wherein, the deformation processing method includes at least one of scaling, translation and rotation.

[0072] Specific to the embodiment, as shown in the left graph of Figure 3 , the average contour is scaled and translated, and the deformed contour preliminarily matches the real boundary of the prostate on the prostate ultrasound image (as shown in the right graph of Figure 3 ).

[0073] Step S2 can be manually performed or realized by an automatic algorithm. The specific principle of the automatic algorithm is that a pyramid method from coarse to fine is adopted, in which the average shape of the trained model is taken as a seed contour, which is scaled to adapt to different resolution levels, and the contour points of the model are moved along the normal direction, using the transition from light to dark; After all the points are moved, part of the significant contour x s can be detected. By using x s and the corresponding average shape part , the similarity transformation T s can be calculated:

[0074]

[0075] In the formula, m is the number of significant contour points, and the transformed shape is used as the initialization result of the current level. The shape obtained at the optimal level is considered as the final initialization result.

[0076] S3, adjusting each point on the deformed contour to the gray value mutation on the prostate ultrasound image by the normal vector contrast boundary algorithm to obtain a rough segmentation contour;

[0077] In order to facilitate understanding, the normal vector contrast boundary algorithm will be introduced first as follows:

[0078] As shown in Figure 4As shown, first, the target object in the ultrasound image is initially located using the method in steps S1 and S2 to form a wide contour curve, which can be discretized into a point set:

[0079] S = {(x1, y1), (x2, y2),..., (x n , y n )}

[0080] A normal vector of the segmentation line at each point on the contour curve is set to be a point set with the same density as the pixel points, and the normal vector direction is defined to point outward. Then each point on the normal vector has a fixed index value, and we take an arbitrary point as p(x, y) and assume that the index value of the point on the normal vector is ind. Then the gray scale values of m points in the positive and negative directions of the p point are taken, and a point set for representing the gray scale of each point is defined:

[0081] f = [f ind+m , f ind+m-1 ,... f ind ..., f ind-m+1 , f ind-m ]

[0082] where f i represents the gray scale value of the point with row index value i. The prostate contour boundary is expanded into a strip chart and a local enlarged view is shown as Figure 8 .

[0083] Next, the gray scale difference between the upper and lower regions of a point in the direction is calculated by the following formula:

[0084]

[0085] ind = argmax C ind

[0086] (3-2)

[0087] where i represents the value range of the upper and lower region boundary algorithm, ind represents the row index value of the point, and C is a set composed of all C ind . Formula (3-1) calculates the gray scale difference between the upper and lower regions of a point in the direction, and the point is moved on the normal vector to form a set C ind . Formula (3-2) is to determine the best segmentation point, and the row index corresponding to the maximum gray scale difference in C can be considered as the index value of the target boundary point. By traversing all points on the discretized curve in this way, the approximate contour boundary can be obtained.

[0088] For better understanding of the present application, the following will introduce in detail how the normal vector contrast boundary algorithm is applied to prostate ultrasound image segmentation.

[0089] Step S3 specifically comprises the following steps:

[0090] S31, normal vectors are made for each point on the deformed contour (as shown in Figure 5 ), and the gray values of each pixel point passed by the normal vector corresponding to each point on the prostate ultrasound image are extracted and combined into a gray value list in order;

[0091] S32, a point is selected, and the first gray value sum within the preset length range on one side of the selected point is calculated, and the second gray value sum within the preset length range on the other side of the selected point is calculated, to obtain the gray value difference on both sides of the selected point;

[0092] S33, the selected point is moved along the direction of the normal vector, and the gray value difference on both sides of the selected point is calculated each time the selected point is moved to the position, and the selected point is adjusted to the position with the maximum gray value difference;

[0093] S34, other points on the deformed contour are processed in turn according to steps S32 and S33 to obtain a rough segmentation contour (as shown in Figure 6 ).

[0094] S4, detecting and removing position abnormal points on the rough segmentation contour;

[0095] Step S4 specifically comprises the following steps:

[0096] S41, according to the rough segmentation contour, the coordinates of the center point of the rough segmentation contour are obtained;

[0097] The coordinates of the center point of the rough segmentation contour are:

[0098] center_x=(x min +x max ) / 2

[0099] center_y=(y min +y max ) / 2

[0100] Wherein, center_x is the horizontal coordinate of the center point of the rough segmentation contour, center_y is the vertical coordinate of the center point of the rough segmentation contour, x max is the maximum horizontal coordinate value corresponding to the rough segmentation contour, x min is the minimum horizontal coordinate value corresponding to the rough segmentation contour, y max is the maximum vertical coordinate value corresponding to the rough segmentation contour, and y minThis represents the minimum ordinate value corresponding to the coarsely segmented contour.

[0101] S42. Calculate the Euclidean distance between each point on the coarse segmentation contour and the center point in a certain order to obtain the functional relationship between the Euclidean distance and the number of the contour points;

[0102] like Figure 7 As shown, in step S42, the functional relationship between the Euclidean distance and the numbering of the contour points is as follows:

[0103]

[0104] Among them, l i Let i be the Euclidean distance between the contour point numbered i and the center point, point_x[i] be the x-coordinate of the contour point numbered i, and point_y[i] be the y-coordinate of the contour point numbered i.

[0105] By iterating through each point on the coarsely segmented contour, a set of points representing the distances from each point to the center point is obtained, as shown in the following figure. Figure 8 As shown.

[0106] S43. Calculate the gradient of the function relating Euclidean distance and the number of contour points at each number. When the gradient satisfies a preset relationship, the contour point at that number is identified as a positional anomaly.

[0107] Step S43 specifically includes:

[0108] S431. Calculate the gradient of the function relating Euclidean distance and the numbering of contour points at each number using the following formula:

[0109] f i =(l i -l i-1 ) / dis

[0110] Among them, l i f is the Euclidean distance between the contour point numbered i and the center point. i Let i be the gradient at the contour point numbered i, and dis be the distance between two adjacent contour points. Since the contour points are uniformly sampled, the distance between any two adjacent contour points is equal. Therefore, for ease of calculation, we can set dis = 1.

[0111] The gradient of the Euclidean distance function with respect to the numbering of contour points at each number is as follows: Figure 9 As shown.

[0112] S432. When the following relationship is satisfied, the contour point numbered i is a positional anomaly point:

[0113]

[0114] Wherein, f i The gradient at the contour point numbered i is f, and M and m are both preset values, the values of M and m are set by experience, and in the embodiment, M and m can be set as 10 and 3 respectively.

[0115] In the embodiment, all contour points are traversed by the above formula, and the position abnormal points numbered i = 51, 57, 76 and 78 can be detected, and the points are stored in a list, such as list = [51, 57, 76, 78]. Then the list is traversed, and if the following condition is met:

[0116] f [list[j]] *f [list[j+1]] <0

[0117] It is indicated that the points corresponding to list[j] and list[j+1] are both abnormal points, and after list[j] and list[j+1] are taken out, the remaining elements in the list are continuously traversed, and all abnormal points are obtained after the traversal is completed. In the embodiment, the numbers of the abnormal points are 51, 52, 53, 54, 55, 56, 57, 76, 77 and 78.

[0118] S44, the position abnormal points are removed on the rough segmentation contour.

[0119] S5, each point on the rough segmentation contour after the position abnormal points are removed is connected in turn to obtain a final prostate ultrasound image segmentation contour.

[0120] The value j satisfying the above formula f [list[j]] *f [list[j+1]] <0 is extracted, since the points corresponding to list[j] and list[j+1] are abnormal points, the points corresponding to list[j]-1 and list[j+1]+1 are normal points, in the embodiment, 50, 58 and 75, 79 are normal points, and each group of normal points is connected into a line segment, such as the white line in Figure 10 , so that the final prostate ultrasound image segmentation contour (such as Figure 11 ) can be obtained.

[0121] In summary, the implementation process of the application is performed in three stages.

[0122] The first stage: the prostate in the original picture is roughly positioned by step S2, and the target prostate is roughly segmented by the normal vector contour algorithm in step S3.

[0123] The second stage: the prostate contour with noise obtained in the first stage is detected to obtain position abnormal points.

[0124] The third stage: the position abnormal points are deleted, and the remaining normal points are connected to obtain a final prostate ultrasound image segmentation contour.

[0125] The application further provides a prostate ultrasound image segmentation device, comprising a processor, a memory and a communication bus;

[0126] The memory stores a computer readable program which can be executed by the processor;

[0127] The communication bus realizes the connection communication between the processor and the memory;

[0128] The processor realizes the prostate ultrasound image segmentation method based on the abnormal point detection when executing the computer readable program.

[0129] In summary, the application preliminarily frames the prostate contour on the prostate ultrasound image by the average contour representing the prostate shape, then corrects the preliminarily framed contour by the normal vector contrast boundary algorithm, so that the corrected contour conforms to the grayscale information on the prostate ultrasound image, then detects the abnormal points on the corrected contour to eliminate the noise points, then connects the remaining normal points to obtain the final prostate ultrasound image segmentation contour, compared with the prior art, since the application firstly preliminarily frames the prostate contour on the prostate ultrasound image by the average contour, greatly reduces the calculation amount required for subsequent contour correction, at the same time, the abnormal point detection algorithm is used to identify the position abnormal point, which can further reduce the calculation amount, greatly improves the prostate ultrasound image segmentation efficiency.

[0130] The above merely describes the preferred embodiments of the application, but the protection scope of the application is not limited to this, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the application, which should be covered in the protection scope of the application.

Claims

1. A method for prostate ultrasound image segmentation based on outlier detection, characterized in that, The method comprises the following steps: S1, obtaining an average contour representing a shape of a prostate; S2, obtaining an ultrasound image of the prostate to be segmented, displaying the ultrasound image of the prostate and the average contour on the same plane, and deforming the average contour to preliminarily match the deformed contour with a real boundary of the prostate on the ultrasound image of the prostate; S3, adjusting each point on the deformed contour to a position of a sudden change in grayscale value on the ultrasound image of the prostate by using a normal vector contrast boundary algorithm to obtain a rough segmentation contour; S4, detecting and removing a position abnormal point on the rough segmentation contour; S5, connecting each point on the rough segmentation contour after the position abnormal point is removed in sequence to obtain a final segmentation contour of the ultrasound image of the prostate; In the step S1, the average contour representing the shape of the prostate is obtained, and the step specifically comprises the following steps: S11, selecting a certain number of typical prostate boundary contours manually outlined; S12, sampling each boundary contour at equal intervals in a certain order to obtain a plurality of boundary contour point sets, and obtaining the average contour representing the shape of the prostate according to each boundary contour point set; In the step S3, the rough segmentation contour is obtained by adjusting each point on the deformed contour to the position of the sudden change in the grayscale value on the ultrasound image of the prostate by using the normal vector contrast boundary algorithm, and the step specifically comprises the following steps: S31, performing a normal vector on each point on the deformed contour, extracting grayscale values of each pixel point passed by a normal vector corresponding to each point on the ultrasound image of the prostate, and combining the grayscale values in sequence into a grayscale value list; S32, selecting a point, calculating a first grayscale value sum within a preset length range on one side of the selected point in the grayscale value list, calculating a second grayscale value sum within a preset length range on the other side of the selected point, and obtaining a grayscale value difference between the two sides of the selected point; S33, moving the selected point along the direction of the normal vector, calculating the grayscale value difference between the two sides of the selected point after each movement, and adjusting the selected point to a position with the maximum grayscale value difference; S34, processing other points on the deformed contour according to the steps S32 and S33 in sequence to obtain the rough segmentation contour; In the step S4, the position abnormal point on the rough segmentation contour is detected and removed, and the step specifically comprises the following steps: S41, obtaining a coordinate of a center point of the rough segmentation contour according to the rough segmentation contour; S42, calculating Euclidean distances between each point on the rough segmentation contour and the center point in sequence according to a certain order to obtain a function relationship between the Euclidean distances and the numbers of the contour points; S43, obtaining gradients of the function relationship between the Euclidean distances and the numbers of the contour points at each number, and determining that a contour point at the number is a position abnormal point when the gradient satisfies a preset relationship; S44, removing the position abnormal point on the rough segmentation contour; In the step S41, the coordinate of the center point of the rough segmentation contour is: wherein, is a horizontal coordinate of a center point of the coarse segmentation contour, is a vertical coordinate of the center point of the coarse segmentation contour, is a maximum horizontal coordinate value corresponding to the coarse segmentation contour, is a minimum horizontal coordinate value corresponding to the coarse segmentation contour, is a maximum vertical coordinate value corresponding to the coarse segmentation contour, is a minimum vertical coordinate value corresponding to the coarse segmentation contour; In the step S42, the function relationship between the Euclidean distances and the numbers of the contour points is: wherein, is the Euclidean distance of the contour point numbered i from the center point, is the abscissa of the contour point numbered i, is the ordinate of the contour point numbered i; In the step S43, the gradient of the function of the Euclidean distance and the number of the contour point at each number is calculated, and when the gradient satisfies a preset relationship, the contour point at the number is determined as the position abnormal point, and specifically includes: S431, the gradient of the function of the Euclidean distance and the number of the contour point at each number is calculated according to the following formula: wherein, is the Euclidean distance of the contour point numbered i from the center point, is the gradient at the contour point numbered i, is the distance between two adjacent contour points; S432, when the following relationship is satisfied, the contour point with the number i is the position abnormal point: wherein, is the gradient at the contour point numbered i, M, m are both preset values.

2. The method of claim 1, wherein, In the step S12, each boundary contour is sampled at equal intervals in a certain order to obtain a plurality of boundary contour point sets, and an average contour representing the shape of the prostate is obtained according to each boundary contour point set, and specifically includes: S121. For each boundary contour, taking the contour point directly above the center point of the boundary contour as the starting point, select a predetermined number of boundary contour points at equal intervals in a counterclockwise direction to obtain several boundary contour point sets. Let the coordinates of the j-th point on the i-th boundary contour be... ; S122, calculate the coordinates of each point on the average profile according to the following formula : Wherein, the serial number of the i boundary contour, n is the number of the boundary contour, and j is the serial number of the contour point.

3. The method of claim 1, wherein, In the step S2, the method of the deformation processing includes at least one of scaling, translation and rotation.

4. A prostate ultrasound image segmentation apparatus characterized by comprising: It includes: A processor, a memory and a communication bus; The memory stores a computer readable program that can be executed by the processor; The communication bus realizes the connection and communication between the processor and the memory; The processor executes the computer readable program to realize the prostate ultrasound image segmentation method based on the abnormal point detection in any one of claims 1-3.

Images