Image processing method based on artificial intelligence

Abstract

The invention discloses an image processing method based on artificial intelligence, and belongs to the technical field of image processing. The method comprises the following steps: setting a sliding window for an ultrasonic image, and obtaining line and column texture relevance of pixel points; converting and exponentially amplifying the irrelevance, and screening row and column lesion points; extracting same target lesion points to form a first lesion target area and a second lesion target area; segmenting the boundary of the region, obtaining sharpness and branch degrees, and forming a sequence; and lesion region classification is obtained by processing the sequence through a lesion region recognition neural network. According to the method, lesion textures and morphological features are accurately extracted and intelligently analyzed, so that the classification accuracy is improved, and an efficient technical scheme is provided for ultrasonic lesion diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an image processing method based on artificial intelligence. BACKGROUND

[0002] In the field of modern medical diagnosis, ultrasonic imaging technology has become an indispensable means in clinical detection due to its non-invasive, real-time, low-cost and other significant advantages, and is widely used in the screening and diagnosis of lesions in many fields such as abdominal organs, cardiovascular system, gynecology and obstetrics. However, the ultrasonic image itself has the characteristics of low resolution, easy to be disturbed by noise, complex tissue texture, etc., which brings great challenges to the accurate identification and classification of the lesion area.

[0003] Traditional ultrasonic image processing methods rely on manual experience for feature extraction and analysis, which is not only inefficient, but also subjective, and is prone to cause deviations in diagnosis results due to differences in experience of operators, and is difficult to meet the high requirements of clinical diagnosis accuracy and consistency. Especially in the detection of some early lesions or small lesions, since the texture difference between the lesion area and the surrounding normal tissue is not obvious, the traditional method is often difficult to achieve accurate identification, and is easy to cause missed diagnosis or misdiagnosis, and delay the best treatment opportunity of the patient.

[0004] With the rapid development of artificial intelligence technology, the image processing method based on deep learning provides a new idea to solve the above problems. However, the existing ultrasonic image processing technology based on artificial intelligence focuses on the overall gray scale features of the lesion area and the contrast between the lesion area and the normal area, ignores the subtle features of texture changes, and some methods fail to effectively amplify the differences between the lesion and normal tissue in the feature processing process, resulting in insufficient feature discrimination and affecting the accuracy of lesion area division, which leads to low accuracy of lesion area classification. SUMMARY

[0005] In view of the above shortcomings in the prior art, the image processing method based on artificial intelligence provided by the present application solves the problem of low accuracy of lesion area classification in the prior art.

[0006] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows: an image processing method based on artificial intelligence, comprising the following steps: Setting a sliding window for the ultrasonic image to obtain the row texture correlation and column texture correlation of each pixel point; Converting the row and column texture correlation into non-correlation and performing exponential amplification to screen out row lesion points and column lesion points; Extracting the row lesion points and column lesion points belonging to the same target to form a first lesion target area and a second lesion target area; The boundaries of the first lesion target region and the second lesion target region are respectively segmented to obtain the sharpness and the branch degree of each segment, and a sharpness sequence and a branch degree sequence of the first lesion target region and the second lesion target region are constituted. The sharpness sequence and the branch degree sequence of the first lesion target region and the second lesion target region are processed by using a lesion region identification neural network to obtain the classification of the lesion region.

[0007] Further, the process of obtaining the row texture correlation and the column texture correlation of each pixel point comprises: A sliding window is slid on each row of the ultrasound image, and the sliding step is 1 pixel point; On each row, the texture correlation of the i-th sliding window and the i+1-th sliding window is taken as the row texture correlation of a new pixel point in the i+1-th sliding window, and i is the number of sliding times; A sliding window is slid on each column of the ultrasound image, and the sliding step is 1 pixel point; On each column, the texture correlation of the i-th sliding window and the i+1-th sliding window is taken as the column texture correlation of a new pixel point in the i+1-th sliding window.

[0008] Further, the process of screening out the row lesion points and the column lesion points comprises: The row texture correlation and the column texture correlation are respectively subtracted by 1 to obtain row texture disassociation and column texture disassociation; The row texture disassociation and the column texture disassociation are respectively exponentially amplified to obtain row texture disassociation amplification values and column texture disassociation amplification values; The row texture disassociation amplification values of each pixel point on the ultrasound image are averaged to obtain a row disassociation average value; The column texture disassociation amplification values of each pixel point on the ultrasound image are averaged to obtain a column disassociation average value; Pixel points with row texture disassociation amplification values greater than the row disassociation average value are marked as row lesion points; Pixel points with column texture disassociation amplification values greater than the column disassociation average value are marked as column lesion points.

[0009] Further, the process of constituting the first lesion target region and the second lesion target region comprises: Row lesion points belonging to the same region range are grouped into a row association group; Column lesion points belonging to the same region range are grouped into a column association group; The pixel point coordinates of the row association group and the column association group are taken as the intersection over union; When the intersection over union is greater than an intersection over union threshold, the corresponding row association group is taken as the first lesion target region, and the corresponding column association group is taken as the second lesion target region.

[0010] Further, the process of constructing the sequence of sharpness and the sequence of branching degree of the first lesion target region and the second lesion target region comprises: Marking the row lesion points existing in the neighborhood range of the first lesion target region as boundary lesion points; Marking the column lesion points existing in the neighborhood range of the second lesion target region as boundary lesion points; Respectively in the first lesion target region and the second lesion target region, cutting each boundary lesion point into multiple segments with any boundary lesion point as the starting point, to obtain multiple boundary segments; According to the distance of the boundary lesion points on each boundary segment to the straight line, the sharpness and the branching degree are obtained, and each sharpness belonging to the same lesion target region constitutes a sequence of sharpness, and each branching degree belonging to the same lesion target region constitutes a sequence of branching degree.

[0011] Further, the process of obtaining the sharpness and the branching degree comprises: Connecting the starting point and the ending point of each boundary segment to obtain a straight line; Calculating the distance of each boundary lesion point in each boundary segment to the straight line; Selecting the boundary lesion point corresponding to the maximum distance from the distances as the farthest point; Taking the ratio of the distance of the farthest point to the length of the straight line as the sharpness of the boundary segment; Selecting the local maximum value from the distances, and when the local maximum value is greater than the distance threshold, the corresponding boundary lesion point is taken as a branch point; According to the difference between the distance of each branch point in the same boundary segment and the distance threshold, the branching degree is obtained.

[0012] Further, the process of obtaining the branching degree comprises: subtracting the distance of each branch point from the distance threshold to obtain the difference of each branch point, and adding the difference of each branch point corresponding to the same boundary segment, and normalizing the addition result to obtain the branching degree.

[0013] Further, the lesion region recognition neural network comprises: a plurality of feature extraction units, a multiplier M1, a multiplier M2, a multiplier M3, a multiplier M4, a splicing layer, a convolution processing unit and a full connection layer; The input end of each feature extraction unit is used for inputting a sequence; The input end of the multiplier M1 is connected with the second output end of the first feature extraction unit and the second output end of the second feature extraction unit respectively; the input end of the multiplier M2 is connected with the first output end of the first feature extraction unit and the first output end of the second feature extraction unit respectively; the input end of the multiplier M3 is connected with the second output end of the third feature extraction unit and the second output end of the fourth feature extraction unit respectively; the input end of the multiplier M4 is connected with the first output end of the third feature extraction unit and the first output end of the fourth feature extraction unit respectively; The input end of the concatenation layer is connected with the output end of the multiplier M1, the output end of the multiplier M2, the output end of the multiplier M3 and the output end of the multiplier M4 respectively, and the output end thereof is connected with the input end of the convolution processing unit; The input end of the full connection layer is connected with the output end of the convolution processing unit, and the output end thereof serves as the output end of the lesion area recognition neural network.

[0014] Further, the plurality of feature extraction units each comprises a first convolution layer, a maximum pooling layer and an average pooling layer. The input end of the first convolution layer serves as the input end of the feature extraction unit, and the output end thereof is connected with the input end of the maximum pooling layer and the input end of the average pooling layer respectively; The output end of the maximum pooling layer serves as the first output end of the feature extraction unit; The output end of the average pooling layer serves as the second output end of the feature extraction unit.

[0015] Further, the convolution processing unit comprises a first stacked convolution module, a residual convolution module and a second stacked convolution module connected in sequence.

[0016] In conclusion, the present application has the following advantages: 1. The present application can capture the difference in texture between the lesion area and normal tissue from a subtle level by setting a sliding window to obtain the row and column texture correlation of each pixel point, and can extract a more accurate lesion target area, which makes up for the defect of insufficient attention to texture features in the traditional method.

[0017] 2. The present application can significantly amplify the subtle lesion feature difference by converting the row and column texture correlation into non-correlation and performing exponential amplification, so as to make the lesion points more prominent and improve the accuracy of screening the row lesion points and the column lesion points, thereby improving the accuracy of lesion area division.

[0018] 3. After the lesion area boundary is divided, the present application quantifies the boundary morphology through sharpness (boundary prominence) and branching degree (boundary branching feature), which describes the lesion from a morphological point of view (such as more irregular malignant tumor boundary and more significant sharpness / branching degree feature).

[0019] 4、The application adopts the lesion area recognition neural network to process the sharpness sequence and the branch degree sequence of the first lesion target area and the second lesion target area, combines the sharpness sequence and the branch degree sequence of the lesion target area obtained in different directions, and improves the classification precision of the lesion target area. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flowchart of an image processing method based on artificial intelligence; Figure 2 A structural diagram of a lesion area recognition neural network; Figure 3 A structural diagram of a feature extraction unit; Figure 4 A structural diagram of a convolution processing unit; Figure 5 An ultrasound diagram of a benign cyst; Figure 6 An ultrasound diagram of a malignant tumor. DETAILED DESCRIPTION

[0021] The specific embodiments of the present application are described below to facilitate the understanding of the present application by those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all the inventions utilizing the concept of the present application are within the scope of protection.

[0022] As shown in Figure 1 An image processing method based on artificial intelligence, comprising the following steps: Setting a sliding window on an ultrasound image to obtain the row texture correlation and column texture correlation of each pixel point; Converting the row and column texture correlation into incoherence and performing exponential amplification to screen out row lesion points and column lesion points; Extracting the row lesion points and column lesion points belonging to the same target to form a first lesion target area and a second lesion target area; Respectively cutting the boundaries of the first lesion target area and the second lesion target area to obtain the sharpness and branch degree of each segment, and forming a sharpness sequence and a branch degree sequence of the first lesion target area and the second lesion target area; Using a lesion area recognition neural network to process the sharpness sequence and the branch degree sequence of the first lesion target area and the second lesion target area to obtain the classification of the lesion area.

[0023] In the present embodiment, the process of obtaining the row texture correlation and column texture correlation of each pixel point includes: The sliding window is slid on each row of the ultrasonic image with a sliding step of 1 pixel point; On each row, the texture correlation of the i-th sliding window and the i+1-th sliding window is taken as the row texture correlation of a new pixel point in the i+1-th sliding window, and i is the number of sliding times; The sliding window is slid on each column of the ultrasonic image with a sliding step of 1 pixel point; On each column, the texture correlation of the i-th sliding window and the i+1-th sliding window is taken as the column texture correlation of a new pixel point in the i+1-th sliding window.

[0024] In the embodiment, the size of the sliding window is set to 1*5, and the texture correlation can be calculated by using cosine similarity or Pearson correlation coefficient, specifically: the cosine similarity or Pearson correlation coefficient of the pixel value under the i-th sliding window and the pixel value under the i+1-th sliding window is taken as the texture correlation. In the embodiment, on the row texture correlation, the first 5 columns of pixel points cannot obtain the texture correlation, and the texture correlation of these pixel points is assigned as 1, or the pixel points on the edges of the image are not considered.

[0025] The sliding window is slid on each row and column of the ultrasonic image with a sliding step of 1 pixel point, so that each pixel point can be associated with the texture information of the adjacent region, and the row texture correlation and the column texture correlation of each pixel point are calculated according to the distribution of the texture on the same column or the same row, so as to reflect the coincidence degree of the texture in the row or column direction.

[0026] In the embodiment, the process of screening the row lesion points and the column lesion points includes: The row texture inassociation and the column texture inassociation are obtained by subtracting 1 from the row texture correlation and the column texture correlation respectively; The row texture inassociation and the column texture inassociation are respectively exponentially amplified to obtain a row texture inassociation amplification value and a column texture inassociation amplification value; The row texture inassociation amplification values of the pixel points on the ultrasonic image are averaged to obtain a row inassociation average value; The column texture inassociation amplification values of the pixel points on the ultrasonic image are averaged to obtain a column inassociation average value; The pixel points with the row texture inassociation amplification value greater than the row inassociation average value are marked as row lesion points; The pixel points with the column texture inassociation amplification value greater than the column inassociation average value are marked as column lesion points.

[0027] In the embodiment, the formula of the row texture inassociation amplification value is: X L = exp(x L ), wherein, XL is the row texture unassociated magnification value, x L is the row texture irrelevance, exp( ) is the natural exponential function; the formula for the column texture irrelevance magnification value is: X A = exp(x A ), where X A Is the column texture unassociated magnification value, x A Column texture irrelevance.

[0028] The present invention uses 1 minus the row and column texture correlations to obtain the irrelevance, converts the "texture matching degree" into the "texture difference degree", and exponentially amplifies the row and column texture irrelevance. It can utilize the nonlinear characteristics of the exponential function to significantly magnify tiny irrelevance differences - for pixels with high irrelevance (possibly lesions), their amplification values ​​will increase sharply; while for normal pixels with low irrelevance, the amplification values ​​increase slowly, effectively solving the problem of unclear original texture differences.

[0029] In this embodiment, the process of forming the first lesion target area and the second lesion target area includes: The row lesions belonging to the same area are grouped into a row association group; The column lesion points belonging to the same area are grouped into a column association group; Take the intersection and comparison of the pixel coordinates of the row association group and the column association group; When the IoU is greater than the IoU threshold, the corresponding row association group is used as the first lesion target area, and the corresponding column association group is used as the second lesion target area.

[0030] In this embodiment, the intersection-over-union ratio threshold is set to a value greater than 0, that is, if there are some lesion points associated with the row association group and the column association group, they can be determined to belong to the same lesion target.

[0031] In this embodiment, the process of grouping row lesion points belonging to the same area into a row association group includes: A1. Taking any row lesion point as the center, detect whether there are other row lesions within its 5×5 neighborhood. If so, these other row lesions and the central row lesion point are grouped into a row association group. A2. Traverse all the row lesion points in the current row association group that are not the center. Take each point as the center and check whether there are other row lesion points in its 5×5 neighborhood that are not included in the row association group. If there are other row lesion points, add these new row lesion points to the row association group. A3. Repeat the A2 process until all the lesions in the row association group have been tested as centers and no new lesions can be included.

[0032] In the embodiment, the process of grouping the column lesion points belonging to the same regional range into a column correlation group comprises: B1, detecting whether there are other column lesion points in the 5*5 neighborhood range of any column lesion point as the center, if there are, then these other column lesion points and the center column lesion point are grouped into a column correlation group; B2, traversing all column lesion points in the current column correlation group that are not as the center, and sequentially taking each point as the center to detect whether there are other column lesion points not included in the column correlation group in the 5*5 neighborhood range, if there are, then these new column lesion points are included in the column correlation group; B3, repeating the process of B2 until all column lesion points in the column correlation group are detected as the center, and no new column lesion point can be included.

[0033] In the embodiment, the process of constructing the sharpness sequence and the branching degree sequence of the first lesion target region and the second lesion target region comprises: In the first lesion target region, marking the row lesion point in the neighborhood range existing the non-row lesion point as the boundary lesion point; In the second lesion target region, marking the column lesion point in the neighborhood range existing the non-column lesion point as the boundary lesion point; In the first lesion target region and the second lesion target region respectively, taking any boundary lesion point as the starting point, the each boundary lesion point is divided into multiple segments to obtain multiple boundary segments; According to the distance of the boundary lesion point on each boundary segment to the straight line, the sharpness and the branching degree are obtained, and each sharpness belonging to the same lesion target region constitutes a sharpness sequence, and each branching degree belonging to the same lesion target region constitutes a branching degree sequence.

[0034] The boundary of the lesion region is the transition zone of the lesion and the normal tissue, and there must be the feature that the lesion point is adjacent to the non-lesion point, therefore, in the embodiment, in the 3*3 neighborhood range existing the non-row lesion point, the row lesion point at the center is the boundary point, and the second lesion target region is the same.

[0035] In the embodiment, the boundary segment length can be set to 10-15 pixels, that is, the continuous 10-15 boundary lesion points are divided into a boundary segment.

[0036] The process of obtaining the sharpness and the branching degree of the first lesion target region and the second lesion target region is the same, and both comprises: Connecting the starting point and the ending point of each boundary segment to obtain a straight line; Calculating the distance of each boundary lesion point in each boundary segment to the straight line; Selecting the boundary lesion point corresponding to the maximum distance from the distances as the farthest point; The ratio of the distance of the farthest point to the length of the straight line is taken as the sharpness of the boundary segment; Local maximum values are selected from the distances, and when the local maximum value is greater than the distance threshold value, the corresponding boundary lesion point is taken as a branch point; The branching degree is obtained according to the distance of each branch point in the same boundary segment and the distance threshold value.

[0037] The straight line equation is Ax+By+C=0, A=y2-y1, B=x1-x2, C=x2y1-x1y2, wherein the coordinates of the starting point of the boundary segment are (x1, y1), the coordinates of the terminal point of the boundary segment are (x2, y2), x is the independent variable of the straight line equation, and y is the dependent variable of the straight line equation. The calculation formula of the distance of the nth boundary lesion point in the boundary segment to the straight line is: Wherein, d n is the distance of the nth boundary lesion point in the boundary segment to the straight line, (x n , y n ) is the coordinates of the nth boundary lesion point in the boundary segment, and || is the absolute value.

[0038] The local maximum value is a boundary lesion point satisfying the condition: d n-1 <d n and d n > d n+1 , wherein d n is the distance of the nth boundary lesion point in the boundary segment to the straight line, d n-1 is the distance of the n-1th boundary lesion point in the boundary segment to the straight line, and d n+1 is the distance of the n+1th boundary lesion point in the boundary segment to the straight line.

[0039] In the embodiment, the distance threshold value is set according to experiments or experience.

[0040] In the embodiment, the branching degree obtaining process comprises: subtracting the distance of each branch point from the distance threshold value to obtain the difference of each branch point, and adding the differences of the branch points corresponding to the same boundary segment and normalizing the addition result to obtain the branching degree.

[0041] The formula of the branching degree is: Wherein, μ is the branching degree, d r is the distance of the rth branch point, d th is the distance threshold value, D is the normalization parameter, R is the number of branch points, and r is the number of branch points.

[0042] In the embodiment, the ratio of the distance of the farthest point to the length of the straight line is taken as the sharpness of the boundary segment, which can intuitively and effectively reflect the convexity and steepness of the boundary segment.

[0043] The formula for calculating the sharpness is: θ i =d i / S i , wherein θ i is the sharpness of the i-th boundary segment, d i is the distance from the farthest point on the i-th boundary segment to the straight line, and S i is the length of the straight line.

[0044] The present application can accurately identify the boundary bifurcation with practical significance by selecting the point with local maximum value and greater than the distance threshold as the branch point, and eliminates the interference caused by small fluctuations. The difference between the distance of each branch point and the distance threshold is added and normalized to obtain the branch degree, which not only reflects the number of branch points, but also reflects the prominence of each branch point.

[0045] As shown in Figure 2 , the lesion area recognition neural network comprises a plurality of feature extraction units, a multiplier M1, a multiplier M2, a multiplier M3, a multiplier M4, a splicing layer, a convolution processing unit and a full connection layer. The input end of each feature extraction unit is used for inputting a sequence. The input end of the multiplier M1 is connected with the second output end of the first feature extraction unit and the second output end of the second feature extraction unit respectively; the input end of the multiplier M2 is connected with the first output end of the first feature extraction unit and the first output end of the second feature extraction unit respectively; the input end of the multiplier M3 is connected with the second output end of the third feature extraction unit and the second output end of the fourth feature extraction unit respectively; the input end of the multiplier M4 is connected with the first output end of the third feature extraction unit and the first output end of the fourth feature extraction unit respectively. The input end of the splicing layer is connected with the output end of the multiplier M1, the output end of the multiplier M2, the output end of the multiplier M3 and the output end of the multiplier M4, and the output end thereof is connected with the input end of the convolution processing unit. The input end of the full connection layer is connected with the output end of the convolution processing unit, and the output end thereof serves as the output end of the lesion area recognition neural network.

[0046] As shown in Figure 3 , the plurality of feature extraction units each comprises a first convolution layer, a maximum pooling layer and an average pooling layer. The input end of the first convolution layer serves as the input end of the feature extraction unit, and the output end thereof is connected with the input end of the maximum pooling layer and the input end of the average pooling layer respectively. The output end of the maximum pooling layer serves as the first output end of the feature extraction unit. The output end of the average pooling layer serves as the second output end of the feature extraction unit.

[0047] In the application, the first convolution layer is one-dimensional convolution, the maximum pooling layer and the average pooling layer are one-dimensional pooling, the convolution kernel size of the first convolution layer is 1x3, the pooling window of the maximum pooling layer is 1x3, and the pooling window of the average pooling layer is 1x3.

[0048] The application focuses on the peak features in the sequence through the maximum pooling, captures the 'extreme features' of the lesion morphology, and retains the overall trend of the sequence through the average pooling. The application processes one sequence through each feature extraction unit, extracts the peak features and the overall trend features, performs fusion processing on the overall trend features output by the first feature extraction unit and the overall trend features output by the second feature extraction unit through the multiplier M1, performs fusion processing on the peak features output by the first feature extraction unit and the peak features output by the second feature extraction unit through the multiplier M2, and realizes the enhancement processing of the overall trend features and the peak features of the first lesion target region.

[0049] The application performs fusion processing on the overall trend features output by the third feature extraction unit and the overall trend features output by the fourth feature extraction unit through the multiplier M3, performs fusion processing on the peak features output by the third feature extraction unit and the peak features output by the fourth feature extraction unit through the multiplier M4, and realizes the enhancement processing of the overall trend features and the peak features of the second lesion target region.

[0050] The splicing layer is used for splicing the output of the multiplier M1, the output of the multiplier M2, the output of the multiplier M3, and the output of the multiplier M4 to form a feature matrix of 4xM, M being the length of the features output by the multipliers, the convolution processing unit is used for further extracting features from the output of the splicing layer, and the full connection layer is used for classification based on the output of the convolution processing unit.

[0051] As shown in Figure 4 , the convolution processing unit comprises, which are connected in sequence: a first stacked convolution module, a residual convolution module, and a second stacked convolution module.

[0052] The stacked convolution module comprises: a second convolution layer and a third convolution layer, the second convolution layer and the third convolution layer are connected in sequence, and the convolution kernel size of the second convolution layer and the third convolution layer is 3x3.

[0053] The residual convolution module adopts a standard residual block and comprises two convolution layers and an adder.

[0054] As shown in Figure 5 , the benign cyst has a clear boundary, as shown in Figure 6 , the malignant tumor has a more irregular boundary, a higher sharpness, and more branches, and the boundary of the benign lesion is relatively regular. These sequence features provide more detailed and more identifiable basis for lesion classification.

[0055] In the embodiment, the classification types include benign lesions and malignant lesions.

[0056] The present application can capture the difference in texture between the lesion area and normal tissue from a subtle level by setting a sliding window to obtain the row texture correlation and column texture correlation of each pixel point, and can extract a more accurate lesion target area, which makes up for the defect of insufficient attention to texture features in traditional methods.

[0057] The present application can significantly amplify the subtle lesion feature difference by converting the row and column texture correlation into non-correlation and exponentially amplifying it, so that the lesion points are more prominent, and the accuracy of screening the row lesion points and the column lesion points is improved, so that the accuracy of lesion area division is improved.

[0058] After the lesion area boundary is cut, the present application quantifies the boundary shape through sharpness (boundary prominence) and branching degree (boundary branch feature), and describes the lesion from a morphological point of view (such as more irregular malignant tumor boundary, more significant sharpness / branching degree feature).

[0059] The present application processes the sharpness sequence and branching degree sequence of the first lesion target area and the second lesion target area by using a lesion area recognition neural network, and combines the sharpness sequence and branching degree sequence of the lesion target area obtained in different directions to improve the classification accuracy of the lesion target area.

[0060] The above is only a preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An image processing method based on artificial intelligence, characterized in that: The following steps are involved: A sliding window is set on the ultrasound image to obtain the row texture correlation and column texture correlation of each pixel; Convert row and column texture correlation into irrelevance and perform exponential amplification to filter out row and column lesion points; Extracting row lesion points and column lesion points belonging to the same target to form a first lesion target area and a second lesion target area; The boundaries of the first lesion target area and the second lesion target area are divided respectively, and the sharpness and branching degree of each segment are obtained to form a sharpness sequence and a branching degree sequence of the first lesion target area and the second lesion target area; The lesion region recognition neural network is used to process the sharpness sequence and branching degree sequence of the first lesion target region and the second lesion target region to obtain the classification of the lesion region.

2. The image processing method based on artificial intelligence according to claim 1, characterized in that: The process of obtaining the row texture relevance and column texture relevance of each pixel includes: A sliding window is used to slide on each line of the ultrasound image, with a sliding step of 1 pixel; On each row, the texture correlation between the i-th sliding window and the i+1-th sliding window is used as the row texture correlation of the new pixel point in the i+1-th sliding window, where i is the number of sliding times; A sliding window is used to slide on each column of the ultrasound image, with a sliding step of 1 pixel; On each column, the texture association between the i-th sliding window and the i+1-th sliding window is used as the column texture association of the new pixel point in the i+1-th sliding window.

3. The image processing method based on artificial intelligence according to claim 1, characterized in that: The process of screening row and column lesion points includes: The row texture correlation and column texture correlation are respectively subtracted from 1 to obtain the row texture irrelevance and column texture irrelevance. The row texture irrelevance and the column texture irrelevance are exponentially amplified respectively to obtain the row texture irrelevance amplification value and the column texture irrelevance amplification value; The row texture uncorrelated amplification values ​​of each pixel point on the ultrasound image are averaged to obtain the row uncorrelated mean value; The column texture uncorrelated amplification values ​​of each pixel point on the ultrasound image are averaged to obtain the column uncorrelated mean value; The pixel points whose row texture uncorrelated magnification value is greater than the row uncorrelated mean are marked as row lesion points; The pixel points whose column texture uncorrelated amplification value is greater than the column uncorrelated mean are marked as column lesion points.

4. The image processing method based on artificial intelligence according to claim 1, characterized in that: The process of forming the first lesion target area and the second lesion target area includes: The row lesions belonging to the same area are grouped into a row association group; The column lesion points belonging to the same area are grouped into a column association group; Take the intersection and comparison of the pixel coordinates of the row association group and the column association group; When the IoU is greater than the IoU threshold, the corresponding row association group is used as the first lesion target area, and the corresponding column association group is used as the second lesion target area.

5. The image processing method based on artificial intelligence according to claim 1, characterized in that: The process of forming the sharpness sequence and the branching degree sequence of the first lesion target area and the second lesion target area includes: Marking the running lesion points with non-running lesion points in the neighborhood within the first lesion target area as boundary lesion points; Mark the column lesion points that exist in the neighborhood of non-column lesion points in the second lesion target area as boundary lesion points; In the first lesion target area and the second lesion target area, taking any boundary lesion point as a starting point, dividing each boundary lesion point into multiple segments to obtain multiple boundary segments; According to the distance from the boundary lesion point to the straight line on each boundary segment, the sharpness and branching degree are obtained, and the sharpness of each lesion target area is composed of a sharpness sequence, and the branching degree of each lesion target area is composed of a branching degree sequence.

6. The artificial intelligence-based image processing method according to claim 1 or 5, characterized in that: The process of obtaining sharpness and branching degree includes: Connect the start and end points of each boundary segment to form a straight line; Calculate the distance from each boundary lesion point to the straight line in each boundary segment; From each distance, the boundary lesion point corresponding to the maximum distance is selected as the farthest point; The ratio of the distance of the farthest point to the length of the straight line is taken as the sharpness of the boundary segment; Select the local maximum value in each distance, and when the local maximum value is greater than the distance threshold, the corresponding boundary lesion point is used as the branch point; The branch degree is obtained based on the difference between the distance of each branch point in the same boundary segment and the distance threshold.

7. The image processing method based on artificial intelligence according to claim 6, characterized in that: The process of obtaining the branching degree includes: subtracting the distance of each branch point from the distance threshold to obtain the gap of each branch point, adding the gaps of each branch point corresponding to the same boundary segment, and normalizing the addition result to obtain the branching degree.

8. The image processing method based on artificial intelligence according to claim 1, characterized in that: The lesion area recognition neural network includes: multiple feature extraction units, multiplier M1, multiplier M2, multiplier M3, multiplier M4, splicing layer, convolution processing unit and full connection layer; The input end of each feature extraction unit is used to input a sequence; The input end of the multiplier M1 is connected to the second output end of the first feature extraction unit and the second output end of the second feature extraction unit respectively; the input end of the multiplier M2 is connected to the first output end of the first feature extraction unit and the first output end of the second feature extraction unit respectively; the input end of the multiplier M3 is connected to the second output end of the third feature extraction unit and the second output end of the fourth feature extraction unit respectively; the input end of the multiplier M4 is connected to the first output end of the third feature extraction unit and the first output end of the fourth feature extraction unit respectively; The input end of the splicing layer is connected to the output end of multiplier M1, multiplier M2, multiplier M3 and multiplier M4 respectively, and its output end is connected to the input end of the convolution processing unit; The input end of the fully connected layer is connected to the output end of the convolution processing unit, and its output end serves as the output end of the lesion area recognition neural network.

9. The image processing method based on artificial intelligence according to claim 8, characterized in that: The multiple feature extraction units each include: a first convolutional layer, a maximum pooling layer, and an average pooling layer; The input of the first convolutional layer serves as the input of the feature extraction unit, and its output is connected to the input of the maximum pooling layer and the input of the average pooling layer respectively; The output of the maximum pooling layer serves as the first output of the feature extraction unit; The output of the average pooling layer serves as the second output of the feature extraction unit.

10. The image processing method based on artificial intelligence according to claim 8, characterized in that: The convolution processing unit includes: a first stacked convolution module, a residual convolution module and a second stacked convolution module connected in sequence.

Images