CT image focus detection method and system based on artificial intelligence

Abstract

The invention provides a CT image focus detection method and system based on artificial intelligence, and relates to the technical field of medical image processing, and the method comprises the steps: obtaining a plain-scan CT image and an enhanced CT image of a liver; determining first position information of a to-be-determined lesion area in the plain-scan CT image through an image detection model; according to the first position information and the arterial phase image, screening out a to-be-determined tumor focus area from the to-be-determined focus area, and determining second position information of the to-be-determined tumor focus area; according to the second position information and the portal pulse period image, the to-be-determined tumor focus area is classified, and the tumor focus area and a common focus area are determined; determining a warning level score according to the tumor focus area and the delay period image; and generating a liver focus detection report according to the warning level score. According to the invention, a complex screening and evaluation process can be automatically executed, and the probability of misjudgment is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a CT image lesion detection method and system based on artificial intelligence. Background Art

[0002] In the related art, CT (Computed Tomography) images of the liver usually include plain scan CT images and enhanced CT images. When taking plain scan CT images, contrast agents do not need to be injected, but when taking enhanced CT images, contrast agents need to be injected to make the difference in pixel values ​​between the area where the blood flows and the surrounding area more obvious. However, plain scan CT images can only detect lesions, but it is difficult to determine the type of lesions. With enhanced CT images, the type of lesions can be determined by the characteristics of the lesions when blood containing contrast agents flows through them, but the judgment process is complicated and manual judgment is prone to misjudgment.

[0003] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0004] The present invention provides a CT image lesion detection method and system based on artificial intelligence, which can solve the technical problems that the judgment process of the type of lesion is complicated and manual judgment is prone to misjudgment.

[0005] According to a first aspect of the present invention, there is provided a method for detecting lesions in CT images based on artificial intelligence, comprising:

[0006] Acquiring a plain scan CT image and an enhanced CT image of the liver, wherein the enhanced CT image includes an arterial phase image, a portal venous phase image, and a delayed phase image of the liver;

[0007] Determining first position information of a to-be-determined lesion area in the plain scan CT image by using an image detection model;

[0008] According to the first position information and the arterial phase image, a tumor lesion region to be determined is screened out from a plurality of lesion regions to be determined, and second position information of the tumor lesion region to be determined is determined;

[0009] Classifying the undetermined tumor lesion region according to the second position information and the portal venous phase image to determine a tumor lesion region and a common lesion region;

[0010] Determining a warning level score of the tumor lesion area according to the tumor lesion area and the delayed phase image;

[0011] A liver lesion detection report is generated based on the warning level score.

[0012] According to a second aspect of the present invention, there is provided a CT image lesion detection system based on artificial intelligence, comprising:

[0013] An image acquisition module, used to acquire a plain scan CT image and an enhanced CT image of the liver, wherein the enhanced CT image includes an arterial phase image, a portal venous phase image, and a delayed phase image of the liver;

[0014] A detection module, used to determine first position information of a pending lesion area in the plain scan CT image through an image detection model;

[0015] a screening module, configured to screen out a tumor lesion region to be determined from a plurality of lesion regions to be determined according to the first position information and the arterial phase image, and determine second position information of the tumor lesion region to be determined;

[0016] A classification module, used for classifying the undetermined tumor lesion area according to the second position information and the portal phase image, and determining the tumor lesion area and the common lesion area;

[0017] A warning level scoring module, used to determine a warning level score of the tumor lesion area according to the tumor lesion area and the delayed phase image;

[0018] A reporting module is used to generate a liver lesion detection report based on the warning level score.

[0019] Technical effect: According to the present invention, lesions in the liver can be identified based on plain scan CT images, and lesions can be screened by enhancing the arterial phase images in the CT images, thereby screening out the areas of suspected tumor lesions, and then classifying them through portal phase images, thereby accurately determining the tumor areas. Furthermore, the severity of the lesions in the tumor area can be evaluated through delayed phase images, thereby generating warning level scores to prompt patients or doctors. Complex screening and evaluation processes can be automatically performed to reduce the probability of misjudgment. When determining the area of ​​the tumor to be determined, the area of ​​the tumor to be determined can be determined from multiple areas of the arterial phase image based on the blood supply mechanism of the liver and the tumor, that is, the area of ​​the tumor to be determined can be found in which the pixel value of the central area of ​​the lesion in the area of ​​the tumor to be determined is close to the pixel value of the pixel point in the area where the abdominal aorta is located, and the contrast between the area of ​​the tumor to be determined and the first liver background area is high. The area of ​​the tumor can be accurately and objectively found, and the accuracy of identifying the tumor lesions can be improved. When determining the first common lesion area, the first common lesion area can be determined from multiple lesion areas to be classified in the portal phase image based on the blood supply mode of benign tumors and liver tissues, that is, lesion tissue with a small contrast with liver tissue is found as a benign tumor lesion, and the area where the lesion is located is determined as the first common lesion area, which can accurately and objectively find the first common lesion area, thereby improving the recognition accuracy of benign tumor lesions. When determining the tumor lesion area, based on the characteristics of the existence of necrotic areas in malignant tumors in portal phase images, a first marking line segment can be set, and the gradient of the pixel values ​​of the pixels on the first marking line segment can be used to determine whether the first marking line segment crosses the necrotic area, thereby determining whether there is a necrotic area in the lesion area to be classified, thereby improving the recognition accuracy of the tumor lesion area where the malignant tumor is located. When determining the warning level score, the integrity of the film can be used to describe the severity of the tumor in the tumor area, and the gradient of the pixel values ​​of the pixels on the second marking line segment can be used to determine whether the second marking line segment crosses the capsule. The capsule discrimination conditions can be set according to the characteristics of the pixel values ​​of the capsule, liver tissue and tumor tissue to determine whether there is a capsule, and then determine the integrity of the capsule. The warning level score can be determined by the maximum value of the warning score of each tumor area, thereby attracting the attention of patients and doctors, helping to promptly discover more serious tumor lesions, and improving the accuracy of identifying more serious malignant tumors.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only and do not limit the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other embodiments can be obtained based on these drawings without creative work.

[0022] Figure 1 A schematic diagram of a process flow of a CT image lesion detection method based on artificial intelligence according to an embodiment of the present invention is exemplarily shown;

[0023] Figure 2 An artificial intelligence-based CT image lesion detection system according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0026] Figure 1 A flow chart of a method for detecting lesions in CT images based on artificial intelligence according to an embodiment of the present invention is exemplarily shown, and the method comprises:

[0027] Step S101, obtaining a plain scan CT image and an enhanced CT image of the liver, wherein the enhanced CT image includes an arterial phase image, a portal venous phase image, and a delayed phase image of the liver;

[0028] Step S102, determining first position information of a pending lesion area in the plain scan CT image by using an image detection model;

[0029] Step S103, screening out a tumor lesion region to be determined from a plurality of lesion regions to be determined, and second position information of the tumor lesion region to be determined, according to the first position information and the arterial phase image;

[0030] Step S104, classifying the undetermined tumor lesion region according to the second position information and the portal phase image to determine the tumor lesion region and the common lesion region;

[0031] Step S105, determining a warning level score of the tumor lesion area according to the tumor lesion area and the delayed phase image;

[0032] Step S106, generating a liver lesion detection report according to the warning level score.

[0033] According to the artificial intelligence-based CT image lesion detection method of an embodiment of the present invention, lesions in the liver can be identified based on plain scan CT images, and lesions can be screened by enhancing the arterial phase images in the CT images, thereby screening out areas of pending tumor lesions suspected of having tumors, and then classifying them through portal venous phase images, thereby accurately determining the tumor areas therein, and further, evaluating the severity of lesions in the tumor area through delayed phase images, thereby generating a warning level score to prompt patients or doctors, and automatically executing complex screening and evaluation processes to reduce the probability of misjudgment.

[0034] According to one embodiment of the present invention, in step S101, a plain scan CT image and an enhanced CT image of the patient's liver area can be obtained. For example, a plain scan CT image can be obtained before the injection of a contrast agent, and an arterial phase image in the enhanced CT image can be obtained about 35 seconds after the injection of the contrast agent, and a portal phase image in the enhanced CT image can be obtained about 75 seconds after the injection of the contrast agent. A delayed phase image in the enhanced CT image can be obtained about 10 minutes after the injection of the contrast agent, wherein the contrast agent is injected into the patient's artery.

[0035] According to one embodiment of the present invention, in step S102, the image detection model is a neural network model based on artificial intelligence, for example, a convolutional neural network model that can be used to process images, and the present invention does not limit the specific type of the image detection model. Through the image detection model, the plain scan CT image can be processed to obtain the first position information of the pending lesion area in the plain scan CT image, and the first position information can be the position information of the rectangular frame that selects the pending lesion area. The pending lesion area is the area where the lesion site on the liver is located, and the lesion may be a tumor (for example, a benign tumor or a malignant tumor), or it may be a lesion such as a hemangioma.

[0036] According to one embodiment of the present invention, in step S103, the image detection model can scan the pending lesion area in the CT image, that is, the area where there may be diseases and other problems, but it is difficult to determine the type and dangerousness of the lesion. Therefore, the type and dangerousness of the lesion in each pending lesion area can be determined by enhancing the CT image.

[0037] According to one embodiment of the present invention, based on the first position information and the arterial phase image, a to-be-determined tumor lesion area is screened out from multiple to-be-determined lesion areas, and the second position information of the to-be-determined tumor lesion area is determined, including: in the arterial phase image, according to the first position information, determining a to-be-determined area corresponding to the to-be-determined lesion area in the plain scan CT image; determining the area where the abdominal aorta is located in the arterial phase image through the image detection model; determining the centroid of the to-be-determined area, and determining the lesion center area in the to-be-determined area with the centroid of the to-be-determined area as the center; determining a first area where the liver is located in the arterial phase image; in the first area where the liver is located, determining a first liver background area outside the to-be-determined area; based on the pixel values ​​of the pixel points in the lesion center area, the pixel values ​​of the pixel points in the first liver background area, and the pixel values ​​of the pixel points in the area where the abdominal aorta is located, screening out the to-be-determined tumor lesion area from multiple to-be-determined lesion areas, and determining the second position information of the to-be-determined tumor lesion area.

[0038] According to one embodiment of the present invention, the arterial phase image is a CT image obtained when arterial blood flows through an artery near the liver. In the arterial phase image, since the contrast agent is injected into the patient's artery, the position where the arterial blood flows has a higher pixel value in the arterial phase image, and the pixel values ​​of other positions are lower, that is, the position where the arterial blood flows has a higher contrast with other positions. In the liver, the liver tissue is supplied by veins, so in the arterial phase image, the pixel value of the area where the liver is located is lower, while the tumor is supplied by arteries. Therefore, if the lesion is a tumor, the pixel value of the area where the lesion is located in the arterial phase image is higher, and the contrast with other areas is higher. If, in the arterial phase image, the area where the lesion is located is similar to other areas of the liver, that is, the pixel value is lower, then the lesion is not a tumor, but may be other lesions, such as hemangioma.

[0039] According to an embodiment of the present invention, the position information corresponding to the first position information in the plain scan CT image can be determined in the arterial phase image, that is, the pending region corresponding to the pending lesion region. Based on the above analysis, the various pending regions in the arterial phase image are screened to screen out the pending tumor lesion region, and the lesions in other regions of the pending region except the pending tumor lesion region are other lesions, such as lesions such as hemangiomas.

[0040] According to one embodiment of the present invention, the region where the abdominal aorta is located in the arterial phase image can be determined by an image detection model. In the arterial phase image, the pixel value of the region where arterial blood flows is higher, so the pixel value of the region where the abdominal aorta is located is higher, and if there is a tumor lesion in the to-be-determined region, the pixel value of the region where the tumor lesion is located is close to the pixel value of the region where the abdominal aorta is located.

[0041] According to one embodiment of the present invention, if the lesion in the area to be determined is a tumor lesion, the boundary between it and the normal tissue is usually not obvious, and the pixel value at the edge position may not be high, that is, the pixel value of the area closer to the liver tissue, rather than the pixel value of the core area closer to the lesion. Therefore, the centroid of the area to be determined can be determined, and the center area of ​​the lesion in the area to be determined can be determined with the centroid of the area to be determined as the center. For example, a circular area with a radius of 0.5 or 0.2 times the short side length of the area to be determined (rectangular frame) and the centroid of the area to be determined as the center, the pixel value in this area is greatly different from the pixel value at the edge, which is more in line with the characteristics of the pixel value of the core area of ​​the tumor lesion. Further, the first area where the liver is located in the arterial phase image can be determined, and the area outside the area to be determined in the first area is the first liver background area, and its pixel value is low.

[0042] According to one embodiment of the present invention, based on the pixel values ​​of the pixel points in the lesion center area, the pixel values ​​of the pixel points in the first liver background area, and the pixel values ​​of the pixel points in the area where the abdominal aorta is located, a tumor lesion area to be determined and the second position information of the tumor lesion area to be determined are screened out from multiple lesion areas to be determined, including: obtaining the determination coefficient J of the i-th area to be determined according to formula (1): i ,

[0043]

[0044] Among them, P i，j is the pixel value of the jth pixel in the center area of ​​the i-th lesion, n i is the number of pixels in the center area of ​​the i-th lesion, P a，k is the pixel value of the kth pixel in the area where the abdominal aorta is located, n a is the number of pixels in the area where the abdominal aorta is located, P 1，b，t is the pixel value of the tth pixel in the first liver background area, n 1，b is the number of pixels in the first liver background area, i, j, k, t, n i 、n 1，b and n a are all positive integers, and j≤n i , k≤n a , t≤n 1，b ; Determine the pending area with a determination coefficient greater than or equal to a preset determination threshold as the pending tumor lesion area; and determine the position information of the pending tumor lesion area as the second position information.

[0045] According to one embodiment of the present invention, in formula (1), is the average pixel value of each pixel in the center area of ​​the i-th lesion, is the average pixel value of each pixel in the first liver background area, so is the average difference in pixel values ​​between the pixel points in the i-th lesion center area and the first liver background area. is the average pixel value of each pixel in the area where the abdominal aorta is located, so is the average difference in pixel values ​​between the central area of ​​the ith lesion and the area where the abdominal aorta is located. As described above, the pixel value of the area where the tumor lesion is located is close to the pixel value of the area where the abdominal aorta is located. Therefore, if the lesion in the ith undetermined area is a tumor, the average difference in pixel values ​​between the central area of ​​the ith lesion and the first liver background area is greater than the average difference in pixel values ​​between the central area of ​​the ith lesion and the first liver background area. Therefore, the greater the ratio of the two, the higher the probability that the lesion in the ith undetermined area is a tumor. The ratio of the two can be used as the determination coefficient of the ith undetermined area. Therefore, the greater the determination coefficient of the ith undetermined area, the greater the possibility that the lesion in the ith undetermined area is a tumor.

[0046] According to one embodiment of the present invention, a preset determination threshold may be set. If the determination coefficient of a certain area to be determined is greater than or equal to the preset determination threshold, the area to be determined may be determined as a tumor lesion area to be determined, and the position information of the tumor lesion area to be determined is the second position information.

[0047] In this way, based on the blood supply mechanism of the liver and the tumor, the pending tumor area can be determined from multiple pending areas of the arterial phase image, that is, the pending area whose median pixel value in the center area of ​​the lesion in the pending area is close to the pixel value of the pixel point in the area where the abdominal aorta is located and whose contrast with the first liver background area is high is found as the pending tumor area. The tumor area can be found accurately and objectively, thereby improving the accuracy of identifying tumor lesions.

[0048] According to one embodiment of the present invention, in step S104, undetermined tumor lesion areas are screened out in the arterial phase image, but these undetermined tumor lesion areas may be benign tumors or malignant tumors. In order to distinguish between benign tumors and malignant tumors, further judgment can be made through portal venous phase images (i.e., CT images obtained when the venous blood in the portal vein supplies blood to the liver).

[0049] According to one embodiment of the present invention, the tumor lesion region to be determined is classified according to the second position information and the portal venous phase image to determine the tumor lesion region and the common lesion region, including: determining the lesion region to be classified corresponding to the tumor lesion region to be determined in the portal venous phase image according to the second position information; determining the centroid of the lesion region to be classified, and determining the lesion center region to be classified with the centroid of the lesion region to be classified as the center; determining a second region in the portal venous phase image where the liver is located; determining a second liver background region outside the lesion region to be classified in the second region where the liver is located; and determining the second liver background region according to the second liver background region. The first common lesion area is determined from multiple lesion areas to be classified based on the pixel values ​​of the pixel points in the domain and the pixel values ​​of the pixel points in the central area of ​​the lesion to be classified; a plurality of first marking points are evenly set on the edges of the lesion areas to be classified except the first common lesion area; the first marking points are connected with the centroid of the lesion areas to be classified to obtain a plurality of first marking line segments; the lesion areas to be classified except the first common lesion area are classified according to the pixel values ​​of the pixel points on the first marking line segments to obtain the tumor lesion area and the second common lesion area; the first common lesion area and the second common lesion area are determined as the common lesion areas.

[0050] According to an embodiment of the present invention, the corresponding lesion region to be classified can be determined in the portal venous phase image based on the second position information of the tumor lesion region to be determined in the arterial phase image. The center region of the lesion to be classified in each lesion region to be classified can be obtained in a manner similar to the above lesion center region, and the second liver background region in the portal venous phase image can be obtained in a manner similar to the above acquisition of the first liver background region, and the acquisition method will not be repeated here.

[0051] According to one embodiment of the present invention, according to the pixel values ​​of the pixels in the second liver background area and the pixel values ​​of the pixels in the central area of ​​the lesion to be classified, determining a first common lesion area from multiple lesion areas to be classified includes: obtaining a discrimination score S of the sth tumor lesion area to be determined according to formula (2): s ,

[0052]

[0053] Among them, P s，v is the pixel value of the vth pixel in the central area of ​​the sth lesion to be classified, n s is the number of pixels in the central area of ​​the sth lesion to be classified, P 2，b，w is the pixel value of the w-th pixel in the second liver background area, n 2，b is the number of pixels in the second liver background area, v≤n s,w≤n 2，b , and v, w, n 2，b and n s are all positive integers; the lesion area to be classified whose discrimination score is less than the preset discrimination threshold is determined as the first common lesion area.

[0054] According to one embodiment of the present invention, since the liver is supplied by veins, in the portal phase image, the pixel values ​​of the pixels in the second liver background area are higher, while the malignant tumor is supplied by arteries. Therefore, the pixel values ​​of the area where the malignant tumor is located are lower, that is, the contrast between the liver tissue and the malignant tumor is still large, while in the portal phase image, the contrast between the benign tumor and the liver tissue is smaller. Therefore, in formula (2), is the average value of the pixel values ​​of each pixel in the central area of ​​the sth lesion to be classified, is the average value of the pixel values ​​of each pixel in the second liver background area, is the difference between the average pixel value of each pixel point in the central area of ​​the sth lesion to be classified and the average pixel value of each pixel point in the second liver background area. The discrimination score shown in formula (2) is the relative difference between the average pixel value of each pixel point in the central area of ​​the sth lesion to be classified and the average pixel value of each pixel point in the second liver background area. As analyzed above, if the lesion in the sth lesion area to be classified is a benign tumor, the relative difference is small, and a preset discrimination threshold can be set. If the discrimination score corresponding to the sth lesion area to be classified is less than the preset discrimination threshold, the lesion in the sth lesion area to be classified can be determined as a benign tumor, and the sth lesion area to be classified can be determined as the first common lesion area. Further, it can be judged whether each lesion area to be classified is the first common lesion area in the above manner.

[0055] In this way, based on the blood supply mode of benign tumors and liver tissues, the first common lesion area can be determined from multiple lesion areas to be classified in the portal phase image. That is, lesion tissue with smaller contrast with liver tissue is found as a benign tumor lesion, and the area where the lesion is located is determined as the first common lesion area. The first common lesion area can be found accurately and objectively, thereby improving the recognition accuracy of benign tumor lesions.

[0056] According to one embodiment of the present invention, the first common lesion area where the benign tumor lesion is located is found from the multiple lesion areas to be classified above, and the lesion areas to be classified except the first common lesion area can be further classified to determine the types of lesions in other lesion areas to be classified.

[0057] According to one embodiment of the present invention, multiple first marking points can be evenly set on the edges of other lesion areas to be classified, that is, the spacing between adjacent first marking points is fixed. For example, the number of first marking points can be determined, for example, 20. The edge of a lesion area to be classified is the edge of a rectangular frame, and the total perimeter of the edge is 200 pixels, then the adjacent first marking points are spaced 9 pixels apart.

[0058] According to an embodiment of the present invention, the first marking point and the centroid of the lesion region to be classified may be connected to obtain a plurality of first marking line segments, and then the lesion region to be classified may be classified using the first marking line segments.

[0059] According to one embodiment of the present invention, according to the pixel values ​​of the pixels on the first marking line segment, the lesion regions to be classified other than the first common lesion region are classified to obtain the tumor lesion region and the second common lesion region, including: obtaining the classification score S of the uth lesion region to be classified among the lesion regions to be classified other than the first common lesion region according to formula (3): C，u ,

[0060]

[0061] Among them, (x u，h，g ,y u，h，g ) is the coordinate of the g-th pixel point on the h-th first marked line segment of the u-th lesion area to be classified, p(x u，h，g ,y u，h，g ) is the pixel value of the g-th pixel point on the h-th first marking line segment of the u-th lesion area to be classified, is the direction vector of the hth first marked line segment of the uth lesion area to be classified, is the pixel value of the g-th pixel on the h-th first marked line segment of the u-th lesion area to be classified along the direction vector The gradient of G, α1 is the first change rate threshold, G u，h is the number of pixels on the hth first marked line segment of the uth lesion area to be classified, Hu is the number of marked line segments of the uth lesion area to be classified, g≤G u，h , h≤H u , and g, h, G u，h and H u are all positive integers, and if is a conditional function; the lesion area to be classified whose classification score is greater than or equal to the preset classification threshold is determined as the tumor lesion area, and the lesion area to be classified whose classification score is less than the preset classification threshold is determined as the second common lesion area.

[0062] According to one embodiment of the present invention, in formula (3), is the modulus of the gradient, which can be used to represent the rate of change of the pixel value of the pixel point on the first marking line segment along the direction of the first marking line segment. If the lesion in the lesion area to be classified is a malignant tumor, there may be multiple necrotic areas in the area, and the pixel values ​​of these necrotic areas are different from the pixel values ​​of the tumor tissue and the pixel values ​​of the liver tissue. Therefore, if the first marking line segment passes through the position of the necrotic area, the pixel value of the pixel point on the first marking line segment will change multiple times along the direction of the first marking line segment. For example, at the boundary between the necrotic area and the area where the tumor tissue is located, the pixel value of the pixel point on the first marking line segment will change, and at the boundary between the tumor tissue and the liver tissue, the pixel value of the pixel point on the first marking line segment will change. Furthermore, since a plurality of radially distributed first marking line segments are set with the centroid of the lesion region to be classified as the center, if there is a necrotic region, one or more first marking line segments will cross the necrotic region, that is, on these first marking line segments, the pixel values ​​of the pixels will change multiple times. In other words, on these first marking line segments, the modulus (i.e., the rate of change) of the gradient at the positions of some pixels will be higher than or equal to the first rate of change threshold. In this case, the conditional function The value is 1 if yes, otherwise it is 0.

[0063] therefore, Then it means the number of times the pixel value of the pixel point on the hth first marked line segment undergoes a significant change, that is, the number of times the boundary between the necrotic area and the tumor tissue area or the boundary between the tumor tissue and the liver tissue is crossed. Formula (3) takes the average of these numbers, that is, the average of the number of times each first marked line segment crosses the above boundary. If there is no necrotic area, each first marked line segment crosses the boundary between the tumor tissue and the liver tissue only once, that is, the classification score is 1. Therefore, the preset classification threshold can be set to a value greater than 1. If the classification score is greater than or equal to the preset classification threshold, it means that there is a necrotic area in the lesion area to be classified, and the area to be classified can be classified as a tumor lesion area. Otherwise, the area to be classified can be classified as a second common lesion area. Further, both the first common lesion area and the second common lesion area can be regarded as common lesion areas where benign tumors exist, while the tumor lesions in the tumor lesion area are malignant tumors.

[0064] In this way, based on the characteristic of the existence of necrotic areas in malignant tumors in portal venous phase images, a first marking line segment can be set, and the gradient of the pixel values ​​of the pixels on the first marking line segment can be used to determine whether the first marking line segment crosses the necrotic area, thereby determining whether there is a necrotic area in the lesion area to be classified, thereby improving the recognition accuracy of the tumor lesion area where the malignant tumor is located.

[0065] According to an embodiment of the present invention, in step S105, after the tumor lesion area is determined, the delayed phase image can be used to determine the severity of the malignant tumor lesion and obtain a corresponding warning level score, thereby prompting the patient or doctor.

[0066] According to one embodiment of the present invention, a warning level score of the tumor lesion area is determined based on the tumor lesion area and the delayed phase image, including: determining the tumor area in the delayed phase image based on the position information of the tumor lesion area in the portal phase image; determining the centroid of the tumor area and evenly setting a plurality of second marking points on the edge of the tumor area; connecting the second marking points and the centroid of the tumor area to obtain a plurality of second marking line segments; and determining the warning level score of the tumor lesion area based on the pixel values ​​of the pixel points on the second marking line segments.

[0067] According to one embodiment of the present invention, the tumor region in the delayed phase image can be determined based on the position information of the tumor lesion region in the portal phase image, that is, the region where the tumor lesion is located can be found in the delayed phase image. A plurality of second marking line segments can be set in a manner similar to the first marking line segment, which will not be described in detail herein.

[0068] According to one embodiment of the present invention, determining the warning level score of the tumor lesion area according to the pixel value of the pixel point on the second marking line segment includes: obtaining the capsule number F on the qth second marking line segment of the pth tumor area according to formula (4): p，q ,

[0069]

[0070] Among them, (x p，q，f ,y p，q，f ) is the coordinate of the fth pixel point on the qth second marking line segment of the pth tumor region, p(x p，q，f ,y p，q，f ) is the pixel value of the f-th pixel on the q-th second marking line segment of the p-th tumor region, is the direction vector of the qth second marking line segment of the pth tumor region, is the pixel value of the f-th pixel on the q-th second marking line segment of the p-th tumor region along the direction vector , α2 is the second preset change rate, if is the conditional function; determine whether the number of non-zero capsule numbers on the qth second marking line segment of the pth tumor region is equal to 2; if the number of non-zero capsule numbers on the qth second marking line segment of the pth tumor region is equal to 2, then determine the capsule discrimination condition of the qth second marking line segment of the pth tumor region according to formula (5),

[0071]

[0072] Among them, F p，q，1 is a non-zero capsule number on the qth second marking line segment of the pth tumor region, F p，q，2 is another capsule number among the non-zero capsule numbers on the qth second marking line segment of the pth tumor region, and F p，q，1 <F p，q，2 , n p.q is the number of pixel points of the qth second marked line segment of the pth tumor region; if the qth second marked line segment of the pth tumor region meets the capsule discrimination condition of the qth second marked line segment of the pth tumor region, then it is determined that the qth second marked line segment of the pth tumor region crosses the capsule; the number of second marked line segments crossing the capsule is counted; the ratio of the number of second marked line segments crossing the capsule to the total number of second marked line segments in the pth tumor region is determined as the warning score of the pth tumor region; the maximum value of the warning scores of each tumor region is determined as the warning level score.

[0073] According to one embodiment of the present invention, if the malignant tumor is hepatocellular carcinoma, adenoma, cystadenoma, cystadenocarcinoma or the like, these tumors all have a capsule, which is a fibrous tissue and is not obvious in the arterial phase image and the portal phase image, but has a higher pixel value in the delayed phase image. Therefore, the severity of the tumor can be determined based on the presence or absence of a capsule and the integrity of the capsule. Tumors with higher capsule integrity exist for a longer time and are more dangerous, and therefore require key monitoring and prompting.

[0074] According to one embodiment of the present invention, in formula (4), the conditional function Indicates that if it satisfies If the condition is met, the conditional function value is f, otherwise it is 0. That is, when the pixel value change rate at the position of the fth pixel on the second marking line segment is large, the serial number f of the pixel is recorded. Since the capsule has a certain thickness, if there is a capsule, the second marking line segment will cross the boundary between the capsule and the tumor tissue, and then cross the boundary between the capsule and the liver tissue, that is, the number of times the pixel value of the pixel on the second marking line segment exceeds the second preset change rate is 2 times. Moreover, in the delayed phase image, the contrast agent flows out of the area where the liver is located, and the outflow time of the area where the tumor is located is earlier than the outflow time of the area where the liver is located. Therefore, a part of the contrast agent may remain in the liver tissue, while the tumor tissue has already discharged the contrast agent. Therefore, the pixel value of the pixel in the tumor tissue is lower than the pixel value of the pixel in the liver tissue, and the pixel value at the position of the capsule is higher than the pixel value of the pixel in the tumor tissue and the pixel value of the pixel in the liver tissue. Therefore, the capsule discrimination condition can be set based on formula (5), that is, the average value of the pixel values ​​of the pixels between the first pixel point on the second marking line segment and the pixel point corresponding to the first capsule number on the second marking line segment Less than the average value of the pixel values ​​between the pixel corresponding to the second envelope number on the second marking line segment and the last pixel and is smaller than the average value of the pixel points between the pixel point corresponding to the first capsule number and the pixel point corresponding to the second capsule number on the second marking line segment. If a second marking line segment meets the capsule discrimination condition described in formula (5), then in each tissue passed by the second marking line segment, if the pixel value of the pixel point in the tumor tissue is lower than the pixel value of the pixel point in the liver tissue, and the pixel value at the position where the capsule is located is higher than the pixel value of the pixel point in the tumor tissue and the pixel value of the pixel point in the liver tissue, it can be determined that the second marking line segment passes through the capsule, and it can also be determined that the capsule exists at the position of the second marking line segment.

[0075] According to one embodiment of the present invention, the number of second marker line segments that meet the capsule discrimination condition can be determined from multiple second marker line segments, and the ratio of this number to the total number of second marker line segments in the pth tumor region can be used to describe the integrity of the capsule in the pth tumor region, that is, the closer the ratio is to 1, the more complete the capsule is, and the ratio can also be used as a warning score for the pth tumor region to alert patients and doctors. Further, the maximum value of the warning score of each tumor region can be determined as the warning level score, that is, the warning score of the most serious tumor region is used to determine the warning level score, thereby attracting the attention of patients and doctors and helping to promptly discover more serious tumor lesions.

[0076] In this way, the integrity of the film can be used to describe the severity of the tumor in the tumor area, and the gradient of the pixel values ​​of the pixels on the second marking line segment can be used to determine whether the second marking line segment crosses the capsule. The capsule discrimination conditions can be set according to the characteristics of the pixel values ​​of the capsule, liver tissue and tumor tissue to determine whether there is a capsule, and then the integrity of the capsule can be determined. The warning level score can be determined by the maximum value of the warning score of each tumor area, thereby attracting the attention of patients and doctors, helping to timely discover more serious tumor lesions, and improving the recognition accuracy of more serious malignant tumors.

[0077] According to an embodiment of the present invention, in step S106, a liver lesion detection report may be generated based on the warning level score, which may be used to describe whether a malignant tumor exists and the severity of the malignant tumor.

[0078] According to the artificial intelligence-based CT image lesion detection method of the embodiment of the present invention, lesions in the liver can be identified based on plain scan CT images, and lesions can be screened by enhancing the arterial phase images in the CT images, thereby screening out the suspected tumor lesion areas, and then classifying them through the portal phase images, thereby accurately determining the tumor areas therein. Furthermore, the severity of the lesions in the tumor area can be evaluated through the delayed phase images, thereby generating a warning level score to prompt the patient or doctor, and the complex screening and evaluation process can be automatically performed to reduce the probability of misjudgment. When determining the tumor area to be determined, the tumor area to be determined can be determined from multiple areas to be determined in the arterial phase image based on the blood supply mechanism of the liver and the tumor, that is, the center pixel value of the lesion center area in the area to be determined is close to the pixel value of the pixel point in the area where the abdominal aorta is located, and the contrast between the center pixel value and the first liver background area is high. As the tumor area to be determined, the tumor area can be accurately and objectively found, and the accuracy of identifying the tumor lesions can be improved. When determining the first common lesion area, the first common lesion area can be determined from multiple lesion areas to be classified in the portal phase image based on the blood supply mode of benign tumors and liver tissues, that is, lesion tissue with a small contrast with liver tissue is found as a benign tumor lesion, and the area where the lesion is located is determined as the first common lesion area, which can accurately and objectively find the first common lesion area, thereby improving the recognition accuracy of benign tumor lesions. When determining the tumor lesion area, based on the characteristics of the existence of necrotic areas in malignant tumors in portal phase images, a first marking line segment can be set, and the gradient of the pixel values ​​of the pixels on the first marking line segment can be used to determine whether the first marking line segment crosses the necrotic area, thereby determining whether there is a necrotic area in the lesion area to be classified, thereby improving the recognition accuracy of the tumor lesion area where the malignant tumor is located. When determining the warning level score, the integrity of the film can be used to describe the severity of the tumor in the tumor area, and the gradient of the pixel values ​​of the pixels on the second marking line segment can be used to determine whether the second marking line segment crosses the capsule. The capsule discrimination conditions can be set according to the characteristics of the pixel values ​​of the capsule, liver tissue and tumor tissue to determine whether there is a capsule, and then determine the integrity of the capsule. The warning level score can be determined by the maximum value of the warning score of each tumor area, thereby attracting the attention of patients and doctors, helping to promptly discover more serious tumor lesions, and improving the accuracy of identifying more serious malignant tumors.

[0079] Figure 2 The following is an exemplary illustration of a CT image lesion detection system based on artificial intelligence according to an embodiment of the present invention, comprising:

[0080] An image acquisition module, used to acquire a plain scan CT image and an enhanced CT image of the liver, wherein the enhanced CT image includes an arterial phase image, a portal venous phase image, and a delayed phase image of the liver;

[0081] A detection module, used to determine first position information of a pending lesion area in the plain scan CT image through an image detection model;

[0082] a screening module, configured to screen out a tumor lesion region to be determined from a plurality of lesion regions to be determined, and second position information of the tumor lesion region to be determined, according to the first position information and the arterial phase image;

[0083] A classification module, used for classifying the undetermined tumor lesion area according to the second position information and the portal phase image, and determining the tumor lesion area and the common lesion area;

[0084] A warning level scoring module, used to determine a warning level score of the tumor lesion area according to the tumor lesion area and the delayed phase image;

[0085] A reporting module is used to generate a liver lesion detection report based on the warning level score.

[0086] It should be understood by those skilled in the art that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A CT image lesion detection method based on artificial intelligence, characterized in that: include: Acquiring a plain scan CT image and an enhanced CT image of the liver, wherein the enhanced CT image includes an arterial phase image, a portal venous phase image, and a delayed phase image of the liver; Determining first position information of a to-be-determined lesion area in the plain scan CT image by using an image detection model; According to the first position information and the arterial phase image, a tumor lesion region to be determined is screened out from a plurality of lesion regions to be determined, and second position information of the tumor lesion region to be determined is determined; Classifying the undetermined tumor lesion region according to the second position information and the portal venous phase image to determine a tumor lesion region and a common lesion region; Determining a warning level score of the tumor lesion area according to the tumor lesion area and the delayed phase image; A liver lesion detection report is generated based on the warning level score.

2. The method for detecting lesions in CT images based on artificial intelligence according to claim 1, characterized in that: According to the first position information and the arterial phase image, a tumor lesion region to be determined is screened out from a plurality of lesion regions to be determined, and second position information of the tumor lesion region to be determined is determined, including: In the arterial phase image, determining a pending area corresponding to a pending lesion area in the plain scan CT image according to the first position information; Determining the region where the abdominal aorta is located in the arterial phase image by using the image detection model; Determine the centroid of the area to be determined, and determine the center area of ​​the lesion in the area to be determined with the centroid of the area to be determined as the center; determining a first region in the arterial phase image where the liver is located; In the first region where the liver is located, determining a first liver background region outside the region to be determined; According to the pixel values ​​of the pixel points in the lesion center area, the pixel values ​​of the pixel points in the first liver background area and the pixel values ​​of the pixel points in the area where the abdominal aorta is located, a pending tumor lesion area is screened out from multiple pending lesion areas, and the second position information of the pending tumor lesion area is determined.

3. The method for detecting lesions in CT images based on artificial intelligence according to claim 2, characterized in that: According to the pixel values ​​of the pixels in the lesion center area, the pixel values ​​of the pixels in the first liver background area, and the pixel values ​​of the pixels in the area where the abdominal aorta is located, a pending tumor lesion area is screened out from a plurality of pending lesion areas, and second position information of the pending tumor lesion area is determined, including: According to the formula Get the determination coefficient J of the i-th undetermined area i , where P i，j is the pixel value of the jth pixel in the center area of ​​the i-th lesion, n i is the number of pixels in the center area of ​​the i-th lesion, P a，k is the pixel value of the kth pixel in the area where the abdominal aorta is located, n a is the number of pixels in the area where the abdominal aorta is located, P 1，b，t is the pixel value of the tth pixel in the first liver background area, n 1，b is the number of pixels in the first liver background area, i, j, k, t, n i 、n 1，b and n a are all positive integers, and j≤n i , k≤n a , t≤n 1，b ; Determine the undetermined area whose determination coefficient is greater than or equal to the preset determination threshold as the undetermined tumor lesion area; The position information of the undetermined tumor focus area is determined as the second position information.

4. The method for detecting lesions in CT images based on artificial intelligence according to claim 1, characterized in that: Classifying the undetermined tumor lesion region according to the second position information and the portal venous phase image to determine the tumor lesion region and the common lesion region includes: Determine, in the portal phase image, a lesion region to be classified corresponding to the tumor lesion region to be determined according to the second position information; Determine the centroid of the lesion area to be classified, and determine the central area of ​​the lesion to be classified with the centroid of the lesion area to be classified as the center; determining a second region where the liver is located in the portal venous phase image; In the second region where the liver is located, determining a second liver background region outside the lesion region to be classified; Determining a first common lesion region from a plurality of lesion regions to be classified according to pixel values ​​of pixel points in the second liver background region and pixel values ​​of pixel points in the central region of the lesion to be classified; Evenly setting a plurality of first marking points on the edge of the lesion region to be classified except the first common lesion region; Connecting the first marking point and the centroid of the lesion area to be classified to obtain a plurality of first marking line segments; Classifying the lesion regions to be classified except the first common lesion region according to the pixel values ​​of the pixels on the first marking line segment to obtain the tumor lesion region and the second common lesion region; The first common lesion area and the second common lesion area are determined as the common lesion area.

5. The method for detecting lesions in CT images based on artificial intelligence according to claim 4, characterized in that: Determining a first common lesion region from a plurality of lesion regions to be classified according to the pixel values ​​of the pixel points in the second liver background region and the pixel values ​​of the pixel points in the central region of the lesion to be classified comprises: According to the formula Get the discrimination score S of the sth undetermined tumor lesion area s , where P s，v is the pixel value of the vth pixel in the central area of ​​the sth lesion to be classified, n s is the number of pixels in the central area of ​​the sth lesion to be classified, P 2，b，w is the pixel value of the w-th pixel in the second liver background area, n 2，b is the number of pixels in the second liver background area, v≤n s ,w≤n 2，b , and v, w, n 2，b and n s All are positive integers; The lesion region to be classified whose discrimination score is less than the preset discrimination threshold is determined as the first common lesion region.

6. The method for detecting lesions in CT images based on artificial intelligence according to claim 4, characterized in that: Classifying the lesion regions to be classified except the first common lesion region according to the pixel values ​​of the pixels on the first marking line segment to obtain the tumor lesion region and the second common lesion region, including: According to the formula Obtain the classification score S of the uth lesion area to be classified among the lesion areas to be classified except the first common lesion area C，u , where (x u，h，g ,y u，h，g ) is the coordinate of the g-th pixel point on the h-th first marked line segment of the u-th lesion area to be classified, p(x u，h，g ,y u，h，g ) is the pixel value of the g-th pixel on the h-th first marking line segment of the u-th lesion area to be classified, is the direction vector of the hth first marked line segment of the uth lesion area to be classified, is the pixel value of the g-th pixel on the h-th first marked line segment of the u-th lesion area to be classified along the direction vector The gradient of G, α1 is the first change rate threshold, G u，h is the number of pixels on the hth first marked line segment of the uth lesion area to be classified, H u is the number of marked line segments in the uth lesion area to be classified, g≤G u，h , h≤H u , and g, h, G u，h and H u All are positive integers, if is a conditional function; The lesion region to be classified whose classification score is greater than or equal to the preset classification threshold is determined as a tumor lesion region, and the lesion region to be classified whose classification score is less than the preset classification threshold is determined as a second common lesion region.

7. The method for detecting lesions in CT images based on artificial intelligence according to claim 1, characterized in that: Determining a warning level score of the tumor lesion area according to the tumor lesion area and the delayed phase image includes: Determining the tumor region in the delayed phase image according to the position information of the tumor lesion region in the portal phase image; Determining the centroid of the tumor region, and evenly setting a plurality of second marking points on the edge of the tumor region; connecting the second marking point and the centroid of the tumor region to obtain a plurality of second marking line segments; The warning level score of the tumor lesion area is determined according to the pixel values ​​of the pixel points on the second marking line segment.

8. The method for detecting lesions in CT images based on artificial intelligence according to claim 7, characterized in that: Determining the warning level score of the tumor lesion area according to the pixel values ​​of the pixels on the second marking line segment includes: According to the formula Get the capsule number F on the qth second marking line segment of the pth tumor area p，q , where (x p，q，f ,y p，q，f ) is the coordinate of the fth pixel point on the qth second marking line segment of the pth tumor region, p(x p，q，f ,y p，q，f ) is the pixel value of the f-th pixel on the q-th second marking line segment of the p-th tumor region, is the direction vector of the qth second marking line segment of the pth tumor region, is the pixel value of the f-th pixel on the q-th second marking line segment of the p-th tumor region along the direction vector The gradient of , α2 is the second preset change rate, and if is the conditional function; Determine whether the number of capsule serial numbers that are not 0 on the qth second marking line segment of the pth tumor region is equal to 2; If the number of non-zero capsule numbers on the qth second marking line segment of the pth tumor region is equal to 2, then according to the formula Determine the capsule discrimination condition of the qth second marking line segment of the pth tumor region, where F p，q，1 is a non-zero capsule number on the qth second marking line segment of the pth tumor region, F p，q，2 is another capsule number among the non-zero capsule numbers on the qth second marking line segment of the pth tumor region, and F p，q，1 <F p，q，2 , n p，q is the number of pixels of the qth second marking line segment of the pth tumor region; If the qth second marking line segment of the pth tumor region meets the capsule discrimination condition of the qth second marking line segment of the pth tumor region, then determining that the qth second marking line segment of the pth tumor region crosses the capsule; Counting the number of second marking line segments crossing the envelope; Determine the ratio of the number of second marking line segments crossing the capsule to the total number of second marking line segments in the p-th tumor region as the warning score of the p-th tumor region; The maximum value of the warning scores of each tumor region is determined as the warning level score.

9. An artificial intelligence-based CT image lesion detection system for executing the method according to any one of claims 1 to 8, characterized in that: include: An image acquisition module, used to acquire a plain scan CT image and an enhanced CT image of the liver, wherein the enhanced CT image includes an arterial phase image, a portal venous phase image, and a delayed phase image of the liver; A detection module, used to determine first position information of a pending lesion area in the plain scan CT image through an image detection model; a screening module, configured to screen out a tumor lesion region to be determined from a plurality of lesion regions to be determined according to the first position information and the arterial phase image, and determine second position information of the tumor lesion region to be determined; A classification module, used for classifying the undetermined tumor lesion area according to the second position information and the portal phase image, and determining the tumor lesion area and the common lesion area; A warning level scoring module, used to determine a warning level score of the tumor lesion area according to the tumor lesion area and the delayed phase image; A reporting module is used to generate a liver lesion detection report based on the warning level score.

Images