Prostate MRI image auxiliary diagnosis method and system based on deep learning

Abstract

The invention relates to the technical field of prostate MRI image segmentation, in particular to a prostate MRI image auxiliary diagnosis method and system based on deep learning. The method comprises the following steps: acquiring all angular points in a prostate MRI image; obtaining a screening retention index according to the gradient and gray level distribution of the angular points, and further obtaining to-be-analyzed angular points; establishing an angular point window of each angular point; obtaining a connection coefficient according to the position relationship between the nearest to-be-analyzed angular points of the target angular point and the angular point window difference; obtaining all local areas in the image; according to the distance between each to-be-analyzed angular point and the local area, obtaining the local area to which each to-be-analyzed angular point belongs; combining the connection coefficient to obtain a prostate region; and obtaining a prostate training model, and carrying out auxiliary diagnosis on the image according to the model. According to the invention, the prostate region can be accurately identified, so that related personnel can carry out accurate auxiliary diagnosis on the prostate symptom of a patient according to the prostate MRI image.

Description

Technical Field

[0001] The present invention relates to the technical field of prostate MRI image segmentation, and in particular to a prostate MRI image-assisted diagnosis method and system based on deep learning. Background Art

[0002] Early diagnosis of prostate disease is crucial to treatment effectiveness and patient prognosis. The application of magnetic resonance imaging, which is commonly used in clinical practice, in prostate detection has gradually become popular in recent years, especially in the screening and evaluation of prostate cancer. Magnetic resonance imaging can evaluate the local invasion of tumors (such as whether it invades the seminal vesicles, surrounding tissues, etc.) and lymph node metastasis. Deep learning algorithms can be used for automated image segmentation and feature extraction. Through training models, deep learning technology can identify the presence of tumors and their malignancy in magnetic resonance images. Magnetic resonance imaging combined with deep learning provides new possibilities for the detection and diagnosis of prostate diseases. With the continuous advancement of technology, the combination of the two will open up broader prospects for the early detection and precise treatment of prostate cancer, and assist doctors in making more accurate judgments.

[0003] In the prior art, neural networks are often used to capture multi-scale features in prostate MRI images. However, the prostate MRI images input into the neural network are usually not properly preprocessed and cannot effectively capture certain scale features of the prostate MRI images, thereby affecting the identification of the prostate area. Ultimately, relevant personnel are unable to make accurate auxiliary diagnosis of the patient's prostate symptoms based on the prostate MRI images. Summary of the invention

[0004] In order to solve the technical problem that if the prostate MRI image input into the neural network is not properly preprocessed, the features of certain scales of the prostate MRI image cannot be effectively captured, thereby affecting the recognition of the prostate area, and ultimately the relevant personnel cannot accurately assist in the diagnosis of the patient's prostate symptoms based on the prostate MRI image, the purpose of the present invention is to provide a prostate MRI image-assisted diagnosis method and system based on deep learning, and the technical solutions adopted are as follows: A prostate MRI image-assisted diagnosis method based on deep learning, the method comprising: Obtaining prostate MRI images; Performing corner point detection on the prostate MRI image to obtain all corner points in the prostate MRI image; obtaining a screening and retention index for each corner point according to a gradient distribution and a grayscale distribution within a preset neighborhood of each corner point; screening all corner points according to the screening and retention index of all corner points to obtain all corner points to be analyzed among the corner points; selecting any corner point to be analyzed as a target corner point; adjusting an initial preset window of the target corner point according to the screening and retention index of the target corner point to obtain a corner point window of the target corner point; obtaining a connection coefficient of the target corner point according to a positional relationship between the target corner point and the nearest corner point to be analyzed and a difference in the corner point window; Perform edge detection on the prostate MRI image to obtain all local areas; obtain the local area to which each corner point to be analyzed belongs based on the distance between each corner point to be analyzed and the local area; obtain the prostate area based on the connection coefficient of the corner point to be analyzed in each local area; obtain a prostate training model based on all prostate areas; and perform auxiliary diagnosis on the prostate MRI image based on the prostate training model.

[0005] Furthermore, the method for obtaining the screening retention index includes: The screening retention index is obtained according to the screening retention index calculation formula, and the screening retention index calculation formula is as follows: In the formula, Represents the screening retention index of each corner point; Represents the gradient value of each corner point; Indicates the number of other pixels in the preset neighborhood of each corner point; Indicates the first The gradient value of each pixel; Represents the gray value of each corner point; Indicates the first The gray value of each pixel; represents the absolute value function.

[0006] Furthermore, the method for obtaining all corner points to be analyzed includes: Sort each corner point in the prostate MRI image from large to small according to the screening retention index to obtain a corner point sequence; In the corner point sequence, corner points are selected from left to right to form a first corner point sequence; When the ratio between the number of corner points in the first corner point sequence and the number of corner points in the first corner point sequence is a preset first threshold, all corner points in the first corner point sequence are used as corner points to be analyzed.

[0007] Furthermore, the method for obtaining the corner point window includes: Calculate the ratio between the screening retention index of the target corner point and the screening retention index and value of all the corner points to be analyzed as the corner point importance of the target corner point; The initial preset window is set to a square; the side length of the initial preset window is adjusted according to the corner point importance of the target corner point to obtain the side length of the corner point window of the target corner point; Get the corner point window of the target corner point according to the side length of the corner point window.

[0008] Further, adjusting the side length of the initial preset window according to the corner point importance of the target corner point to obtain the side length of the corner point window of the target corner point includes: The corner point window side length is obtained according to the corner point window side length formula, and the corner point window side length formula is as follows: In the formula, The length of the corner window of the target corner; A preset initial window representing the target corner point; Indicates the importance of the target corner point; Represents the ceiling function.

[0009] Furthermore, the method for obtaining the connection coefficient includes: The connection coefficient is obtained according to the connection coefficient calculation formula, and the connection coefficient calculation formula is as follows: In the formula, Represents the connection coefficient of the target corner point; The length of the corner point window representing the target corner point; The length of the corner point window representing the corner point to be analyzed that is closest to the target corner point; Indicates the distance between the target corner point and the nearest corner point to be analyzed; Represents Iverson brackets, where if the condition in the brackets is met, the value in the brackets is 1, and if the condition in the brackets is not met, the value in the brackets is 0.

[0010] Furthermore, a method for obtaining the local area to which each corner point to be analyzed belongs includes: Obtain the centroid points in all local areas; calculate the distance between the target corner point and each centroid point as the corner point distance between the target corner point and each local area; take the local area with the closest corner point distance as the local area to which the target corner point belongs; Traverse all corner points to be analyzed and obtain the local area to which each corner point to be analyzed belongs.

[0011] Furthermore, the method for acquiring the prostate region includes: Accumulate the connection coefficients of the corner points to be analyzed in each local area to obtain a comprehensive evaluation index for each local area; The local area with the largest comprehensive evaluation index is taken as the prostate area.

[0012] Furthermore, the method for obtaining the prostate training model includes: Obtaining a prostate thermal distribution map according to the prostate target area, matching the prostate MRI image with the prostate thermal distribution map, and obtaining a prostate training data set; A prostate training model is obtained according to the prostate training data set, and the prostate MRI image is segmented using the prostate training model.

[0013] A prostate MRI image-assisted diagnosis system based on deep learning, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above-mentioned prostate MRI image-assisted diagnosis method based on deep learning are implemented.

[0014] The present invention has the following beneficial effects: The present invention obtains a prostate MRI image; since there are a large number of important feature points in the prostate MRI image, all corner points are obtained; since other tissue regions may exist in the prostate MRI image in addition to the prostate region, and the prostate MRI image has a certain complexity, the influence of factors such as tissue density changes may cause some pixels existing on non-edges to be mistakenly identified as corner points, so it is necessary to screen the corner points, therefore, the screening retention index of each corner point is analyzed, and then all the corner points to be analyzed are obtained, wherein the screened-out corner points are considered to have nothing to do with the prostate region; since the corner points with different screening retention indexes are scored for their corresponding corner point windows, the corner point corresponding to the corner point window with a higher score has a closer connection with other corner points, and is more likely to be a corner point constituting the edge of the prostate region, therefore the initial prediction of the target corner point is performed according to the screening retention index of the target corner point. Set the window to adjust, and obtain the corner point window of the target corner point; because the corner point windows of different corner points may overlap in a certain area, a local area is regarded as composed of the corner point windows of the corner points. When there are still a large number of corner points in a local area after screening, and the larger the corner point windows of these corner points are, the more likely the corner point is the corner point of the prostate area, so the positional relationship between the corner point windows of the target corner point and other corner points is studied, and then the connection coefficient of the corner point is obtained; because the connection coefficient can reflect the local density of the corner point distribution, first the local area in the prostate MRI image is found, and the local area to which each corner point belongs is found, and then all the prostate areas are obtained according to the connection coefficient of the corner point to be analyzed in each local area; a prostate training model is obtained according to all the prostate areas; and an auxiliary diagnosis is performed on the prostate MRI image according to the prostate training model. The present invention can accurately identify the prostate area, so that relevant personnel can accurately perform auxiliary diagnosis on the patient's prostate symptoms according to the prostate MRI image. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flow chart of a prostate MRI image-assisted diagnosis method based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation, structure, features and effects of a prostate MRI image-assisted diagnosis method and system based on deep learning proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0018] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0019] The following is a detailed description of a prostate MRI image-assisted diagnosis method and system based on deep learning provided by the present invention in conjunction with the accompanying drawings.

[0020] See also Figure 1 , which shows a prostate MRI image-assisted diagnosis method based on deep learning provided by an embodiment of the present invention, the method comprising: Step S1: Acquire a prostate MRI image.

[0021] The embodiment of the present invention is mainly used in the auxiliary diagnosis scenario of lesions in the prostate area of ​​the patient. In order to facilitate medical personnel to judge the prostate condition of the patient, it is necessary to first obtain a medical image of the prostate. Since MRI images themselves have many advantages such as multi-parameter imaging, non-invasiveness and high soft tissue resolution, which can reduce unnecessary biopsies, in the embodiment of the present invention, a prostate MRI image of the patient is obtained.

[0022] In an embodiment of the present invention, a prostate MRI image is provided: the prostate MRI image in the data set is converted from a medical image format to a jpg image format and the image is grayed out. The specific process is to use pydicom to read The image is extracted and the pixel array is extracted, and then the pixel array is saved as a jpg format using pillow, and finally the jpg image is converted into a grayscale image using OpenCV and saved to obtain a prostate MRI image for subsequent steps. It should be noted that the process of obtaining a prostate MRI image is a technical means well known to those skilled in the art and will not be described in detail here.

[0023] Step S2: perform corner point detection on the prostate MRI image to obtain all corner points in the prostate MRI image; obtain the screening and retention index of each corner point according to the gradient distribution and grayscale distribution in the preset neighborhood of each corner point; screen all corner points according to the screening and retention index of all corner points to obtain all the corner points to be analyzed among the corner points; select any corner point to be analyzed as the target corner point; adjust the initial preset window of the target corner point according to the screening and retention index of the target corner point to obtain the corner point window of the target corner point; obtain the connection coefficient of the target corner point according to the positional relationship between the target corner point and the nearest corner point to be analyzed and the difference in the corner point window.

[0024] Since the prostate MRI image contains a large number of image features of different regions, it is necessary to first obtain important feature points in the prostate MRI image in order to facilitate the subsequent segmentation of the prostate region. In one embodiment of the present invention, corner point detection is performed on the prostate MRI image to obtain all corner points in the prostate MRI image, wherein the corner point detection algorithm adopts the ORB algorithm. In addition to this algorithm, other corner point detection algorithms can also be used. The algorithm is a technical means well known to those skilled in the art and is not limited or elaborated herein.

[0025] In one embodiment of the present invention, The neural network is used to segment the prostate region, and the following processes are all operated using this neural network.

[0026] Since there may be other tissue regions in addition to the prostate region in the prostate MRI image, and the prostate MRI image has a certain degree of complexity, the influence of factors such as tissue density changes will cause some pixels existing on non-edges to be mistakenly identified as corner points. Therefore, it is impossible to confirm the prostate region by directly using edge detection. Therefore, after obtaining all corner points, it is necessary to first screen the corner points. Since the gradient and grayscale of the corner points in different tissue regions are quite different, in the embodiment of the present invention, the screening and retention index of each corner point is obtained according to the gradient distribution and grayscale distribution in the preset neighborhood of each corner point, and the corner points are screened by the screening and retention index.

[0027] Preferably, in one embodiment of the present invention, the method for obtaining the screening retention index includes: The screening retention index is obtained according to the screening retention index calculation formula. The screening retention index calculation formula is as follows: In the formula, Represents the screening retention index of each corner point; Represents the gradient value of each corner point; Indicates the number of other pixels in the preset neighborhood of each corner point; Indicates the first The gradient value of each pixel; Represents the gray value of each corner point; Indicates the first The gray value of each pixel; represents the absolute value function.

[0028] In the calculation formula of the screening retention index, the gradient of the corner point The larger the value, the greater the edge strength of the corner point, and the more likely it is a corner point at the edge of the tissue area; the corner point presets the gradient accumulation value of other pixel points in the neighborhood The larger the value is, the greater the local edge strength around the corner point is. At this time, it is more likely to be a corner point at the edge of the tissue area and less likely to be a non-edge pixel point. Since the grayscale change of the boundary of the prostate area in the prostate MRI image is more intense, and the boundaries of other tissue areas often have lower grayscale local contrast, the grayscale change value within the preset neighborhood of the corner point is The larger the value is, the greater the grayscale variation around the corner point is, and the more likely the corner point is the prostate area corner point.

[0029] Corner points are screened using the screening retention index to obtain all corner points to be analyzed, and the screened-out corner points are considered to have nothing to do with the prostate region.

[0030] Preferably, in one embodiment of the present invention, the method for acquiring all corner points to be analyzed includes: Each corner point in the prostate MRI image is sorted from large to small according to the screening retention index to obtain a corner point sequence.

[0031] In the corner point sequence, corner points are selected from left to right to form a first corner point sequence; when the ratio between the number of corner points in the first corner point sequence and the number of corner points in the corner point sequence is a preset first threshold, all corner points in the first corner point sequence are used as corner points to be analyzed. In the embodiment of the present invention, the preset first threshold is set to 10%. It should be noted that the preset first threshold can be set voluntarily and is not limited here.

[0032] After obtaining the corner points to be analyzed, different corner point windows are established for each corner point to be analyzed, and the corresponding corner point windows are scored by the corner points with different screening and retention indicators. The corner point corresponding to the corner point window with a higher score has a closer connection with other corner points, and is more likely to be a corner point constituting the edge of the prostate region. Therefore, in the embodiment of the present invention, the initial preset window of the target corner point is first adjusted according to the screening and retention indicator of the target corner point to obtain the corner point window of the target corner point.

[0033] Preferably, in one embodiment of the present invention, the method for obtaining the corner point window includes: Calculate the ratio between the screening retention index of the target corner point and the screening retention index and value of all the corner points to be analyzed as the corner point importance of the target corner point; The initial preset window is set to a square, and the side length of the initial preset window is adjusted according to the corner point importance of the target corner point to obtain the side length of the corner point window of the target corner point. In one embodiment of the present invention, the formula for the side length of the corner point window is as follows: In the formula, The length of the corner window of the target corner; A preset initial window representing the target corner point; Indicates the importance of the target corner point; Represents the ceiling function.

[0034] In the corner point window side length formula, in order to ensure that the target corner point is always located in the middle of the corner point window, As , the corner window side length is set to ,in, The value of must be an odd number, in which case the target corner point is always located in the middle of the corner point window; It can ensure that the importance of corner points changes slightly, and can produce a sudden change in the length of the corner point window of the corner point, so that the corner point windows with different screening and retention indicators can produce greater differences.

[0035] In the embodiment of the present invention, the preset initial window of the target corner point is set to It should be noted that, in other embodiments of the present invention, the preset initial window can be set voluntarily, and is not limited here.

[0036] Get the corner point window of the target corner point according to the side length of the corner point window.

[0037] After adjusting the preset initial window of the target corner point, the corner point window of the target corner point is obtained. The corner point windows of different corner points may overlap to a certain extent. A local area is regarded as being composed of the corner point windows of the corner points. When there are still a large number of corner points in a local area after screening, and the larger the corner point windows of these corner points are, the more likely the corner point is a corner point of the prostate region, and the more the local area tends to be a prostate region. Therefore, it is necessary to study the positional relationship between the corner point windows of the target corner point and other corner points. Therefore, in the embodiment of the present invention, the connection coefficient of the target corner point is obtained according to the position distribution and corner point window difference between the target corner point and the nearest corner point to be analyzed.

[0038] Preferably, in one embodiment of the present invention, the method for obtaining the connection coefficient includes: The connection coefficient is obtained according to the connection coefficient calculation formula. The connection coefficient calculation formula is as follows: In the formula, Represents the connection coefficient of the target corner point; The length of the corner point window representing the target corner point; The length of the corner point window representing the corner point to be analyzed that is closest to the target corner point; Indicates the distance between the target corner point and the nearest corner point to be analyzed; Represents Iverson brackets, where if the condition in the brackets is met, the value in the brackets is 1, and if the condition in the brackets is not met, the value in the brackets is 0.

[0039] In the connection coefficient calculation formula, when the corner point window side length of the target corner point is larger, it means that the screening retention index of the target corner point is larger, and the target corner point is more likely to belong to the prostate region. The smaller the difference between the corner point window side length of the target corner point and the corner point window side length of the nearest corner point to be analyzed, the more similar the target corner point is to the nearest corner point to be analyzed, and the more likely it is to belong to the prostate region. At this time, the local density of the target corner point is greater, that is, the connection coefficient of the target corner point is larger; when the conditions in the Iverson brackets are met, that is, At this time, there is an overlapping area between the corner point windows of the two corner points. The larger the overlapping area, the closer the positional relationship between the target corner point and the nearest corner point to be analyzed is, and the larger the connection coefficient of the target corner point is.

[0040] Step S3: Perform edge detection on the prostate MRI image to obtain all local areas; obtain the local area to which each corner point to be analyzed belongs based on the distance between each corner point to be analyzed and the local area; obtain the prostate area based on the connection coefficient of the corner point to be analyzed in each local area; obtain a prostate training model based on all prostate areas; and perform auxiliary diagnosis on the prostate MRI image based on the prostate training model.

[0041] Because the connection coefficient of a corner point tends to be larger when the corner point window is larger and there are corner points with similar corner point windows and close positions around it, the connection coefficient can reflect the density of the corner point distribution in the local area. If there are many corner points with large connection coefficients in a local area, then the local area is likely to be a prostate area. Therefore, in the embodiment of the present invention, all local areas in the prostate MRI image are first obtained, and then the local area to which each corner point to be analyzed belongs is obtained according to the distance distribution between each corner point to be analyzed and the local area.

[0042] In the embodiment of the present invention, Sobel edge detection is performed on the prostate MRI image. Since the complexity of the prostate MRI image may lead to discontinuous edges, contour fitting is performed on the edge detection result to obtain all local areas. It should be noted that the edge detection algorithm is a technical means well known to those skilled in the art and is not limited or elaborated herein.

[0043] Preferably, in one embodiment of the present invention, a method for obtaining the local area to which each corner point to be analyzed belongs includes: Obtain the centroid points in all local regions; calculate the distance between the target corner point and each centroid point as the corner point distance between the target corner point and each local region; and take the local region with the shortest corner point distance as the local region to which the target corner point belongs.

[0044] Traverse all corner points to be analyzed and obtain the local area to which each corner point to be analyzed belongs.

[0045] Since the number of retained corner points in the prostate region is greater and the corner point window of the corner points is relatively larger, the connection coefficient of the corner points in the prostate region is greater, and the cumulative value of the connection coefficient is greater; while other tissue regions have fewer corner points and smaller corner point windows than the prostate region, the smaller the connection coefficient of the corner points, the smaller the cumulative value of the connection coefficient. Therefore, in the embodiment of the present invention, all prostate regions are obtained according to the connection coefficients of the corner points to be analyzed in each local region.

[0046] Preferably, in one embodiment of the present invention, the method for acquiring the prostate region includes: The connection coefficients of the corner points to be analyzed in each local area are accumulated to obtain a comprehensive evaluation index of each local area; the local area with the largest comprehensive evaluation index is taken as the prostate area.

[0047] Preferably, in one embodiment of the present invention, the method for obtaining the prostate training model includes: exist In the neural network, as the thermal value of the prostate area increases, the thermal value of the area extending outward from the prostate area gradually decreases. Effectively utilizing the thermal distribution information in the prostate MRI image can improve the recognition accuracy of the prostate area. Therefore, a prostate thermal distribution map is obtained according to the prostate target area, and the prostate MRI image is matched with the prostate thermal distribution map to obtain a prostate training data set; a prostate training model is obtained according to the prostate training data set, wherein the methods for obtaining the prostate training data set and the prostate training model are technical means well known to those skilled in the art and will not be elaborated here.

[0048] In other embodiments of the present invention, after completing the positioning of the prostate region, the prostate training model can also be obtained by using adaptive multi-scale training and transfer learning methods. The specific process is a technical means well known to those skilled in the art and will not be described in detail here.

[0049] In one embodiment of the present invention, after receiving the prostate MRI image, the computer device uses the prostate MRI image as input, and may pre-process the image if necessary, including resizing operations, and then input the image into the model, and then the model passes through its multiple layers. The structure extracts features from the image and uses the self-attention mechanism to capture local and global features in the image. The model gradually integrates feature information of different scales, constructs a hierarchical feature representation of the prostate region, outputs the classification probability of each pixel, and generates a segmentation map. The segmentation map distinguishes the prostate region from the surrounding tissue, annotates the segmentation result as the prostate region, and clearly marks the boundary of the prostate on the image, providing an intuitive visual reference for the doctor's subsequent analysis and diagnosis, thereby assisting the diagnosis.

[0050] At this point, the auxiliary diagnosis of prostate MRI images is completed.

[0051] In summary, a prostate MRI image is acquired; corner point detection is performed on the prostate MRI image to obtain all corner points in the prostate MRI image; a screening and retention index of each corner point is obtained according to the gradient distribution and grayscale distribution in the preset neighborhood of each corner point; all corner points are screened according to the screening and retention index of all corner points to obtain all corner points to be analyzed in the corner points; any corner point to be analyzed is selected as the target corner point; the initial preset window of the target corner point is adjusted according to the screening and retention index of the target corner point to obtain the corner point window of the target corner point; the connection coefficient of the target corner point is obtained according to the position distribution and corner point window difference between the target corner point and the nearest corner point to be analyzed; edge detection is performed on the prostate MRI image to obtain all local areas; the local area to which each corner point to be analyzed belongs is obtained according to the distance distribution between each corner point to be analyzed and the local area; all prostate areas are obtained according to the connection coefficient of the corner point to be analyzed in each local area; a prostate training model is obtained according to all prostate areas; and auxiliary diagnosis of the prostate MRI image is performed according to the prostate training model.

[0052] An embodiment of the present invention provides a prostate MRI image-assisted diagnosis system based on deep learning, which includes a memory, a processor and a computer program, wherein the memory is used to store the corresponding computer program, and the processor is used to run the corresponding computer program. When the computer program runs in the processor, it can implement the method described in steps S1-S3.

[0053] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A deep learning-based prostate MRI image-assisted diagnosis method, characterized in that: The method comprises: Obtaining prostate MRI images; Performing corner point detection on the prostate MRI image to obtain all corner points in the prostate MRI image; obtaining a screening and retention index for each corner point according to a gradient distribution and a grayscale distribution within a preset neighborhood of each corner point; screening all corner points according to the screening and retention index of all corner points to obtain all corner points to be analyzed among the corner points; selecting any corner point to be analyzed as a target corner point; adjusting an initial preset window of the target corner point according to the screening and retention index of the target corner point to obtain a corner point window of the target corner point; obtaining a connection coefficient of the target corner point according to a positional relationship between the target corner point and the nearest corner point to be analyzed and a difference in the corner point window; Perform edge detection on the prostate MRI image to obtain all local areas; obtain the local area to which each corner point to be analyzed belongs based on the distance between each corner point to be analyzed and the local area; obtain the prostate area based on the connection coefficient of the corner point to be analyzed in each local area; obtain a prostate training model based on all prostate areas; and perform auxiliary diagnosis on the prostate MRI image based on the prostate training model.

2. The deep learning-based prostate MRI image-assisted diagnosis method according to claim 1, characterized in that: The method for obtaining the screening retention index includes: The screening retention index is obtained according to the screening retention index calculation formula, and the screening retention index calculation formula is as follows: In the formula, Represents the screening retention index of each corner point; Represents the gradient value of each corner point; Indicates the number of other pixels in the preset neighborhood of each corner point; Indicates the first The gradient value of each pixel; Represents the gray value of each corner point; Indicates the first The gray value of each pixel; represents the absolute value function.

3. The deep learning-based prostate MRI image-assisted diagnosis method according to claim 1, characterized in that: The methods for obtaining all corner points to be analyzed include: Sort each corner point in the prostate MRI image from large to small according to the screening retention index to obtain a corner point sequence; In the corner point sequence, corner points are selected from left to right to form a first corner point sequence; When the ratio between the number of corner points in the first corner point sequence and the number of corner points in the first corner point sequence is a preset first threshold, all corner points in the first corner point sequence are used as corner points to be analyzed.

4. The deep learning-based prostate MRI image-assisted diagnosis method according to claim 1, characterized in that: The method for obtaining the corner point window includes: Calculate the ratio between the screening retention index of the target corner point and the screening retention index and value of all the corner points to be analyzed as the corner point importance of the target corner point; The initial preset window is set to a square; the side length of the initial preset window is adjusted according to the corner point importance of the target corner point to obtain the side length of the corner point window of the target corner point; Get the corner point window of the target corner point according to the side length of the corner point window.

5. The deep learning-based prostate MRI image-assisted diagnosis method according to claim 4, characterized in that: The side length of the initial preset window is adjusted according to the corner point importance of the target corner point to obtain the side length of the corner point window of the target corner point, including: The corner point window side length is obtained according to the corner point window side length formula, and the corner point window side length formula is as follows: In the formula, The length of the corner window of the target corner; A preset initial window representing the target corner point; Indicates the importance of the target corner point; Represents the ceiling function.

6. The deep learning-based prostate MRI image-assisted diagnosis method according to claim 1, characterized in that: The method for obtaining the connection coefficient includes: The connection coefficient is obtained according to the connection coefficient calculation formula, and the connection coefficient calculation formula is as follows: In the formula, Represents the connection coefficient of the target corner point; The length of the corner point window representing the target corner point; The length of the corner point window representing the corner point to be analyzed that is closest to the target corner point; Indicates the distance between the target corner point and the nearest corner point to be analyzed; Represents Iverson brackets, where if the condition in the brackets is met, the value in the brackets is 1, and if the condition in the brackets is not met, the value in the brackets is 0.

7. The deep learning-based prostate MRI image-assisted diagnosis method according to claim 1, characterized in that: The method for obtaining the local area to which each corner point to be analyzed belongs includes: Obtain the centroid points in all local areas; calculate the distance between the target corner point and each centroid point as the corner point distance between the target corner point and each local area; take the local area with the closest corner point distance as the local area to which the target corner point belongs; Traverse all corner points to be analyzed and obtain the local area to which each corner point to be analyzed belongs.

8. The deep learning-based prostate MRI image-assisted diagnosis method according to claim 1, characterized in that: The method for acquiring the prostate region comprises: The connection coefficients of the corner points to be analyzed in each local area are accumulated to obtain a comprehensive evaluation index of each local area; The local area with the largest comprehensive evaluation index is taken as the prostate area.

9. The deep learning-based prostate MRI image-assisted diagnosis method according to claim 1, characterized in that: The method for obtaining the prostate training model comprises: Obtaining a prostate thermal distribution map according to the prostate target area, matching the prostate MRI image with the prostate thermal distribution map, and obtaining a prostate training data set; A prostate training model is obtained according to the prostate training data set, and the prostate MRI image is segmented using the prostate training model.

10. A prostate MRI image-assisted diagnosis system based on deep learning, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images