A device, equipment, and medium for lesion detection

Abstract

This application discloses a device, equipment, and medium for lesion detection, relating to the field of medical technology. It includes: an acquisition module for acquiring images of the area to be detected by a user using an image acquisition device; a first acquisition module for acquiring anchor frames containing a first lesion and a second lesion from the image using a target detection algorithm; a second acquisition module for acquiring the location of blood vessels in the image; and a detection module for determining whether the lesion detection result is a first lesion or a second lesion based on the anchor frames, or the relationship between the anchor frames and blood vessels. In this device, anchor frames containing a first lesion and a second lesion are first acquired, and then the first and second lesions are further determined based on the anchor frames, or the relationship between the anchor frames and blood vessels. Therefore, this device achieves lesion differentiation and improves the accuracy of lesion detection.

Description

Technical Field

[0001] This application relates to the field of medical technology, and in particular to a device, equipment and medium for lesion detection. Background Technology

[0002] Diabetic retinopathy (DR) is one of the most serious microvascular complications of diabetes. In its early stages, DR often presents with no obvious symptoms, making it easy for patients to overlook. However, as the disease progresses, lesions such as microaneurysms and hemorrhages will appear in the fundus, and the patient's vision will gradually decline. If the optimal treatment period is missed, irreversible vision damage will occur, potentially leading to blindness. Early intervention and treatment of DR can reduce the occurrence of complications, making early detection of fundus lesions in DR patients crucial.

[0003] In actual testing, because bleeding points are similar to microaneurysms and retinal tissues, and the differences between the two are small, it is easy to cause false detection, which reduces the accuracy of lesion detection.

[0004] How to differentiate lesions and improve the accuracy of lesion detection is a technical problem that urgently needs to be solved by people in this field. Summary of the Invention

[0005] The purpose of this application is to provide a device, equipment, and medium for lesion detection, which can be used to distinguish lesions and thereby improve the accuracy of lesion detection.

[0006] To address the aforementioned technical problems, this application provides a device for lesion detection, comprising:

[0007] The acquisition module is used to acquire images of the area to be detected by the user through an image acquisition device;

[0008] The first acquisition module is used to acquire an anchor box containing a first lesion and a second lesion from the image using a target detection algorithm;

[0009] The second acquisition module is used to acquire the location of blood vessels in the image;

[0010] The detection module is used to determine whether the lesion detection result is the first lesion or the second lesion based on the anchor frame or the relationship between the anchor frame and the blood vessel.

[0011] Preferably, the first acquisition module includes:

[0012] The second acquisition module is used to acquire all target box sets and the scores corresponding to each target box in the target box set through the Faster R-CNN detection network;

[0013] The selection module is used to select the candidate box with the highest score from the target box;

[0014] The removal and placement module is used to remove the candidate box with the highest score from the target box set and place it into the final detection box result set, so as to obtain the anchor box containing the first lesion and the second lesion from the final detection box result set.

[0015] Preferably, the second acquisition module includes:

[0016] The third acquisition module is used to acquire all the target box sets and the initial score corresponding to each target box set through the Faster R-CNN detection network;

[0017] The fourth acquisition module is used to acquire the overlap rate between each target box and the candidate box with the highest score;

[0018] The fifth acquisition module is used to acquire the score corresponding to the target box based on the initial score corresponding to the target box and the intersection-union ratio of the target box and the candidate box with the highest score when the overlap rate between the target box and the candidate box with the highest score is greater than a first threshold.

[0019] As a module, it is used to take the initial score corresponding to the target box as the score of the target box when the overlap rate between the target box and the candidate box with the highest score is less than or equal to the first threshold.

[0020] Preferably, the detection module includes:

[0021] The sixth acquisition module is used to acquire the area of ​​the anchor frame;

[0022] The judgment module is used to determine whether the area is greater than the second threshold.

[0023] If yes, the first determination module is triggered; if no, the second determination module is triggered.

[0024] The first determining module is used to determine that the result of the lesion detection is the first lesion;

[0025] The second determining module is used to determine that the result of the lesion detection is the second lesion.

[0026] Preferably, the detection module includes:

[0027] The first segmentation module is used to segment the image using a Res_Unet network to obtain a blood vessel segmentation image;

[0028] The first conversion module is used to convert the blood vessel segmentation image into a binary image;

[0029] The seventh acquisition module is used to acquire the number of binary image pixels with a value of 255 within the anchor frame area;

[0030] The third determining module is used to determine the result of the lesion detection as the first lesion when the quantity is 0;

[0031] The fourth determining module is used to determine the result of the lesion detection as the second lesion when the number is not zero.

[0032] Preferably, the detection module includes:

[0033] The second segmentation module is used to segment the image using a Res_Unet network to obtain a blood vessel segmentation image;

[0034] The second conversion module is used to convert the blood vessel segmentation image into a binary image;

[0035] The eighth acquisition module is used to acquire the first number of binary image pixels with a value of 255 within the current anchor frame region;

[0036] The adjustment and acquisition module is used to adjust the size of the current anchor frame area according to preset rules and acquire the adjusted anchor frame area.

[0037] The ninth acquisition module is used to acquire a second number of binary image pixel values ​​of 255 within the adjusted anchor frame area;

[0038] The fifth determining module is used to determine the lesion detection result as the first lesion when the second quantity is greater than the first quantity;

[0039] The sixth determining module is used to determine the result of the lesion detection as the second lesion when the second quantity is less than or equal to the first quantity.

[0040] Preferably, the device further includes:

[0041] The differentiation module is used to distinguish the image into images of different disease stages using a convolutional neural network model.

[0042] Preferably, the convolutional neural network model is a ResNet50 network, and the ResNet50 network includes a 3x3 convolutional kernel with a stride of 2 and two 3x3 convolutional kernels with a stride of 1; the last layer of the ResNet50 network introduces a CBAM model with an attention mechanism.

[0043] To address the aforementioned technical problems, this application also provides a device for lesion detection, comprising:

[0044] Memory, used to store computer programs;

[0045] The processor, when executing the computer program, implements the following steps: acquiring an image of the user's area to be detected through an image acquisition device; obtaining an anchor frame containing a first lesion and a second lesion from the image through a target detection algorithm; obtaining the location of blood vessels in the image; and determining the result of the lesion detection as the first lesion or the second lesion based on the anchor frame or the relationship between the anchor frame and the blood vessel.

[0046] To address the aforementioned technical problems, this application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the following steps: acquiring an image of a user's area to be detected using an image acquisition device; obtaining an anchor frame containing a first lesion and a second lesion from the image using a target detection algorithm; obtaining the location of blood vessels in the image; and determining the lesion detection result as either the first lesion or the second lesion based on the anchor frame or the relationship between the anchor frame and the blood vessel.

[0047] The lesion detection apparatus provided in this application includes: an acquisition module for acquiring an image of the area to be detected by a user using an image acquisition device; a first acquisition module for acquiring anchor frames containing a first lesion and a second lesion from the image using a target detection algorithm; a second acquisition module for acquiring the location of blood vessels in the image; and a detection module for determining whether the lesion detection result is a first lesion or a second lesion based on the anchor frames or the relationship between the anchor frames and blood vessels. In this apparatus, anchor frames containing a first lesion and a second lesion are first acquired, and then the first lesion and the second lesion are further determined based on the anchor frames or the relationship between the anchor frames and blood vessels. Therefore, this apparatus achieves lesion differentiation and improves the accuracy of lesion detection.

[0048] In addition, this application also provides a device for lesion detection and a computer-readable storage medium, which have the same or corresponding technical features as the aforementioned lesion detection device and have the same effect. Attached Figure Description

[0049] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. The accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A structural diagram of a lesion detection device provided in an embodiment of this application;

[0051] Figure 2 A structural diagram of an improved ResNet50 network provided in an embodiment of this application;

[0052] Figure 3 This is a structural diagram of a lesion detection device provided in one embodiment of this application;

[0053] Figure 4 This is an application scenario diagram of a lesion detection device provided in an embodiment of this application. Detailed Implementation

[0054] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0055] The core of this application is to provide a device, equipment, and medium for lesion detection, which can be used to distinguish lesions and thereby improve the accuracy of lesion detection.

[0056] It should be noted that the lesion differentiation device of this application can be applied to detect lesions in different locations and to differentiate between multiple different lesions, not just two types. This application mainly targets the detection of fundus lesions in patients with diabetic retinopathy (DR). DR is one of the most serious microvascular complications of diabetes. In its early stages, DR has no obvious symptoms and is not easily noticed by patients. However, as the disease progresses, lesions such as microaneurysms and hemorrhages will appear in the fundus, and the patient's vision will gradually decline. If the optimal treatment period is missed, treatment becomes more difficult, and in severe cases, it may lead to blindness. Early intervention and treatment of DR can reduce the occurrence of complications, so the detection of fundus lesions in DR patients is very important. Since hemorrhages are similar to microaneurysms and parts of the retina, and the differences between the two are small, this application uses a deep learning algorithm to detect fundus images and differentiate between hemorrhages and microaneurysms to improve the accuracy of fundus image detection.

[0057] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Figure 1 A structural diagram of a lesion detection device provided in an embodiment of this application is shown below. Figure 1 As shown, the device includes:

[0058] Acquisition module 1 is used to acquire images of the user's area to be detected through an image acquisition device.

[0059] The first acquisition module 2 is used to acquire anchor boxes containing the first lesion and the second lesion from the image through a target detection algorithm;

[0060] The second acquisition module 3 is used to acquire the location of blood vessels in the image;

[0061] The detection module 4 is used to determine whether the lesion detection result is the first lesion or the second lesion based on the anchor frame or the relationship between the anchor frame and the blood vessel.

[0062] When detecting lesions in a user, images of the area to be detected need to be acquired via an image acquisition module. The area to be detected is the site of the lesion caused by the disease. The image acquisition device used is not limited and is determined based on the actual situation. For example, as the disease progresses in DR patients, lesions such as hemorrhages and microaneurysms may appear in the fundus; therefore, images of the patient's fundus need to be acquired. To obtain a larger area of ​​fundus image, this embodiment uses an ultra-wide-angle fundus image. This imaging technology uses red-green dual-color laser dual-channel scanning imaging, and a single image can display approximately 80% of the retinal area. When combined with ultra-wide-angle fundus images in clinical practice, it not only broadens the ophthalmologist's field of vision but also facilitates the doctor's assessment of the DR patient's condition.

[0063] After acquiring images of the user's target area via the acquisition module, lesions can be detected based on the images. To distinguish between the first and second lesions, this application first needs to detect the targets (first and second lesions) from the acquired images. Target detection is performed using a target detection algorithm to obtain anchor boxes containing both the first and second lesions from the image. When detecting fundus images of DR patients, the first lesion can be considered a hemorrhage point, and the second lesion can be considered a microaneurysm.

[0064] After obtaining the anchor boxes containing the first and second lesions, the location of blood vessels in the image is further obtained through a second acquisition module. When acquiring the blood vessel image, a Res_Unet network can be used to segment the blood vessels in the image to obtain a segmented blood vessel image. Finally, the first and second lesions can be distinguished by the detection module based on the area of ​​the anchor boxes, or based on the positional relationship between the anchor boxes and the blood vessels.

[0065] The lesion detection device provided in this embodiment includes: an acquisition module for acquiring an image of the area to be detected by a user through an image acquisition device; a first acquisition module for acquiring anchor frames containing a first lesion and a second lesion from the image using a target detection algorithm; a second acquisition module for acquiring the location of blood vessels in the image; and a detection module for determining whether the lesion detection result is a first lesion or a second lesion based on the anchor frames or the relationship between the anchor frames and blood vessels. In this device, anchor frames containing a first lesion and a second lesion are first acquired, and then the first lesion and the second lesion are further determined based on the anchor frames or the relationship between the anchor frames and blood vessels. Therefore, this device achieves lesion differentiation and improves the accuracy of lesion detection.

[0066] When acquiring an image containing a first lesion and a second lesion using a target detection algorithm, in order to improve the detection accuracy, in a preferred implementation, the first acquisition module includes:

[0067] The second acquisition module is used to acquire all the target box sets and the scores corresponding to each target box in the target box set through the Faster R-CNN detection network;

[0068] The selection module is used to select the candidate box with the highest score from the target box;

[0069] The Remove and Place module is used to remove the candidate box with the highest score from the target box set and place it into the final detection box result set, so as to obtain the anchor box containing the first lesion and the second lesion from the final detection box result set.

[0070] The object detection algorithm used is not limited. Faster R-CNN, through its two-stage network and Region Proposal Network (RPN), achieves high-precision object detection performance. Compared to other one-stage networks, the two-stage network is more accurate, especially for high-precision, multi-scale, and small object problems. Faster R-CNN performs well on multiple datasets and object tasks. For personal datasets, fine-tuning often yields good results. Therefore, this embodiment uses the Faster R-CNN detection algorithm for object detection. The Faster R-CNN detection network mainly includes feature box extraction, feature map extraction, RoI pooling layers, and object classification and bounding box regression. Specifically, the Faster R-CNN feature extraction network is a Deep Residual Network 50 (ResNet 50) combined with a Feature Pyramid Network (FPN) structure.

[0071] (1) First, obtain the set B (bi∈B) of all object boxes generated by the Faster R-CNN detection network and the corresponding scores s. i .

[0072] (2) Select the candidate box M with the highest score from set B, remove M from set B and put it into the final detection box result set D.

[0073] In this embodiment, the target bounding box refers to the feature box containing the first lesion and the second lesion. Currently, the target bounding box is usually scored using formula (1).

[0074]

[0075] In formula (1) s i N represents the score of the target box. IOU represents the intersection-union ratio (IoU) between the candidate box with the highest score and the target box. t This represents the threshold. The threshold is not fixed and is determined based on the specific circumstances. When the intersection-union ratio (IU) is less than the threshold, the score of the corresponding bounding box is recorded as s. i When the difference between 1 and the intersection-union ratio is greater than or equal to the threshold, the score of the corresponding target box is recorded as 0.

[0076] The anchor boxes containing the first and second lesions obtained by the Faster RCNN detection network provided in this embodiment can improve the accuracy of lesion detection.

[0077] In the scoring mechanism of the above embodiment, when the difference between 1 and the intersection-union ratio is greater than or equal to a threshold, the score of the corresponding target box is recorded as 0. However, in practice, some lesions may overlap. If a region that is too large is directly set to 0, the box will be removed, resulting in the inability to retain target boxes containing both the first and second lesions as much as possible, thus reducing the detection accuracy. Therefore, in practice, the preferred implementation is that the second acquisition module includes:

[0078] The third acquisition module is used to obtain all sets of bounding boxes and the initial score corresponding to each set of bounding boxes through the Faster R-CNN detection network;

[0079] The fourth acquisition module is used to obtain the overlap rate between each target box and the candidate box with the highest score;

[0080] The fifth acquisition module is used to acquire the score corresponding to the target box based on the initial score of the target box and the intersection-union ratio of the target box and the candidate box with the highest score when the overlap rate between the target box and the candidate box with the highest score is greater than the first threshold.

[0081] As a module, it is used to take the initial score of the target box as the score of the target box when the overlap rate between the target box and the candidate box with the highest score is less than or equal to a first threshold.

[0082] In this embodiment, the scoring mechanism in the above embodiments is improved. The RPN in the Faster RCNN network is improved, and the Soft Non-maximum Suppression (NMS) algorithm is used to detect lesions in the ultra-wide-angle fundus image of diabetic retinopathy. The improved scoring method is shown in formula (2):

[0083]

[0084] In formula (2), to distinguish it from the threshold described below, N is used here. t The first threshold is described as follows. To distinguish between the scores of bounding boxes with an intersection-union ratio (IU) greater than the first threshold and the scores of bounding boxes with an IU less than the first threshold, this embodiment uses s... i This is called the initial score. When the overlap ratio (also known as the intersection-union ratio) between the target box and the candidate box with the highest score is greater than the first threshold, the score of the target box is the product of the initial score and (1 - intersection-union ratio).

[0085] Table 2 shows the anchor box sizes and ratios in the RPN of Faster R-CNN.

[0086] Table 2 Anchor frame dimensions and proportions in RPN

[0087]

[0088]

[0089] The improved scoring mechanism used in this embodiment to filter target boxes differs from the previous mechanism that directly sets the score of the corresponding target box to 0 when the intersection-union ratio (IU) exceeds a threshold, effectively removing the target box. In this embodiment, the device assigns a certain score to target boxes with an IU exceeding the threshold, allowing for the preservation of small first and second lesions and improving the detection accuracy of subtle features. Taking fundus lesion detection as an example, the device in this embodiment can preserve small hemorrhages and microaneurysms, facilitating subsequent detection and improving detection accuracy.

[0090] In practice, to distinguish between the first lesion and the second lesion, a preferred implementation is that the detection module includes:

[0091] The sixth acquisition module is used to obtain the area of ​​the anchor frame;

[0092] The judgment module is used to determine whether the area is greater than the second threshold.

[0093] If yes, the first determination module is triggered; if no, the second determination module is triggered.

[0094] The first determination module is used to determine the result of the lesion detection as the first lesion;

[0095] The second determination module is used to determine that the result of the lesion detection is the second lesion.

[0096] Taking fundus lesion detection as an example, after obtaining the anchor boxes {Anchor|x1, y1, x2, y2} containing hemorrhages and microaneurysms using the Faster R-CNN detection algorithm, the coordinates of the upper left and lower right corners of the anchor boxes (x1, y1) and (x2, y2) are used. The area S enclosed by the anchor boxes {Anchor|x1, y1, x2, y2} is calculated. If the area S enclosed by {x1, y1, x2, y2} is greater than a second threshold, the lesion is directly identified as a hemorrhage. The second threshold is not limited and is determined according to the actual situation. In this embodiment, after extensive testing, the preferred second threshold is 2000, that is, if the area S is greater than 2000, the lesion is directly identified as a hemorrhage; otherwise, the lesion is identified as a microaneurysm.

[0097] The method provided in this embodiment for determining lesions based on the area of ​​the anchor frame containing the first lesion and the second lesion can quickly distinguish between the first lesion and the second lesion.

[0098] In the above embodiments, the first lesion and the second lesion are distinguished based on the area of ​​the anchor frame. In practice, they can also be distinguished based on the relationship between the anchor frame and the blood vessel. A preferred embodiment includes a detection module comprising:

[0099] The first segmentation module is used to segment the image using the Res_Unet network in order to obtain a blood vessel segmentation image;

[0100] The first conversion module is used to convert the blood vessel segmentation image into a binary image;

[0101] The seventh acquisition module is used to acquire the number of binary image pixels with a value of 255 within the anchor frame region;

[0102] The third determination module is used to determine the lesion detection result as the first lesion when the quantity is 0;

[0103] The fourth determination module is used to determine the result of lesion detection as a second lesion when the number is not zero.

[0104] Taking fundus lesion detection as an example, after obtaining anchor boxes {Anchor|x1, y1, x2, y2} containing hemorrhages and microaneurysms using the Faster R-CNN detection algorithm, the image is segmented using the Res_Unet network to obtain a blood vessel segmentation image. The Res_Unet network is a Unet with ResNet34 as the feature extraction network. The blood vessel segmentation image is converted into a binary image, and the number of pixels with a value of 255 within the region enclosed by the coordinates in the binary image is calculated (SUM). If the number of pixels with a value of 255 within the region enclosed by the coordinates {x1, y1, x2, y2} in the binary image is 0 (SUM), it is determined to be a hemorrhage (because microaneurysms appear on blood vessels, and the enclosed region must contain blood vessels, and the blood vessel segmentation result is represented as white in the binary image); otherwise, it is determined to be a microaneurysm.

[0105] The present embodiment provides a method for distinguishing between the first lesion and the second lesion based on the relationship between the anchor frame and the blood vessel.

[0106] In addition to the method for distinguishing between the first lesion and the second lesion described in the above embodiments, this embodiment also provides a method for distinguishing between the first lesion and the second lesion. Specifically, the detection module includes:

[0107] The second segmentation module is used to segment the image using the Res_Unet network in order to obtain a blood vessel segmentation image;

[0108] The second conversion module is used to convert the blood vessel segmentation image into a binary image;

[0109] The eighth acquisition module is used to acquire the first number of binary image pixels with a value of 255 within the current anchor frame region;

[0110] The adjustment and acquisition module is used to adjust the size of the current anchor frame area according to preset rules and acquire the adjusted anchor frame area.

[0111] The ninth acquisition module is used to acquire the second number of binary image pixel values ​​of 255 within the adjusted anchor frame area;

[0112] The fifth determination module is used to determine the lesion detection result as the first lesion when the second quantity is greater than the first quantity;

[0113] The sixth determination module is used to determine the lesion detection result as a second lesion when the second quantity is less than or equal to the first quantity.

[0114] Here, taking fundus lesion detection as an example, after obtaining the anchor boxes {Anchor|x1, y1, x2, y2} containing hemorrhage points and microaneurysms through the Faster RCNN detection algorithm, the Res_Unet network is used to segment the image to obtain the vascular segmentation image; the vascular segmentation image is converted into a binary image; in this embodiment, the anchor box area is adjusted according to {x1 - i, x2 + i, y1 - i, y2 + i|0 <= i <= 20, i % 5 = 0}, where i is the adjustment step length, and 0 <= i <= 20, i % 5 = 0 here is obtained based on experience. Calculate the number sumi of pixels with a value of 255 in the area enclosed by the coordinates {x1 - i, x2 + i, y1 - i, y2 + i|0 <= i <= 20, i % 5 = 0} in the binary image. If the number of sumi does not increase continuously, it is determined as a hemorrhage point, otherwise, it is determined as a microaneurysm.

[0115] The size of the adjusted anchor box area provided in this embodiment can distinguish the first lesion from the second lesion according to the positional relationship between the anchor box area and the blood vessels.

[0116] In summary, three methods for distinguishing the first lesion from the second lesion are described above, making the method for distinguishing the first lesion from the second lesion more flexible.

[0117] Since the images of the lesions are different in different cycles, in order to accurately distinguish the lesions, a preferred implementation is that the device for lesion detection further includes:

[0118] A distinguishing module for distinguishing the image into images of different lesion cycles through a convolutional neural network model.

[0119] For DR patients, in this embodiment, the fundus image is divided into two parts, one part is the fundus image of stages 0 - 3, and the other part is the fundus image of stage 4. In this embodiment, the lesions in the fundus images of stages 0 - 3 are detected, and the hemorrhage points and microaneurysms therein are detected.

[0120] There is no limitation on the specific convolutional neural network model. In this embodiment, the adopted convolutional neural network model is the ResNet50 network, and the ResNet50 network includes a 3X3 convolutional kernel with a stride of 2, and two 3X3 convolutional kernels with a stride of 1; an attention mechanism convolutional block attention module (Convolutional Block Attention Module, CBAM) model is introduced in the last layer of the ResNet50 network. Figure 2This document presents a structural diagram of an improved ResNet50 network provided in an embodiment of this application. The difference from the previous ResNet50 network lies in the modification of the convolutional kernels. The 7x7 kernel with a stride of 2 is changed to a 3x3 kernel with a stride of 2, and two additional 3x3 kernels with a stride of 1 are added, allowing the model to focus more on subtle lesions in the image. Furthermore, a CBAM model is added to the last layer of the ResNet to improve the accuracy of network staging.

[0121] Figure 3 This is a structural diagram of a lesion detection device provided in an embodiment of this application. This embodiment is based on a hardware perspective, such as... Figure 3 As shown, the equipment for lesion detection includes:

[0122] Memory 20 is used to store computer programs;

[0123] The processor 21 is configured to execute a computer program to implement the steps described in the above embodiments: acquiring an image of the user's target area using an image acquisition device; obtaining an anchor frame containing a first lesion and a second lesion from the image using a target detection algorithm; obtaining the location of blood vessels in the image; and determining the lesion detection result as a first lesion or a second lesion based on the anchor frame or the relationship between the anchor frame and the blood vessel.

[0124] The device for lesion detection provided in this embodiment may include, but is not limited to, smartphones, tablets, laptops, or desktop computers.

[0125] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one of the following hardware forms: Digital Signal Processor (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as the Central Processing Unit (CPU), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a Graphics Processing Unit (GPU), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an Artificial Intelligence (AI) processor, which is used to handle computational operations related to machine learning.

[0126] The memory 20 may include one or more computer-readable storage media, which may be non-transitory. The memory 20 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In this embodiment, the memory 20 is used to store at least the following computer program, which, after being loaded and executed by the processor 21, is capable of implementing the relevant functions of a lesion detection device disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 20 may also include an operating system and data, and the storage method may be temporary or permanent storage. The operating system may include Windows, Unix, Linux, etc.

[0127] In some embodiments, the lesion detection device may further include a display screen, an input / output interface, a communication interface, a power supply, and a communication bus.

[0128] Those skilled in the art will understand that Figure 3 The structure shown does not constitute a limitation on the device for lesion detection and may include more or fewer components than shown.

[0129] The lesion detection device provided in this application includes a memory and a processor. When the processor executes the program stored in the memory, it can realize the relevant functions of a lesion detection device, with the same effect as above.

[0130] Finally, this application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, performs the following steps: acquiring an image of the user's area to be detected via an image acquisition device; obtaining an anchor frame containing a first lesion and a second lesion from the image using a target detection algorithm; obtaining the location of blood vessels in the image; and determining the lesion detection result as either a first lesion or a second lesion based on the anchor frame, or the relationship between the anchor frame and the blood vessel.

[0131] It is understood that if the methods in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and executes all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] To enable those skilled in the art to better understand the present application, the following description is provided in conjunction with the appendix. Figure 4 The present application will be further described in detail with reference to specific embodiments. Figure 4 This is an application scenario diagram of a lesion detection device provided in an embodiment of this application. For example... Figure 4 As shown, after acquiring the ultra-wide-angle fundus image, the image is processed in three stages.

[0133] In the first stage, the images are preprocessed; an improved ResNet50 is used to divide the images into 0-3 stage images and four stage images.

[0134] In the second stage, the image is segmented to obtain a segmentation map, and the improved Faster R-CNN algorithm is used to obtain preliminary results of lesion detection.

[0135] In the third stage, the blood vessels are segmented to obtain a blood vessel segmentation map. The lesion detection results are output by using the algorithm to distinguish between bleeding points and microaneurysms and combining the preliminary results of lesion detection.

[0136] This embodiment provides the acquisition of ultra-wide-angle fundus images of DR patients, using a Faster R-CNN network for preliminary detection of hemorrhages and microaneurysms. By comparing the anchor frame coordinates of the preliminarily detected hemorrhages and microaneurysms with the vascular segmentation map of the ultra-wide-angle fundus image, the positional relationship between the anchor frame and the blood vessels is calculated, thereby distinguishing between hemorrhages and microaneurysms and improving the accuracy of lesion detection.

[0137] The foregoing has provided a detailed description of the apparatus, device, and medium for lesion detection provided in this application. The various embodiments in the specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0138] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A device for lesion detection, characterized in that, include: The acquisition module is used to acquire images of the area to be detected by the user through an image acquisition device; The first acquisition module is used to acquire an anchor box containing a first lesion and a second lesion from the image using a target detection algorithm; The acquisition module is used to acquire the location of blood vessels in the image; The detection module is used to determine whether the lesion detection result is the first lesion or the second lesion based on the relationship between the anchor frame and the blood vessel; The detection module includes: The second segmentation module is used to segment the image using a Res_Unet network to obtain a blood vessel segmentation image; The second conversion module is used to convert the blood vessel segmentation image into a binary image; The eighth acquisition module is used to acquire the first number of binary image pixels with a value of 255 within the current anchor frame region; The adjustment and acquisition module is used to adjust the size of the current anchor frame area according to preset rules and acquire the adjusted anchor frame area. The ninth acquisition module is used to acquire a second number of binary image pixel values ​​of 255 within the adjusted anchor frame area; The fifth determining module is used to determine the lesion detection result as the first lesion when the second quantity is greater than the first quantity; The sixth determining module is used to determine the result of the lesion detection as the second lesion when the second quantity is less than or equal to the first quantity.

2. The lesion detection device according to claim 1, characterized in that, The first acquisition module includes: The second acquisition module is used to acquire all target box sets and the scores corresponding to each target box in the target box set through the Faster R-CNN detection network; The selection module is used to select the candidate box with the highest score from the target box; The removal and placement module is used to remove the candidate box with the highest score from the target box set and place it into the final detection box result set, so as to obtain the anchor box containing the first lesion and the second lesion from the final detection box result set.

3. The lesion detection device according to claim 2, characterized in that, The second acquisition module includes: The third acquisition module is used to acquire all the target box sets and the initial score corresponding to each target box set through the Faster R-CNN detection network; The fourth acquisition module is used to acquire the overlap rate between each target box and the candidate box with the highest score; The fifth acquisition module is used to acquire the score corresponding to the target box based on the initial score corresponding to the target box and the intersection-union ratio of the target box and the candidate box with the highest score when the overlap rate between the target box and the candidate box with the highest score is greater than a first threshold. As a module, it is used to take the initial score corresponding to the target box as the score of the target box when the overlap rate between the target box and the candidate box with the highest score is less than or equal to the first threshold.

4. The lesion detection apparatus according to any one of claims 1 to 3, characterized in that, The device further includes: The differentiation module is used to distinguish the image into images of different disease stages using a convolutional neural network model.

5. The lesion detection device according to claim 4, characterized in that, The convolutional neural network model is a ResNet50 network, which contains a 3x3 convolutional kernel with a stride of 2 and two 3x3 convolutional kernels with a stride of 1. The last layer of the ResNet50 network introduces a CBAM model with an attention mechanism.

6. A device for detecting lesions, characterized in that, include: Memory, used to store computer programs; The processor is configured to execute the computer program to acquire images of the user's area to be detected via an image acquisition device. The steps are: obtaining anchor frames containing a first lesion and a second lesion from the image using a target detection algorithm; obtaining the location of blood vessels in the image; and determining whether the lesion detection result is the first lesion or the second lesion based on the relationship between the anchor frames and the blood vessels. The determination of whether the lesion detection result is the first lesion or the second lesion based on the relationship between the anchor frame and the blood vessel includes: The image is segmented using a Res_Unet network to obtain a segmented blood vessel image; the segmented blood vessel image is converted into a binary image; a first number of binary image pixel values ​​of 255 within the current anchor frame region is obtained; the size of the current anchor frame region is adjusted according to a preset rule and the adjusted anchor frame region is obtained; a second number of binary image pixel values ​​of 255 within the adjusted anchor frame region is obtained; if the second number is greater than the first number, the lesion detection result is determined to be the first lesion; if the second number is less than or equal to the first number, the lesion detection result is determined to be the second lesion.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the following steps: acquiring an image of the user's area to be detected through an image acquisition device; obtaining an anchor frame containing a first lesion and a second lesion from the image through a target detection algorithm; obtaining the location of blood vessels in the image; and determining the lesion detection result as either the first lesion or the second lesion based on the relationship between the anchor frame and the blood vessels. The determination of whether the lesion detection result is the first lesion or the second lesion based on the relationship between the anchor frame and the blood vessel includes: The image is segmented using a Res_Unet network to obtain a segmented blood vessel image; the segmented blood vessel image is converted into a binary image; a first number of binary image pixel values ​​of 255 within the current anchor frame region is obtained; the size of the current anchor frame region is adjusted according to a preset rule and the adjusted anchor frame region is obtained; a second number of binary image pixel values ​​of 255 within the adjusted anchor frame region is obtained; if the second number is greater than the first number, the lesion detection result is determined to be the first lesion; if the second number is less than or equal to the first number, the lesion detection result is determined to be the second lesion.

Images