Method, device, and storage medium for training a disc hemorrhage detection model

Abstract

Embodiments of the present disclosure relate to a method, device, and storage medium for training a disc hemorrhage detection model. In the method for training a disc hemorrhage detection model provided by the embodiments of the present disclosure, fundus images are acquired, and the fundus images include true hemorrhage information of disc hemorrhage; image analysis is performed on the fundus images to obtain disc edge information that at least indicates the edge of the disc in the fundus images; and the disc hemorrhage detection model is trained using the fundus images and the disc edge information. In this way, a disc hemorrhage detection model that can reduce the false detection rate of disc hemorrhage can be provided.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of medical image processing, and more particularly, to methods, devices, and storage media for training a disc hemorrhage detection model. Background Art

[0002] Disc hemorrhages are often located in the superotemporal or inferotemporal regions of the optic disc. In fundus images, a disc hemorrhage is an isolated flame-shaped or fragmented hemorrhage area on the optic disc that is roughly perpendicular to the edge of the optic disc. The situation of disc hemorrhages has characteristics such as being similar to blood vessels and adjacent to blood vessels, which poses a great challenge to detecting disc hemorrhages in fundus images. Summary of the Invention

[0003] In a first aspect of the present disclosure, a method for training a disc hemorrhage detection model is provided. The method includes obtaining a fundus image that includes true hemorrhage information of a disc hemorrhage. The method further includes performing image analysis on the fundus image to obtain disc edge information that at least indicates the edge of the optic disc in the fundus image. The method further includes training a disc hemorrhage detection model using the fundus image and the disc edge information.

[0004] In a second aspect of the present disclosure, a method for processing a fundus image is provided. The method includes performing image analysis on the fundus image to obtain disc edge information that at least indicates the edge of the optic disc in the fundus image. The method further includes detecting disc hemorrhage information based on the fundus image and the disc edge information according to a trained disc hemorrhage detection model for detecting disc hemorrhages. The method further includes outputting the disc hemorrhage information.

[0005] According to the second aspect of the present disclosure, the method further includes: after detecting the disc hemorrhage information according to the trained disc hemorrhage detection model, determining whether the detected disc hemorrhage information meets the disc hemorrhage condition. The disc hemorrhage condition at least indicates one or more of the following: the angle between the hemorrhage area and the edge of the optic disc, the shape of the hemorrhage area, and the overlap between the hemorrhage area and the optic disc. The method further includes: in response to the detected disc hemorrhage information meeting the disc hemorrhage condition, outputting the disc hemorrhage information.

[0006] In a third aspect of the present disclosure, an electronic device is provided. The electronic device includes a processor and a memory coupled to the processor. The memory has instructions stored therein, and the instructions, when executed by the processor, cause the device to perform actions. The actions include obtaining a fundus image that includes true hemorrhage information of a disc hemorrhage. The actions further include performing image analysis on the fundus image to obtain disc edge information that at least indicates the edge of the optic disc in the fundus image. The actions further include training a disc hemorrhage detection model using the fundus image and the disc edge information.

[0007] In a fourth aspect of the present disclosure, an electronic device is provided. The electronic device includes a processor and a memory coupled to the processor. The memory has instructions stored therein, and when the instructions are executed by the processor, the device performs operations. The operations include performing image analysis on a fundus image to obtain optic disc edge information that at least indicates the edge of the optic disc in the fundus image. The operations further include detecting optic disc hemorrhage information based on the fundus image and the optic disc edge information according to a trained optic disc hemorrhage detection model for detecting optic disc hemorrhage. The operations further include outputting the optic disc hemorrhage information.

[0008] In a fifth aspect of the present disclosure, a computer program product is provided. The computer program product is tangibly stored on a computer-readable medium and includes machine-executable instructions that, when executed, cause the machine to perform the methods according to the first aspect and the second aspect.

[0009] The summary of the invention is provided to introduce, in a simplified form, a selection of concepts that will be further described in the detailed description below. The summary of the invention is not intended to identify the key features or main features of the present disclosure, nor is it intended to limit the scope of the present disclosure. Brief Description of the Drawings

[0010] By describing the exemplary embodiments of the present disclosure in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present disclosure will become more apparent. In the exemplary embodiments of the present disclosure, the same reference numerals generally represent the same components. In the drawings:

[0011] Figure 1 A schematic diagram showing optic disc hemorrhage in a fundus image is shown;

[0012] Figure 2 A flowchart showing an example method for training an optic disc hemorrhage detection model according to an embodiment of the present disclosure is shown;

[0013] Figure 3 A flowchart showing an example method for training an optic disc hemorrhage detection model based on optic disc edge information according to some embodiments of the present disclosure is shown;

[0014] Figure 4 A flowchart showing an example method for detecting optic disc hemorrhage information in a fundus image according to an optic disc hemorrhage detection model according to another embodiment of the present disclosure is shown;

[0015] Figure 5 A flowchart showing an example method for determining whether the detected optic disc hemorrhage information meets the optic disc hemorrhage condition according to some embodiments of the present disclosure; and

[0016] Figure 6 A block diagram of an example device that can be used to implement the embodiments of the present disclosure is shown. Detailed implementation manners

[0017] The principles of the present disclosure will be described below with reference to several exemplary embodiments shown in the accompanying drawings. Although the preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the description of these embodiments is only for enabling those skilled in the art to better understand and thus implement the present disclosure, rather than limiting the scope of the present disclosure in any way.

[0018] The term "including" and its variants used herein mean open inclusion, that is, "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "at least partially based on". The term "an exemplary embodiment" and "an embodiment" mean "at least one exemplary embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc. may refer to different or the same objects. There may also be other explicit and implicit definitions hereinafter.

[0019] Figure 1 A schematic diagram showing a hemorrhage in the optic disc 120 in the fundus image 100 is shown. As Figure 1 shown, the hemorrhage areas 110-1 and 110-2 of the optic disc 120 (which may also be collectively or individually referred to as "hemorrhage area 110") are substantially perpendicular to the optic disc edge and are in isolated flame shapes (as shown by the hemorrhage area 110-2) or fragmented shapes (as shown by the hemorrhage area 110-1). In known traditional solutions, the detection of optic disc hemorrhage is based on image processing methods such as using histograms, edge detection, and color analysis, and the optic disc hemorrhage is detected in the region of interest after removing blood vessels. However, although such methods can detect optic disc hemorrhage, in processes such as setting the edge detection threshold and setting the size of the post-processed optic disc hemorrhage, strong professional knowledge is required and the false detection rate is high. In addition, in known traditional solutions, in one detection, at most one detection result of optic disc hemorrhage can be retained in the fundus image.

[0020] Embodiments of the present disclosure propose a solution for training an optic disc hemorrhage detection model, enabling automatic detection of optic disc hemorrhage without manual parameter setting.

[0021] According to various embodiments of the present disclosure, a fundus image including the true hemorrhage information of the optic disc hemorrhage is obtained. Image analysis is performed on the fundus image to obtain optic disc edge information. The optic disc edge information at least indicates the edge of the optic disc in the fundus image. Then, the fundus image and the optic disc edge information are used to train the optic disc hemorrhage detection model.

[0022] According to the embodiments described herein, detecting disc hemorrhage in fundus images using disc edge information can better reduce the false detection rate. In addition, by using a trained disc hemorrhage detection model to detect disc hemorrhage, automatic detection can be achieved because there is no need to manually set detection parameters. Moreover, in one detection, the detection results of multiple disc hemorrhages can be retained in the fundus image.

[0023] The following refers to Figures 2 to 6 to illustrate the basic principles and several exemplary implementation manners of the present disclosure. It should be understood that these exemplary embodiments are only provided to enable those skilled in the art to better understand and then implement the embodiments of the present disclosure, rather than limiting the scope of the present disclosure in any way.

[0024] Figure 2 FIG. shows a flowchart of an exemplary method 200 for training a disc hemorrhage detection model according to an embodiment of the present disclosure. It should be understood that method 200 may further include additional actions not shown and / or may omit the shown actions, and the scope of the present disclosure is not limited in this regard. The following describes method 200 in detail with reference to Figure 1 this.

[0025] At block 210, a fundus image is acquired, and the fundus image includes ground truth hemorrhage information of a disc hemorrhage. The ground truth hemorrhage information may be information that has been pre-determined in the acquired fundus image. For example, the ground truth hemorrhage information may be marked by a professional physician in the fundus image. For example, through a fundus photography device, such as Figure 1 the fundus image 100 shown can be acquired. In the fundus image 100, the hemorrhage area 110 indicates the ground truth hemorrhage information. The ground truth hemorrhage information is used as a target for comparison with the predicted result in the subsequent training of the disc hemorrhage detection model, which will be described in detail later.

[0026] At block 220, image analysis is performed on the fundus image to obtain disc edge information that at least indicates the edge of the disc in the fundus image. For example, known or future-developed image processing methods can be used to perform image analysis on the fundus image 100 acquired at block 210.

[0027] First, an optic disc border indicating the approximate position of the optic disc in the fundus image is detected in the fundus image. In some embodiments, a known conventional edge detection technique may be used to detect the optic disc border in the fundus image. In some embodiments, a deep learning-based method may also be used, such as using a known target detection model to detect the optic disc border. The target detection model may be, for example, FasterR-CNN, Single Shot MultiBox Detector (SSD), YOLO (You only look once), etc. However, this is not limited to this, and a customized target detection model may also be used to detect the optic disc border.

[0028] Then, based on the detected optic disc border, a suitable region of interest is determined from the fundus image that has not been image processed. In some embodiments, the shape of the region of interest is a square. In some other embodiments, the shape of the region of interest may also be a rectangle or other shapes, and the scope of the present disclosure is not limited in this respect. A portion corresponding to the region of interest is intercepted from the fundus image as a region of interest image. The region of interest image includes the detected optic disc border.

[0029] In some embodiments, the center of the optic disc may be calculated based on the detected optic disc border, or the center of the optic disc may be detected using the above object detection model, and then a suitable region of interest may be determined based on the calculated or detected center of the optic disc.

[0030] For example, the radius of the region of interest ROI radius It can be expressed as follows:

[0031] ROI radius =λ*max(disc ver_diameter ,disc hor_diameter ) (1)

[0032] λ represents the range factor and may be, for example, 1.5. ver_diameter Indicates the vertical diameter of the optic disc. hor_diameter Indicates the horizontal diameter of the optic disc. max(disc ver_diameter ,disc hor_diameter ) means taking the larger value of the item in the brackets.

[0033] Next, the image of the region of interest is analyzed to obtain the optic disc edge information.

[0034] In some embodiments, a hierarchical attention network, HANet (Ding, F., Yang, G., Liu, J., Wu, J., & Li, X. (2019). Hierarchical attention networks for medical image segmentation) can be utilized to segment the region of interest image to extract the edge contour of the optic disc. Other segmentation models (such as UNet, Mnet, CENet, etc.) can also be used to segment the region of interest image. In some embodiments, traditional image segmentation algorithms can also be used to extract the edge contour of the optic disc. In actual operation, any method or algorithm that can segment images can be used, and the scope of the present disclosure is not limited in this regard.

[0035] Next, polar coordinate transformation is performed on the extracted edge contour. Based on the Gaussian distribution, the edge contour after polar coordinate transformation is expanded. Finally, inverse polar coordinate transformation is performed on the expanded edge contour after polar coordinate transformation to obtain the optic disc edge intensity image. At least this optic disc edge intensity image is used as optic disc edge information for the subsequent process of training the optic disc hemorrhage detection model.

[0036] By using the gradient calculated by the edge detection algorithm, the edge of the object in the image can be found because there is a large gradient value at the edge of the object. Therefore, in some embodiments, an optic disc edge gradient image can also be generated as optic disc edge information. For example, using the edge detection algorithm, the gradient is calculated within the above-mentioned region of interest image. After obtaining the gradient within the entire region of interest image, the gradient information within the annular range of 0.25 times to 0.75 times the radius ROI radius of the region of interest is retained, and the edge of the optic disc is included in this annular region. Then, an optic disc edge gradient image including the retained gradient information is generated as optic disc edge information.

[0037] In some embodiments, other known edge expansion methods can also be adopted to generate optic disc edge information based on the extracted edge contour. In some embodiments, an optic disc edge information can be generated based on the extracted edge contour according to a trained customized edge expansion model.

[0038] At block 230, using the fundus image and the optic disc edge information, the optic disc hemorrhage detection model is trained. For example, the fundus image 100 as shown Figure 1 and the optic disc edge information obtained at block 220 can be used to train the optic disc hemorrhage detection model.

[0039] In some embodiments, the above-mentioned region of interest image and the above-mentioned optic disc edge information can be used to train the optic disc hemorrhage detection model. The following will be combined Figure 3Describe in detail the steps of training a disc hemorrhage detection model.

[0040] In this way, since the disc hemorrhage detection model has been pre-trained and no manual setting of detection parameters is required during detection, automatic detection of disc hemorrhage can be achieved. Moreover, since the model is trained using the disc edge information of the optic disc, the false detection rate during detection can be reduced.

[0041] Figure 3 A flowchart of an example method 300 for training a disc hemorrhage detection model based on disc edge information according to some embodiments of the present disclosure is shown. Method 300 can be regarded as an example implementation of block 230 in method 200. It should be understood that method 300 may further include additional actions not shown and / or the actions shown may be omitted, and the scope of the present disclosure is not limited in this regard.

[0042] At block 310, based on the fundus image and the disc edge information, predict the disc hemorrhage information according to the disc hemorrhage detection model. For example, it can be based on the fundus image 100 as Figure 1 shown and the disc edge information described in combination with Figure 2 to predict the disc hemorrhage information according to the disc hemorrhage detection model.

[0043] In some embodiments, it can be based on the region of interest image and the disc edge information described in combination with Figure 2 to predict the disc hemorrhage information according to the disc hemorrhage detection model.

[0044] For example, the region of interest image and the disc edge information can be fused to obtain a fused information image in which the original information of the fundus image and the disc edge information are fused.

[0045] In some embodiments, the fused information image can be obtained by splicing the 3 channels of the RGB image of the region of interest image and the 1 channel of the disc edge intensity image into 4 channels. In some embodiments, other fusion methods can also be used for fusion, such as pixel-level addition and multiplication operations, etc.

[0046] Next, the fused information image is input into the disc hemorrhage detection model, and image features are extracted through the feature encoder therein. The feature encoder can be based on ResNext101 with the fully connected layer removed. The original input size of the fused information image can be 512 pixels in height, 512 pixels in width, and 4 channels. After 4 times of downsampling, image features with a height of 32 pixels, a width of 32 pixels, and 2048 channels are finally extracted.

[0047] Then, in the optic disc hemorrhage detection model, by performing operations such as convolution and upsampling on the image features extracted by the feature encoder, the optic disc hemorrhage region is predicted as at least a part of the predicted optic disc hemorrhage information. The convolution can pass through convolutional layers with kernel sizes of 1×1 and 3×3 respectively. The upsampling operation can upsample the extracted features by 4 times.

[0048] In some embodiments, the optic disc hemorrhage center can also be predicted according to the optic disc hemorrhage detection model. For example, the above-mentioned image features extracted by the feature encoder can also be used to regress the center of the optic disc hemorrhage. Specifically, in the optic disc hemorrhage detection model, the image features extracted by the feature encoder can pass through convolutional layers with kernel sizes of 1×1 and 3×3 respectively, a pooling layer, a flattening layer, and a fully connected layer, so that the predicted coordinates of the optic disc hemorrhage center

[0049] are predicted.

[0050] In some embodiments, the optic disc hemorrhage detection model can also be trained based on the weighted sum of one or more of the following: the difference between the predicted optic disc hemorrhage region and the true hemorrhage region in the true hemorrhage information, and the difference between the predicted optic disc hemorrhage center and the true hemorrhage center in the true hemorrhage information.

[0051] For example, the difference Loss seg between the predicted optic disc hemorrhage region and the true hemorrhage region in the true hemorrhage information

[0052]

[0053] N represents the number of pixels in the region of interest image, k∈{0,1} represents the segmentation category, 0 represents the background, and 1 represents the foreground (optic disc hemorrhage). y ik represents the true category of the i-th pixel, where y ik ∈{0,1}. p ik represents the probability that the i-th pixel is predicted to be the k-th category, where p ik ∈[0,1].

[0054] The difference Loss reg between the predicted optic disc hemorrhage center and the true hemorrhage center in the true hemorrhage information

[0055]

[0056] x and y represent the true coordinates of the true bleeding center. and represents the predicted coordinates of the predicted optic disc bleeding center.

[0057] Loss seg and Loss reg The weighted sum Loss of can be expressed as follows:

[0058] Loss = α * Loss seg + β * Loss reg (4)

[0059] α and β represent weight coefficients, which can take 0.5 respectively, or other values.

[0060] Then, by iteratively backpropagating the difference between the predicted optic disc bleeding information and the true bleeding information, the difference is gradually reduced (i.e., the predicted optic disc bleeding information is getting closer and closer to the true bleeding information), so as to obtain a trained optic disc bleeding detection model. The number of iterations can be preset, for example, 100 times.

[0061] Figure 4 FIG. shows a flowchart of an example method 400 for detecting optic disc bleeding information in a fundus image according to another embodiment of the present disclosure. It should be understood that method 400 may further include additional actions not shown and / or may omit the actions shown, and the scope of the present disclosure is not limited in this regard.

[0062] At block 410, image analysis is performed on the fundus image to obtain optic disc edge information that at least indicates the edge of the optic disc in the fundus image. The description of the embodiment related to block 410 can be referred to in conjunction with Figure 2 the description of the embodiment related to block 220.

[0063] At block 420, based on the fundus image and the optic disc edge information, according to the trained optic disc bleeding detection model for detecting optic disc bleeding, the optic disc bleeding information is detected. For example, based on the fundus image 100 as shown in Figure 1 and the optic disc edge information obtained at block 410, according to the optic disc bleeding detection model, the optic disc bleeding information is detected. The optic disc bleeding detection model is a model trained according to the method described in conjunction with Figure 2 and Figure 3 described.

[0064] In some embodiments, based on the region of interest image and the optic disc edge information, according to the optic disc bleeding detection model, the optic disc bleeding information can be detected. The description of the region of interest image can also be referred to in conjunction with Figure 2 the description of the embodiment related to block 220.

[0065] After block 420, method 400 may proceed to block 430. At block 430, it is determined whether the optic disc hemorrhage information detected at block 420 meets the optic disc hemorrhage condition. This will be described in detail below in conjunction with Figure 5 such embodiments.

[0066] At block 440, the optic disc hemorrhage information is output.

[0067] In this way, by utilizing the optic disc edge information of the optic disc, according to the trained optic disc hemorrhage detection model, the false detection rate of optic disc hemorrhage can be reduced. Moreover, compared with the method of detecting optic disc hemorrhage by performing image analysis on fundus images using traditional image processing algorithms, one or more detection results can be retained in the fundus image in one detection. The detection efficiency is improved.

[0068] Figure 5 A flowchart of an example method 500 for determining whether the detected optic disc hemorrhage information meets the optic disc hemorrhage condition according to some embodiments of the present disclosure is shown. Method 500 can be regarded as an example implementation of block 430 in method 400. It should be understood that method 500 may also include additional actions not shown and / or actions shown may be omitted, and the scope of the present disclosure is not limited in this regard.

[0069] At block 510, it is determined whether the included angle between the hemorrhage area included in the optic disc hemorrhage information detected at block 420 and the edge of the optic disc is within the included angle threshold range. The included angle threshold range varies according to the shape of the hemorrhage area included in the optic disc hemorrhage information.

[0070] Specifically, the detected hemorrhage area is ellipse-fitted, and the central position of the ellipse is calculated. The included angles θ and π - θ between the major axis of the ellipse and the line connecting the ellipse center and the optic disc center are calculated. If |θ| ≤ γ or |π - θ| ≤ γ, it indicates that the detected hemorrhage area is substantially perpendicular to the edge of the optic disc and meets the optic disc hemorrhage condition. γ represents the included angle threshold range.

[0071] Next, a method for calculating the included angle threshold range γ will be described.

[0072]

[0073]

[0074]

[0075] rss perimeter represents the perimeter of the fitted ellipse. res area represents the area of the fitted ellipse. ρ and τ represent intermediate variables.

[0076] According to the theory that "among shapes with the same perimeter, a circle encloses the largest area", when the fitted contour is a circle, the minimum value of the intermediate variable τ is 4π. In the limit case, when the fitted contour is a line segment, the area of the fitted contour is 0. At this time, the intermediate variable τ reaches the maximum value of +∞.

[0077] According to equations (5) to (7), the value range of γ is (0°, 90°]. The closer the contour is to a circle, the closer the value of γ is to 90°. The more elongated the contour is, the closer the value of γ is to 0°. In other words, the more elongated the detected bleeding area is, the stricter the judgment of whether it is substantially perpendicular to the edge of the optic disc is, otherwise it is more relaxed.

[0078] If the result at block 510 is "yes", then method 500 proceeds to block 520. If the result at block 510 is "no", then the detected optic disc bleeding information is discarded.

[0079] At block 520, it is determined whether the shape of the bleeding area included in the detected optic disc bleeding information at block 420 is the shape indicated by the optic disc bleeding condition, such as flame-shaped or fragmented. It can be determined by using known or future-developed image analysis methods. For example, traditional geometric shape detection methods can be used, or methods based on deep learning can be used. The scope of the present disclosure is not limited in this regard.

[0080] The method based on deep learning can be, for example, training a machine learning model using real bleeding information, such as SVM and decision trees. Then, classify and model the shape of the bleeding area. The classification can be, for example, flame-shaped, fragmented, and other shapes. The bleeding area detected at block 420 passes through the trained machine learning model to determine whether its shape is the shape indicated by the optic disc bleeding condition. If it is determined that the shape of the bleeding area is flame-shaped or fragmented (the "yes" at block 520), then method 500 proceeds to block 530. Otherwise, the detected optic disc bleeding information is discarded.

[0081] At block 530, it is determined whether the bleeding area included in the detected optic disc bleeding information at block 420 overlaps with the optic disc. For example, it can be determined by comparing whether the pixel positions of the detected bleeding area overlap with the pixel positions of the optic disc area. In some embodiments, other image analysis methods can also be used for determination. The scope of the present disclosure is not limited in this regard.

[0082] If the result at block 530 is "yes", then return to Figure 4 block 440 in. If the result at block 530 is "no", then the detected optic disc bleeding information is discarded.

[0083] In this way, since the detected optic disc hemorrhage information is further processed and screened after being detected by the optic disc hemorrhage detection model, the false detection rate can be further reduced, making the detection result more accurate.

[0084] Figure 6 FIG. shows a schematic block diagram of an exemplary device 600 that can be used to implement embodiments of the present disclosure. For example, as Figure 2 shown in method 200 and as Figure 4 shown in method 400 can be implemented by device 600 respectively. As Figure 6 shown, device 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 602 or computer program instructions loaded from a storage unit 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of device 600 can also be stored. The CPU 601, ROM 602, and RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0085] A plurality of components in device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a magnetic disk, an optical disc, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0086] The various processes and treatments described above, such as method 200 and method 400, can be executed by the processing unit 601. For example, in some embodiments, method 200 and method 400 can be implemented as computer software programs, which are tangibly included in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed onto device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the CPU 601, one or more actions of method 200 and method 400 described above can be executed.

[0087] The present disclosure can be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for performing various aspects of the present disclosure.

[0088] A computer-readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example—but not limited to—an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanically encoded device, such as a punched card or raised structures in grooves storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage medium as used herein is not construed as an instantaneous signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0089] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to respective computing / processing devices, or downloaded to an external computer or external storage device through a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include a copper transmission cable, an optical fiber transmission, a wireless transmission, a router, a firewall, a switch, a gateway computer, and / or an edge server. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0090] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine - related instructions, microcode, firmware instructions, state - setting data, or source code or object code written in any combination of one or more programming languages, including object - oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer - readable program instructions may be executed entirely on the user's computer, partially on the user's computer, executed as a stand - alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or, alternatively, may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, by using the state information of the computer - readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field - programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer - readable program instructions to implement various aspects of the present disclosure.

[0091] Aspects of the present disclosure are described herein with reference to the flowchart and / or block diagram of a method, apparatus (system), and computer program product according to embodiments of the present disclosure. It should be understood that each block of the flowchart and / or block diagram, and the combinations of blocks in the flowchart and / or block diagram, can be implemented by computer - readable program instructions.

[0092] These computer - readable program instructions can be provided to a processing unit of a general - purpose computer, a special - purpose computer, or other programmable data - processing apparatus to produce a machine such that, when the instructions are executed by the processing unit of the computer or other programmable data - processing apparatus, a device is produced that implements the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer - readable program instructions can also be stored in a computer - readable storage medium, which causes a computer, a programmable data - processing apparatus, and / or other devices to operate in a particular manner, so that the computer - readable medium storing the instructions includes a manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0093] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device, causing a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process such that the instructions executed on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in one or more boxes of the flowchart and / or block diagram.

[0094] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of code, or a portion of an instruction, and the module, segment of code, or portion of an instruction includes one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two consecutive blocks may in fact be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block of the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented by special-purpose hardware-based systems that perform the specified functions or acts, or combinations of special-purpose hardware and computer instructions.

[0095] The embodiments of the present disclosure have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable other ordinary skilled artisans in the art to understand the embodiments disclosed herein.

Claims

1. A method for training a disc hemorrhage detection model, comprising: obtaining a fundus image, the fundus image including true hemorrhage information of disc hemorrhage; performing image analysis on the fundus image to obtain disc edge information indicating at least the edge of the disc in the fundus image; predicting disc hemorrhage information based on the fundus image and the disc edge information according to the disc hemorrhage detection model; and training the disc hemorrhage detection model based on the difference between the predicted disc hemorrhage information and the true hemorrhage information, and detecting disc hemorrhage information according to the trained disc hemorrhage detection model; after detecting the disc hemorrhage information according to the trained disc hemorrhage detection model, determining whether the detected disc hemorrhage information meets the disc hemorrhage condition, the disc hemorrhage condition indicating an angle between a hemorrhage area included in the disc hemorrhage information and the edge of the disc; and in response to the detected disc hemorrhage information meeting the disc hemorrhage condition, outputting the disc hemorrhage information; in response to the angle between the hemorrhage area and the edge of the disc being within an angle threshold range, determining that the disc hemorrhage information meets the disc hemorrhage condition, wherein the angle threshold range varies according to the shape of the hemorrhage area; wherein the method further comprises: Perform an elliptical fit on the bleeding area and calculate the center position of the ellipse to determine the included angle, which is the included angle between the major axis of the ellipse and the line connecting the center of the ellipse and the center of the optic disc. and If or , the detected bleeding area is substantially perpendicular to the edge of the optic disc, meeting the condition of optic disc bleeding. Indicates the included angle threshold range.

2. The method according to claim 1, wherein, predicting the disc hemorrhage information according to the disc hemorrhage detection model further comprises: predicting one or more of the following: a disc hemorrhage area and a disc hemorrhage center; and training the disc hemorrhage detection model based on a weighted sum of one or more of the following: a difference between the predicted disc hemorrhage area and a true hemorrhage area in the true hemorrhage information, and a difference between the predicted disc hemorrhage center and a true hemorrhage center in the true hemorrhage information.

3. The method according to claim 1, wherein the disc hemorrhage condition further indicates at least one or more of the following: the shape of the hemorrhage area, and the overlap between the hemorrhage area and the disc; wherein determining whether the detected disc hemorrhage information meets the disc hemorrhage condition further comprises: in response to one or more of the following, determining that the disc hemorrhage information meets the disc hemorrhage condition: the shape of the hemorrhage area is the shape indicated by the disc hemorrhage condition; and the hemorrhage area overlaps with the disc.

4. A method for processing a fundus image, comprising: performing image analysis on the fundus image to obtain disc edge information indicating at least the edge of the disc in the fundus image; detecting disc hemorrhage information based on the fundus image and the disc edge information according to a trained disc hemorrhage detection model for detecting disc hemorrhage; and outputting the disc hemorrhage information, wherein the disc hemorrhage detection model is trained based on the difference between the predicted disc hemorrhage information and the true hemorrhage information, and disc hemorrhage information is detected according to the trained disc hemorrhage detection model; After detecting the optic disc hemorrhage information according to the trained optic disc hemorrhage detection model, determining whether the detected optic disc hemorrhage information meets the optic disc hemorrhage condition, where the optic disc hemorrhage condition indicates the angle between the hemorrhage area included in the optic disc hemorrhage information and the edge of the optic disc; and In response to the detected optic disc hemorrhage information meeting the optic disc hemorrhage condition, outputting the optic disc hemorrhage information; In response to the angle between the hemorrhage area and the edge of the optic disc being within the angle threshold range, determining that the optic disc hemorrhage information meets the optic disc hemorrhage condition, where the angle threshold range varies according to the shape of the hemorrhage area; wherein the method further includes: Perform an elliptical fit on the bleeding area, and calculate the center position of the ellipse to determine the included angle, and calculate the included angle between the major axis of the ellipse and the line connecting the center of the ellipse and the center of the optic disc and , if or , then the detected bleeding area is substantially perpendicular to the edge of the optic disc, meeting the optic disc bleeding condition, indicating the included angle threshold range.

5. The method according to claim 4, wherein the optic disc hemorrhage condition at least further indicates one or more of the following: the shape of the hemorrhage area, and the overlap between the hemorrhage area and the optic disc.

6. The method according to claim 5, wherein determining whether the detected optic disc hemorrhage information meets the optic disc hemorrhage condition further includes: In response to one or more of the following, determining that the optic disc hemorrhage information meets the optic disc hemorrhage condition: the shape of the hemorrhage area is the shape indicated by the optic disc hemorrhage condition; and the hemorrhage area overlaps with the optic disc.

7. The method according to claim 4, wherein the trained optic disc hemorrhage detection model is trained according to the method for training the optic disc hemorrhage detection model described in claims 1-3.

8. An electronic device, comprising: a processor; and a memory coupled to the processor, the memory having instructions stored therein that, when executed by the processor, cause the electronic device to perform actions, the actions including: acquiring a fundus image, the fundus image including the true hemorrhage information of the optic disc hemorrhage; performing image analysis on the fundus image to obtain optic disc edge information that at least indicates the edge of the optic disc in the fundus image; predicting optic disc hemorrhage information based on the fundus image and the optic disc edge information according to the optic disc hemorrhage detection model; and training the optic disc hemorrhage detection model based on the difference between the predicted optic disc hemorrhage information and the true hemorrhage information, and detecting the optic disc hemorrhage information according to the trained optic disc hemorrhage detection model; After detecting the optic disc hemorrhage information according to the trained optic disc hemorrhage detection model, determining whether the detected optic disc hemorrhage information meets the optic disc hemorrhage condition, where the optic disc hemorrhage condition indicates the angle between the hemorrhage area included in the optic disc hemorrhage information and the edge of the optic disc; and In response to the detected optic disc hemorrhage information meeting the optic disc hemorrhage condition, outputting the optic disc hemorrhage information; In response to the angle between the hemorrhage area and the edge of the optic disc being within the angle threshold range, determining that the optic disc hemorrhage information meets the optic disc hemorrhage condition, where the angle threshold range varies according to the shape of the hemorrhage area; wherein the actions further include: Perform an elliptical fitting on the bleeding area, and calculate the center position of the ellipse to determine the included angle, and calculate the included angle between the major axis of the ellipse and the line connecting the center of the ellipse and the center of the optic disc and , if or , then the detected bleeding area is substantially perpendicular to the edge of the optic disc, meeting the optic disc bleeding condition, indicating the included angle threshold range.

9. The electronic device according to claim 8, wherein, predicting the optic disc hemorrhage information according to the optic disc hemorrhage detection model further includes: Predict one or more of the following: the area of the disc hemorrhage and the center of the disc hemorrhage; and Train the disc hemorrhage detection model based on a weighted sum of one or more of the following: the difference between the predicted area of the disc hemorrhage and the true hemorrhage area in the true hemorrhage information, and the difference between the predicted center of the disc hemorrhage and the true center of the disc hemorrhage in the true hemorrhage information.

10. The electronic device according to claim 8, wherein the disc hemorrhage condition further indicates at least one or more of the following: the shape of the hemorrhage area, and the overlap of the hemorrhage area with the optic disc; wherein determining whether the detected disc hemorrhage information meets the disc hemorrhage condition further includes: Determine that the disc hemorrhage information meets the disc hemorrhage condition in response to one or more of the following: The shape of the hemorrhage area is the shape indicated by the disc hemorrhage condition; and The hemorrhage area overlaps with the optic disc.

11. An electronic device, comprising: A processor; and A memory coupled to the processor, the memory having instructions stored therein that, when executed by the processor, cause the electronic device to perform actions, the actions including: Perform image analysis on the fundus image to obtain disc edge information indicating at least the edge of the optic disc in the fundus image; Based on the fundus image and the disc edge information, detect disc hemorrhage information according to a trained disc hemorrhage detection model for detecting disc hemorrhage; and Output the disc hemorrhage information, wherein the disc hemorrhage detection model is trained based on the difference between the predicted disc hemorrhage information and the true hemorrhage information, and the disc hemorrhage information is detected according to the trained disc hemorrhage detection model; After detecting the disc hemorrhage information according to the trained disc hemorrhage detection model, determine whether the detected disc hemorrhage information meets the disc hemorrhage condition, the disc hemorrhage condition indicating the angle between the hemorrhage area included in the disc hemorrhage information and the edge of the optic disc; and In response to the detected disc hemorrhage information meeting the disc hemorrhage condition, output the disc hemorrhage information; Determine that the disc hemorrhage information meets the disc hemorrhage condition in response to the angle between the hemorrhage area and the edge of the optic disc being within the angle threshold range, The angle threshold range varies according to the shape of the hemorrhage area; wherein the actions further include: Perform an elliptical fitting on the bleeding area, and calculate the center position of the ellipse to determine the included angle, and calculate the included angle between the major axis of the ellipse and the line connecting the center of the ellipse and the center of the optic disc and , if or , then the detected bleeding area is substantially perpendicular to the edge of the optic disc, meeting the optic disc bleeding condition, represents the included angle threshold range.

12. The electronic device according to claim 11, wherein the disc hemorrhage condition further indicates at least one or more of the following: the shape of the hemorrhage area, and the overlap of the hemorrhage area with the optic disc.

13. The electronic device according to claim 12, wherein determining whether the detected disc hemorrhage information meets the disc hemorrhage condition further includes: Determine that the disc hemorrhage information meets the disc hemorrhage condition in response to one or more of the following: The shape of the hemorrhage area is the shape indicated by the disc hemorrhage condition; and The hemorrhage area overlaps with the optic disc.

14. The electronic device according to claim 11, wherein the trained optic disc hemorrhage detection model is trained according to the method for training an optic disc hemorrhage detection model as described in claims 1-3.

15. A computer-readable storage medium having computer-readable program instructions stored thereon, the computer-readable program instructions being for performing the method according to any one of claims 1-7.

Images