Bleeding area detection method and device based on layer peeling and local features

Abstract

The present invention belongs to the technical field of eye image recognition. An embodiment thereof provides a method and apparatus for detecting hemorrhage areas based on layered peeling and local features. The method comprises: peeling the optic disc and blood vessels from the original retinal image to obtain a processed image; constructing a T-distribution difference pyramid on the processed image, wherein the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing; selecting key points from the local extreme points in the T-distribution difference pyramid space and obtaining descriptions of the key points; and positioning and marking each key point on the original retinal image based on the description of the key point, with the marked area being the hemorrhage area. Embodiments of the present invention can improve recognition accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of eye image recognition, and in particular to a bleeding area detection method based on layer peeling and local features, a bleeding area detection method based on layer peeling and local features, an electronic device, and a computer-readable storage medium. Background Art

[0002] Using optical coherence tomography (OCT) technology to generate retinal images is crucial for the manual diagnosis of eye diseases. Numerous diseases, such as diabetes, hypertension, tumors, or diseases of the ocular neovascularization, can cause retinal lesions and result in retinal hemorrhages (retinal hemorrhages, often called fundus hemorrhages. These lesions typically occur in the capillaries within the retina, primarily due to damage to the capillary lining). Manual diagnosis relies primarily on the physician's diagnostic experience and subjective judgment. However, the ratio of eye disease patients to professional physicians in my country is extremely unbalanced, and a significant gap remains. With the increasing number of patients with eye diseases and the dual pressures of overloaded ophthalmologists, the probability of misdiagnosis and missed diagnoses is also increasing. Some eye diseases often worsen rapidly within a short period of time after diagnosis. Missing the optimal time for diagnosis and treatment of retinal diseases can cause irreversible damage to patients. Therefore, early diagnosis and timely treatment of retinal diseases are crucial.

[0003] With the popularization of deep learning technology, it has been widely applied in the imaging field. Using deep learning technology, retinal hemorrhage areas can be quickly and accurately identified and located, assisting doctors in diagnosing symptoms. This effectively reduces misdiagnoses caused by manual diagnosis and alleviates the pressure on doctors. However, due to differences in medical equipment, shooting angles, lighting, and other external factors, as well as the characteristics of the patient's retinal disc, retinal vessel curvature, overlapping vessels, lesion areas, and complex structures, the collected retinal images may have blurred hemorrhage areas and the influence of the retinal disc, blood vessels, and other background factors, which seriously affect the effectiveness of auxiliary diagnosis. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a method and device for detecting hemorrhage areas based on layer peeling and local features. The layer peeling method is used to remove explicit targets such as the optic disc and blood vessels to reduce the impact on the identification and detection of implicit targets such as hemorrhage areas. At the same time, local feature extraction is used to search and identify hemorrhage areas. Compared with traditional target detection algorithms, this method not only eliminates the impact of explicit targets on implicit target detection, but also enhances the understanding of implicit targets, thereby better identifying hemorrhage areas.

[0005] To achieve the above-mentioned objectives, the first aspect of the present invention provides a hemorrhage area detection method based on hierarchical peeling and local features, the method comprising: peeling the optic disc and blood vessels from the original retinal image to obtain a processed image; constructing a T-distribution difference pyramid on the processed image, the T-distribution difference pyramid being obtained through multiple cycles of convolution and smoothing; selecting key points from the local extreme points in the T-distribution difference pyramid space and obtaining a description of the key points; positioning and marking each key point on the original retinal image according to the description of the key points, and using the marked area as the hemorrhage area.

[0006] Preferably, stripping the optic disc and blood vessels from the original retinal image to obtain a processed image includes: identifying the optic disc area from the original retinal image by an optic disc identification function; identifying the blood vessel area from the original retinal image by a blood vessel identification function; stripping the optic disc area and the blood vessel area from the original retinal image, and filling the stripped areas to obtain the processed image.

[0007] Preferably, the optic disc recognition function and the blood vessel recognition function are integrated into a hierarchical peeling network model; the hierarchical peeling network model uses labeled optic disc and blood vessel images as training samples, adopts a twelve-fold cross-validation method for training, and uses images in a test set to verify whether the trained hierarchical peeling network model is accurate.

[0008] Preferably, a T-distribution difference pyramid is constructed on the processed image, and the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing, including: taking the processed image as the first layer of the first group; performing T-distribution convolution on the first layer of the first group, and using the obtained image as the second layer of the first group; updating the smoothing factor in the T-distribution convolution, and sequentially obtaining subsequent layers of images in the first group; determining the first layer in each subsequent group, and obtaining subsequent layers in each group by T-distribution convolution with different smoothing factors; repeating the above steps to obtain multiple groups of images containing multiple layers of images in each group, and the multiple groups of images constitute a T-distribution pyramid; and constructing the T-distribution difference pyramid based on the T-distribution pyramid.

[0009] Preferably, selecting a key point from the local extreme points in the T-distributed differential pyramid space includes: in the middle layer of each group, determining whether a certain pixel point is an extreme point based on the comparison result between the certain pixel point and the surrounding pixel points; if the certain pixel point is an extreme point, correcting the position of the extreme point using Taylor expansion; and using the extreme point after position correction as the key point.

[0010] Preferably, obtaining a description of the key point includes: assigning a reference direction to each key point; collecting the gradient distribution characteristics and direction distribution characteristics of pixels in the image domain window of the T-distributed difference pyramid where the key point is located in the reference direction; and describing the key point according to the position information, scale information and direction information of the key point.

[0011] Preferably, each key point is positioned and marked on the original retinal image according to the description of the key point, and the marked area is used as the hemorrhage area, including: matching key points according to different collected images of the same retina; assigning a weight to each key point according to the number of matches; positioning and marking on the original retinal image according to the weight of each key point; and using the marked area as the hemorrhage area.

[0012] In a second aspect of the present invention, a hemorrhage area detection device based on hierarchical peeling and local features is also provided, which includes: an image peeling module, used to peel off the optic disc and blood vessels from the original retinal image to obtain a processed image; a pyramid processing module, used to construct a T-distribution difference pyramid on the processed image, and the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing; a key point determination module, used to select key points from local extreme points in the T-distribution difference pyramid space and obtain a description of the key points; and a positioning marking module, used to position and mark each key point on the original retinal image according to the description of the key point, and use the marked area as the hemorrhage area.

[0013] Preferably, stripping the optic disc and blood vessels from the original retinal image to obtain a processed image includes: identifying the optic disc area from the original retinal image by an optic disc identification function; identifying the blood vessel area from the original retinal image by a blood vessel identification function; stripping the optic disc area and the blood vessel area from the original retinal image, and filling the stripped areas to obtain the processed image.

[0014] Preferably, the optic disc recognition function and the blood vessel recognition function are integrated into a hierarchical peeling network model; the hierarchical peeling network model uses labeled optic disc and blood vessel images as training samples, adopts a twelve-fold cross-validation method for training, and uses images in a test set to verify whether the trained hierarchical peeling network model is accurate.

[0015] Preferably, a T-distribution difference pyramid is constructed on the processed image, and the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing, including: taking the processed image as the first layer of the first group; performing T-distribution convolution on the first layer of the first group, and using the obtained image as the second layer of the first group; updating the smoothing factor in the T-distribution convolution, and sequentially obtaining subsequent layers of images in the first group; determining the first layer in each subsequent group, and obtaining subsequent layers in each group by T-distribution convolution with different smoothing factors; repeating the above steps to obtain multiple groups of images containing multiple layers of images in each group, and the multiple groups of images constitute a T-distribution pyramid; and constructing the T-distribution difference pyramid based on the T-distribution pyramid.

[0016] Preferably, selecting a key point from the local extreme points in the T-distributed differential pyramid space includes: in the middle layer of each group, determining whether a certain pixel point is an extreme point based on the comparison result between the certain pixel point and the surrounding pixel points; if the certain pixel point is an extreme point, correcting the position of the extreme point using Taylor expansion; and using the extreme point after position correction as the key point.

[0017] Preferably, obtaining a description of the key point includes: assigning a reference direction to each key point; collecting the gradient distribution characteristics and direction distribution characteristics of pixels in the image domain window of the T-distributed difference pyramid where the key point is located in the reference direction; and describing the key point according to the position information, scale information and direction information of the key point.

[0018] Preferably, each key point is positioned and marked on the original retinal image according to the description of the key point, and the marked area is used as the hemorrhage area, including: matching key points according to different collected images of the same retina; assigning a weight to each key point according to the number of matches; positioning and marking on the original retinal image according to the weight of each key point; and using the marked area as the hemorrhage area.

[0019] In the third aspect of the present invention, an electronic device is also provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the aforementioned method for detecting bleeding areas based on hierarchical peeling and local features are implemented.

[0020] In a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein instructions are stored in the storage medium. When the instructions are executed on a computer, the computer executes the steps of the aforementioned method for detecting bleeding areas based on layer peeling and local features.

[0021] A fifth aspect of the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the aforementioned method for detecting bleeding areas based on layer peeling and local features.

[0022] The above technical solution has at least the following beneficial effects:

[0023] (1) Compared with the traditional method, the method of hemorrhage area detection based on layered peeling and local features directly removes the influence of obvious targets such as optic disc and blood vessels on the hemorrhage area by using layered peeling technology, solves the problem of high similarity between the hemorrhage area and the retinal image background, effectively captures the characteristics of the hemorrhage area, and accurately identifies the location and image of the hemorrhage area.

[0024] (2) The embodiment of the present invention uses the T-distribution differential pyramid for bleeding point detection, which improves the accuracy and efficiency in bleeding area identification.

[0025] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings:

[0027] Figure 1 Schematically shows the steps of a bleeding area detection method based on layer peeling and local features according to an embodiment of the present invention;

[0028] Figure 2 Schematically shows an implementation diagram of a bleeding area detection method based on layer peeling and local features according to an embodiment of the present invention;

[0029] Figure 3 The structure diagram of the bleeding area detection device based on layer peeling and local features in an embodiment of the present invention is schematically shown. DETAILED DESCRIPTION

[0030] The following describes the specific implementation of the embodiment of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the embodiment of the present invention and is not used to limit the embodiment of the present invention.

[0031] Figure 1 The following schematically shows the steps of the bleeding area detection method based on layer peeling and local features according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0032] S01, stripping the optic disc and blood vessels from the original retinal image to obtain a processed image;

[0033] S02. Constructing a T-distribution difference pyramid on the processed image, wherein the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing;

[0034] S03, selecting key points from local extreme value points in the T-distribution difference pyramid space, and obtaining descriptions of the key points;

[0035] S04. Mark each key point on the original retinal image according to the description of the key point, and use the marked area as the hemorrhage area.

[0036] The above implementation utilizes layered peeling technology to directly remove the influence of visible targets such as the optic disc and blood vessels on the hemorrhage area. It also utilizes a T-distributed differential pyramid to identify local features, improving recognition accuracy. Compared to traditional object detection algorithms, this not only eliminates the influence of visible targets on latent target detection but also enhances understanding of latent targets, enabling better identification of hemorrhage areas.

[0037] In some optional embodiments, the original retinal image needs to be preprocessed to adjust the brightness and contrast of the image data. The purpose is to enhance the distinction between the background of the retinal image and the dominant targets such as the optic disc and blood vessels, thereby improving the separation effect of the dominant targets.

[0038] In some optional embodiments, the optic disc recognition function and the vascular recognition function are integrated into a hierarchical peeling network model. The hierarchical peeling network model uses labeled optic disc and vascular images as training samples, employing a twelve-fold cross-validation method for training, and the accuracy of the trained hierarchical peeling network model is verified using images from a test set. Based on an explicit target detection model, labeled optic disc and vascular images are trained using a twelve-fold cross-validation method for training and prediction. The collected dataset is divided into twelve equal parts, eleven of which serve as training sets. After the model is trained, the remaining part serves as a test set for prediction. This cycle is repeated twelve times, ensuring that each dataset can serve as both a training set and a prediction set. Finally, the twelve results are averaged as the final result. This embodiment employs a hierarchical peeling method for training and testing data, and the accuracy of the hierarchical peeling network model is verified using images from the test set. The twelve-fold cross-validation method aims to reduce overfitting to a certain extent and to extract as much valid information as possible from limited data before outputting the model. In some scenarios, the layered peeling model will judge the recognition target and check whether the recognition target is the specified target. If not, it will be re-recognized. If the recognition is successful three times, the recognition will be canceled.

[0039] In some optional embodiments, stripping the optic disc and blood vessels from the original retinal image to obtain a processed image includes: identifying the optic disc area from the original retinal image using an optic disc identification function; identifying the blood vessel area from the original retinal image using a blood vessel identification function; stripping the optic disc area and the blood vessel area from the original retinal image, and filling the stripped areas to obtain the processed image. The specific steps are as follows:

[0040] As the name suggests, layer peeling involves sequentially peeling off the explicit target layer by layer. The first step is to peel off the optic disc. The optic disc recognition function is defined as: S = f1(I); where I is the input pre-processed image, S is the optic disc recognition area, and f1 is the function that marks the optic disc.

[0041] The second step is to strip off the blood vessel area and define the blood vessel identification function as: X=f2(I); where X is the blood vessel identification area and f2 is the function for marking blood vessels.

[0042] Then, the image is stripped according to the identified optic disc and blood vessel areas, and the resulting image is represented by the following formula: I(S, X) = STR(S, X); wherein, I(S, X) represents the image after stripping the optic disc and blood vessels, and the STR function is the stripping function. The principle is to remove the pixels of the identified optic disc and blood vessel areas, and then fill them with the surrounding pixels. The filling method is to average the surrounding pixels. The formula is as follows: I(S, X) = STR(S, X). wherein, P(n) is the pixel value filled in for the pixel, n represents the number of pixels around that have not been stripped, and p i represents the i-th pixel around it, where i∈[1,8].

[0043] Finally, the image after stripping and filling is represented as follows: Wherein, I(x, y) represents the final image after stripping the optic disc and blood vessels and filling the blanks in the stripped area.

[0044] In the above steps, the embodiment of the present invention uses the layered peeling method to identify the optic disc and blood vessels as explicit targets. The output result is the original image after removing the explicit targets, which is then used as input for the subsequent steps. The subsequent steps are to detect the bleeding area using the local feature search and recognition method.

[0045] In some embodiments, a T-distribution difference pyramid is constructed on the processed image, and the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing, including: using the processed image as the first layer of the first group; performing T-distribution convolution on the first layer of the first group, and using the obtained image as the second layer of the first group; updating the smoothing factor in the T-distribution convolution to obtain subsequent layers of images in the first group in sequence; determining the first layer in each subsequent group, and convolving the subsequent layers in each group with T-distributions of different smoothing factors; repeating the above steps to obtain multiple groups of images containing multiple layers of images in each group, and the multiple groups of images constitute the T-distribution difference pyramid. Specifically, constructing a T-distribution pyramid with respect to the original image includes the following steps:

[0046] S021. After doubling the original image, use it as the first layer of the first group and then pass it through the T-distributed convolution. The function is as follows: L(x, y) = T(x, y) * I(x, y). Where T(x, y) represents the T-distributed probability density function, which is expressed as: * represents the convolution operation, (x, y) represents the pixel position of the image, n represents the degree of freedom, and λ is a parameter.

[0047] S022. Multiply λ by a proportional coefficient to obtain a new smoothing factor λ=K*λ, and use it to smooth the first group of second-layer images, and the result is used as the first group of third-layer images.

[0048] S023. Repeat the above process to finally obtain L layers of images. In the same group, the image size of each layer is the same, but the smoothing coefficient λ is different.

[0049] S024. Downsample the third-to-last layer image of the first group by a scaling factor of 2 to obtain the first layer of the second group of images. Then, smooth the first layer image of the second group by a smoothing factor of λ to obtain the second layer of the second group. Just like step 2, the L layer images of the second group are obtained in this way, and the same size as the same group is obtained.

[0050] Repeating the above steps yields O groups, each with L layers, for a total of O*L images. These images together form the image T distribution pyramid. Here, O = [log3 min(M, N)] - 3. M is the row height of the original image, N is the column width of the original image, O is the number of groups in the image T distribution pyramid, c is the number of images to extract image features, and L is the number of layers in each group of the image T distribution pyramid, where L = c + 3.

[0051] Finally, a T-distribution difference pyramid (DoT) is constructed based on the T-distribution pyramid. This T-distribution difference pyramid is crucial for later detection of extreme points. Its construction is achieved by subtracting adjacent layers. Each layer in the T-distribution difference pyramid has one less layer than the T-distribution pyramid, and subsequent feature point extraction is performed within the T-distribution difference pyramid.

[0052] In some embodiments of the present invention, selecting key points from local extreme points in the T-distributed difference pyramid space includes: determining whether a pixel is an extreme point within the middle layer of each group based on a comparison result with surrounding pixels; if the pixel is an extreme point, correcting the position of the extreme point using a Taylor expansion; and using the extreme point after position correction as the key point. Specifically, feature points are composed of local extreme points in the T-distributed difference pyramid space. Therefore, it is necessary to search for extreme points in the T-distributed difference pyramid space by comparing each pixel with its neighbors to determine whether it is larger or smaller than its neighbors in the image domain and scale domain. When searching for extreme points in the T-distributed difference pyramid space, in addition to considering the x and y directions, points in different layer directions must also be considered. Therefore, determining whether a pixel is an extreme point requires comparison with 26 points in the surrounding space. If a group has three layers, the extreme point can only be searched in the middle layer. If a group has five layers, the extreme point is searched in the middle three layers, and so on. After the extreme point is detected, because the scale space and pixel points are discrete, the differential T distribution pyramid is also discrete, so the extreme point found may not be accurate. The real extreme point is likely to be nearby. Therefore, in order to find the extreme point with higher sub-pixel position accuracy, the Taylor expansion is needed to correct the position of the extreme point.

[0053] In some embodiments of the present invention, obtaining a description of the key point includes: assigning a reference direction to each key point; collecting the gradient distribution characteristics and direction distribution characteristics of the pixels in the image domain window of the T-distribution difference pyramid where the key point is located in the reference direction; and describing the key point based on the position information, scale information and direction information of the key point. Specifically, to find the extreme point through scale invariance, it is necessary to use the local features of the image to assign a reference direction to each key point so that the descriptor is invariant to image rotation. For the key point detected in the T-distribution difference pyramid, the gradient and direction distribution characteristics of the pixels in the image domain window of the T-distribution pyramid where the key point is located are collected. The modulus and direction of the gradient are as follows:

[0054]

[0055]

[0056] Through the above steps, we can obtain three pieces of information about a keypoint: location, scale, and orientation. Next, we create a set of vectors to describe this keypoint, ensuring that it remains stable despite various changes (such as lighting and viewing angle). This descriptor includes not only the keypoint but also the surrounding pixels that contribute to it. The descriptor should be highly unique to increase the probability of correct feature point matching.

[0057] In some embodiments of the present invention, each key point is located and marked on the original retinal image based on the key point description, and the marked area is defined as the hemorrhage area. This includes: performing key point matching on different captured images of the same retina; assigning a weight to each key point based on the number of matches; performing a location and marking on the original retinal image based on the weight of each key point; and defining the marked area as the hemorrhage area. Specifically, the descriptions of each key point are derived as described above. Key point matching is performed on different captured images of the same retina, that is, by comparing the descriptors between the two point sets. An exhaustive matching method can then be employed. Retinal image acquisition can introduce errors due to environmental influences, so multiple acquisitions may be performed based on factors such as angle and brightness. The primary purpose of image matching is to reduce errors caused by the image acquisition process. Hemorrhage points should be present in all captured images. Based on the number of matches, the importance of each key point can be determined and a weight assigned. Based on the response weights obtained for each key point, each key point can be located and marked on the original image. The key point is marked with a rectangular border and output to assist ophthalmologists in diagnosis.

[0058] Through the above implementation method, compared with the traditional target detection algorithm, not only the influence of the explicit target on the implicit target detection is eliminated, but also the understanding of the implicit target is enhanced, and the bleeding area is better identified.

[0059] Figure 2 The following schematically shows an implementation diagram of a bleeding area detection method based on layer peeling and local features according to an embodiment of the present invention. Figure 2 As shown, this embodiment includes steps such as data preprocessing, layer stripping, and local feature recognition. The layer stripping step includes identifying explicit targets and determining whether the identified target is the optic disc or a blood vessel. The local feature recognition step includes extracting features to construct a T-distribution difference pyramid, locating spatial extreme points, and describing key points.

[0060] Based on the same inventive concept, some embodiments of the present invention further provide a bleeding area detection device based on layer peeling and local features. Figure 3The schematic diagram of the structure of the bleeding area detection device based on layer peeling and local features in an embodiment of the present invention is shown as follows: Figure 3 As shown, the device includes: an image stripping module, used to strip the optic disc and blood vessels from the original retinal image to obtain a processed image; a pyramid processing module, used to construct a T distribution difference pyramid on the processed image, wherein the T distribution difference pyramid is obtained through multiple cycles of convolution and smoothing; a key point determination module, used to select key points from local extreme points in the T distribution difference pyramid space and obtain descriptions of the key points; and a positioning marking module, used to position and mark each key point on the original retinal image according to the description of the key point, and use the marked area as the hemorrhage area.

[0061] In some optional embodiments, stripping the optic disc and blood vessels from the original retinal image to obtain a processed image includes: identifying the optic disc area from the original retinal image using an optic disc identification function; identifying the blood vessel area from the original retinal image using a blood vessel identification function; stripping the optic disc area and the blood vessel area from the original retinal image, and filling the stripped areas to obtain the processed image.

[0062] In some optional embodiments, the optic disc recognition function and the blood vessel recognition function are integrated into a hierarchical peeling network model; the hierarchical peeling network model uses labeled optic disc and blood vessel images as training samples, adopts a twelve-fold cross-validation method for training, and uses images in a test set to verify whether the trained hierarchical peeling network model is accurate.

[0063] In some optional embodiments, a T-distribution difference pyramid is constructed on the processed image, and the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing, including: taking the processed image as the first layer of the first group; performing T-distribution convolution on the first layer of the first group, and taking the obtained image as the second layer of the first group; updating the smoothing factor in the T-distribution convolution, and sequentially obtaining subsequent layers of images in the first group; determining the first layer in each subsequent group, and obtaining subsequent layers in each group by T-distribution convolution with different smoothing factors; repeating the above steps to obtain multiple groups of images containing multiple layers of images in each group, and the multiple groups of images constitute a T-distribution pyramid; and constructing the T-distribution difference pyramid based on the T-distribution pyramid.

[0064] In some optional embodiments, selecting key points from local extreme points in the T-distributed differential pyramid space includes: in the middle layer of each group, determining whether a certain pixel point is an extreme point based on a comparison result between the certain pixel point and surrounding pixel points; if the certain pixel point is an extreme point, correcting the position of the extreme point using Taylor expansion; and using the extreme point after position correction as the key point.

[0065] In some optional embodiments, obtaining a description of the key point includes: assigning a reference direction to each key point; collecting the gradient distribution characteristics and direction distribution characteristics of pixels in the image domain window of the T-distributed difference pyramid where the key point is located in the reference direction; and describing the key point based on the position information, scale information and direction information of the key point.

[0066] In some optional embodiments, each key point is positioned and marked on the original retinal image according to the description of the key point, and the marked area is used as the hemorrhage area, including: matching key points based on different collected images of the same retina; assigning a weight to each key point according to the number of matches; positioning and marking on the original retinal image according to the weight of each key point; and using the marked area as the hemorrhage area.

[0067] The specific definitions of the various functional modules in the aforementioned bleeding area detection device based on layer peeling and local features can be found in the above-mentioned definitions of the bleeding area detection method based on layer peeling and local features, and will not be repeated here. The various modules in the aforementioned device can be implemented in whole or in part through software, hardware, or a combination thereof. The aforementioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the aforementioned modules.

[0068] In some embodiments provided herein, an electronic device is also provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the aforementioned method for detecting bleeding areas based on layer peeling and local features are implemented. The processor herein has the functions of numerical calculation and logical operations, and comprises at least a central processing unit (CPU) with data processing capabilities, a random access memory (RAM), a read-only memory (ROM), multiple I / O ports, and an interrupt system. The processor includes a kernel, which retrieves corresponding program units from the memory. One or more kernels may be provided, and the aforementioned method is implemented by adjusting kernel parameters. The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. The memory includes at least one memory chip.

[0069] In one embodiment of the present invention, a computer-readable storage medium is also provided, which stores instructions. When the storage medium is run on a computer, the instructions, when executed by a processor, configure the processor to perform the steps of the above-mentioned bleeding area detection method based on layer peeling and local features.

[0070] In one embodiment of the present invention, a computer program product is provided, including a computer program. When executed by a processor, the computer program implements the steps of the above-mentioned bleeding area detection method based on layer peeling and local features.

[0071] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0072] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0075] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0076] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0077] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0078] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0079] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A bleeding area detection method based on layer peeling and local features, characterized in that: The method includes: The optic disc and blood vessels are stripped from the original retinal image to obtain the processed image; Constructing a T-distribution difference pyramid on the processed image, wherein the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing; Selecting key points from local extreme value points in the T-distribution difference pyramid space and obtaining descriptions of the key points; Each key point is positioned and marked on the original retinal image according to the description of the key point, and the marked area is used as the hemorrhage area.

2. The method according to claim 1, characterized in that The optic disc and blood vessels are stripped from the original retinal image to obtain the processed image, including: identifying an optic disc area from the original retinal image using an optic disc identification function; identifying a blood vessel area from the original retinal image using a blood vessel identification function; The optic disc area and the blood vessel area are stripped from the original retinal image, and the stripped areas are filled to obtain a processed image.

3. The method according to claim 2, characterized in that The optic disc recognition function and the blood vessel recognition function are integrated into a layered peeling network model; The hierarchical peeling network model uses labeled optic disc and blood vessel images as training samples, adopts a twelve-fold cross-validation method for training, and uses images in a test set to verify whether the trained hierarchical peeling network model is accurate.

4. The method according to claim 1, wherein A T-distribution difference pyramid is constructed on the processed image, wherein the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing, including: Using the processed image as a first group of first layers; Performing T-distributed convolution on the first group of first layers, and using the resulting image as the first group of second layers; Updating the smoothing factor in the T-distributed convolution to obtain subsequent layers of images in the first group in sequence; Determine the first layer in each subsequent group, and obtain the subsequent layers in each group by convolution with T distribution with different smoothing factors; Repeat the above steps to obtain multiple groups of images, each group containing multiple layers of images, and the multiple groups of images constitute a T distribution pyramid; The T distribution difference pyramid is constructed according to the T distribution pyramid.

5. The method according to claim 4, characterized in that Selecting key points from local extreme points in the T-distribution difference pyramid space includes: In the middle layer of each group, determining whether a certain pixel point is an extreme point based on the comparison result between the certain pixel point and the surrounding pixels; If a certain pixel point is an extreme point, the Taylor expansion is used to correct the position of the extreme point; The extreme point after position correction is taken as the key point.

6. The method according to claim 1, characterized in that Get a description of the key points, including: Assign a reference direction to each key point; Collecting gradient distribution features and direction distribution features of pixels in the image domain window of the T-distribution difference pyramid where the key point is located in the reference direction; The key points are described according to the position information, scale information and direction information of the key points.

7. The method according to claim 6, characterized in that Marking each key point on the original retinal image according to the description of the key point, and using the marked area as the hemorrhage area, including: Key point matching is performed based on different collected images of the same retina; Assign a weight to each key point based on the number of matches; Performing positioning marking on the original retinal image according to the weight of each key point; The marked area is used as the bleeding area.

8. A bleeding area detection device based on layer peeling and local features, characterized in that: The device includes: An image stripping module is used to strip the optic disc and blood vessels from the original retinal image to obtain a processed image; a pyramid processing module, configured to construct a T-distribution difference pyramid on the processed image, wherein the T-distribution difference pyramid is obtained through multiple cycles of convolution and smoothing; a key point determination module, configured to select key points from local extreme value points in the T-distribution difference pyramid space and obtain descriptions of the key points; and A positioning marking module is used to mark each key point on the original retinal image according to the description of the key point, and use the marked area as the hemorrhage area.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the bleeding area detection method based on layer peeling and local features according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing instructions which, when executed on a computer, enable the computer to execute the steps of the method for detecting bleeding areas based on layer peeling and local features according to any one of claims 1 to 7.

Images