Method for acquiring edema feature in retinal image, and electronic device

Abstract

The embodiments of the present application relate to the technical field of image recognition. Provided are a method for acquiring an edema feature in a retinal image, and an electronic device. The method for acquiring an edema feature in a retinal image comprises: acquiring, from among a plurality of retinal images, a current retinal image to be identified; determining an edema brightness threshold value of the current retinal image on the basis of brightness values of pixels within a vitreous region of the current retinal image; and on the basis of the brightness values of the pixels in the current retinal image and the edema brightness threshold value of the current retinal image, determining an edema region in the current retinal image, wherein brightness values of pixels in the edema region are less than the edema brightness threshold value. By means of the present application, feature extraction can be performed on an edema region in an image on the basis of brightness values of pixels in a retinal image.

Description

Methods and electronic devices for obtaining edema features in retinal images

[0001] Cross-referencing related applications

[0002] This patent application claims priority to Chinese Patent Application No. 2025101813445, filed on February 18, 2025, entitled “Method for obtaining macular edema features in retinal images and electronic device”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application relates to the field of image recognition technology, specifically to a method for obtaining edema features in retinal images and an electronic device. Background Technology

[0004] The retina is a thin layer of cells located at the back of the eye. It consists of photoreceptor cells (receptor cells) and pigment epithelial cells. When stimulated by light, the retina converts the light signal into a nerve signal and transmits it to the brain, thus enabling us to see. The retina is a very sensitive, thin, and complex structure, composed of photoreceptor cells, bipolar cells, ganglion cells, and more. Not only can eye diseases be reflected in the state of the retina, but some metabolic-related psychosomatic diseases can also cause changes in the retina, such as diabetic retinopathy.

[0005] Retinal edema is a pathological swelling caused by fluid accumulation between the layers of the retina. It is one of the common causes of blindness. The core pathogenesis of retinal edema is damage to the blood-retinal barrier, increased vascular permeability, and leakage of fluid into the retinal neuroepithelial layer or pigment epithelium. Common related diseases that can cause retinal edema include diabetic retinopathy, age-related macular degeneration, and uveitis. These diseases can severely impair central vision and affect patients' daily lives.

[0006] For retinal edema, ophthalmologists need to utilize advanced diagnostic equipment and techniques, such as optical coherence tomography (OCT) and fluorescein fundus angiography (FFA), to achieve early diagnosis and precise treatment. Treatment for retinal edema caused by disease involves addressing the underlying condition, along with intravitreal injections of anti-vascular endothelial growth factor (anti-VEGF) drugs, corticosteroid therapy, and, if necessary, laser photocoagulation or surgery. Most patients experience significant visual improvement after standardized treatment; however, delayed treatment leading to retinal structural damage can result in permanent visual impairment. Prognosis is closely related to the cause and timing of treatment.

[0007] Currently, the diagnosis of retinal edema primarily relies on trained radiologists performing eye examinations. Experienced ophthalmologists then interpret the acquired images to determine if a patient has retinal edema-related conditions. However, there is a severe shortage of ophthalmologists and radiologists, far from meeting clinical needs. Young ophthalmologists also require extensive training and development to accumulate experience and improve their image interpretation skills. This traditional method of manually interpreting images—selecting those with pathological tissue from large datasets and drawing diagnostic conclusions from subtle tissue lesions—is insufficient for large-scale, precise, and personalized services in routine healthcare. It cannot accurately identify, automatically analyze, or quantitatively track and evaluate lesions, and thus falls short of the goals of future precision medicine.

[0008] Meanwhile, due to the rapid development of OCT technology, its clinical promotion and application have become more widespread, leading to a large and rapid accumulation of OCT image data. With technological iteration, the image quality, imaging range, and resolution presented by OCT have also greatly improved. Currently available diagnostic analysis software for OCT images can only roughly assess the thickness of the center of retinal lesions. It cannot achieve effective accuracy in automatically identifying retinal layers or assessing the thickness and volume of peripheral retinal edema. This results in poor accuracy in the identification and assessment of retinal edema; it also cannot accurately identify and locate lesions, precisely identify the affected tissue area, and perform quantitative statistics such as edema area and subretinal fluid area. Furthermore, it cannot objectively monitor and record the specific changes in lesion size during patient follow-up, let alone obtain effective data through patient follow-up and treatment to support or predict prognosis. The current predicament not only hinders future in-depth research in the field of retinal diseases but also fails to meet the objective quantitative analysis needs of patients, hindering the transition from qualitative to quantitative precision and thus failing to contribute to precision medicine.

[0009] Therefore, if intelligent identification and quantitative analysis of retinal edema lesions can be achieved, and the acquired OCT images can be conveniently processed through an effective computer-aided image analysis system, providing clinicians with quantitative and objective measurements, it can greatly assist doctors in making clinical decisions, predicting disease prognosis, analyzing disease outcomes through big data, and accurately and personally developing patient treatment plans.

[0010] However, the accuracy of existing clinical diagnostic analysis software for OCT images is often limited by the turbidity of the refractive media and the ability of technicians to capture images, making it impossible to effectively provide objective quantitative results of lesions. In addition, describing and quantifying changes in lesion areas by using central thickness is not completely consistent and cannot fully explain the nature of lesion changes.

[0011] Thanks to the rapid development of artificial intelligence technology, it is now possible to automatically identify and label lesion areas by inputting a large number of manually labeled OCT images using artificial intelligence. This data can then be used to collect quantitative information about the lesions. Most commonly used retinal edema disease identification systems rely on machine learning models, which are trained using a sufficient number of labeled OCT images to obtain machine learning models with theoretically high accuracy.

[0012] However, this is only theoretical, and there are many problems in implementing it. For example, machine learning training requires a large number of labeled OCT images, but OCT image labeling requires specialized training from ophthalmologists. Professional ophthalmologists often don't have the time to focus on OCT image labeling, making it extremely difficult and time-consuming to obtain high-precision labeled data. This high labor cost is one of the barriers to using AI technology for OCT image recognition. Furthermore, the lesions in different diseases are different, requiring separate training for each disease. Using the same model for lesion identification across all diseases would result in poor accuracy, further increasing the difficulty of AI for OCT image recognition. Also, machine learning models have high requirements for training and testing data. If the model encounters special cases or the OCT imaging equipment is changed, the robustness and accuracy of the trained model will decrease.

[0013] The aforementioned problems demonstrate the immense difficulty of applying artificial intelligence technology to OCT image recognition. Furthermore, even if artificial intelligence technology is applied to OCT image recognition, it remains uncertain whether it can achieve the desired recognition results. Summary of the Invention

[0014] The purpose of this application is to provide a method, electronic device, and storage medium for acquiring edema features in retinal images. Based on the brightness values ​​of pixels within the vitreous region of the retinal image, an edema brightness threshold is determined. Then, based on this edema brightness threshold, pixels in the retinal image are binary-coded to distinguish between edema and noise, thus removing non-edema noise and achieving feature extraction of edema from the retinal image. In other words, edema features can be automatically extracted based on pixel-level information in the retinal image. Through automatic encoding and compilation techniques, the edema region of the retinal image can be obtained using only a computer. Compared to machine learning for edema recognition, it does not require training with data or high-performance hardware, making it simpler and easier to use. It achieves a high signal-to-noise ratio, high accuracy, fast recognition speed, and low hardware requirements, making it suitable for various edema feature extraction scenarios.

[0015] To achieve the above objective, this application provides a method for obtaining edema features in a retinal image, comprising: acquiring a current retinal image to be identified from multiple retinal images; determining an edema brightness threshold of the current retinal image based on the brightness values ​​of pixels in the vitreous region of the current retinal image; determining an edema region in the current retinal image based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, wherein the brightness values ​​of pixels in the edema region are less than the edema brightness threshold.

[0016] This application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the edema feature acquisition method in retinal images as described above.

[0017] This application also provides a computer-readable storage medium, which is a non-volatile or non-transient storage medium, on which a computer program is stored, characterized in that the computer program is executed by a processor to perform the above-described method for obtaining edema features in retinal images.

[0018] In one embodiment, the method further includes:

[0019] The current retinal image is subjected to various preprocessing steps to obtain multiple reference retinal images after various preprocessing steps;

[0020] Based on the brightness value of each pixel in each of the reference retinal images and the edema brightness threshold of each of the reference retinal images, the edema region in each of the reference retinal images is determined;

[0021] Based on the edema regions of each of the reference retinal images and the edema region of the current retinal image, as well as preset weight parameters, the effective edema region in the current retinal image is determined. The preset weight parameters include the weights of each of the reference retinal images and the current retinal image.

[0022] In one embodiment, before determining the edema region in the current retinal image based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, the method further includes:

[0023] The RPE layer of the current retinal image is identified based on the brightness values ​​of each pixel within the current retinal image;

[0024] Based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, the edema region in the current retinal image is determined, including:

[0025] Based on the RPE layer of the current retinal image, the brightness value of each pixel in the current retinal image, and the edema brightness threshold of the current retinal image, the edema region in the current retinal image is determined, and the pixels in the edema region are all located above the RPE layer.

[0026] In one embodiment, determining the edema brightness threshold of the current retinal image based on the brightness values ​​of pixels within the vitreous region of the current retinal image includes:

[0027] An initial brightness threshold is obtained for the brightness values ​​of pixels that are greater than a preset percentage of all pixels in the vitreous region, wherein the preset percentage is greater than 50%.

[0028] Within the vitreous region, target pixels with brightness values ​​within a set brightness value range are located, and based on the target brightness value of each target pixel, the number of target pixels with each target brightness value is counted; wherein the set brightness value range is the brightness value range with the initial brightness threshold as the midpoint.

[0029] Among all the target brightness values, find the target brightness value with the smallest difference in the number of target pixels between it and the initial brightness threshold, and use it as the minimum change brightness threshold;

[0030] The edema brightness threshold is determined based on the initial brightness threshold and the minimum change brightness threshold.

[0031] In one embodiment, determining the edema brightness threshold based on the initial brightness threshold and the minimum change brightness threshold includes:

[0032] The product of the average of the initial brightness threshold and the minimum change brightness threshold and a preset correction coefficient is obtained as the edema brightness threshold, wherein the preset correction coefficient is greater than 0.5.

[0033] In one embodiment, determining the edema region in the current retinal image based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image includes:

[0034] In the current retinal image, among the closed regions composed of pixels whose brightness values ​​are less than the edema brightness threshold, a closed region whose area meets the preset conditions is selected as the edema region.

[0035] The preset condition is that the area of ​​the closed region is less than the product of the effective area in the current retinal image and the set threshold, and is greater than the area of ​​the preset number of pixels. The set threshold is greater than 0 and less than 0.5.

[0036] In one embodiment, after determining the edema region in the current retinal image based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, the method further includes:

[0037] For each edema region, the area of ​​the edema region is determined based on the pixel spacing information of the current retinal image and the number of pixels within the edema region.

[0038] In one embodiment, after determining the area of ​​each edema region based on the pixel spacing information of the current retinal image and the number of pixels within the edema region, the method further includes:

[0039] The volume of each edema region in the current retinal image is obtained based on the area of ​​each edema region in the current retinal image and the image spacing information of the current retinal image.

[0040] In one embodiment, before determining the area of ​​each edema region based on the pixel spacing information of the current retinal image and the number of pixels within the edema region, the method further includes:

[0041] The current retinal image is partitioned into ETDRS (Early Treatment Diabetic Retinopathy Study) regions to obtain the edema areas located within the ETDRS regions;

[0042] For each edema region, the area of ​​the edema region is determined based on the pixel spacing information of the current retinal image and the number of pixels within the edema region, including:

[0043] For each edema region located within an ETDRS partition, the area of ​​the edema region is determined based on the pixel spacing information of the current retinal image and the number of pixels within the edema region.

[0044] In one embodiment, identifying the RPE layer of the current retinal image based on the brightness values ​​of each pixel within the current retinal image includes:

[0045] Within the effective area of ​​the current retinal image, the specified pixel with the highest brightness is searched column by column. For the specified pixel in the current column within the effective area of ​​the current retinal image, if the difference in the vertical axis position between the specified pixel in the current column and the specified pixel in the adjacent previous column is greater than a first difference threshold, the vertical axis position of the specified pixel in the current column is determined based on the vertical axis position of the specified pixel in the adjacent previous column.

[0046] The RPE layer of the current retinal image is obtained by using the specified pixel combinations in each column within the effective area of ​​the current retinal image.

[0047] In one embodiment, searching for the brightest target pixel column-wise within the effective area of ​​the current retinal image includes:

[0048] Starting from any middle column of the effective area of ​​the current retinal image, search to the left and right respectively for the target pixel with the highest brightness in each column.

[0049] In one embodiment, after obtaining the RPE layer of the current retinal image using the specified pixel combinations in each column within the effective area of ​​the current retinal image, the method further includes:

[0050] Acquire multiple specified retinal images associated with the current retinal image, and determine the pixel reference height of the RPE layer based on the average height of the pixels contained in the RPE layer of the multiple specified retinal images;

[0051] If the difference between the average pixel height of the pixels contained in the RPE layer of the current retinal image and the pixel reference height is greater than a second difference threshold, then the RPE layer of the current retinal image is determined based on the vertical axis position of the pixels contained in the RPE layer of a specified retinal image adjacent to the current retinal image.

[0052] In one embodiment, the vitreous region of the current retinal image is determined as follows:

[0053] The effective area of ​​the current retinal image is sampled sequentially using a sampling frame of a set size. If the average brightness value of the pixels in the sampling area currently selected by the sampling frame is within a preset brightness range, then the sampling area currently selected by the sampling frame is determined to be the vitreous region. Attached Figure Description

[0054] Figure 1 is a schematic diagram of a method for obtaining edema features in a retinal image according to the first embodiment of this application;

[0055] Figure 2 is a schematic diagram of the original retinal image according to the first embodiment of this application;

[0056] Figure 3 is a schematic diagram of the layered structure of the retinal image in Figure 2;

[0057] Figure 4 is a flowchart of step 102 of the method for obtaining edema features in the retinal image in Figure 1.

[0058] Figure 5 is a schematic diagram of the edema region identified in the original retinal image by the edema feature acquisition method in the first embodiment of this application.

[0059] Figure 6 is a schematic diagram of the current retinal image partitioned by ETDRS according to the first embodiment of this application;

[0060] Figure 7 is a schematic diagram of a method for obtaining edema features in a retinal image according to a second embodiment of this application;

[0061] Figure 8 is a schematic diagram of the current retinal image before and after non-local mean processing and sharpening processing according to the second embodiment of this application;

[0062] Figure 9 is a schematic diagram of the current retinal image before and after automatic contrast adjustment processing according to the second embodiment of this application;

[0063] Figure 10 is a schematic diagram of the current retinal image before and after sharpening according to the second embodiment of this application;

[0064] Figure 11 is a schematic diagram of the current retinal image before and after processing by the Otsu method according to the second embodiment of this application;

[0065] Figure 12 is a schematic diagram of a method for obtaining edema features in a retinal image according to a third embodiment of this application;

[0066] Figure 13 is a schematic diagram of the portion of the current retinal image located below the RPE layer before and after being adjusted to black according to the third embodiment of this application;

[0067] Figure 14 is a schematic diagram of the edema region identified in the current retinal image with the portion below the RPE layer adjusted to black according to the edema feature acquisition method in the retinal image according to the third embodiment of this application.

[0068] Figure 15 is a schematic diagram of the missed labeling rate of edema regions after verification, which is obtained by the edema feature acquisition method in retinal images according to the third embodiment of this application.

[0069] Figure 16 is a schematic diagram of the multi-labeling rate of edema regions after verification, which is obtained by identifying edema regions in multiple example retinal images according to the edema feature acquisition method in the third embodiment of this application. Specific Implementation

[0070] The embodiments of this application will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the purpose, features, and advantages of this application. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of this application, but are merely for illustrating the essential spirit of the technical solution of this application.

[0071] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0072] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0073] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0074] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to include the meaning of “or / and” unless otherwise expressly stated herein.

[0075] In the following description, in order to clearly demonstrate the structure and working method of this application, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0076] The first embodiment of this application relates to a method for obtaining edema features in retinal images, applied to electronic devices, such as common computer devices like desktop computers and laptops, or mobile phones. The method for obtaining edema features in retinal images using this embodiment can extract edema features caused by retinal diseases (with macular degeneration as a typical example); that is, this embodiment can detect edema appearing within the range of retinal images obtained from standard OCT examinations, with the detection range covering the entire retinal OCT image.

[0077] Figure 1 shows a flowchart of the method for obtaining edema features in retinal images according to this embodiment.

[0078] Step 101: Obtain the current retinal image to be identified from multiple retinal images. The retinal image is an OCT image of the fundus.

[0079] Specifically, retinal images are OCT images of the user's fundus acquired by an OCT device (Optical Coherence Tomography). A computer device can directly connect to the OCT device to acquire DICOM format files sent by the OCT device, or it can connect to other intermediate data collection devices or storage to acquire DICOM format files acquired by the OCT device. The DICOM format file includes the retinal OCT image and various other information related to that image, such as: capture date, serial number, patient ID, device serial number, pixel spacing, and image spacing. The computer device can use DICOM file recognition software to parse the acquired DICOM format file to obtain the retinal image and information about each retinal image. The retinal images can be stored in common formats (such as JPG, PNG, etc.) for easy subsequent image processing. The information of each retinal image is converted into elements accessible by tag name. Figure 2 shows a retinal OCT image extracted from a DICOM format file, which is the original retinal image. Please refer to Figure 3, which is a schematic diagram of the layered structure of the eye in a retinal image. From top to bottom, they are: ILM (Internal Limiting Membrane), RNFL (Retinal Nerve Fibre Layer), GCL (Ganglion Cell Layer), IPL (Inner Plexiform Layer), INL (Inner Nuclear Layer), OPL (Outer Plexiform Layer), ONL (Outer Nuclear Layer), ELM (External Limiting Membrane), PR (Photoreceptor Layers), RPE (Retinal Pigment Epithelium), BM (Bruch's Membrane), CC (Choriocapillaris), and CS (Choroidal Stroma).

[0080] Therefore, the multiple retinal images acquired by the computer device are obtained by slicing a single three-dimensional image of the user. Image recognition is performed on these multiple retinal images sequentially. After the recognition of each retinal image is completed, the next retinal image to be recognized (i.e., the next image obtained from slicing) is determined, which is the current retinal image. The brightness values ​​of each pixel in the current retinal image can be read first. Among the multiple retinal images obtained after slicing a single three-dimensional image of the user, the image spacing between any two adjacent retinal images is also known. This image spacing can be a fixed value or a non-fixed value; that is, the image spacing between any two adjacent retinal images can be the same or different.

[0081] Step 102: Determine the edema brightness threshold of the current retinal image based on the brightness values ​​of the pixels in the vitreous region of the current retinal image.

[0082] Specifically, based on the similarity between the brightness of the edematous region of the retina and the brightness of the vitreous region in the retina, the brightness threshold of the edematous region in the retina can be determined by referring to the brightness of the vitreous region in the retina, and is denoted as the edema brightness threshold.

[0083] The vitreous region in the current retinal image is the region on the ILM layer. The specific method for determining the vitreous region in the current retinal image is as follows:

[0084] First, obtain the effective area of ​​the current retinal image to remove the black or white borders at the edges of the current retinal image caused by the OCT capturing area exceeding the imaging area.

[0085] The specific process involves starting from the edge of the current retinal image and searching row by row and column by column to find black or white borders that extend beyond the imaging area, thus obtaining the effective area of ​​the current retinal image. Taking the top edge of the current retinal image as an example, starting from the first row of the top edge, the average brightness of all pixels contained in that row is calculated. If the average brightness value of that row is greater than 50 or less than 5, then that row is determined to be a black or white border extending beyond the imaging area. This process continues, checking whether the average brightness of all pixels in the second row is greater than 50 or less than 5, until a row with an average brightness of no more than 50 or less than 5 is reached. Similarly, rows and / or columns where the four edges of the current retinal image extend beyond the imaging area (i.e., the effective area) can be determined. These rows and columns form the borders of the effective area, and the image within the borders is the effective area of ​​the current retinal image.

[0086] Then, the vitreous region is determined within the effective area of ​​the current retinal image. The determination method is as follows: the effective area of ​​the current retinal image is sampled sequentially using a sampling frame of a set size. If the average brightness value of the pixels within the sampling area currently selected by the sampling frame is within a preset brightness range, then the sampling area currently selected by the sampling frame is determined to be the vitreous region. The sampling frame can be set as needed, generally a square or rectangular frame, such as a square frame with a side length of 50 pixels. The size of the sampling frame should not be too large to avoid sampling white pixels from non-vitreous regions, which would affect the recognition of the vitreous region.

[0087] The specific process is as follows: The sampling frame traverses the effective area of ​​the current retinal image sequentially from right to left and from top to bottom. During each sampling, the sampling frame selects a sampling region within the effective area of ​​the current retinal image. Then, the average brightness value of all pixels within the sampling region is calculated. If the average brightness value is within a preset brightness range, the sampling region is determined to be the vitreous region; if the average brightness value is outside the preset brightness range, the sampling region is determined not to be the vitreous region. The process then moves to the next position for sampling until a vitreous region is determined. The preset brightness range is, for example, 30-40.

[0088] It should be noted that the vitreous region determined in this embodiment is only a part of the vitreous region, and not the complete vitreous region.

[0089] Subsequently, based on the brightness values ​​of pixels within the vitreous region of the current retinal image, the edema brightness threshold of the current retinal image can be determined. Please refer to Figure 4. Step 102 specifically includes the following sub-steps:

[0090] Sub-step 1021: Obtain an initial brightness threshold for pixels whose brightness values ​​are greater than a preset percentage of all pixels in the glass region, where the preset percentage is greater than 50%.

[0091] Sub-step 1022: Find target pixels in the vitreous region whose brightness values ​​are within the set brightness value range, and count the number of target pixels with each target brightness value based on the target brightness value of each target pixel; wherein the set brightness value range is the brightness value range with the initial brightness threshold as the midpoint.

[0092] Sub-step 1023: Find the target brightness value with the smallest difference in the number of target pixels between all target brightness values ​​and the initial brightness threshold as the minimum change brightness threshold.

[0093] Sub-step 1024: Determine the edema brightness threshold based on the initial brightness threshold and the minimum change brightness threshold.

[0094] Specifically, in step 102, a glass region is determined. All pixels in the glass region can be arranged in ascending order of brightness. The brightness value of a preset percentage of pixels that can cover the glass region is set as the initial brightness threshold. For example, if the preset percentage is 95%, then the determined initial brightness threshold is greater than the brightness value of 95% of the pixels in the glass region.

[0095] A set brightness value range is obtained with the initial brightness threshold as the middle. For example, the brightness range of ±10 brightness of the initial brightness threshold is used as the set brightness value range; for example, if the initial brightness threshold is 35, then the set brightness value range is (25, 45).

[0096] Then, among all pixels in the vitreous region, find pixels whose brightness values ​​are within the set brightness value range and record them as target pixels. Each target pixel has a corresponding brightness value, recorded as the target brightness value. Then, count the number of target pixels corresponding to each target brightness value. The initial brightness threshold also has its corresponding number of target pixels. Then, calculate the difference between the number of target pixels at the initial brightness threshold and the number of target pixels at each target brightness value. Select the target brightness value corresponding to the smallest difference as the minimum change brightness threshold.

[0097] The product of the average of the initial brightness threshold and the minimum brightness change threshold and the preset correction coefficient is then obtained as the edema brightness threshold. The preset correction coefficient is greater than 0.5 to reduce excessive identification during subsequent edema region identification. For example, if the preset correction coefficient is 0.8, the edema brightness threshold P = 0.8 × (L1 + L2) / 2, where L1 represents the initial brightness threshold and L2 represents the minimum brightness change threshold. The edema brightness threshold obtained by setting the preset correction coefficient to 0.8 can improve the accuracy of identifying edema regions in the retinal image to a certain extent.

[0098] Step 103: Based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, determine the edema region in the current retinal image, where the brightness values ​​of the pixels in the edema region are less than the edema brightness threshold.

[0099] Specifically, based on the edema brightness threshold, the system finds the locations of feature pixels in the current retinal image whose brightness values ​​are lower than the threshold. Then, the region formed by these feature pixels is considered the edema region. In other words, the system performs a binary classification of the current retinal image based on the edema brightness threshold, distinguishing between edematous feature pixels and non-edematous noise pixels. This allows for feature extraction of the edema region in the current retinal image. Feature extraction here can be understood as determining pixels belonging to the edema region. For example, by binarizing the pixels in the current retinal image using the edema brightness threshold as a boundary, pixels with brightness values ​​greater than the threshold are considered background regions and assigned a value of 0; pixels with brightness values ​​less than or equal to the threshold are considered potential edema regions and assigned a value of 1. This yields a binary image of the current retinal image. Alternatively, when extracting features of the edema region from the current retinal image, the extraction can be performed only on the effective regions within the current retinal image.

[0100] The pixel position mentioned in this embodiment and subsequent embodiments refers to the pixel's coordinates in the image. A coordinate system is constructed with the pixel in the lower left corner of the image as the origin, the bottom row of pixels as the X-axis, and the leftmost column of pixels as the Y-axis, so that the position of all pixels in the image can be coordinated.

[0101] In one example, based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, the edema region in the current retinal image is determined. This includes: selecting a closed region whose area meets preset conditions from the closed regions composed of pixels whose brightness values ​​are less than the edema brightness threshold in the current retinal image as the edema region; the preset conditions are that the area of ​​the closed region is less than the product of the effective area in the current retinal image and a set threshold, and greater than the area of ​​a preset number of pixels; the set threshold is greater than 0 and less than 0.5, for example, 0.2; the preset number of pixels is the area of ​​a certain number of pixels, such as the area of ​​two pixels; this can reduce false recognition of the vitreous region and achieve noise reduction. Please refer to Figure 5, where the pure black pixels marked within the circular and elliptical dashed boxes in the current retinal image form the identified edema region.

[0102] In one example, after step 103, the following is also included:

[0103] Step 104: For each edema region, determine the area of ​​the edema region based on the pixel spacing information of the current retinal image and the number of pixels within the edema region.

[0104] Specifically, after identifying the edema regions contained in the current retinal image, the number of pixels within each edema region is counted. Based on the pixel spacing information of the current retinal image, the area of ​​a single pixel in the current retinal image can be calculated. Then, the number of pixels in each edema region is multiplied by the area of ​​a single pixel to obtain the area of ​​each edema region. Thus, by summing the areas of all edema regions, the total area of ​​the edema regions in the current retinal image can be obtained.

[0105] The pixel spacing information of the current retinal image can be extracted from the DICOM file of the current retinal image. The pixel spacing information represents the length of each pixel in actual space, including the two pixel spacings (x1, y1) on the x-axis and y-axis. x1 can be equal to y1, representing a square pixel, or x1 can be equal to y1, representing a rectangular pixel. For example, if the pixel spacing is (0.5, 0.5) mm, and a certain edema region includes K pixels, then the area of ​​the edema region is 0.25K, in square millimeters.

[0106] Step 105: Based on the area of ​​each edema region in the current retinal image and the image spacing information between the current retinal image and adjacent retinal images, obtain the volume of each edema region in the current retinal image.

[0107] Specifically, image spacing information can be extracted from the DICOM file of the current retinal image. Image spacing information represents the distance between the current retinal image and the adjacent slice of the previous retinal image. The area of ​​each edema region on the current retinal image has been obtained in step 104. For each edema region, the area of ​​the edema region is integrated using the above-mentioned image spacing information to obtain the volume of the edema region. The integration method is, for example, Simpson's integral. By repeating the above process, the volume of all edema regions on the current retinal image relative to the previous retinal image can be obtained.

[0108] By repeating the above process, the volume of the edema region in the second to last retinal images relative to the corresponding previous retinal image can be obtained. The volume of the current user's edema can be obtained by adding up the volumes of all these edema regions. In other words, the area of ​​each edema region can be integrated separately using the image spacing information, so that the calculated edema volume has a high accuracy.

[0109] Furthermore, after identifying all edema areas in the retinal images, the current retinal image is first divided into ETDRS partitions to obtain the edema areas located within the ETDRS partitions. For example, taking retinal edema caused by macular degeneration as an example, multiple circles are drawn with the fovea as the center to further segment the retinal edema area. Specifically, circles with diameters of 1mm, 3mm, and 6mm are drawn with the fovea as the center. Then, the 6mm diameter circle is divided into four equal parts. This divides the edema area into fan-shaped regions with diameters of 1mm, 1-3mm, and 3-6mm. Furthermore, the bisector can be extended to divide the area outside the 6mm diameter circle into four regions, and the edema in these regions will also be included in the subsequent edema region calculation. Taking Figure 6 as an example, region 0 is a circle with a diameter of 1mm, regions 0 to 4 form a circle with a diameter of 3mm, regions 0 to 8 form a circle with a diameter of 6mm, and the rectangular frame is the extension area of ​​the 6mm diameter circle, which is a 6mm×6mm square. Regions 9-12 in this square will also be included in the subsequent edema region calculation. In some examples, the edema region markings outside the 6mm diameter circle can also be deleted.

[0110] Subsequent calculations of the area and volume of edema regions are performed on the edema regions within the ETDRS partition (circular area) and its extended area (6mm × 6mm square). If the edema region markers outside the circle are removed, the area and volume calculations can be performed only on the edema regions within the ETDRS partition (circular area). The ETDRS partitioning method is a partitioning method defined by the Institute for Early Treatment of Diabetic Retinopathy. By partitioning retinal images into ETDRS partitions, edema regions with certain clinical significance can be obtained. Therefore, calculating the area and volume of edema regions within the ETDRS partitions allows the calculated edema area and volume to have more prominent clinical indicative significance.

[0111] In steps 104 to 105, after determining all edema regions in the current retinal image, the area and volume of the edema regions in the current retinal image can be calculated by combining the pixel spacing information and image spacing information of the current retinal image. This quantitatively characterizes the features of the edema regions in the retinal image, realizing quantitative analysis of the edema regions in the retina and providing accurate quantitative analysis results for reference.

[0112] After calculating the area and volume of the edema region in the current retinal image, the annotation information of the current retinal image can be further enriched. The output is an OCT image set with the final edema region annotated on the current retinal image, including the shooting date, serial number, patient ID, device serial number, shooting date, edema area array, edema volume file, the original OCT image of the current retinal image, and the OCT image with the final edema region marked.

[0113] In this embodiment, the edema brightness threshold of the retinal image can be determined based on the brightness value of pixels in the vitreous region of the retinal image. Then, based on the edema brightness threshold, the pixels in the retinal image are binary divided into edema and noise, which can remove non-edema noise in the retinal image and realize the feature extraction of edema in the retinal image. In other words, edema features can be automatically extracted based on pixel-level information of the retinal image. Through automatic encoding and compilation technology, the edema region of the retinal image can be obtained using only a computer. Compared with machine learning for edema recognition, it does not require training with data or high-performance hardware, making it simpler and easier to use. It achieves a high signal-to-noise ratio, with high accuracy, fast recognition speed and low hardware requirements, and is suitable for various edema feature extraction scenarios.

[0114] The second embodiment of this application relates to a method for obtaining edema features in a retinal image. Compared with the first embodiment, this embodiment adds a correction for the edema region in the current retinal image.

[0115] Figure 7 shows a flowchart of the method for obtaining edema features in retinal images according to this embodiment.

[0116] Step 201: Obtain the current retinal image to be identified from multiple retinal images. This is largely the same as step 101 in the first embodiment, and will not be described again here.

[0117] Step 202: Perform various preprocessing steps on the current retinal image to obtain multiple reference retinal images after various preprocessing steps.

[0118] Specifically, the current retinal image is preprocessed in different ways, including but not limited to: non-local mean processing, automatic contrast adjustment processing, sharpening processing, and Otsu's method processing.

[0119] Taking the current retinal image after each of the above four preprocessing steps as an example:

[0120] Method 1: Non-local means processing, such as fast non-local means denoising, considers the self-similarity of the image and makes full use of redundant information in the image, preserving the image's detailed features to the greatest extent while denoising. However, non-local means processing loses edge information, so the image obtained after non-local means processing is usually sharpened to enhance the edge information. Sharpening methods include image convolution operations, such as using filter2D convolution, which uses a 3*3 matrix operator (e.g., [0, -1, 0], [-1, 5, -1], [0, -1, 0]) to convolve the image. As shown in Figure 8, Figure 8a is the original image of the current retinal image, Figure 8b is the reference retinal image obtained after non-local means processing of the current retinal image, and Figure 8c is the image obtained after sharpening the reference retinal image in Figure 8b.

[0121] Method 2: Automatic contrast adjustment processing. This method automatically enhances the contrast of the current retinal image, increasing the contrast between bright and dark areas. This makes the brightness difference between the edematous area and the surrounding medium more obvious, resulting in a corresponding reference retinal image. As shown in Figure 9, Figure 9a is the original image of the current retinal image, and Figure 9b is the reference retinal image obtained after contrast enhancement processing.

[0122] Method 3: Sharpening. This method directly sharpens the current retinal image, enhancing the edge information between bright and dark areas. For details, refer to the sharpening process in Method 1; it will not be elaborated further here. As shown in Figure 10, Figure 10a is the original image of the current retinal image, and Figure 10b is the reference retinal image obtained after sharpening the current retinal image.

[0123] Method 4: Otsu's method. This method applies Otsu's method to the current retinal image, using the Otsu threshold. Based on the distribution of pixels with different brightness levels, darker peaks are defined as background, and brighter peaks as foreground. The brightness threshold is selected based on the brightness that maximizes the inter-class variance between the foreground and background images. Otsu's method is unaffected by brightness and contrast, effectively highlighting foreground information in the current retinal image. As shown in Figure 11, Figure 11a is the original image of the current retinal image, and Figure 11b is the reference retinal image obtained after processing the current retinal image using Otsu's method.

[0124] Step 203: Based on the brightness value of each pixel in each reference retinal image and the edema brightness threshold of each reference retinal image, determine the edema region in each reference retinal image.

[0125] Specifically, the edema regions in each reference retinal image are determined. The specific methods for determining the edema brightness threshold of the reference retinal image and the edema regions in the reference retinal image are similar to the methods for determining the edema regions of the current retinal image in steps 102 and 103 of the first embodiment, and will not be repeated here.

[0126] Step 204: Determine the edema brightness threshold of the current retinal image based on the brightness values ​​of pixels within the vitreous region of the current retinal image. This is largely the same as step 102 in the first embodiment and will not be described again here.

[0127] Step 205: Based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, determine the edema region in the current retinal image. This is largely the same as step 103 in the first embodiment and will not be described again here.

[0128] Step 206: Based on the edema regions of each reference retinal image and the edema region of the current retinal image, as well as preset weight parameters, determine the effective edema region in the current retinal image. The preset weight parameters include the weights of each reference retinal image and the current retinal image.

[0129] After the above process, the current retinal image undergoes the four preprocessing steps described above, resulting in four reference retinal images. These four reference retinal images, along with the current retinal image, are then combined with edema region identification to obtain five images containing potential edema regions. The pixels in these five images are binarized, with pixels in edema regions having a value of 1 and pixels in background regions having a value of 0. Then, the pixel values ​​at each location in the five images are multiplied by the corresponding weight for each image, and a sum is calculated to obtain the score for each pixel location. For each pixel location, if the pixel's score is greater than a preset value (e.g., 70), the pixel is determined to belong to a valid edema region; otherwise, it is determined not to belong to a valid edema region.

[0130] Each of the four reference retinal images and the current retinal image has a corresponding weight. For example, the reference retinal image obtained by method 1 (non-local mean processing) has a weight of 40; the reference retinal image obtained by method 2 (automatic contrast adjustment processing) has a weight of 20; the reference retinal image obtained by method 3 (sharpening processing) has a weight of 50; and the reference retinal image obtained by method 4 (Otsu's method processing) has a weight of 20. The current retinal image has a weight of 40. For instance, if the pixel values ​​at a certain location in the five images are 0, 1, 1, 0, and 1 respectively, then the score for that pixel is (40×0) + (20×1) + (50×1) + (20×0) + (40×1) = 110. Since 110 is greater than the preset score of 70, the pixel is determined to belong to an effective edema area.

[0131] Based on the above process, it is possible to determine whether each pixel in the current retinal image belongs to the effective edema region, and thus obtain the effective edema region in the current retinal image.

[0132] Step 207: For each edema region, determine the area of ​​the edema region based on the pixel spacing information of the current retinal image and the number of pixels within the edema region.

[0133] Step 208: Based on the area of ​​each edema region in the current retinal image and the image spacing information of the current retinal image, obtain the volume of each edema region in the current retinal image.

[0134] Steps 207 and 208 are similar to steps 104 and 105 in the first embodiment, and will not be described again here. The main difference is that steps 207 and 208 are aimed at the effective edema area determined in step 206.

[0135] Subsequently, the reference retinal images obtained from the four preprocessing methods can be saved, and the identified edema areas, sampling areas, etc. can be marked; similarly, the current retinal image marked with the final edema areas can be saved.

[0136] In this embodiment, the current retinal image is subjected to various preprocessing methods. Different preprocessing methods can highlight different image features. Then, the edema region in the current retinal image is adjusted by combining the edema region in the reference retinal image obtained after preprocessing the current retinal image, so as to obtain the final effective edema region, achieve better noise reduction effect, and further improve the accuracy of edema region feature extraction, that is, further improve the segmentation accuracy of edema region in retinal image.

[0137] The third embodiment of this application relates to a method for obtaining edema features in retinal images. Compared with the first embodiment, this embodiment adds denoising processing based on RPE layer recognition.

[0138] Figure 12 shows a flowchart of the method for obtaining edema features in retinal images according to this embodiment.

[0139] Step 301: Acquire the current retinal image to be identified from multiple retinal images. This is largely the same as step 101 in the first embodiment, and will not be described again here.

[0140] Step 302: Determine the edema brightness threshold of the current retinal image based on the brightness values ​​of pixels within the vitreous region of the current retinal image. This is largely the same as step 102 in the first embodiment and will not be described again here.

[0141] Step 303: Identify the RPE layer of the current retinal image based on the brightness values ​​of each pixel in the current retinal image.

[0142] Specifically, the effective area of ​​the current retinal image is first obtained, as can be found in the relevant content of the first embodiment. Then, the target pixel with the highest brightness is searched in columns within the effective area of ​​the current retinal image. For the target pixel in the current column within the effective area of ​​the current retinal image, if the difference in the vertical axis position between the target pixel in the current column and the target pixel in the previous column adjacent to the current column is greater than the first difference threshold, the vertical axis position of the target pixel in the previous column adjacent to the current column is determined as the vertical axis position of the target pixel in the current column.

[0143] In one example, starting from any middle column of the effective area of ​​the current retinal image, the brightest target pixel in each column is searched to the left and right. For example, from multiple columns of the effective area of ​​the current retinal image, the middle column is selected (if the total number of columns is even, either of the two middle columns is selected). Starting from the middle column, the identification proceeds to the right column by column. For the current column, the brightest target pixel in the current column is found. Then, the Y-axis position of the target pixel is compared with the Y-axis position of the target pixel in the previous column. If the difference between the two Y-axis positions is less than a first difference threshold, the target pixel in the current column is determined to belong to the RPE layer. If the difference between the two Y-axis positions is greater than the first difference threshold (e.g., 20 pixels), it indicates that there may be other bright positions in the image causing interference. Therefore, the target pixel in the current column is determined not to belong to the RPE layer. The Y-axis position of the target pixel in the previous column is used as the Y-axis position of the target pixel in the current column, while the X-axis position of the target pixel in the current column remains unchanged. Then, starting from the middle column of the effective area of ​​the current retinal image and proceeding column by column to the left, the above process is repeated to determine the target pixels belonging to the RPE layer in each column to the left of the middle column. Starting pixel identification from the middle column of the effective area of ​​the current retinal image can avoid the impact of image edge blurring on the accuracy of RPE layer identification.

[0144] From the above, the target pixels belonging to the RPE layer in each column of the effective area of ​​the current retinal image can be determined. It should be noted that during the recognition process, if there are multiple columns (the specific number can be determined based on the total number of columns in the effective area, such as one-quarter of the total number of columns) in the effective area of ​​the current retinal image where the difference between the vertical axis position of the target pixel and the vertical axis position of the target pixel in the previous column is greater than the first difference threshold, it is necessary to consider whether there is a problem with the vertical axis position of the target pixel in the column that was first identified; or whether there is too much bright interference in the effective area of ​​the current retinal image.

[0145] Then, the RPE layer of the current retinal image is obtained by combining the target pixels in each column within the effective area of ​​the current retinal image. Subsequently, the positions of all target pixels in the RPE layer are recorded to form the corresponding coordinate array.

[0146] In addition, if the current retinal image has multiple associated designated retinal images, the position of the RPE layer of the current retinal image can be corrected based on the position of the RPE layer of these multiple designated retinal images. The specific process is as follows: Multiple designated retinal images associated with the current retinal image refer to multiple OCT images (OCT images in the same group) acquired consecutively during one OCT image acquisition for the same patient. Other OCT images in these multiple OCT images besides the current retinal image can be used as designated retinal images; the positions of the RPE layers in these multiple OCT images should be similar.

[0147] For these multiple specified retinal images, based on a similar process described above, the position of the RPE layer in the effective region of each specified retinal image is determined; then, based on the average height of the pixels contained in the RPE layers of the multiple specified retinal images, the pixel reference height of the RPE layer is determined; that is, for the RPE layer of each specified retinal image, the mean of the vertical axis position coordinates of all pixels in the RPE layer of that specified retinal image is calculated as the average pixel height of the RPE layer of that specified retinal image; thus, the average pixel height of the RPE layers of all specified retinal images can be calculated. The mean of the average pixel heights of the RPE layers of all specified retinal images is then calculated as the pixel reference height of the RPE layer; however, this is not limited to this, the median of the average pixel heights of the RPE layers of all specified retinal images can also be selected as the pixel reference height of the RPE layer.

[0148] After determining the pixel reference height of the RPE layer, the average pixel height of the pixels contained in the RPE layer of the current retinal image is compared with this pixel reference height. If the difference between the average pixel height of the pixels contained in the RPE layer of the current retinal image and the pixel reference height is greater than a second difference threshold, it indicates that the determined position of the RPE layer in the effective region of the current retinal image is inaccurate. The RPE layer of the current retinal image is then determined based on the vertical axis position of the pixels contained in the RPE layer of a specified retinal image adjacent to the current retinal image. In other words, the vertical axis position of the pixels contained in the RPE layer of a specified retinal image adjacent to the current retinal image is used as the vertical axis position of the pixels in the RPE layer of the current retinal image, thereby allowing the RPE layer to be re-determined in the current retinal image. The specified retinal image adjacent to the current retinal image can be the specified retinal image whose acquisition time is closest to that of the current retinal image.

[0149] After the above process is completed and the RPE layer of the current retinal image is determined, the RPE layer can be marked on the current retinal image. There cannot be edema areas in the area below the RPE layer. Therefore, based on the location of the RPE layer in the retinal image, the area below the RPE layer can be avoided from being misidentified as an edema area.

[0150] Furthermore, step 303 also includes:

[0151] Step 304: Obtain multiple specified retinal images associated with the current retinal image, and determine the pixel reference height of the RPE layer based on the average height of the pixels contained in the RPE layer of the multiple specified retinal images; if the difference between the average pixel height of the pixels contained in the RPE layer of the current retinal image and the pixel reference height is greater than the second difference threshold, then determine the RPE layer of the current retinal image based on the vertical axis position of the pixels contained in the RPE layer of a specified retinal image adjacent to the current retinal image.

[0152] In other words, by slicing a user's 3D image into multiple retinal images, and then identifying the RPE layers of several designated retinal images adjacent to the current retinal image (e.g., five images to the left and five to the right of the current retinal image), the RPE layers of these designated retinal images are obtained. Then, the average pixel height of the pixels contained in the RPE layers of each designated retinal image is calculated, resulting in the average pixel height of the RPE layers across the multiple designated retinal images. Based on these average pixel heights, a pixel reference height is determined. For example, the median of the average pixel heights of the RPE layers across the multiple designated retinal images can be selected as the pixel reference height, or the mean can be calculated as the pixel reference height. The reference height is then used; the average pixel height of the pixels in the RPE layer of the current retinal image is compared with the reference height. If the difference between the two is greater than a second difference threshold (e.g., 20 pixels), it is determined that the RPE layer of the current retinal image has been identified incorrectly. If the RPE layer of the current retinal image has been identified incorrectly, the coordinates of each pixel in the RPE layer of the previous or next specified retinal image adjacent to the current retinal image are used as the coordinates of the RPE layer in the current retinal image, thereby re-determining an RPE layer in the current retinal image. If the difference between the two is less than the second difference threshold, it is determined that the RPE layer of the current retinal image has not been identified incorrectly.

[0153] The height of a pixel is its Y-axis coordinate. The coordinate system has the pixel at the bottom left corner of the retina image as its origin, with the Y-axis pointing upwards from the origin and the X-axis pointing to the right of the origin.

[0154] Step 305: Based on the RPE layer of the current retinal image, the brightness value of each pixel in the current retinal image, and the edema brightness threshold of the current retinal image, determine the edema region in the current retinal image. Pixels in the edema region are all located above the RPE layer.

[0155] Step 103 in the first embodiment is largely the same and will not be repeated here. The main difference is that the edema region features in the current retinal image can be extracted and denoised based on the position of the RPE layer in the current retinal image. The specific method is as follows:

[0156] Method 1: After determining the location of the RPE layer in the current retinal image, the portion of the current retinal image below the RPE layer is adjusted to black. Since the area below the RPE layer is similar to the edema area in brightness and shape, blacking it out prevents the area below the RPE layer from being misidentified as an edema area. Refer to Figure 13, where Figure 12a is the original forward retinal image; Figure 12b is the current retinal image with the portion below the RPE layer adjusted to black. Here, the edema area in the current retinal image is determined based on the brightness values ​​of each pixel in the current retinal image with the portion below the RPE layer adjusted to black and the edema brightness threshold of the current retinal image. That is, when determining the edema area in the current retinal image, the focus is on the current retinal image below the RPE layer that has been adjusted to black. Refer to Figure 14, which shows the current retinal image with the portion below the RPE layer adjusted to black. The pure black pixels within the marked circular and elliptical dashed boxes form the identified edema area, demonstrating the elimination of noise below the RPE layer.

[0157] Method 2 involves first extracting features of the edema region from the current retinal image to identify the edema region. Then, based on the position of the RPE layer in the current retinal image and the position of each pixel in the identified edema region, the edema region below the RPE layer is removed. This also avoids the region below the RPE layer being misidentified as an edema region.

[0158] For example, please refer to Figures 15 and 16. Based on the edema feature acquisition method in retinal images of this embodiment, after identifying edema regions in multiple retinal example images, professional doctor 1 and doctor 2 manually review the edema regions in all these retinal example images, resulting in the schematic diagram of the missed labeling rate of edema regions shown in Figure 15 and the schematic diagram of the multi-labeling rate of edema regions shown in Figure 16.

[0159] In the schematic diagram of the missed labeling rate in the edema area in Figure 15, the missed labeling rates marked by Doctor 1 are: median 0.00%, 25th percentile 0.00%, and 75th percentile 2.00%, with more than 90% of the retinal example images having a missed labeling rate of less than 10%; the missed labeling rates marked by Doctor 2 are: median 0.00%, 25th percentile 0.00%, and 75th percentile 3.00%, with more than 90% of the retinal example images having a missed labeling rate of less than 8%.

[0160] In Figure 16, the multi-scale rate of edema areas is shown in the diagram. Among the multi-scale rates marked by Doctor 1, the median is 0.00%, the 25th percentile is 0.00%, and the 75th percentile is 2.00%. More than 90% of the retinal example images have a multi-scale rate of less than 13%. Among the multi-scale rates marked by Doctor 2, the median is 0.00%, the 25th percentile is 0.00%, and the 75th percentile is 3.00%. More than 90% of the retinal example images have a multi-scale rate of less than 9%.

[0161] Step 306: For each edema region, determine the area of ​​the edema region based on the pixel spacing information of the current retinal image and the number of pixels within the edema region. This is largely the same as step 104 in the first embodiment and will not be described again here.

[0162] Step 307: Based on the area of ​​each edema region in the current retinal image and the image spacing information between the current retinal image and adjacent retinal images, the volume of each edema region in the current retinal image is obtained. This is largely the same as step 105 in the first embodiment and will not be described again here.

[0163] It should be noted that this embodiment can also be used as an improvement on the second embodiment. Specifically, after obtaining multiple reference retinal images after various preprocessing steps, the RPE layer of each reference retinal image can be determined based on the method of this embodiment. Based on the position of the RPE layer of each reference retinal image, the edema region in each determined reference retinal image is denoised to avoid the region below the RPE layer being misidentified as an edema region.

[0164] The fourth embodiment of this application relates to an electronic device, which can be a common computer device, such as a desktop computer, a laptop computer, etc., or a mobile phone.

[0165] The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the edema feature acquisition method in a retinal image according to any one of the first to third embodiments.

[0166] Since the first embodiment corresponds to this embodiment, this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment remain valid in this embodiment, and the technical effects achievable in the first embodiment can also be achieved in this embodiment. To reduce repetition, they will not be repeated here. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0167] It should be noted that the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Similarly, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0168] The fifth embodiment of this application relates to a computer-readable storage medium, which is a non-volatile or non-transient storage medium, and stores a computer program thereon. The computer program is executed by a processor to perform the edema feature acquisition method in any one of the first to third embodiments.

[0169] Since the first embodiment corresponds to this embodiment, this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment remain valid in this embodiment, and the technical effects achievable in the first embodiment can also be achieved in this embodiment. To reduce repetition, they will not be repeated here. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0170] The preferred embodiments of this application have been described in detail above, but it should be understood that, if necessary, aspects of the embodiments can be modified to utilize aspects, features, and concepts from various patents, applications, and publications to provide other embodiments.

[0171] In light of the detailed description above, these and other changes can be made to the embodiments. Generally, the terminology used in the claims should not be considered limited to the specific embodiments disclosed in the specification and claims, but should be understood to include all possible embodiments together with the full scope of equivalents enjoyed by these claims.

Claims

1. A method for obtaining edema features in retinal images, characterized in that, include: Acquire the current retinal image to be identified from multiple retinal images, wherein the retinal image is an OCT image of the fundus; Based on the brightness values ​​of pixels in the vitreous region of the current retinal image, determine the edema brightness threshold of the current retinal image; Based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, an edema region in the current retinal image is determined, wherein the brightness value of the pixels in the edema region is less than the edema brightness threshold.

2. The method for obtaining edema features in retinal images according to claim 1, characterized in that, The method further includes: The current retinal image is subjected to various preprocessing steps to obtain multiple reference retinal images after various preprocessing steps; Based on the brightness value of each pixel in each of the reference retinal images and the edema brightness threshold of each of the reference retinal images, the edema region in each of the reference retinal images is determined; Based on the edema regions of each of the reference retinal images and the edema region of the current retinal image, as well as preset weight parameters, the effective edema region in the current retinal image is determined. The preset weight parameters include the weights of each of the reference retinal images and the current retinal image.

3. The method for obtaining edema features in retinal images according to claim 1, characterized in that, Before determining the edema region in the current retinal image based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, the method further includes: The RPE layer of the current retinal image is identified based on the brightness values ​​of each pixel within the current retinal image; Based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, the edema region in the current retinal image is determined, including: Based on the RPE layer of the current retinal image, the brightness value of each pixel in the current retinal image, and the edema brightness threshold of the current retinal image, the edema region in the current retinal image is determined, and the pixels in the edema region are all located above the RPE layer.

4. The method for obtaining edema features in retinal images according to claim 1, characterized in that, Determining the edema brightness threshold of the current retinal image based on the brightness values ​​of pixels within the vitreous region of the current retinal image includes: An initial brightness threshold is obtained for the brightness values ​​of pixels that are greater than a preset percentage of all pixels in the glass region, wherein the preset percentage is greater than 50%. Within the vitreous region, target pixels with brightness values ​​within a set brightness value range are located, and based on the target brightness value of each target pixel, the number of target pixels with each target brightness value is counted; wherein the set brightness value range is the brightness value range with the initial brightness threshold as the midpoint. Among all the target brightness values, find the target brightness value with the smallest difference in the number of target pixels between it and the initial brightness threshold, and use it as the minimum change brightness threshold; The edema brightness threshold is determined based on the initial brightness threshold and the minimum change brightness threshold.

5. The method for obtaining edema features in retinal images according to claim 4, characterized in that, Determining the edema brightness threshold based on the initial brightness threshold and the minimum change brightness threshold includes: The product of the average of the initial brightness threshold and the minimum change brightness threshold and a preset correction coefficient is obtained as the edema brightness threshold, wherein the preset correction coefficient is greater than 0.

5.

6. The method for obtaining edema features in retinal images according to claim 1, characterized in that, Based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, the edema region in the current retinal image is determined, including: In the current retinal image, among the closed regions composed of pixels whose brightness values ​​are less than the edema brightness threshold, a closed region whose area meets the preset conditions is selected as the edema region. The preset condition is that the area of ​​the closed region is less than the product of the effective area in the current retinal image and the set threshold, and is greater than the area of ​​the preset number of pixels. The set threshold is greater than 0 and less than 0.

5.

7. The method for obtaining edema features in retinal images according to claim 1, characterized in that, After determining the edema region in the current retinal image based on the brightness values ​​of each pixel in the current retinal image and the edema brightness threshold of the current retinal image, the method further includes: For each edema region, the area of ​​the edema region is determined based on the pixel spacing information of the current retinal image and the number of pixels within the edema region.

8. The method for obtaining edema features in retinal images according to claim 7, characterized in that, After determining the area of ​​each edema region based on the pixel spacing information of the current retinal image and the number of pixels within the edema region, the method further includes: The volume of each edema region in the current retinal image is obtained based on the area of ​​each edema region in the current retinal image and the image spacing information between the current retinal image and adjacent retinal images.

9. The method for obtaining edema features in retinal images according to claim 7 or 8, characterized in that, Before determining the area of ​​each edema region based on the pixel spacing information of the current retinal image and the number of pixels within the edema region, the method further includes: Perform ETDRS partitioning on the current retinal image to obtain the edema region located within the ETDRS partition; For each edema region, the area of ​​the edema region is determined based on the pixel spacing information of the current retinal image and the number of pixels within the edema region, including: For each edema region located within an ETDRS partition, the area of ​​the edema region is determined based on the pixel spacing information of the current retinal image and the number of pixels within the edema region.

10. The method for obtaining edema features in retinal images according to claim 3, characterized in that, Identifying the RPE layer of the current retinal image based on the brightness values ​​of each pixel within the current retinal image includes: Within the effective area of ​​the current retinal image, the specified pixel with the highest brightness is searched column by column. For the specified pixel in the current column within the effective area of ​​the current retinal image, if the difference in the vertical axis position between the specified pixel in the current column and the specified pixel in the adjacent previous column is greater than a first difference threshold, the vertical axis position of the specified pixel in the current column is determined based on the vertical axis position of the specified pixel in the adjacent previous column. The RPE layer of the current retinal image is obtained by using the specified pixel combinations in each column within the effective area of ​​the current retinal image.

11. The method for obtaining edema features in retinal images according to claim 10, characterized in that, Search for the brightest target pixel column-wise within the effective area of ​​the current retinal image, including: Starting from any middle column of the effective area of ​​the current retinal image, search to the left and right respectively for the target pixel with the highest brightness in each column.

12. The method for obtaining edema features in retinal images according to claim 10, characterized in that, After obtaining the RPE layer of the current retinal image using the specified pixel combinations in each column within the effective area of ​​the current retinal image, the method further includes: Acquire multiple specified retinal images associated with the current retinal image, and determine the pixel reference height of the RPE layer based on the average height of the pixels contained in the RPE layer of the multiple specified retinal images; If the difference between the average pixel height of the pixels contained in the RPE layer of the current retinal image and the pixel reference height is greater than a second difference threshold, then the RPE layer of the current retinal image is determined based on the vertical axis position of the pixels contained in the RPE layer of a specified retinal image adjacent to the current retinal image.

13. The method for obtaining edema features in retinal images according to claim 1 or 4, characterized in that, The method for determining the vitreous region in the current retinal image is as follows: The effective area of ​​the current retinal image is sampled sequentially using a sampling frame of a set size. If the average brightness value of the pixels in the sampling area currently selected by the sampling frame is within a preset brightness range, then the sampling area currently selected by the sampling frame is determined to be the vitreous region.

14. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method for acquiring edema features in a retinal image as described in any one of claims 1 to 13.

15. A computer-readable storage medium, said computer-readable storage medium being a non-volatile storage medium or a non-transient storage medium, having stored thereon a computer program, characterized in that, The computer program is executed by the processor to perform the method for obtaining edema features in retinal images as described in any one of claims 1 to 13.

Images