Retinal vessel segmentation system based on pixel-level screening strategy

Abstract

The invention discloses a retinal vessel segmentation system based on a pixel-level screening strategy, and the system comprises a fundus image obtaining unit which is used for obtaining a fundus image to be segmented; the image preprocessing unit is used for performing standardized preprocessing on the eye fundus image to be segmented to obtain a standardized image sample; the dichotomy label output unit is used for inputting the standardized image samples into an advanced segmentation model subjected to secondary supervision training, outputting the blood vessel category probability of each pixel unit, and converting a blood vessel category probability threshold into dichotomy labels of blood vessels or backgrounds; wherein the training of the advanced segmentation model comprises the following steps: executing a loss shielding operation on a low-credibility pixel unit in a second pseudo label sample; according to the method, the dependence on a large amount of labeled data is reduced, the negative influence of error propagation is avoided, and finally, the generated advanced segmentation model shows higher accuracy in a blood vessel segmentation task.

Description

Technical Field

[0001] This invention relates to the field of medical image sample segmentation, specifically to a retinal vessel segmentation system and method based on a pixel-level screening strategy. Background Technology

[0002] With the deep integration of artificial intelligence and medical imaging technology, retinal vessel segmentation is playing an increasingly important role in the early diagnosis of diseases such as diabetic retinopathy and hypertension. Traditional fully supervised segmentation methods rely on large amounts of high-quality labeled data, but these methods have significant limitations due to the high cost, time-consuming nature, and difficulty in ensuring label consistency of manual annotation. To address this issue, semi-supervised learning combines a small amount of labeled data with a large amount of unlabeled data, reducing reliance on manual annotation and effectively lowering annotation costs.

[0003] Patent document CN115115659B discloses an automatic retinal vessel segmentation method based on low-cost noisy data, which trains a retinal vessel segmentation system using supervised and semi-supervised methods. However, existing semi-supervised methods still face the following challenges:

[0004] False label error propagation: The false labels generated from unlabeled data may contain a large number of pixels with incorrect predictions. These errors will gradually accumulate during iterative training, eventually leading to a decline in model performance.

[0005] Low-confidence pixel unit interference: Even if high-quality image samples are retained in the pseudo-labels, local errors can still affect training results. The discarded pseudo-label samples may also contain some correctly predicted pixels, limiting the effectiveness of the training data and leading to interference from low-confidence pixel units during training.

[0006] Therefore, solving the problems of false label error propagation and interference from low-confidence pixels has become a research hotspot in the field of retinal vessel segmentation. To address this, this invention provides a retinal vessel segmentation system based on a pixel-level selection strategy. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a retinal vessel segmentation system based on a pixel-level screening strategy. By introducing high- and low-confidence pixel units from a second pseudo-label and training an advanced segmentation model based on the high-confidence pixel units, the technical problems mentioned in the background art can be solved.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A retinal vessel segmentation system based on a pixel-level screening strategy includes:

[0010] The fundus image acquisition unit is used to acquire fundus images to be segmented;

[0011] The image preprocessing unit is used to perform standardized preprocessing on the fundus image to be segmented to obtain standardized image samples;

[0012] The binary classification label output unit is used to input standardized image samples into the advanced segmentation model trained under secondary supervision, output the blood vessel category probability of each pixel unit, and convert the blood vessel category probability threshold into a binary classification label of blood vessel or background.

[0013] The training of the advanced segmentation model includes: performing binary classification-based supervised training on pixel units of labeled image samples or high-confidence pixel units of the second pseudo-label; and also includes: performing a loss masking operation on low-confidence pixel units in the second pseudo-label samples.

[0014] In some specific embodiments, the secondary supervised training step of the advanced segmentation model includes:

[0015] S1. Load the N fundus image samples to be processed;

[0016] S2. Label K labeled image samples from N fundus image samples, and predict and generate M first pseudo-label samples;

[0017] In the first pseudo-label sample, each pixel unit is labeled with the probability of blood vessel category;

[0018] S3. Filter and label the pixel units of the first pseudo-label sample until M second pseudo-label samples are obtained; wherein, the pixel units of the second pseudo-label samples are labeled as high-confidence pixel units or low-confidence pixel units.

[0019] S4. Merge the M second pseudo-labeled samples with the K labeled image samples to obtain J high-quality image samples; where J = M + K;

[0020] S5. Input J high-quality image samples into the basic segmentation model for secondary supervised training, and generate an advanced segmentation model through iteration.

[0021] In some specific embodiments, K labeled image samples are annotated from N fundus image samples, and M first pseudo-label samples are predicted and generated, including:

[0022] S2-1. Select K fundus image samples from N fundus image samples;

[0023] S2-2. Perform binary classification and labeling on the K selected fundus image samples, defining them as K labeled image samples; and define the fundus image samples that are not selected for labeling as unlabeled image samples;

[0024] Among them, the pixel units of the labeled image samples are all labeled with binary true labels. The binary true labels indicate that the category of the pixel unit in the fundus image sample is actually labeled as retinal vessels or retinal background.

[0025] S2-3. Input K labeled image samples into the U-Net network for initial supervised training, and generate a basic segmentation model through iteration; wherein, the output target of the basic segmentation model is defined as the blood vessel category probability of pixel unit;

[0026] S2-4. Input the unlabeled image samples into the basic segmentation model for forward propagation inference to predict and generate M first pseudo-label samples.

[0027] In some specific embodiments, K selected fundus image samples are binary-classified and labeled, defined as K labeled image samples; and unlabeled fundus image samples are defined as unlabeled image samples, including:

[0028] S2-2-1. Perform standardized preprocessing on N fundus image samples to obtain N standardized image samples;

[0029] S2-2-2. Perform data augmentation on N standardized image samples to obtain M unlabeled image samples; M > N;

[0030] S2-2-3. Select K unlabeled image samples from M unlabeled image samples;

[0031] S2-2-4. Perform binary classification labeling on the pixel units of the K unlabeled image samples to obtain K labeled image samples.

[0032] In some specific embodiments, the pixel units of the first pseudo-label sample are filtered and marked until M second pseudo-label samples are obtained, including:

[0033] S3-1. Perform median filtering on the first pseudo-label sample to generate a noise-free image sample;

[0034] S3-2. Perform edge detection on noise-free image samples to label pixel units as contour pixel units or non-contour pixel units.

[0035] S3-3. Perform dynamic threshold determination on non-contour pixel units of noise-free image samples to mark non-contour pixel units as high-confidence pixel units or low-confidence pixel units.

[0036] S3-4. Traverse all pixel units of the M first pseudo-label samples, and repeatedly perform median filtering, edge detection and dynamic threshold determination until the M first pseudo-label samples are selected and marked as M second pseudo-label samples.

[0037] In some specific embodiments, median filtering is performed on the first pseudo-label sample to generate noise-free image samples, including:

[0038] S3-1-1, Anchor the pixel unit to be labeled in the first pseudo-label sample;

[0039] S3-1-2. Construct a square neighborhood window with the pixel unit to be labeled as the center of the window;

[0040] The neighborhood window covers several adjacent pixel units in the first pseudo-label sample;

[0041] S3-1-3. Obtain the pixel values ​​of several adjacent pixel units within the neighborhood window, and construct a pixel value sequence based on the sorted pixel values;

[0042] S3-1-4. Select the median value of the pixels in the pixel value sequence and replace the pixel value of the pixel unit to be marked with the median value;

[0043] S3-1-5. Traverse all pixel units of the first pseudo-label sample until the pixel value of all pixel units is replaced by the median value of the pixels in its neighborhood window to generate the noise-free image sample.

[0044] In some specific embodiments, edge detection is performed on noise-free image samples to label pixel units as contour pixel units or non-contour pixel units, including:

[0045] S3-2-1. Obtain the gray value of each pixel unit in a noise-free image sample;

[0046] S3-2-2, In a noise-free image sample, mark adjacent pixel units and calculate the gray-level difference between adjacent pixel units;

[0047] S3-2-3. If the grayscale difference is not 0, then the adjacent pixel units are determined to be contour pixel units; otherwise, they are determined to be non-contour pixel units.

[0048] S3-2-4. Traverse all pixel units of the noise-free image sample and mark the pixel units as contour pixel units or non-contour pixel units.

[0049] In some specific embodiments, dynamic thresholding is performed on non-contour pixel units of noise-free image samples to mark them as high-confidence or low-confidence pixel units, including:

[0050] S3-3-1. Select a non-contour pixel unit in a noise-free image sample and mark it as the current pixel unit;

[0051] S3-3-2, Obtain the current pixel unit to the current iteration number;

[0052] S3-3-3. Calculate the dynamic probability threshold of the current pixel unit based on the current iteration number;

[0053] S3-3-4. Obtain the probability of blood vessel category for the current pixel unit;

[0054] S3-3-5. If the probability of the blood vessel category is greater than the dynamic probability threshold, then the current pixel unit is marked as a high-confidence pixel unit; otherwise, it is marked as a low-confidence pixel unit.

[0055] In some specific embodiments, J high-quality image samples are input into the basic segmentation model for secondary supervised training, and an advanced segmentation model is generated iteratively, including:

[0056] S5-1, Mark the current iteration pixel unit of the high-quality image sample;

[0057] S5-2. Based on the unit category inherited by the current iteration pixel unit, perform binary classification supervised training or loss masking operation;

[0058] S5-3. If the unit category is a pixel unit of a labeled image sample or a highly reliable pixel unit of the second pseudo-label, then perform binary classification-based supervised training on the pixel unit of the labeled image sample or the highly reliable pixel unit of the second pseudo-label.

[0059] S5-4. If the sample category is a low-confidence pixel unit of the second pseudo-label, then perform a loss masking operation on the low-confidence pixel units in the second pseudo-label sample.

[0060] In some specific embodiments, a masking operation for loss calculation is performed on low-confidence pixel units in the second pseudo-label sample, including:

[0061] S5-4-1, Anchoring low-confidence pixel units;

[0062] S5-4-2, Forward propagation of low-confidence pixel units to predict the probability of blood vessel category that generates low-confidence pixel units;

[0063] S5-4-3. Skip the cross-entropy loss calculation for low-confidence pixel units and directly proceed to the secondary supervised training of the next iteration pixel units.

[0064] This invention provides a retinal vessel segmentation system based on a pixel-level screening strategy, which has the following advantages:

[0065] This invention effectively expands the scale of training samples by generating second pseudo-label samples. Even with few samples, it improves the model's training performance on limited labeled fundus images by generating high-quality pseudo-labels from unlabeled images, reducing reliance on large amounts of labeled data and still improving the accuracy of retinal vessel segmentation even with scarce data. Furthermore, when dealing with low-confidence pixel units in the second pseudo-label samples, a loss masking operation is used to skip the loss calculation of these low-quality pixel units, thereby avoiding interference from low-confidence data in the training process. This allows the training process to focus more on optimizing high-quality, high-confidence pixel units, avoiding the negative impact of error propagation. Ultimately, the generated advanced segmentation model exhibits higher accuracy in vessel segmentation tasks. Attached Figure Description

[0066] Figure 1 This is a structural block diagram of a retinal vessel segmentation system based on a pixel-level screening strategy according to the present invention.

[0067] Figure 2 This is a schematic diagram of a retinal vessel segmentation process based on a pixel-level screening strategy according to the present invention.

[0068] Figure 3 This is a schematic diagram illustrating the training of the advanced segmentation model described in this invention;

[0069] Figure 4 This is a schematic diagram of the labeling process for the second pseudo-label sample of the present invention;

[0070] Figure 5 This is a schematic diagram of the masking operation for calculating the loss of low-confidence pixel units as described in this invention;

[0071] Figure 6 This is a comparative schematic diagram of the fundus image described in one embodiment of the present invention; Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Please see Figures 1 to 6 This invention provides a retinal vessel segmentation system based on a pixel-level screening strategy, comprising:

[0074] The fundus image acquisition unit is used to acquire fundus images to be segmented;

[0075] The image preprocessing unit is used to perform standardized preprocessing on the fundus image to be segmented to obtain standardized image samples;

[0076] The binary classification label output unit is used to input standardized image samples into the advanced segmentation model trained under secondary supervision, output the blood vessel category probability of each pixel unit, and convert the blood vessel category probability threshold into a binary classification label of blood vessel or background.

[0077] The training of the advanced segmentation model includes: performing binary classification-based supervised training on pixel units of labeled image samples or high-confidence pixel units of the second pseudo-label; and also includes: performing a loss masking operation on low-confidence pixel units in the second pseudo-label samples.

[0078] In this embodiment, the secondary supervised training step of the advanced segmentation model includes:

[0079] S1. Load the N fundus image samples to be processed;

[0080] S2. Label K labeled image samples from N fundus image samples, and predict and generate M first pseudo-label samples;

[0081] In the first pseudo-label sample, each pixel unit is labeled with the probability of blood vessel category;

[0082] S3. Filter and label the pixel units of the first pseudo-label sample until M second pseudo-label samples are obtained; wherein, the pixel units of the second pseudo-label samples are labeled as high-confidence pixel units or low-confidence pixel units.

[0083] S4. Merge the M second pseudo-labeled samples with the K labeled image samples to obtain J high-quality image samples; where J = M + K;

[0084] S5. Input J high-quality image samples into the basic segmentation model for secondary supervised training, and generate an advanced segmentation model through iteration.

[0085] In this embodiment, secondary supervised training optimizes segmentation performance when processing high-confidence pixel units in labeled samples and second pseudo-labeled samples. During training, the model focuses on learning accurately labeled pixel units, improving the accuracy of blood vessel segmentation. For low-confidence pixel units in the second pseudo-labeled samples, a loss masking operation is employed. In this operation, although low-confidence pixel units generate prediction results, their cross-entropy loss calculation is skipped, avoiding the negative impact of low-quality pixel units on model training. This effectively prevents low-confidence data from interfering with the training process, ensuring accurate and stable training. The resulting advanced segmentation model can more accurately segment retinal vessels in fundus images, improving the image segmentation level of retinal vessels.

[0086] For example, step S2 specifically includes:

[0087] S2-1. Select K fundus image samples from N fundus image samples;

[0088] S2-2. Perform binary classification and labeling on the K selected fundus image samples, defining them as K labeled image samples; and define the fundus image samples that are not selected for labeling as unlabeled image samples;

[0089] Among them, the pixel units of the labeled image samples are all labeled with binary true labels. The binary true labels indicate that the category of the pixel unit in the fundus image sample is actually labeled as retinal vessels or retinal background.

[0090] S2-3. Input K labeled image samples into the U-Net network for initial supervised training, and generate a basic segmentation model through iteration; wherein, the output target of the basic segmentation model is defined as the blood vessel category probability of pixel unit;

[0091] Specifically, the U-Net network is a classic convolutional neural network architecture widely used in medical image segmentation tasks. U-Net can effectively handle image segmentation tasks, especially when processing medical images (such as fundus images), and can accurately segment vascular regions.

[0092] S2-4. Input the unlabeled image samples into the basic segmentation model for forward propagation inference to predict and generate M first pseudo-label samples. Specifically, each unlabeled image sample is assigned a sample index, and the first pseudo-label samples share the same sample index with the unlabeled image samples to form a mapping between the unlabeled image samples and the first pseudo-label samples.

[0093] In this embodiment, firstly, the pixel units of K fundus image samples are explicitly labeled in binary classification, and then a basic segmentation model is generated through initial supervised training using a U-Net network. The difference is that this basic segmentation model outputs the blood vessel category probability for each pixel unit. Then, unlabeled image samples are used as input to the basic segmentation model, ultimately outputting a first pseudo-label sample with the blood vessel category probability. Overall, by using a small number of labeled image samples, which cover a large number of pixel units, the basic segmentation model can be generated through supervised training, thereby improving the quality of the first pseudo-label in the semi-supervised process and ultimately increasing the accuracy of the blood vessel segmentation model.

[0094] Specifically, step S2-2 further includes:

[0095] S2-2-1. Perform standardized preprocessing on N fundus image samples to obtain N standardized image samples;

[0096] Specifically, standardized preprocessing includes:

[0097] The pixel grayscale values ​​of each image are uniformly normalized to adjust them to the range of [0,1] to eliminate brightness differences between images. Histogram equalization or adaptive contrast enhancement methods are used to enhance the contrast of the images, making the contrast between blood vessels and the background more obvious and improving the segmentability of blood vessels in the image.

[0098] S2-2-2. Perform data augmentation on N standardized image samples to obtain M unlabeled image samples; M > N;

[0099] Specifically, data enhancement includes:

[0100] Rotation: The image is randomly rotated between -30° and 30° to enhance the model's adaptability to images at different angles.

[0101] Flip: Randomly flip the image horizontally and vertically to simulate the morphology of blood vessels under different shooting directions.

[0102] Scaling: Randomly scale the image within a certain range to help the model identify blood vessels at different scales.

[0103] Translation: Randomly translate the image to make the location of blood vessels in the image more diverse, thus improving the generalization ability of the model.

[0104] Brightness and contrast adjustment: Randomly adjusts the brightness and contrast of the image to simulate images under different lighting conditions.

[0105] S2-2-3. Select K unlabeled image samples from M unlabeled image samples;

[0106] Specifically, unlabeled image samples should be uniformly sized sub-blocks (e.g., 96×96 pixels). Such uniformly sized image sub-blocks help reduce the complexity of data processing and improve the stability of training.

[0107] S2-2-4. Perform binary classification labeling on the pixel units of the K unlabeled image samples to obtain K labeled image samples.

[0108] This embodiment enhances the diversity of training data through standardized preprocessing and data augmentation. Standardized preprocessing unifies the grayscale values ​​of image pixels, eliminating brightness differences, while contrast enhancement improves the separability of blood vessels from the background. Data augmentation, such as rotation, flipping, and scaling, generates more diverse unlabeled image samples, enhancing the model's adaptability to blood vessels under different conditions. Finally, K unlabeled samples are selected and processed for binary classification labeling to obtain labeled image samples for supervised training of the U-Net network and sample generation during the inference phase.

[0109] For example, step S3 specifically includes:

[0110] S3-1. Perform median filtering on the first pseudo-label sample to generate a noise-free image sample;

[0111] S3-2. Perform edge detection on noise-free image samples to label pixel units as contour pixel units or non-contour pixel units.

[0112] S3-3. Perform dynamic threshold determination on non-contour pixel units of noise-free image samples to mark non-contour pixel units as high-confidence pixel units or low-confidence pixel units.

[0113] S3-4. Traverse all pixel units of the M first pseudo-label samples, and repeatedly perform median filtering, edge detection and dynamic threshold determination until the M first pseudo-label samples are selected and marked as M second pseudo-label samples.

[0114] Specifically, step S3-1 further includes:

[0115] S3-1-1, Anchor the pixel unit to be labeled in the first pseudo-label sample;

[0116] S3-1-2. Construct a square neighborhood window with the pixel unit to be labeled as the center of the window;

[0117] The neighborhood window covers several adjacent pixel units in the first pseudo-label sample;

[0118] S3-1-3. Obtain the pixel values ​​of several adjacent pixel units within the neighborhood window, and construct a pixel value sequence based on the sorted pixel values;

[0119] S3-1-4. Select the median value of the pixels in the pixel value sequence and replace the pixel value of the pixel unit to be marked with the median value;

[0120] S3-1-5. Traverse all pixel units of the first pseudo-label sample until the pixel value of all pixel units is replaced by the median value of the pixels in its neighborhood window to generate the noise-free image sample.

[0121] This embodiment generates noise-free image samples by performing median filtering on the first pseudo-labeled samples. Specifically, firstly, an operation is performed on each pixel unit to be labeled, constructing a square neighborhood window centered on the pixel unit, covering multiple adjacent pixel units. Then, the pixel values ​​within the neighborhood window are acquired and sorted to generate a pixel value sequence. The median value in the sequence is selected and replaced with the value of the pixel unit to be labeled. By traversing the entire first pseudo-labeled sample, it is ensured that the value of each pixel unit is replaced by the median value in the neighborhood window, thereby eliminating noise and generating noise-free image samples. This embodiment effectively removes noise from the image, improves image quality, and thus enhances the accuracy of blood vessel segmentation model training.

[0122] Specifically, step S3-2 further includes:

[0123] S3-2-1. Obtain the gray value of each pixel unit in a noise-free image sample;

[0124] The grayscale value is 0 or 255, which respectively represent the background or the foreground (blood vessels).

[0125] S3-2-2, In a noise-free image sample, mark adjacent pixel units and calculate the gray-level difference between adjacent pixel units;

[0126] S3-2-3. If the grayscale difference is not 0, then the adjacent pixel units are determined to be contour pixel units; otherwise, they are determined to be non-contour pixel units.

[0127] S3-2-4. Traverse all pixel units of the noise-free image sample and mark the pixel units as contour pixel units or non-contour pixel units.

[0128] This embodiment effectively distinguishes between contour pixel units and non-contour pixel units in an image by performing edge detection on noise-free image samples. First, the grayscale value of each pixel unit is obtained, with 0 representing the background and 255 representing the blood vessel foreground. Boundaries are identified by the difference in grayscale values; if the grayscale difference between adjacent pixels is not zero, it is marked as a contour pixel unit; otherwise, it is marked as a non-contour pixel unit. This process ensures that the boundary between blood vessels and the background is effectively captured. By traversing all pixel units and determining whether they are contour pixel units based on grayscale differences, this embodiment can clearly identify boundary regions in the image, providing accurate edge information for blood vessel segmentation.

[0129] Specifically, step S3-3 further includes:

[0130] S3-3-1. Select a non-contour pixel unit in a noise-free image sample and mark it as the current pixel unit;

[0131] S3-3-2, Obtain the current pixel unit to the current iteration number;

[0132] S3-3-3. Calculate the dynamic probability threshold of the current pixel unit based on the current iteration number;

[0133] The formula for calculating the dynamic probability threshold is:

[0134]

[0135] Where T represents the dynamic probability threshold, I represents the number of iterations, and z represents the number of fundus image samples;

[0136] Specifically, when the iteration count I is 1 or 2, the dynamic probability threshold is set to a baseline value of 0.5. When the iteration count is greater than 2 and the number of fundus image samples is less than 20, the dynamic probability threshold of the current pixel unit is calculated according to the above formula. Therefore, in this embodiment, the number of fundus image samples is required to be kept below 20 during each training process. The dynamic probability threshold is introduced because during training, the proportion of background pixels is usually higher than that of blood vessel pixels. As the number of iterations increases, the model's response to the background gradually increases, thus incorrectly predicting pixels that should belong to blood vessels as background. As the iteration progresses, the prediction probability of some blood vessel pixels gradually decreases, even falling below the predetermined baseline threshold. To solve this problem, through experimental design, the above-mentioned dynamic probability threshold calculation method is proposed, which can effectively suppress the decrease in the prediction probability of blood vessel pixels caused by the increase of the number of iterations, thereby improving the accuracy of blood vessel segmentation.

[0137] S3-3-4. Obtain the probability of blood vessel category for the current pixel unit;

[0138] S3-3-5. If the probability of the blood vessel category is greater than the dynamic probability threshold, then the current pixel unit is marked as a high-confidence pixel unit; otherwise, it is marked as a low-confidence pixel unit.

[0139] This embodiment addresses the issue of increased background pixel proportion during training by employing a dynamic probability threshold, thus avoiding misclassification of blood vessel pixels and enhancing the model's accuracy in blood vessel segmentation at different iteration stages. This embodiment precisely distinguishes between high-confidence and low-confidence pixel units, improving the accuracy of blood vessel segmentation.

[0140] For example, step S5 specifically includes:

[0141] S5-1, Mark the current iteration pixel unit of the high-quality image sample;

[0142] S5-2. Based on the unit category inherited by the current iteration pixel unit, perform binary classification supervised training or loss masking operation;

[0143] Specifically, each pixel unit in the high-quality image samples has its unit category explicitly labeled before training begins. In this embodiment, the pixel units of the high-quality image samples inherit their unit categories before merging; that is, the category of each pixel (such as blood vessel or background) will remain unchanged after merging into the high-quality samples.

[0144] In other words, in high-quality image samples, the categories of pixel units have already been labeled through previous steps (such as probability-based filtering or pseudo-label generation). This category information will be directly inherited as attributes of pixel units without additional processing. Therefore, during training, further supervised training or filtering operations can be performed directly based on this labeled category information.

[0145] S5-3. If the unit category is a pixel unit of a labeled image sample or a highly reliable pixel unit of the second pseudo-label, then perform binary classification-based supervised training on the pixel unit of the labeled image sample or the highly reliable pixel unit of the second pseudo-label.

[0146] S5-4. If the sample category is a low-confidence pixel unit of the second pseudo-label, then perform a loss masking operation on the low-confidence pixel units in the second pseudo-label sample.

[0147] In this embodiment, J high-quality image samples are input into the base segmentation model for secondary supervised training, and an advanced segmentation model is generated iteratively. During training, each pixel unit in the high-quality image samples inherits its class information before merging, ensuring that no additional annotation is needed when merging high-quality samples; instead, the class information generated in the previous steps is used directly.

[0148] During training, standard binary classification supervised training is performed on pixel units of labeled image samples and high-confidence pixel units of the second pseudo-label. For low-confidence pixel units of the second pseudo-label, a loss masking operation is used to skip the loss calculation of these pixel units, thereby avoiding the impact of low-confidence pixel units on model training.

[0149] In this embodiment, step S5-3 further includes:

[0150] S5-3-1. Anchor the current pixel unit in the pixel unit of the currently identified labeled image sample or the high-confidence pixel unit of the second pseudo-label;

[0151] S5-3-2, Perform forward propagation on the current pixel unit to predict the probability of generating the blood vessel category for the current pixel unit;

[0152] S5-3-3, Calculate the cross-entropy loss between the blood vessel category probability and the true label of the current pixel unit;

[0153] S5-3-4. Calculate the gradient of the cross-entropy loss with respect to the model parameters;

[0154] S5-3-5. Update the model parameters based on the gradient to generate updated parameters;

[0155] S5-3-6. Anchor the next iteration pixel unit in the currently identified labeled image samples;

[0156] S5-3-7. Perform forward propagation on the next iteration pixel unit based on the updated parameters to predict the probability of generating the blood vessel category for the next iteration pixel unit.

[0157] S5-3-8. Repeat the calculation of cross-entropy loss and gradient in the next iteration until the updated parameters cause the basic segmentation model to converge to the advanced segmentation model.

[0158] This embodiment uses standard binary classification supervised training to continuously optimize and adjust the parameters of the basic segmentation model in order to improve the segmentation accuracy of blood vessel images.

[0159] In this embodiment, step S5-4 further includes:

[0160] S5-4-1, Anchoring low-confidence pixel units;

[0161] S5-4-2, Forward propagation of low-confidence pixel units to predict the probability of blood vessel category that generates low-confidence pixel units;

[0162] S5-4-3. Skip the cross-entropy loss calculation for low-confidence pixel units and directly proceed to the secondary supervised training of the next iteration pixel units.

[0163] In this embodiment, for low-confidence pixel units in the second pseudo-label sample, the impact of low-confidence pixel units on training is effectively avoided through a loss masking operation. Specifically, although these low-confidence pixel units generate vessel class probabilities through forward propagation, their cross-entropy loss calculation is skipped and they do not participate in gradient updates. This method avoids the negative impact of low-confidence pixel units on parameter updates, allowing for greater focus on training optimization of high-confidence pixel units. This embodiment ensures that optimization during training focuses on samples of high-quality pixel units, avoiding training bias caused by unreliable data, improving the accuracy of the vessel segmentation model. The resulting advanced segmentation model can more accurately segment retinal vessels in fundus images, improving the segmentation accuracy of retinal vessels in fundus images.

[0164] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means.

[0165] The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives (SSDs).

[0166] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Furthermore, the mutual couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0167] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A pixel-level screening strategy based retinal blood vessel segmentation system, characterized in that, The application relates to an eye fundus image segmentation method and device. An eye fundus image acquisition unit is configured to acquire an eye fundus image to be segmented. An image preprocessing unit is configured to perform standardization preprocessing on the eye fundus image to be segmented to obtain a standardized image sample. A binary classification label output unit is configured to input the standardized image sample into an advanced segmentation model subjected to secondary supervision training, output a blood vessel category probability of each pixel unit, and convert the blood vessel category probability threshold into a binary classification label of blood vessels or background. The training of the advanced segmentation model comprises performing binary classification supervision training on pixel units of the labeled image sample or high-confidence pixel units of the second pseudo label; and further comprises performing a loss masking operation on low-confidence pixel units in the second pseudo label sample.

2. The retinal blood vessel segmentation system based on pixel-level screening strategy according to claim 1, wherein, The secondary supervision training step of the advanced segmentation model comprises: S1, loading N eye fundus image samples to be processed; S2, labeling K labeled image samples in the N eye fundus image samples and predicting M first pseudo label samples; Each pixel unit of the first pseudo label sample is marked with a blood vessel category probability. S3, screening and marking the pixel units of the first pseudo label sample until M second pseudo label samples are obtained; wherein the pixel units of the second pseudo label sample are marked as high-confidence pixel units or low-confidence pixel units; S4, merging the M second pseudo label samples and the K labeled image samples to obtain J high-quality image samples; wherein J=M+K; S5, inputting the J high-quality image samples into a basic segmentation model for secondary supervision training to generate an advanced segmentation model through iteration.

3. The pixel-level screening policy based retinal blood vessel segmentation system according to claim 2, wherein, In the N eye fundus image samples, K labeled image samples are labeled, and M first pseudo label samples are predicted, comprising: S2-1, selecting K eye fundus image samples from the N eye fundus image samples; S2-2, performing binary classification labeling on the K selected eye fundus image samples, defining the K labeled image samples, and defining the eye fundus image samples not selected for labeling as unlabeled image samples; The pixel units of the labeled image sample are labeled with a binary classification true label, and the binary classification true label represents that the category of the pixel unit in the eye fundus image sample is truly labeled as retinal blood vessels or retinal background; S2-3, inputting the K labeled image samples into a U-Net network for primary supervision training to generate a basic segmentation model through iteration; wherein the output target of the basic segmentation model is defined as a blood vessel category probability of a pixel unit; S2-4, inputting the unlabeled image samples into the basic segmentation model for forward propagation inference to predict M first pseudo label samples.

4. The pixel-level screening policy based retinal blood vessel segmentation system according to claim 3, wherein, In the K selected eye fundus image samples, binary classification labeling is performed to define K labeled image samples, and eye fundus image samples not selected for labeling are defined as unlabeled image samples, comprising: S2-2-1, performing standardization preprocessing on the N eye fundus image samples to obtain N standardized image samples; S2-2-2, performing data enhancement on the N standardized image samples to obtain M unlabeled image samples; M>N; S2-2-3, selecting K unlabeled image samples from the M unlabeled image samples; S2-2-4, performing binary classification labeling on the pixel units of the K unlabeled image samples to obtain K labeled image samples.

5. The pixel-level screening policy based retinal blood vessel segmentation system according to claim 4, wherein, S3-1, performing median filtering on the first pseudo-label sample to generate a noise-free image sample; S3-2, performing edge detection on the noise-free image sample to label the pixel units as contour pixel units or non-contour pixel units; S3-3, performing dynamic threshold judgment on the non-contour pixel units of the noise-free image sample to label the non-contour pixel units as high-confidence pixel units or low-confidence pixel units; S3-4, traversing all pixel units of the M first pseudo-label samples, repeatedly performing median filtering, edge detection and dynamic threshold judgment until the M first pseudo-label samples are marked as M second pseudo-label samples. S3-1-1, anchoring a pixel unit to be marked of the first pseudo-label sample; 6. The pixel-level screening policy based retinal blood vessel segmentation system according to claim 5, wherein, S3-1-2, constructing a square neighborhood window with the pixel unit to be marked as the center of the window; S3-1-3, obtaining pixel values of the neighboring pixel units in the neighborhood window, and constructing a pixel value sequence based on the pixel value sorting; S3-1-4, selecting a pixel median value in the pixel value sequence, and replacing the pixel value of the pixel unit to be marked with the pixel median value; S3-1-5, traversing all pixel units of the first pseudo-label sample until the pixel values of all pixel units are replaced by the pixel median values in their neighborhood windows to generate the noise-free image sample. S3-2-1, obtaining a gray value of each pixel unit in the noise-free image sample; S3-2-2, marking neighboring pixel units in the noise-free image sample and calculating the gray difference value of the neighboring pixel units; S3-2-3, if the gray difference value is not 0, it is determined that the neighboring pixel units are all contour pixel units; otherwise, it is determined to be non-contour pixel units; 7. The pixel-level screening policy based retinal blood vessel segmentation system according to claim 5, wherein, S3-2-4, traversing all pixel units of the noise-free image sample to label the pixel units as contour pixel units or non-contour pixel units. S3-3-1, selecting a non-contour pixel unit in the noise-free image sample as a current pixel unit; S3-3-2, obtaining the current iteration number of the current pixel unit; S3-3-3, calculating a dynamic probability threshold value of the current pixel unit according to the current iteration number; S3-3-4, obtaining a blood vessel class probability of the current pixel unit; 8. The pixel-level screening policy based retinal blood vessel segmentation system according to claim 5, wherein, S3-3-5, if the blood vessel class probability is greater than the dynamic probability threshold value, the current pixel unit is labeled as a high-confidence pixel unit, otherwise it is labeled as a low-confidence pixel unit. ​ ​ ​ ​ ​ 9. The pixel-level screening policy based retinal blood vessel segmentation system according to claim 2, wherein, The J high-quality image samples are input into the basic segmentation model for secondary supervision training, and an advanced segmentation model is generated through iteration, including: S5-1, marking the current iteration pixel unit of the high-quality image sample; S5-2, performing binary classification supervision training or loss shielding operation according to the unit category inherited by the current iteration pixel unit; S5-3, if the unit category is the pixel unit of the labeled image sample or the high-confidence pixel unit of the second pseudo label, performing binary classification supervision training based on the pixel unit of the labeled image sample or the high-confidence pixel unit of the second pseudo label; S5-4, if the sample category is the low-confidence pixel unit of the second pseudo label, performing loss shielding operation on the low-confidence pixel unit in the second pseudo label sample.

10. The pixel-level screening policy based retinal blood vessel segmentation system according to claim 8, wherein, The shielding operation of loss calculation on the low-confidence pixel unit in the second pseudo label sample includes: S5-4-1, anchoring the low-confidence pixel unit; S5-4-2, forward propagation of the low-confidence pixel unit, predicting the blood vessel category probability of the low-confidence pixel unit; S5-4-3, skipping the cross-entropy loss calculation of the low-confidence pixel unit, and directly entering the secondary supervision training of the next iteration pixel unit.

Images