A quality assessment system for retinal laser photocoagulation surgery

Abstract

The present invention discloses a retinal laser photocoagulation surgery quality assessment system, electronic equipment, and computer-readable storage medium, belonging to the field of laser spot detection technology. The present invention establishes an assessment system based on three dimensions: laser spot level, laser spot spacing, and whether blood vessels are damaged. It also utilizes an artificial intelligence algorithm for automatic assessment, thereby resolving the large workload, subjectivity, and instability inherent in manual assessments. The system can accurately and rapidly assess the quality of retinal laser photocoagulation surgery, reducing reliance on manual labor and the influence of subjective factors. This resolves the existing technical issues of "large workload, subjectivity, and instability inherent in manual assessments of retinal laser photocoagulation surgery quality."

Description

Technical Field

[0001] The present application relates to the technical field of laser spot detection, and in particular to a retinal laser photocoagulation surgery quality assessment system. Background Art

[0002] The statements in this section merely mention the background technology related to this application and do not necessarily constitute the prior art.

[0003] Common retinal blinding diseases such as diabetic retinopathy, retinal vein occlusion, and retinopathy of prematurity have high morbidity and blindness rates, placing a heavy medical and economic burden on society. Retinal laser photocoagulation surgery is an important treatment for these common blinding fundus diseases. High-quality retinal laser photocoagulation surgery can effectively slow disease progression and save patients' vision. However, substandard retinal laser photocoagulation fails to slow disease progression and can lead to serious complications. Therefore, there is an urgent need to evaluate the quality of retinal laser photocoagulation surgery.

[0004] The parameters of laser photocoagulation surgery include laser power, spot size, duration, laser spot spacing, etc. Studies have shown that various diseases require a certain laser energy to exert a therapeutic effect, and the laser energy acts on the retina and is manifested as a laser spot formed after the laser is irradiated on the retina. Therefore, researchers divide laser spots into four levels according to the color and shape of the laser spots formed by different energies. Clinically, the level of laser spots is often used to judge whether retinal laser surgery meets the standards. For example, diabetic retinopathy, retinal vein occlusion, and retinopathy of prematurity all require level 3 laser spots to exert the therapeutic effect of laser photocoagulation. But the current problems are:

[0005] (1) The evaluation of retinal laser photocoagulation surgery is labor-intensive, and manual evaluation is subjective and unstable. In addition, junior doctors lack experience. Therefore, it is difficult to conduct an effective and efficient evaluation of retinal laser photocoagulation surgery.

[0006] (2) Currently, there is no intelligent system that can automatically evaluate the quality of retinal laser photocoagulation surgery. Summary of the Invention

[0007] In order to address the shortcomings of the existing technology, the present application provides a retinal laser photocoagulation surgery quality assessment system, electronic equipment and computer-readable storage medium, establishes an assessment system based on three dimensions: laser spot level, laser spot spacing, and whether blood vessels are damaged, and uses artificial intelligence algorithms for automatic assessment, thereby solving the problems of large workload and subjectivity and instability of manual assessment.

[0008] In a first aspect, the present application provides a retinal laser photocoagulation surgery quality assessment system;

[0009] A retinal laser photocoagulation surgery quality assessment system, comprising:

[0010] The image acquisition module is configured to: acquire a color fundus photograph of a patient after retinal laser photocoagulation surgery;

[0011] The light spot target detection module is configured to: input the fundus color photograph of the patient after retinal laser photocoagulation surgery into the light spot target detection model to obtain the light spot rectangular block image and the light spot level;

[0012] The retinal blood vessel damage prediction module is configured to: input the rectangular block image of the light spot into the light spot and retinal blood vessel relationship model to predict whether the light spot damages the retinal blood vessels;

[0013] The light spot distribution detection module is configured to: determine whether the light spot distribution is uniform based on the light spot rectangular block image, the variance analysis method based on the light spot spacing, and the 3sigma anomaly detection principle.

[0014] Furthermore, the step of inputting the fundus color photograph of the patient after retinal laser photocoagulation surgery into the spot target detection model to obtain the spot rectangular block image and the spot level specifically includes:

[0015] Receive color fundus photos of patients after retinal laser photocoagulation surgery and pre-process the color fundus photos of patients after retinal laser photocoagulation surgery;

[0016] According to the pre-processed fundus color photograph, the light spot features in the image are extracted and fused;

[0017] According to the fused spot features, the position, size and level of the spot are determined, the rectangular block image of the spot is obtained and the category confidence is output.

[0018] Furthermore, the pre-processing of the fundus color photograph after retinal laser photocoagulation surgery includes:

[0019] Perform cropping and zooming operations on the fundus color photograph after retinal laser photocoagulation surgery;

[0020] The cropped and scaled fundus color photograph was resized to 608x608x3.

[0021] Furthermore, the spot target detection model is a YOLO V5 model, and the spot target detection model includes an input end, a Backone network, a Neck network, and a Prediction network connected in sequence;

[0022] The input end is used to receive the fundus color photographs of patients after retinal laser photocoagulation surgery, and preprocess the fundus color photographs of patients after retinal laser photocoagulation surgery. The Backone network is used to extract the spot features of the preprocessed fundus color photographs. The Neck network is used to fuse the spot features. The Prediction network is used to determine the position, size and level of the spot based on the spot features and output the category confidence and the spot rectangular block image.

[0023] Furthermore, the rectangular block image of the light spot is input into the light spot and retinal blood vessel relationship model to predict whether the light spot damages the retinal blood vessels. Specifically, the following steps are involved:

[0024] Preprocessing the light spot rectangular block image;

[0025] Extract high-dimensional feature maps of the preprocessed spot rectangular block image and adjust the number of high-dimensional feature maps;

[0026] The high-dimensional feature map is reduced in dimensionality, and classification is completed through the output feature map, and the classification result is output, which is whether retinal blood vessels are contained.

[0027] Furthermore, the light spot and retinal blood vessel relationship model is an EfficientNet network model.

[0028] Furthermore, the EfficientNet network model includes a Conv3x3 layer, an MBConv module, a Conv 1x1 layer and an FC fully connected module connected in sequence. The Conv3x3 layer is used to preprocess the light spot rectangular block image, and the MBConv module is used to obtain the image high-dimensional feature map based on the preprocessed light spot rectangular block image; the Conv 1x1 layer is used to adjust the number of high-dimensional feature maps of the image; the FC fully connected module is used to reduce the dimensionality of the high-dimensional feature map, complete classification through the output feature map, and output the classification result.

[0029] Furthermore, the implementation process of the light spot distribution detection module is as follows:

[0030] Extract the center of the light spot rectangular frame in the light spot rectangular block image and calculate the Euclidean distance between the centers of each light spot;

[0031] The mean and variance of the spot spacing are calculated and compared with the parameters of the standard spot distribution. Based on the 3sigma anomaly detection principle, it is determined whether the spot distribution is uniform.

[0032] In a second aspect, the present application provides an electronic device;

[0033] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed by the processor, wherein when the computer instructions are executed by the processor, the following steps are performed:

[0034] Obtain color fundus photographs of patients after retinal laser photocoagulation surgery;

[0035] Input the fundus color photo of the patient after retinal laser photocoagulation surgery into the spot target detection model to obtain the spot rectangular block image and the spot category;

[0036] The rectangular block image of the light spot is input into the light spot and retinal blood vessel relationship model to predict whether the light spot will damage the retinal blood vessels;

[0037] According to the rectangular block image of the light spot, the variance analysis method based on the light spot spacing and the 3sigma anomaly detection principle are used to determine whether the light spot distribution is uniform.

[0038] In a third aspect, the present application provides a computer-readable storage medium;

[0039] A computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the following steps are performed:

[0040] Obtain color fundus photographs of patients after retinal laser photocoagulation surgery;

[0041] Input the fundus color photograph of the patient after retinal laser photocoagulation surgery into the spot target detection model to obtain the spot rectangular block image and the spot level;

[0042] The rectangular block image of the light spot is input into the light spot and retinal blood vessel relationship model to predict whether the light spot will damage the retinal blood vessels;

[0043] According to the rectangular block image of the light spot, the variance analysis method based on the light spot spacing and the 3sigma anomaly detection principle are used to determine whether the light spot distribution is uniform.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] (1) This application focuses on the difficulty of evaluating the quality of retinal laser photocoagulation surgery. An evaluation system is established based on three dimensions: laser spot level, laser spot spacing, and whether blood vessels are damaged. An artificial intelligence algorithm is used for automatic evaluation. This solves the problems of large workload, subjectivity, and instability in manual evaluation of the quality of retinal laser photocoagulation surgery, thereby improving the accuracy and efficiency of the quality evaluation of retinal laser photocoagulation surgery.

[0046] (2) The technical solution provided by this application can efficiently and accurately evaluate the quality of retinal laser photocoagulation surgery, reduce the adverse reactions of retinal laser photocoagulation surgery, and improve the efficacy. At the same time, it can help to quickly improve the quality of retinal laser photocoagulation surgery performed by junior doctors and reduce the occurrence of adverse medical events. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings in the specification, which constitute a part of this application, are used to provide further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute improper limitations on this application.

[0048] Figure 1 A schematic diagram of the system framework provided in an embodiment of the present application;

[0049] Figure 2 A schematic diagram of the YOLO v5 network structure provided in an embodiment of the present application;

[0050] Figure 3 The EfficientNet network structure provided in the embodiments of this application;

[0051] Figure 4 Schematic diagram of the 3sigma anomaly detection principle provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this application have the same meaning as commonly understood by those skilled in the art to which this application belongs.

[0053] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0054] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0055] Example 1

[0056] In the existing technology, manual evaluation of retinal laser photocoagulation surgery is very labor-intensive and subjective and unstable; therefore, the present application provides a retinal laser photocoagulation surgery quality evaluation system, which uses artificial intelligence algorithms to establish an evaluation system for the quality of retinal laser photocoagulation surgery from three dimensions: laser spot level, laser spot spacing, and whether blood vessels are damaged.

[0057] Next, combine Figure 1-4 A retinal laser photocoagulation surgery quality assessment system disclosed in this embodiment is described in detail.

[0058] A retinal laser photocoagulation surgery quality assessment system includes an image acquisition module, a light spot target detection module, a retinal vascular damage prediction module and a light spot distribution detection module; the image acquisition module is used to obtain a color fundus photograph of a patient after retinal laser photocoagulation surgery; the light spot target detection module is used to input the color fundus photograph of the patient after retinal laser photocoagulation surgery into a light spot target detection model to obtain a light spot rectangular block image and the level of the light spot (i.e., to grade the light spot); the retinal vascular damage prediction module is used to input the light spot rectangular block image into a light spot and retinal vascular relationship model to predict whether the light spot damages the retinal blood vessels; the light spot distribution detection module is used to determine whether the light spot distribution is uniform based on the light spot rectangular block image, a variance analysis method based on the light spot spacing, and the 3sigma anomaly detection principle.

[0059] The implementation of the image acquisition module is as follows:

[0060] Obtain color fundus photographs of patients immediately after retinal laser photocoagulation surgery.

[0061] The implementation of the spot target detection module is as follows:

[0062] The fundus color photograph of the patient after retinal laser photocoagulation surgery is input into the spot target detection model to obtain the spot rectangular block image and spot category. The specific steps include:

[0063] (1) Receive a color fundus photograph of a patient after retinal laser photocoagulation surgery, and pre-process the color fundus photograph of the patient after retinal laser photocoagulation surgery; specifically, perform cropping and scaling operations on the color fundus photograph after retinal laser photocoagulation surgery in sequence; and adjust the size of the cropped and scaled color fundus photograph to 608x608x3.

[0064] (2) Extracting the light spot features in the image based on the preprocessed fundus color photograph and fusing the light spot features;

[0065] (3) According to the fused light spot features, the position, size and category of the light spot are determined, the rectangular block image of the light spot is obtained and the category confidence is output; wherein, the category of the light spot is the classification result of the light spot by the light spot target detection model according to the light spot classification standard.

[0066] For example, the target detection model of the light spot adopts the YOLO v5 model, and its network structure is as follows Figure 2As shown in the figure, it specifically consists of four parts: the input, the Backbone network, the Neck network, and the Prediction network. The input receives pre-processed color fundus images after retinal laser photocoagulation surgery, which undergo a series of transformations such as cropping and scaling to a size of 608x608x3. The Backbone network, which includes the Focus and CSP structures, serves as the backbone network and is used to extract the light spot features from the pre-processed color fundus images. The Neck network further integrates the light spot features through a more complex combination of modules. The Prediction network is the classifier and regressor of the YOLO v5 model, which is used to determine the location, size, and level of the light spot and output its category confidence. Once the YOLO v5 model is trained, it can detect light spots and their levels based on the input of post-operative retinal images.

[0067] The implementation of the retinal vascular damage prediction module is as follows:

[0068] The rectangular block image of the light spot is pre-processed and input into the light spot and retinal blood vessel relationship model to predict whether the light spot will damage the retinal blood vessels. Specifically, it includes:

[0069] (1) Preprocessing the light spot rectangular block image;

[0070] (2) extracting high-dimensional feature maps of the preprocessed spot rectangular block image and adjusting the number of high-dimensional feature maps;

[0071] (3) Dimensionality reduction is performed on the high-dimensional feature map, and classification is completed through the output feature map, and the classification result is output, which is whether retinal blood vessels are contained.

[0072] For example, the light spot and retinal blood vessel relationship model uses the advanced convolutional neural network EfficientNet, and adjusts the convolution input and FC output of the network results according to the characteristics of this task, as shown in the following example: Figure 3 As shown in the figure, the EfficientNet network model uses a composite model scaling method to balance multiple dimensions of resolution, depth, and width to optimize the network's efficiency and accuracy.

[0073] The workflow of the EfficientNet network model is as follows: 1. The rectangular block image of the light spot detected by the YOLO v5 model is preprocessed to obtain an image of size 32x32x3 as the network input; 2. The input image is converted into the input dimension required by the MBConv module through the first Conv3x3 layer; 3. The image is passed through a series of MBConv modules to obtain high-dimensional feature maps of the image to achieve high-dimensional feature extraction. Among them, the parameters of each MBConv module are fine-tuned to adapt to the current task. The combined scale optimization method enables the network to obtain a better receptive field; 4. The number of high-dimensional feature maps is adjusted using the Conv 1x1 layer; 5. The feature dimension is reduced using a series of FC fully connected modules; 5. Finally, the output feature map is used to complete the classification and accurately predict whether the light spot has damaged the retinal blood vessels. If the classification result shows that retinal blood vessels are present, the light spot has damaged the retinal blood vessels. If the classification result shows that retinal blood vessels are not present, the light spot has not damaged the retinal blood vessels.

[0074] The implementation of the spot distribution detection module is as follows:

[0075] (1) Extract the center of the spot rectangle in the spot rectangle image and calculate the Euclidean distance between the centers of each spot. Assume that the mean z of the spot distribution is independent and conforms to the normal distribution:

[0076]

[0077] (2) The mean and variance of the spot spacing are calculated and compared with the parameters of the standard spot distribution. Based on the 3sigma anomaly detection principle, it is determined whether the spot distribution is uniform.

[0078] Anomaly detection based on the 3sigma statistical method does not require an empirical threshold. Instead, it adopts the 3sigma principle, assuming that the detection data only contains random errors. The original data is calculated and processed to obtain the standard deviation, and then an interval is determined according to a certain probability. It is considered that errors exceeding this interval are outliers. Figure 4 This is the principle diagram of 3sigma anomaly detection, and the values ​​are distributed in the interval (μ-3σ，μ+3σ) The proportion is as high as 99.7%, and the points before this range are outliers, that is, the light spot distribution is uneven.

[0079] Finally, the quality of retinal laser photocoagulation surgery was evaluated based on the level of light spot, whether the light spot damaged retinal blood vessels, and whether the light spot was evenly distributed.

[0080] In addition to the Yolo series, the light spot target detection model can also be replaced by other deep learning target detection networks; in addition to EfficientNet, the light spot and retinal blood vessel relationship model can also be replaced by other deep learning classification networks, such as VGG, ResNet, GoogleNet, and ViT.

[0081] In some embodiments, a surgery quality assessment module is further included, which is used to assess the quality of retinal laser photocoagulation surgery based on the light spot level, whether the light spot damages retinal blood vessels, and whether the light spot distribution is uniform.

[0082] Example 2

[0083] A second embodiment of the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the following steps are performed: obtaining a color fundus photograph of a patient after retinal laser photocoagulation surgery;

[0084] Input the fundus color photo of the patient after retinal laser photocoagulation surgery into the spot target detection model to obtain the spot rectangular block image and the spot category;

[0085] The rectangular block image of the light spot is input into the light spot and retinal blood vessel relationship model to predict whether the light spot will damage the retinal blood vessels;

[0086] According to the rectangular block image of the light spot, the variance analysis method based on the light spot spacing and the 3sigma anomaly detection principle are used to determine whether the light spot distribution is uniform.

[0087] Example 3

[0088] A third embodiment of the present invention provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the following steps are performed:

[0089] Obtain color fundus photographs of patients after retinal laser photocoagulation surgery;

[0090] Input the fundus color photo of the patient after retinal laser photocoagulation surgery into the spot target detection model to obtain the spot rectangular block image and the spot category;

[0091] The rectangular block image of the light spot is input into the light spot and retinal blood vessel relationship model to predict whether the light spot will damage the retinal blood vessels;

[0092] According to the rectangular block image of the light spot, the variance analysis method based on the light spot spacing and the 3sigma anomaly detection principle are used to determine whether the light spot distribution is uniform.

[0093] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0094] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide the functions for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0096] The descriptions of the various embodiments in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0097] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A retinal laser photocoagulation surgery quality assessment system, characterized by: include: The image acquisition module is configured to: acquire a color fundus photograph of a patient after retinal laser photocoagulation surgery; The light spot target detection module is configured to: input the fundus color photograph of the patient after retinal laser photocoagulation surgery into the light spot target detection model to obtain the light spot rectangular block image and the light spot level; The retinal blood vessel damage prediction module is configured to: input the rectangular block image of the light spot into the light spot and retinal blood vessel relationship model to predict whether the light spot damages the retinal blood vessels; The light spot distribution detection module is configured to: determine whether the light spot distribution is uniform based on the light spot rectangular block image, the variance analysis method based on the light spot spacing, and the 3sigma anomaly detection principle.

2. The retinal laser photocoagulation surgery quality assessment system according to claim 1, wherein: The method of inputting the fundus color photograph of the patient after retinal laser photocoagulation surgery into the spot target detection model to obtain the spot rectangular block image and the spot category specifically includes: Receive color fundus photos of patients after retinal laser photocoagulation surgery and pre-process the color fundus photos of patients after retinal laser photocoagulation surgery; According to the pre-processed fundus color photograph, the light spot features in the image are extracted and fused; According to the fused light spot features, the position, size and category of the light spot are determined, the rectangular block image of the light spot is obtained and the category confidence is output.

3. The retinal laser photocoagulation surgery quality assessment system according to claim 2, wherein: Pre-processing of fundus color photographs after retinal laser photocoagulation surgery includes: Perform cropping and zooming operations on the fundus color photograph after retinal laser photocoagulation surgery; The cropped and scaled fundus color photograph was resized to 608x608x3.

4. The retinal laser photocoagulation surgery quality assessment system according to claim 1, wherein: The spot target detection model is a YOLO V5 model, which includes an input end, a Backone network, a Neck network, and a Prediction network connected in sequence; The input end is used to receive the fundus color photographs of patients after retinal laser photocoagulation surgery and preprocess the fundus color photographs of patients after retinal laser photocoagulation surgery. The Backone network is used to extract the spot features of the preprocessed fundus color photographs. The Neck network is used to fuse the spot features. The Prediction network is used to determine the position, size and category of the spot based on the spot features and output the category confidence and the spot rectangular block image.

5. The retinal laser photocoagulation surgery quality assessment system according to claim 1, wherein: Inputting the rectangular block image of the light spot into the light spot and retinal blood vessel relationship model to predict whether the light spot damages the retinal blood vessels specifically includes: Preprocessing the light spot rectangular block image; Extract high-dimensional feature maps of the preprocessed spot rectangular block image and adjust the number of high-dimensional feature maps; The high-dimensional feature map is reduced in dimensionality, and classification is completed through the output feature map, and the classification result is output, which is whether retinal blood vessels are contained.

6. The retinal laser photocoagulation surgery quality assessment system according to claim 1, wherein: The relationship model between the light spot and the retinal blood vessels is an EfficientNet network model.

7. The retinal laser photocoagulation surgery quality assessment system according to claim 6, wherein: The EfficientNet network model includes a Conv3x3 layer, an MBConv module, a Conv 1x1 layer, and an FC fully connected module connected in sequence. The Conv3x3 layer is used to preprocess the light spot rectangular block image. The MBConv module is used to obtain the image high-dimensional feature map based on the preprocessed light spot rectangular block image; the Conv 1x1 layer is used to adjust the number of high-dimensional feature maps of the image; the FC fully connected module is used to reduce the dimensionality of the high-dimensional feature map, complete classification through the output feature map, and output the classification result.

8. The retinal laser photocoagulation surgery quality assessment system according to claim 1, wherein: The implementation process of the light spot distribution detection module is as follows: Extract the center of the light spot rectangular frame in the light spot rectangular block image and calculate the Euclidean distance between the centers of each light spot; The mean and variance of the spot spacing are calculated and compared with the parameters of the standard spot distribution. Based on the 3sigma anomaly detection principle, it is determined whether the spot distribution is uniform.

9. An electronic device, characterized in that: The system comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein the computer instructions are executed by the processor to complete the following steps: Obtain color fundus photographs after retinal laser photocoagulation surgery and perform preprocessing; Input the pre-processed fundus color photo into the spot target detection model to obtain the spot rectangular block image and the spot category; The rectangular block image of the light spot is pre-processed and input into the light spot and retinal blood vessel relationship model to predict whether the light spot will damage the retinal blood vessels; According to the rectangular block image of the light spot, the variance analysis method based on the light spot spacing and the 3sigma anomaly detection principle are used to determine whether the light spot distribution is uniform.

10. A computer-readable storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, complete the following steps: Obtain color fundus photographs after retinal laser photocoagulation surgery and perform preprocessing; Input the pre-processed fundus color photo into the spot target detection model to obtain the spot rectangular block image and the spot category; The rectangular block image of the light spot is pre-processed and input into the light spot and retinal blood vessel relationship model to predict whether the light spot will damage the retinal blood vessels; According to the rectangular block image of the light spot, the variance analysis method based on the light spot spacing and the 3sigma anomaly detection principle are used to determine whether the light spot distribution is uniform.

Images