Fundus recognition hyperspectral imaging chip design method, fundus imaging device and equipment

Abstract

The application discloses a fundus recognition hyperspectral imaging chip design method, a fundus imaging device and equipment, and relates to the field of intelligent medical treatment. The method comprises the following steps: collecting a hyperspectral image sample of a clinical hyperspectral leopard-shaped fundus in a set wave band range, and performing wave band splitting to obtain a single-channel spectral image; performing image scoring on the single-channel spectral image to form a training sample, so as to train a neural network model; randomly combining a plurality of single-channel spectral images corresponding to each patient to obtain a plurality of single-channel spectral image combinations; scoring each single-channel spectral image combination based on the trained neural network model, and screening a set number of single-channel spectral image combinations with high scores; constructing a corresponding hyperspectral imaging chip for each single-channel spectral image combination, and performing clinical evaluation to obtain an optimal hyperspectral imaging chip. The application can effectively improve the recognition efficiency of the leopard-shaped fundus in clinical treatment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent medical treatment, in particular to a fundus recognition hyperspectral imaging chip design method, a fundus imaging device and equipment. BACKGROUND

[0002] With the continuous development of ophthalmology and artificial intelligence technology, fundus image analysis plays an increasingly important role in early disease screening, diagnosis and treatment monitoring. Fundus images can reflect the health status of human vascular, neural and retinal structures, and precise image analysis is of great significance, especially in the diagnosis of ophthalmic diseases such as diabetic retinopathy, glaucoma and macular degeneration. In recent years, with the continuous progress of deep learning and computer vision technology, automatic diagnosis technology based on fundus images has gradually been applied.

[0003] Leopard-like fundus is a relatively special fundus lesion pattern, which is often seen in specific genetic eye diseases or systemic diseases. Its characteristic is that there are irregularly distributed pigmentations or atrophies on the retinal fundus. Due to the particularity of leopard-like fundus, it has certain challenges in early diagnosis and clinical monitoring, especially in the process of differential diagnosis with other lesions. Traditional fundus image analysis methods often have difficulty in achieving high-precision and high-efficiency automatic recognition when dealing with such complex images, so there is an urgent need for further technological development to achieve accurate leopard-like fundus recognition.

[0004] In recent years, hyperspectral imaging technology has been widely used in medical imaging field because it can capture image information at different spectral bands, and has shown great potential in ophthalmic diagnosis. For accurate leopard-like fundus recognition, the application of hyperspectral imaging chip is expected to break through the limitations of traditional band image analysis and provide more information. Hyperspectral imaging chip can acquire fundus images at multiple bands such as visible light and near-infrared. This multi-angle imaging method can effectively reveal different levels and subtle structures of fundus lesions, especially the complex features of leopard-like fundus such as pigment deposition and vascular changes. This means that through the design and construction of hyperspectral chip, fine recognition and real-time analysis of fundus images can be achieved, which is expected to achieve accurate leopard-like fundus recognition. Therefore, how to construct a hyperspectral imaging chip for accurate leopard-like fundus recognition has become a problem to be solved. SUMMARY

[0005] The present application provides a fundus recognition hyperspectral imaging chip design method, a fundus imaging device and equipment, which can effectively improve the clinical leopard-like fundus recognition efficiency.

[0006] In a first aspect, embodiments of this application provide a method for designing a fundus recognition hyperspectral imaging chip, the method comprising:

[0007] Hyperspectral images of clinical leopard-print fundus under a set band range were acquired, and the hyperspectral images were split into bands to obtain single-channel spectral images.

[0008] Image scoring is performed on single-channel spectral images to form training samples for training the neural network model. Multiple single-channel spectral images corresponding to each patient are randomly combined to obtain multiple single-channel spectral image combinations.

[0009] The trained neural network model scores each single-channel spectral image combination, and the top-scoring single-channel spectral image combinations are selected based on the scoring results.

[0010] For each single-channel spectral image combination obtained from the screening, a corresponding hyperspectral imaging chip was constructed and integrated into the fundus imaging device for clinical evaluation to obtain the optimal hyperspectral imaging chip.

[0011] In conjunction with the first aspect, in one embodiment, the acquisition of hyperspectral image samples of clinical hyperspectral leopard-print fundus under a set wavelength range, and the band splitting of the hyperspectral image samples to obtain single-channel spectral images, specifically includes:

[0012] Collect hyperspectral image samples of clinical hyperspectral leopard-spot fundus, wherein the hyperspectral image samples are imaging data of each specific band within a set band range;

[0013] The acquired hyperspectral image samples are divided into bands to obtain single-channel spectral images of specific bands.

[0014] In conjunction with the first aspect, in one implementation, the step of performing image scoring on single-channel spectral images to form training samples for training a neural network model specifically includes:

[0015] The current single-channel spectral image is scored based on Sobel gradient, Laplace variance and subjective scoring methods respectively, and the scores of each scoring method are weighted and calculated to obtain the score of the current single-channel spectral image.

[0016] Training samples are constructed based on each single-channel spectral image and its score to train the neural network model.

[0017] In conjunction with the first aspect, in one implementation, the step of scoring each single-channel spectral image combination based on the trained neural network model, and selecting a predetermined number of single-channel spectral image combinations with the highest scores based on the scoring results, specifically includes:

[0018] The trained neural network model scores each single-channel spectral image within the single-channel spectral image combination, and the sum of the scores of each single-channel spectral image is used as the score of the current single-channel spectral image combination.

[0019] Based on the scores of each single-channel spectral image combination, they are sorted from highest to lowest, and a set number of single-channel spectral image combinations with the highest scores are selected.

[0020] In conjunction with the first aspect, in one implementation, the step of constructing a corresponding hyperspectral imaging chip for each selected single-channel spectral image combination and integrating it into a fundus imaging device for clinical evaluation to obtain the optimal hyperspectral imaging chip specifically includes:

[0021] For each single-channel spectral image combination obtained from the screening, a corresponding hyperspectral imaging chip is constructed;

[0022] Hyperspectral imaging chips were integrated into fundus imaging devices and clinical assessments of leopard-pattern fundus recognition were performed sequentially. The optimal hyperspectral imaging chip was determined based on the clinical assessment results.

[0023] Secondly, embodiments of this application provide a fundus imaging device for leopard-pattern fundus recognition, comprising a fundus camera and a hyperspectral imaging chip camera working in concert, wherein the hyperspectral imaging chip camera is constructed based on the optimal hyperspectral imaging chip obtained by the design method described above.

[0024] In conjunction with the second aspect, in one embodiment, the specific process of the fundus imaging device performing leopard-pattern fundus recognition includes:

[0025] Establish a physical connection between the fundus camera and the hyperspectral imaging chip camera, and maintain consistency in the field of view and focus of the fundus camera and the hyperspectral imaging chip camera;

[0026] The system controls a fundus camera and a hyperspectral imaging chip camera to acquire fundus images, and then fuses the fundus images acquired by the two cameras to generate a comprehensive image containing structural features and spectral information, thereby enabling the recognition of leopard-patterned fundus.

[0027] In conjunction with the second aspect, in one implementation, establishing a physical connection between the fundus camera and the hyperspectral imaging chip camera, and maintaining consistency in the field of view and focus of the fundus camera and the hyperspectral imaging chip camera, specifically includes:

[0028] A physical connection is established between the fundus camera and the hyperspectral imaging chip camera via an adapter or standard interface.

[0029] Adjust the field of view and focus of the fundus camera and the hyperspectral imaging chip camera to keep them consistent.

[0030] In conjunction with the second aspect, in one embodiment, the control of the fundus camera and the hyperspectral imaging chip camera to acquire fundus images, and the fusion of the fundus images acquired by the fundus camera and the hyperspectral imaging chip camera to generate a comprehensive image containing structural features and spectral information, specifically including:

[0031] Synchronous control of the fundus camera and the hyperspectral imaging chip camera is achieved through hardware and software collaboration, so that the hyperspectral imaging chip camera can acquire fundus images of the same area as the fundus camera under different spectral bands.

[0032] The fundus images acquired by the fundus camera and the hyperspectral imaging chip camera are fused using an image fusion algorithm to generate a comprehensive image that includes structural features and spectral information.

[0033] Thirdly, this application provides a fundus recognition hyperspectral imaging chip design device, characterized in that the fundus recognition hyperspectral imaging chip design device includes a processor, a memory, and a fundus recognition hyperspectral imaging chip design program stored in the memory and executable by the processor, wherein when the fundus recognition hyperspectral imaging chip design program is executed by the processor, the steps of the fundus recognition hyperspectral imaging chip design method described above are implemented.

[0034] The beneficial effects of the technical solutions provided in this application include:

[0035] By designing an optimal hyperspectral imaging chip, a hyperspectral imaging chip camera is obtained based on the optimal hyperspectral imaging chip, thereby obtaining a fundus imaging device for clinical patient fundus imaging. This enables accurate identification of leopard-patterned fundus images with a high accuracy rate, providing ophthalmologists with intuitive and convenient diagnostic support. This improves the efficiency of clinical identification of leopard-patterned fundus and provides important support for the accurate identification of leopard-patterned fundus lesions, demonstrating significant clinical value and significance. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the design method of the fundus recognition hyperspectral imaging chip in this application;

[0037] Figure 2 This is a schematic diagram of the functional modules of the fundus imaging device of this application;

[0038] Figure 3 This is a schematic diagram of the hardware structure of the fundus recognition hyperspectral imaging chip design device of this application. Detailed Implementation

[0039] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0040] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0041] In a first aspect, embodiments of this application provide a design method for a hyperspectral imaging chip for fundus recognition.

[0042] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating the design method of the fundus recognition hyperspectral imaging chip for this application. Figure 1 As shown, the design method of the fundus recognition hyperspectral imaging chip includes:

[0043] S1: Acquire hyperspectral image samples of clinical hyperspectral leopard-print fundus under a set band range, and perform band splitting on the hyperspectral image samples to obtain single-channel spectral images;

[0044] S2: Image scoring is performed on single-channel spectral images to form training samples for training the neural network model. Multiple single-channel spectral images corresponding to each patient are randomly combined to obtain multiple single-channel spectral image combinations.

[0045] S3: Based on the trained neural network model, score each single-channel spectral image combination, and select the top-scoring single-channel spectral image combinations based on the scoring results;

[0046] S4: For each single-channel spectral image combination obtained from the screening, a corresponding hyperspectral imaging chip is constructed and integrated into the fundus imaging device for clinical evaluation to obtain the optimal hyperspectral imaging chip.

[0047] Furthermore, in one embodiment, hyperspectral image samples of clinical hyperspectral leopard-print fundus under a set wavelength range are acquired, and the hyperspectral image samples are split into wavelengths to obtain single-channel spectral images, specifically including:

[0048] S101: Collect hyperspectral image samples of clinical hyperspectral leopard-spot fundus, wherein the hyperspectral image samples are imaging data of each specific band within a set band range;

[0049] S102: The acquired hyperspectral image samples are divided into bands to obtain single-channel spectral images of specific bands.

[0050] Specifically, in practical applications, the first step is to collect hyperspectral image samples of clinical hyperspectral leopard-print fundus, covering imaging data in the 623nm~794nm band, specifically including 623nm, 650nm, 663nm, 685nm, 709nm, 730nm, 751nm, 773nm, and 794nm. Then, the collected hyperspectral image samples are split into bands, resulting in single-channel spectral images of 623nm, 650nm, 663nm, 685nm, 709nm, 730nm, 751nm, 773nm, and 794nm.

[0051] Furthermore, in one embodiment, image scoring of single-channel spectral images is performed to form training samples for training the neural network model, specifically including:

[0052] S201: The current single-channel spectral image is scored based on Sobel gradient, Laplace variance and subjective scoring methods respectively, and the scores of each scoring method are weighted to obtain the score of the current single-channel spectral image.

[0053] S202: Based on each single-channel spectral image and its score, training samples are constructed to train the neural network model.

[0054] Specifically, for the multiple single-channel spectral images obtained after splitting, each single-channel spectral image is acquired sequentially. The current single-channel spectral image is scored using methods encompassing Sobel gradient, Laplace variance, and doctor's subjective evaluation. Then, weights are calculated for the Sobel gradient score, Laplace variance score, and doctor's subjective evaluation score. In one possible real-time mode, the weights for the Sobel gradient score and Laplace variance score are 0.25, and the weight for the doctor's subjective evaluation score is 0.5. Finally, the score of the current single-channel spectral image is obtained. Using the above method, the score of each single-channel spectral image is calculated. Then, based on the single-channel spectral images and their corresponding scores, training samples are constructed to train the neural network model, enabling the trained neural network model to score single-channel spectral images.

[0055] Furthermore, in one embodiment, the single-channel spectral image combinations are scored based on the trained neural network model, and a predetermined number of single-channel spectral image combinations with the highest scores are selected based on the scoring results. Specifically, this includes:

[0056] S301: Based on the trained neural network model, score each single-channel spectral image in the single-channel spectral image combination, and use the sum of the scores of each single-channel spectral image as the score of the current single-channel spectral image combination.

[0057] S302: Based on the scores of each single-channel spectral image combination, sort them from high to low, and select the set number of single-channel spectral image combinations with the highest scores.

[0058] Specifically, for the single-channel spectral images obtained after the previous splitting, the multiple single-channel spectral images corresponding to each patient are grouped together. The multiple single-channel spectral images within a single group are randomly recombinated, and each single group yields a set number of combinations of single-channel spectral images. For all patients, multiple combinations of single-channel spectral images are finally obtained. Then, a trained neural network model is used to score each single-channel spectral image within a single combination of single-channel spectral images. The sum of the scores of each single-channel spectral image is used as the score of the current single-channel spectral image combination. The single-channel spectral image combinations are sorted from high to low scores, and a set number of single-channel spectral image combinations with the highest scores are selected. In a possible real-time mode, the top 20 single-channel spectral image combinations are selected.

[0059] Furthermore, in one embodiment, a corresponding hyperspectral imaging chip is constructed for each selected single-channel spectral image combination and integrated into a fundus imaging device for clinical evaluation to obtain the optimal hyperspectral imaging chip, specifically including:

[0060] S401: Construct a corresponding hyperspectral imaging chip for each single-channel spectral image combination obtained from the screening;

[0061] S402: Integrate the hyperspectral imaging chips into the fundus imaging device and perform clinical evaluations of leopard-pattern fundus recognition in sequence. Determine the optimal hyperspectral imaging chip based on the clinical evaluation results.

[0062] Specifically, for each of the selected single-channel spectral image combinations, a hyperspectral imaging chip is designed and customized, resulting in multiple hyperspectral imaging chips. These chips are then integrated into fundus imaging devices, and the fundus imaging devices integrating the current hyperspectral imaging chips are evaluated for clinical use in turn. Finally, a professional physician determines the hyperspectral imaging chip that has the best effect on recognizing leopard-patterned fundus, thus obtaining the optimal hyperspectral imaging chip.

[0063] The following example will be used to illustrate the design method of the fundus recognition hyperspectral imaging chip of this application.

[0064] First, we collected hyperspectral images of clinically observed leopard-patterned fundus images. The collected image data covered the spectral band from 623nm to 794nm, including multiple spectral points such as 623nm, 650nm, 663nm, 685nm, 709nm, 730nm, 751nm, 773nm, and 794nm. By collecting hyperspectral data within this range, we were able to obtain rich information about fundus images from multiple levels and multiple bands, providing comprehensive data support for subsequent image processing and analysis. The collected hyperspectral image samples were then split into nine single-channel spectral images, corresponding to the 623nm, 650nm, 663nm, 685nm, 709nm, 730nm, 751nm, 773nm, and 794nm wavelengths, respectively. Each single-channel spectral image will be evaluated using multiple metrics to ensure its quality and the validity of the information. The image evaluation metrics include Sobel gradient score, Laplace variance score, and physician subjective evaluation score. These evaluation metrics will be used to comprehensively analyze each single-channel spectral image to assess its practicality and effectiveness in the diagnosis of fundus lesions.

[0065] Then, a neural network model was developed using AI technology. This model was trained and optimized for leopard-spot fundus recognition based on image evaluation results. The constructed training samples were then imported into the neural network model for learning and optimization. Multiple single-channel spectral images corresponding to each patient were grouped together. Within each single group, the multiple single-channel spectral images were randomly recombine. Considering the complementarity of information from different spectral bands, each combination of single-channel spectral images must contain information from at least three different bands, and each combination of spectral bands is allowed to appear a maximum of three times.

[0066] During model training, evaluation metrics are crucial for training effectiveness. The score for each combination comprehensively considers the Sobel gradient score, Laplace variance score, and physician evaluation score, with each of the Sobel gradient score and Laplace variance score having a weight of 0.25, while the physician evaluation score has a weight of 0.5. The neural network model scores each single-channel spectral image within a single single-channel spectral image combination. The sum of the scores for each single-channel spectral image is used as the score for the current single-channel spectral image combination. The single-channel spectral image combinations are then sorted from highest to lowest score, and the top 20 single-channel spectral image combinations are selected to obtain the most effective spectral combinations, thereby improving the accuracy of leopard-spot fundus recognition.

[0067] Furthermore, to improve recognition accuracy, the top three single-channel spectral image combinations were selected from the 20 selected combinations. Based on the neural network model and the scoring results of the neural network model, a corresponding hyperspectral imaging chip was developed. Compared with traditional color images and other hyperspectral chips, it has a stronger ability to recognize leopard-patterned fundus.

[0068] Secondly, embodiments of this application also provide a fundus imaging device.

[0069] In one embodiment, reference is made to Figure 2 , Figure 2 This is a schematic diagram of the functional modules of the fundus imaging device of this application. Figure 2 As shown, the fundus imaging device includes a fundus camera and a hyperspectral imaging chip camera that work together, wherein the hyperspectral imaging chip camera is constructed based on the optimal hyperspectral imaging chip obtained by the design method described above. By using the fundus camera and the hyperspectral imaging chip camera together, fundus imaging of clinical patients can be achieved. In one possible implementation, the fundus camera is a TopCon NW8 fundus camera.

[0070] In this application, the specific process of fundus imaging device recognizing leopard-patterned fundus includes:

[0071] a: Establish a physical connection between the fundus camera and the hyperspectral imaging chip camera, and maintain consistency in the field of view and focus of the fundus camera and the hyperspectral imaging chip camera;

[0072] Furthermore, a physical connection is established between the fundus camera and the hyperspectral imaging chip camera, while maintaining consistency in the field of view and focus of both cameras. Specifically, this includes:

[0073] a1: Establish a physical connection between the fundus camera and the hyperspectral imaging chip camera via an adapter or standard interface;

[0074] a2: Adjust the field of view and focus of the fundus camera and the hyperspectral imaging chip camera to keep them consistent.

[0075] b: Control the fundus camera and hyperspectral imaging chip camera to acquire fundus images, and fuse the fundus images acquired by the fundus camera and hyperspectral imaging chip camera to generate a comprehensive image containing structural features and spectral information, thereby realizing the recognition of leopard-patterned fundus.

[0076] Furthermore, the fundus camera and hyperspectral imaging chip camera are controlled to acquire fundus images, and the fundus images acquired by the fundus camera and hyperspectral imaging chip camera are fused to generate a comprehensive image containing structural features and spectral information, specifically including:

[0077] b1: Synchronous control of the fundus camera and the hyperspectral imaging chip camera is achieved through hardware and software collaboration, so that the hyperspectral imaging chip camera can acquire fundus images of the same area as the fundus camera under different spectral bands.

[0078] b2: By using an image fusion algorithm, fundus images acquired by fundus cameras and hyperspectral imaging chip cameras are fused to generate a comprehensive image containing structural features and spectral information, thereby enabling more accurate analysis of fundus lesions.

[0079] Furthermore, in one embodiment, fundus images acquired by a fundus camera and a hyperspectral imaging chip camera are fused to generate a comprehensive image containing structural features and spectral information, specifically including:

[0080] b201: Acquire fundus images obtained by fundus camera, denoted as the first image, and fundus images obtained by hyperspectral imaging chip camera, denoted as the second image. Perform feature extraction on the first image through convolutional neural network to obtain the preliminary structural feature map and spectral information map corresponding to the first image. Perform feature extraction on the second image through convolutional neural network to obtain the preliminary structural feature map and spectral information map corresponding to the second image.

[0081] b202: For the obtained preliminary structural feature map, the edge intensity distribution is calculated based on the edge detection algorithm to generate an edge enhancement feature map. For the generated edge enhancement feature map, the saliency score of each region is calculated through the saliency evaluation model to generate a saliency distribution map.

[0082] b203: Based on the spectral information map and saliency distribution map corresponding to the first image, calculate the ratio between the saliency score and the preset threshold. If the ratio is greater than 1, adjust the structural feature weight coefficients according to the ratio to generate an initial fusion weight matrix. Based on the spectral information map and saliency distribution map corresponding to the second image, calculate the ratio between the saliency score and the preset threshold. If the ratio is greater than 1, adjust the structural feature weight coefficients according to the ratio to generate an initial fusion weight matrix.

[0083] b204: Using the initial fusion weight matrix corresponding to the first image, the preliminary structural feature map and spectral information map corresponding to the first image are weighted and fused to generate preliminary integrated image data. Using the initial fusion weight matrix corresponding to the second image, the preliminary structural feature map and spectral information map corresponding to the second image are weighted and fused to generate preliminary integrated image data.

[0084] b205: Based on the image fusion algorithm, the preliminary integrated image data corresponding to the first image and the preliminary integrated image data corresponding to the second image are fused to obtain a comprehensive image. Then, a generative adversarial network is applied to optimize the texture region of the comprehensive image to generate a texture-enhanced comprehensive image.

[0085] b206: Extract the fusion quality index from the texture-enhanced composite image. If the fusion quality index is lower than the preset standard, adjust the structural feature weight coefficients in the initial fusion weight matrix according to the deviation ratio between the fusion quality index and the preset standard, re-weight the fusion, and generate the composite image again.

[0086] b207: The edge details and spectral information integrity of the regenerated composite image are detected by multi-scale analysis method, and an integrity assessment report is generated. According to the integrity assessment report, if edge details or spectral information deviations are detected, the pixel values ​​in the deviation area are weighted and corrected by a local adaptive adjustment algorithm, and finally the optimized composite image is generated.

[0087] Furthermore, an integrated operating interface can be built, which, combined with the comprehensive images generated by the fundus imaging device, enables image processing, analysis, and display functions, providing ophthalmologists with intuitive and convenient diagnostic support, thereby improving the clinical efficiency of recognizing leopard-patterned fundus.

[0088] The following example illustrates the practical application effect of the fundus imaging device of this application.

[0089] Clinical trials were conducted, with six patients selected for a double-blind test to identify leopard-patterned fundus images. Four senior medical technicians operated the fundus imaging device described in this application, along with a traditional fundus photography system; four intermediate-level medical technicians operated the same device. Results showed that in the senior-level medical technician group, the traditional fundus photography system and the fundus imaging device described in this application achieved accuracy rates of 95.8% and 100% for identifying leopard-patterned fundus images, respectively. In the intermediate-level medical technician group, the traditional fundus photography system and the fundus imaging device described in this application achieved accuracy rates of 83.3% and 95.8% for identifying leopard-patterned fundus images, respectively. This demonstrates that the fundus imaging device described in this application significantly improves the accuracy of ordinary experts in identifying leopard-patterned fundus images.

[0090] Thirdly, embodiments of this application provide a fundus recognition hyperspectral imaging chip design device, which can be a personal computer (PC), laptop computer, server or other device with data processing capabilities.

[0091] Reference Figure 3 , Figure 3 This is a schematic diagram of the hardware structure of the fundus recognition hyperspectral imaging chip design device involved in the embodiments of this application. In the embodiments of this application, the fundus recognition hyperspectral imaging chip design device may include a processor, a memory, a communication interface, and a communication bus.

[0092] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0093] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting internal components of the fundus recognition hyperspectral imaging chip design device, as well as interfaces used for interconnecting the fundus recognition hyperspectral imaging chip design device with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0094] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0095] The processor can be a general-purpose processor, which can call the fundus recognition hyperspectral imaging chip design program stored in the memory and execute the fundus recognition hyperspectral imaging chip design method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the fundus recognition hyperspectral imaging chip design program is called can be referred to in the various embodiments of the fundus recognition hyperspectral imaging chip design method of this application, and will not be repeated here.

[0096] Those skilled in the art will understand that Figure 3 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0097] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0098] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0099] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0100] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0102] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A fundus recognition hyperspectral imaging chip design method, characterized in that, The fundus recognition hyperspectral imaging chip design method comprises: Collecting a clinical hyperspectral leopard-shaped fundus hyperspectral image sample under a set waveband range, and performing waveband splitting on the hyperspectral image sample to obtain a single-channel spectral image; An image score of the single-channel spectral image is formed to form a training sample to train the neural network model, specifically comprising: Based on the Sobel gradient, Laplacian variance and subjective scoring method, the current single-channel spectral image is scored, and the scores of each scoring method are weighted to obtain the score of the current single-channel spectral image; wherein the weight of the Sobel gradient score and the Laplacian variance score is 0.25, and the weight of the doctor's subjective evaluation score is 0.5; Based on each single-channel spectral image and the score of each single-channel spectral image, a training sample is constructed to train the neural network model; Randomly combining a plurality of single-channel spectral images corresponding to each patient to obtain a plurality of single-channel spectral image combinations; Based on the trained neural network model, each single-channel spectral image in each single-channel spectral image combination is scored, and a set number of single-channel spectral image combinations with high scores are selected according to the score results, specifically comprising: Based on the trained neural network model, each single-channel spectral image in the single-channel spectral image combination is scored, and the sum of the scores of each single-channel spectral image is taken as the score of the current single-channel spectral image combination; According to the scores of each single-channel spectral image combination, the single-channel spectral image combinations are sorted from high to low, and a set number of single-channel spectral image combinations with high scores are selected; Each single-channel spectral image combination selected is constructed into a corresponding hyperspectral imaging chip, and integrated into a fundus imaging device for clinical evaluation to obtain the best hyperspectral imaging chip.

2. The fundus recognition hyperspectral imaging chip design method of claim 1, wherein, The collecting of the clinical hyperspectral leopard-shaped fundus hyperspectral image sample under a set waveband range, and the waveband splitting of the hyperspectral image sample to obtain a single-channel spectral image, specifically comprises: Collecting a clinical hyperspectral leopard-shaped fundus hyperspectral image sample, the hyperspectral image sample being imaging data of each specific waveband under a set waveband range; The collected hyperspectral image sample is waveband split to obtain a single-channel spectral image of a specific waveband.

3. The fundus recognition hyperspectral imaging chip design method of claim 1, wherein, The construction of the corresponding hyperspectral imaging chip for each single-channel spectral image combination selected, and the integration into a fundus imaging device for clinical evaluation to obtain the best hyperspectral imaging chip, specifically comprises: Each single-channel spectral image combination selected is constructed into a corresponding hyperspectral imaging chip; The hyperspectral imaging chips are integrated into a fundus imaging device respectively and sequentially evaluated clinically for leopard-shaped fundus recognition, and the best hyperspectral imaging chip is determined according to the clinical evaluation results.

4. An ocular fundus imaging apparatus for leopard-spot fundus identification, characterized by: The fundus imaging device comprises a fundus camera and a hyperspectral imaging chip camera that work cooperatively, wherein the hyperspectral imaging chip camera is constructed based on the best hyperspectral imaging chip obtained by the design method of any one of claims 1 to 3.

5. An ocular fundus imaging apparatus as claimed in claim 4, characterized in that The specific process of the fundus imaging device for leopard-shaped fundus recognition comprises: The physical connection between the fundus camera and the hyperspectral imaging chip camera is established, and the field of view range and focus of the fundus camera and the hyperspectral imaging chip camera are kept consistent; The fundus camera and the hyperspectral imaging chip camera are controlled to collect fundus images, and the fundus images collected by the fundus camera and the hyperspectral imaging chip camera are fused to generate a comprehensive image containing structural features and spectral information, thereby realizing the recognition of the leopard-shaped fundus.

6. An ocular fundus imaging apparatus as claimed in claim 5, characterized in that The physical connection between the fundus camera and the hyperspectral imaging chip camera is established, and the field of view range and focus of the fundus camera and the hyperspectral imaging chip camera are kept consistent, specifically including: The physical connection between the fundus camera and the hyperspectral imaging chip camera is established through an adapter or a standard interface; The field of view range and focus of the fundus camera and the hyperspectral imaging chip camera are adjusted to keep the field of view range and focus of the fundus camera and the hyperspectral imaging chip camera consistent.

7. An apparatus for imaging the fundus as defined in claim 5, wherein The fundus camera and the hyperspectral imaging chip camera are controlled to collect fundus images, and the fundus images collected by the fundus camera and the hyperspectral imaging chip camera are fused to generate a comprehensive image containing structural features and spectral information, specifically including: The synchronization control of the fundus camera and the hyperspectral imaging chip camera is realized based on software and hardware cooperation, so that the hyperspectral imaging chip camera collects fundus images of the same area as the fundus camera under different spectral bands; The fundus images collected by the fundus camera and the hyperspectral imaging chip camera are fused through an image fusion algorithm to generate a comprehensive image containing structural features and spectral information.

8. An apparatus for fundus recognition hyperspectral imaging chip design, characterized in that, The fundus recognition hyperspectral imaging chip design device includes a processor, a memory, and a fundus recognition hyperspectral imaging chip design program stored on the memory and executable by the processor, wherein when the fundus recognition hyperspectral imaging chip design program is executed by the processor, the steps of the fundus recognition hyperspectral imaging chip design method according to any one of claims 1 to 3 are realized.

Images