Method and device for assessing refractive development of neural circuit based on visual development

Abstract

This invention provides a method and device for refractive development assessment based on visual development prediction neural circuits, relating to the field of myopia management technology. The method includes: acquiring color retinal optical imaging signals and refractive ground state quantification values ​​of the subject; performing retinal stabilization preprocessing; extracting multi-level morphological features through a reentrant layered retinal choroid encoder; performing spatial feature aggregation gating on the extracted feature maps to obtain feature vectors; concatenating and integrating the feature vectors with the refractive ground state quantification values ​​to obtain fused feature vectors; constructing a fused feature vector sequence and inputting it into a refractive development dynamics internal model inferrer to generate a future multi-time-point refractive state prediction sequence; converting it into an interpretable refractive state numerical sequence; and providing risk warnings based on preset risk thresholds. This invention explicitly embeds physiological mechanisms prior into the model structure and parameter adaptive strategies, thereby maintaining the stability and interpretability of predictions even in a single medical visit scenario.

Description

Technical Field

[0001] This invention relates to the field of myopia management technology, and in particular to a method and device for assessing refractive development based on visual development prediction neural circuits. Background Technology

[0002] Myopia, especially high myopia, has become a global public health problem. Once high myopia develops, its progression is irreversible; therefore, early intervention remains the most effective prevention strategy.

[0003] The onset of myopia in childhood is closely related to the development of high myopia later in life. Studies have shown that drug intervention can reduce the progression of myopia in children aged 6-13 by more than 70%, but the lack of predictive ability for individual myopia trajectories hinders timely treatment. Although experienced ophthalmologists can identify the risk of high myopia through eye examinations, this assessment is subjective and requires extensive expertise.

[0004] Currently, myopia prediction faces three main challenges: (1) the lack of ability to accurately predict the trajectory of individual myopia development affects the implementation of timely intervention; (2) existing assessment methods are highly subjective and require a high level of professional experience; and (3) the global prevalence of myopia continues to rise, requiring efficient and scalable prediction methods.

[0005] Existing myopia prediction methods include: statistical models based on refractive parameters, with a representative study being the CLEERE (Collaborative Longitudinal Evaluation of Ethnicity and Refractive Error) prediction model developed by Zadnik et al. (Zadnik K, et al. JAMA Ophthalmol. 2015;133(6):683-9). Its technical principle involves establishing a statistical regression model using traditional factors such as spherical equivalent power, age, and parental myopia history. Multidimensional data machine learning models are also used, with a representative study being the electronic medical record data machine learning study by Lin et al. (Lin H, et al. PLoS Med. 2018;15(11):e1002674). Its technical principle involves integrating refractive data from multiple visits and using machine learning algorithms such as random forests and support vector machines. Early deep learning methods based on fundus images are represented by the Singapore children study by Foo et al. (Foo LL, et al. NPJ Digit Med. 2023;6(1):10). The technical principle is to combine fundus images and baseline refractive data and use convolutional neural networks for binary classification. However, this study has limitations; it can only perform classification prediction and cannot provide quantitative prediction of refractive error progression.

[0006] Existing technologies adopt a biomimetic-inspired approach. Specifically, existing myopia / refractive development prediction methods mostly use fundus images as general visual input and directly fit classification labels or single-time-point results through statistical regression or end-to-end deep networks, lacking explicit mechanism modeling for the biological process of "visual development".

[0007] Meanwhile, existing technologies have the following drawbacks: 1. High data requirements and limited practicality: Existing machine learning models require a large amount of data accessed multiple times to achieve good predictive performance, which is difficult to achieve in resource-constrained environments, especially in remote areas where complete longitudinal data is hard to obtain. 2. Limited predictive function and clinical guidance value: Existing deep learning methods mainly focus on binary classification problems, and can only predict whether high myopia will occur, but cannot provide quantitative refractive error progression trajectories, and cannot provide accurate numerical prediction guidance for clinicians. 3. Insufficient temporal prediction capability: Existing methods are often limited to predictions at specific points in time (such as whether myopia will occur one year later), lacking the ability to predict continuously over multiple years, and cannot support long-term treatment planning. 4. Insufficient utilization of medical information: Traditional myopia prediction relies on doctors' experience and cannot fully explore potential biomarkers in medical information, thus limiting the accuracy of myopia prediction. 5. Poor cross-population applicability: Existing models are mostly developed based on single populations, lacking validation across different geographical regions, climate types, urban-rural distributions, and educational burdens, limiting their promotion and application in different populations. Summary of the Invention

[0008] To address the technical problems of existing technologies, such as high data requirements, limited practicality, single predictive function, limited clinical guidance value, insufficient time prediction ability, inadequate utilization of medical information, and poor cross-population applicability, this invention provides a method and device for refractive development assessment based on visual development prediction neural circuits. The objectives are: (1) to develop a prediction system with minimal data requirements: accurate prediction can be achieved using only retinal optical imaging signals and refractive ground state quantification values ​​from a single visit, suitable for resource-constrained environments; (2) to achieve quantitative multifunctional prediction: simultaneously providing myopia risk classification prediction and accurate refractive error prediction, providing comprehensive support for clinical decision-making; (3) to construct the ability to extrapolate the dynamic trajectory of visual development, supporting long-term treatment planning; (4) to maximize the extraction of retinal optical imaging signal information: automatically identifying subtle features and biomarkers related to myopia progression through deep learning technology; (5) to enhance cross-population applicability: developing a model with good cross-population generalization and validation capabilities, applicable to populations of different races and regions. The technical solution is as follows: On the one hand, a refractive development assessment method based on visual development prediction neural circuits is provided. This method is implemented using a pediatric refractive development assessment device and includes: S1. For adolescents and children, collect color retinal optical imaging signals and refractive ground state quantification values ​​for each subject.

[0009] S2. Perform retinal stabilization preprocessing on the acquired color retinal optical imaging signals to obtain preprocessed retinal optical imaging signals; wherein, retinal stabilization preprocessing includes retinal stabilization processing, retinal lateral inhibition enhancement, and receptive field tuning during developmental stages.

[0010] S3. For the preprocessed retinal optical imaging signal, multi-level morphological features are extracted by a reentry-type layered retinal choroid encoder to obtain a feature map.

[0011] S4. Perform a spatial feature aggregation gate operation on the feature map to obtain a feature vector; cascade the feature vector with the corresponding refractive ground state quantization value to obtain a fused feature vector.

[0012] S5. Obtain the fused feature vectors from multiple time points, form a fused feature vector sequence, input it into the refractive development dynamics internal model extrapolator, and generate a future multi-time point refractive state prediction sequence; convert the refractive state prediction sequence into an interpretable refractive state numerical sequence.

[0013] S6. Risk warning is given based on the numerical sequence of refractive state and the preset risk threshold.

[0014] Optionally, the retinal stabilization processing in S2 includes: preserving and enhancing the anatomical structures in the retinal optical imaging signal, simulating optical distortion, and synthesizing pathological enhancement.

[0015] Among them, the preservation and enhancement of anatomical structures in retinal optical imaging signals includes: rotational enhancement with the center of the optic disc in the anatomical structure as the origin, while maintaining the relative positional relationship between the optic disc and the macula; vertical flipping of the anatomical structure; and maintaining the overall morphology of the main vascular arches.

[0016] Optical distortion simulation includes: adding radial gradient blur to simulate different degrees of refractive media opacity; simulating flash illumination of different intensities and angles; and simulating the differences in color response curves of different brands of fundus cameras.

[0017] Synthetic pathological enhancements include: adding controllable choroidal texture to retinal optical imaging signals; simulating different degrees of optic disc tilt through affine transformations; and using elastic deformation to simulate vascular morphological changes in high myopia.

[0018] Optionally, retinal-side inhibition enhancement in S2 includes: For retinal vessels in anatomical structures, spatial frequency filters are used to enhance the information of vessel edges.

[0019] For the optic disc in the anatomical structure, the edge detection operator combined with the spatial frequency filter is used to enhance the pixel gradient of the optic disc boundary.

[0020] For the macula in the anatomical structure, a spatial frequency filter is used to enhance high-frequency components and suppress low-frequency background noise.

[0021] Optionally, developmental receptive field tuning in S2 includes: The spatial frequency response parameters of the spatial frequency filter during the retinal side suppression enhancement process are dynamically adjusted according to different age groups.

[0022] Optionally, the reentrant hierarchical retinal choroid encoder in S3 includes: multiple serially connected feature abstraction levels, with parallel bypass connections between adjacent feature abstraction levels, each feature abstraction level equipped with an independent dual attention gating module, and employing a top-down feature propagation path.

[0023] Optionally, the spatial feature aggregation gate operation in S4 adopts a full spatial domain averaging strategy, averaging all pixel values ​​of each feature map to obtain a one-dimensional feature vector.

[0024] Optionally, the method further includes: The risk warning process is optimized using the cross-entropy loss function, and the process of converting the refractive state prediction sequence into an interpretable refractive state numerical sequence is optimized using the mean square error loss function. The two optimization processes are then jointly optimized using a weighted summation method.

[0025] On the other hand, a refractive development assessment device based on visual development prediction neural circuits is provided. This device is applied to a refractive development assessment method based on visual development prediction neural circuits. The device includes: The retinal optical imaging signal acquisition module is used to acquire color retinal optical imaging signals and refractive ground state quantification values ​​for each subject, specifically for adolescents and children.

[0026] The retinal stabilization preprocessing module is used to perform retinal stabilization preprocessing on the acquired color retinal optical imaging signals to obtain preprocessed retinal optical imaging signals; wherein, retinal stabilization preprocessing includes retinal stabilization processing, retinal lateral inhibition enhancement, and receptive field tuning during developmental stages.

[0027] The reentrant layered retinal choroid encoder module is used to extract multi-level morphological features from the preprocessed retinal optical imaging signal to obtain a feature map.

[0028] The multisensory integration kernel module is used to perform spatial feature aggregation gate operations on the feature map to obtain feature vectors; the feature vectors are then cascaded and integrated with the corresponding refractive ground state quantization values ​​to obtain fused feature vectors.

[0029] The refractive development dynamics internal model inferrer module is used to obtain fused feature vectors from multiple time points, form a fused feature vector sequence, input it into the refractive development dynamics internal model inferrer, generate a future multi-time point refractive state prediction sequence; and convert the refractive state prediction sequence into an interpretable refractive state numerical sequence.

[0030] The threshold warning reflection module is used to provide risk warnings based on the refractive state numerical sequence and preset risk thresholds.

[0031] On the other hand, a child refractive development assessment device is provided, the child refractive development assessment device comprising: a processor; a memory, the memory storing computer-readable instructions, which, when executed by the processor, implement any of the above-described methods for refractive development assessment based on visual development prediction neural circuits.

[0032] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement any of the above-described methods for refractive development assessment based on visual development prediction neural circuits.

[0033] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this invention, compared to traditional methods that require data from multiple medical visits, accurate prediction can be achieved using only retinal optical imaging signals and refractive ground state quantification values ​​from a single visit. This design significantly reduces data collection costs, enabling the technology to be widely applied in resource-constrained environments, and is particularly suitable for myopia screening needs in remote areas.

[0034] This invention simultaneously realizes threshold warning reflex (myopia risk assessment) and developmental trajectory reading (specific refractive error value), providing clinicians with more comprehensive decision support. Compared with existing methods that can only perform binary classification, the quantitative prediction function of this invention (average error 0.25D / year) can provide precise numerical guidance for the formulation of personalized treatment plans.

[0035] This invention achieves an accuracy rate of 95% in predicting myopia risk and 99% in predicting the risk of high myopia, significantly outperforming existing technologies. Furthermore, its ability to extrapolate the dynamic trajectory of visual development provides support for long-term treatment planning, a capability not found in existing technologies.

[0036] This invention constructs a database containing statistics on the prevalence and progression rate of myopia in different regions. After cross-group generalization validation across various geographical regions, climate types, urban-rural distributions, and education burden indices, its excellent generalization ability has been demonstrated. This cross-population applicability allows the technology to be promoted and applied globally, without geographical or racial limitations.

[0037] By explicitly embedding biomimetic mechanisms such as visual homeostasis regulation, lateral inhibition, developmental stage tuning, and reentrant hierarchical coding into the processing chain, this invention is less sensitive to domain shifts when faced with different imaging conditions of different devices, different age structures, and cross-regional population differences. At the same time, vigilance reflexes and attentional retrospection mapping directly align the prediction with clinical decision thresholds, making the model output easier for doctors to interpret, review, and adopt. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of a refractive development assessment method based on visual development prediction neural circuits provided in an embodiment of the present invention; Figure 2 This is a block diagram of a refractive development assessment device based on a visual development prediction neural circuit provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a child refractive development assessment device provided in an embodiment of the present invention. Detailed Implementation

[0040] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0041] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0042] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0043] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0044] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0045] This invention provides a method for assessing refractive development based on visual developmental prediction neural circuits. This method can be implemented using a pediatric refractive developmental assessment device, which can be a terminal or a server. Figure 1 The flowchart shown is for a refractive development assessment method based on visual developmental prediction neural circuits. The processing flow of this method may include the following steps: S1. For adolescents and children, collect color retinal optical imaging signals and refractive ground state quantification values ​​for each subject.

[0046] In one feasible implementation, the present invention is mainly applied to the following products and fields: 1. Intelligent screening system for children's refractive development in ophthalmic medical devices; 2. School health examination equipment (supporting risk classification with uncertainty quantification); 3. Remote medical diagnosis platform (supporting continuous monitoring and trajectory correction); 4. Ophthalmic artificial intelligence-assisted diagnostic software; 5. Eye health screening equipment in community medical centers (supporting regional threshold adaptation).

[0047] In this invention, the retinal optical imaging signal is defined as: the optical image signal of the posterior retina of the eyeball acquired by a fundus camera, including anatomical information such as the optic disc, macula, and retinal vessels.

[0048] The refractive ground state quantification value is defined as: a quantitative parameter characterizing the current optical focusing state of the subject's eyeball, that is, the equivalent spherical power (spherical power plus half of the cylindrical power).

[0049] Specifically, this invention acquires color retinal optical imaging signals using a non-mydriatic fundus camera or equivalent equipment. The signal acquisition targets are school-aged children and adolescents aged 3 to 16, covering key developmental stages of school age. Specific age groups include: 3-8 years old: Critical period for early myopia risk screening (corresponding to the receptive field tuning parameter group in the early developmental stage); 9-11 years old: Rapid development period of myopia (corresponding to the receptive field tuning parameter group in the middle-age development stage); 12-16 years old: the initial stage of high myopia risk (corresponding to the receptive field tuning parameter group of the advanced developmental stage).

[0050] Each subject had color retinal optical imaging signals collected from one or both eyes in JPEG or DICOM format.

[0051] Furthermore, the equipment for detecting the refractive ground state quantification value includes: Refractometer: An automatic refractometer is used to perform cycloplegic refraction and measure the spherical power (S), cylindrical power (C), and axis. Mydriatic method: Standardized mydriasis is performed using 1% cyclopentolate or tropicamide to ensure the accuracy of the refractive ground state quantification value.

[0052] In addition to color retinal optical imaging signals and refractive ground state quantification values, OCT (Optical Coherence Tomography) signals can be added as supplementary inputs, or combined with anterior segment imaging signals for comprehensive analysis, or integrated with additional biometric parameters such as axial length and corneal curvature.

[0053] S2. Perform retinal stabilization preprocessing on the acquired color retinal optical imaging signal to obtain the preprocessed retinal optical imaging signal.

[0054] In one feasible implementation, the retinal stabilization preprocessing of the present invention is defined as: a module for standardizing the input signal, including brightness normalization, noise suppression and contrast equalization, so that the input reaches a stable and consistent physiological ground state, improves signal quality, enhances biological features related to myopia progression, and provides standardized input for subsequent deep learning models.

[0055] Specifically, retinal homeostasis pretreatment includes retinal homeostasis treatment, retinal lateral inhibition enhancement, and receptive field tuning during development.

[0056] Among them, retinal stabilization treatment includes: (1) Enhanced anatomical structure preservation: Rotation enhancement is performed with the center of the optic disc as the origin, while maintaining the relative positional relationship between the optic disc and the macula; Horizontal flipping is prohibited (since the left and right eye discs are in opposite positions, flipping will cause confusion); vertical flipping is permitted only. Protection of vascular course: Enhanced procedures do not alter the overall morphology of the major vascular arches.

[0057] (2) Simulation of optical distortion: Lens optics simulation: Add radial gradient blur to simulate different degrees of refractive media opacity; Changes in lighting conditions: Simulating flash lighting at different intensities and angles; Device characteristic simulation: Simulate the differences in color response curves of fundus cameras from different brands.

[0058] (3) Synthetic pathological enhancement: Leopard-print fundus simulation: Adding controlled choroidal texture to fundus images; Visual disc tilt simulation: Different degrees of visual disc tilt are simulated through affine transformation; Vascular tortuosity simulation: using elastic deformation to simulate the morphological changes of blood vessels in high myopia.

[0059] Furthermore, retinal-side inhibition enhancement is defined as: a processing module that enhances the edge response of structures such as blood vessel contours and optic disc boundaries by borrowing from the retinal ganglion cell-side inhibition mechanism and using spatial frequency filtering.

[0060] This invention has made targeted optimizations to the retinal lateral inhibition enhancement process, drawing on the principle of lateral inhibition mechanism of retinal ganglion cells, aiming to highlight fundus morphological features closely related to the progression of myopia in children, specifically including: Enhanced anatomical regions: Using non-mydriatic color retinal optical imaging signals, the focus is on enhancing three key regions: the optic disc, the macula, and the retinal vascular network.

[0061] Enhanced details and their morphological significance: Retinal vessels: Features include vessel angle, branching pattern, diameter changes, and vessel tortuosity; Morphological significance: during myopia progression, retinal vessels may be stretched and twisted, and changes in vessel density are early biomarkers; Enhancement method: Butterworth spatial frequency filter is used to enhance vessel edge information and highlight the contrast between vessels and background. Optic disc boundary: Features include optic disc outline clarity, tilt, and cup-to-disc ratio; Morphological significance: high myopia is often accompanied by optic disc tilt or rotation, and blurred boundaries may be related to axial myopia; Enhancement method uses the Sobel edge detection operator combined with spatial frequency filtering to enhance the pixel gradient of the optic disc boundary; Cellular structure of the macular region: Features include foveal reflection, peripheral pigment distribution, and retinal hierarchical structure; Morphological significance: the macular region is a high-incidence area of ​​myopic macular degeneration, and its subtle structural changes can predict later complications; Enhancement method uses frequency domain filtering to enhance high-frequency components (such as cellular hierarchical texture) and suppress low-frequency background noise.

[0062] Furthermore, developmental receptive field tuning is defined as: an adaptive mechanism that dynamically adjusts spatial frequency response parameters according to the subject's visual developmental stage to simulate the developmental changes in the receptive field scale of the visual system.

[0063] This invention proposes a developmental stage-based receptive field tuning strategy based on age groups, which dynamically adjusts the spatial frequency response parameters of retinal lateral inhibition enhancement according to the developmental characteristics of fundus structures in children of different ages: In the early developmental stage group (3-8 years old), the fundus characteristics are finer blood vessels, weaker macular reflection, and lower signal contrast. The receptive field tuning parameters are set to the first spatial frequency cutoff parameter (a relatively high value, such as 0.15Hz normalized frequency) to enhance the fine blood vessels and macular texture. In the middle-age developmental stage group (9-11 years old), the fundus characteristics are that the vascular network tends to be stable and the optic disc boundary is clear. The receptive field tuning parameter adopts the second spatial frequency cutoff parameter (moderate value, such as 0.10Hz normalized frequency) to balance detail enhancement and noise control. In the advanced developmental stage group (12-16 years old), the fundus structure is similar to that of adults, and early pathological changes may occur. The receptive field tuning parameter is set to the third spatial frequency cutoff parameter (lower value, such as 0.05Hz normalized frequency) to highlight the overall structural changes and regional abnormalities.

[0064] In practice, the system automatically selects a preset parameter set based on the patient's age and metadata, and fine-tunes it using the signal grayscale histogram to ensure consistent enhancement. This tuning mechanism simulates the natural variation of the receptive field scale in the visual system at different developmental stages.

[0065] S3. For the preprocessed retinal optical imaging signal, multi-level morphological features are extracted by a reentry-type layered retinal choroid encoder to obtain a feature map.

[0066] In one feasible implementation, the visual development prediction neural circuit system proposed in this invention employs a perception-prediction dual-pathway architecture of a reentrant hierarchical retinal choroid encoder and an internal model inferrer for refractive developmental dynamics. This simulates the hierarchical processing mechanism of the visual nervous system, achieving a complete information processing pathway from signal acquisition to developmental trajectory prediction, capable of simultaneously processing visual features and temporal series information. The encoder in the perception pathway is a multi-level feature abstraction encoding module, from retinal surface texture to deep choroidal structure, and uses a parallel bypass integration mechanism to achieve cross-level feature transfer.

[0067] Specifically, the encoder is based on a multi-scale visual feature pyramid fusion mechanism, including: 1. Cross-scale feature pyramid construction employs multi-level feature abstraction, containing multiple cascaded feature abstraction levels to achieve a progressive abstraction of "retinal surface texture features → local morphological features → global structural features": A four-level feature pyramid (P2-P5) is constructed in the reentrant hierarchical retinal choroid encoder, corresponding to feature maps at different spatial resolutions. The size relationship of each level of feature map is as follows: P2 (1 / 4 input size), P3 (1 / 8 input size), P4 (1 / 16 input size), P5 (1 / 32 input size).

[0068] A parallel bypass integration mechanism is adopted: parallel bypass connections (i.e. cross-level feature transfer connections) are set between adjacent layers to directly transfer shallow feature representations to deep layers for fusion, realize cross-level steady-state feature transfer, and prevent the loss of micro-texture information during deep abstraction.

[0069] 2. Channel-Spatial Dual Attention Gating Module: Each pyramid level is equipped with an independent dual attention gating module, including: Channel Attention Branch: Channel statistics are extracted in parallel using global average pooling and global max pooling, and channel weight vectors are generated via a shared multilayer perceptron. This enables adaptive weighting for different feature channels. The spatial attention branch performs average pooling and max pooling along the channel dimension, concatenates the results, and then passes them through a convolutional layer to generate a spatial weight map. This enables adaptive attention to different spatial locations. Gated fusion: final feature This enables joint attention modulation of the channel and spatial dimensions.

[0070] 3. Top-down Path Aggregation: A top-down feature propagation path is adopted to pass high-level semantic information to lower levels step by step: Upsampling Module: Performs 2x bilinear interpolation upsampling on the high-level feature maps. Lateral Connection: Adjusts the number of channels of the corresponding layer's low-level features through 1x1 convolution. Feature Fusion: Adds the upsampled features to the lateral connection features, and then eliminates aliasing effects through 3x3 convolution.

[0071] 4. Multi-scale feature aggregation output: Spatial feature aggregation gate operation is performed on the fused features at each level to obtain feature vectors at four different scales. , , , Weighted fusion is performed using learnable weights: The weighting coefficients are automatically learned through backpropagation.

[0072] The encoder processes each input retinal optical imaging signal sequentially. After processing through multiple convolutional layers, pooling layers, and residual blocks, the network can automatically learn and extract rich spatial features related to myopia progression, such as optic disc tilt, the size of the peripapillary atrophy arc, the course and branching angle of blood vessels, and the morphology of the macula. The final output high-dimensional feature vector of fundus morphology possesses both sensitivity to micro-texture and macro-structural representation capabilities.

[0073] Meanwhile, for reentrant hierarchical retinal choroid encoders, deeper residual connection structures can be used to replace the current architecture, or other advanced convolutional network architectures such as dense connection networks or high-performance networks can be adopted, or a visual encoder structure based on self-attention mechanism can be used.

[0074] S4. Perform a spatial feature aggregation gate operation on the feature map to obtain a feature vector; cascade the feature vector with the corresponding refractive ground state quantization value to obtain a fused feature vector.

[0075] In one feasible implementation, in order to combine visual information with clinical numerical information and prepare for subsequent time series analysis, the system performs key feature fusion steps through a multisensory integration kernel.

[0076] Among them, the spatial feature aggregation gate is defined as: a module that performs aggregation operations on high-dimensional spatial feature maps, compressing the feature responses of two-dimensional spatial distribution into one-dimensional feature vector representations.

[0077] Specifically, a spatial feature aggregation gate operation is applied to the feature maps output by the reentrant hierarchical retinal choroid encoder. This aggregation operation employs a full spatial domain averaging strategy, averaging all pixel values ​​of each feature map and compressing it into a single value. In this way, a set of high-dimensional spatial feature maps is transformed into a one-dimensional feature vector. This vector can be viewed as a highly condensed "fingerprint" of all the key spatial features of the retinal optical imaging signal.

[0078] The multisensory integration kernel is defined as: a module that realizes cross-modal fusion of visual morphological features and refractive values ​​to form a unified multimodal state representation vector.

[0079] Specifically, the feature vector generated from the visual signal is concatenated and integrated with the quantized value of the refractive ground state at that time point. For example, if the length of the visual feature vector is 512, the length of the integrated multimodal fusion feature vector is 513.

[0080] This fusion step is the essence of the invention's architecture. Instead of processing visual features and refractive data as two separate streams, it fuses them into a unified, multimodal state representation before time-series modeling. The advantage of this is that when the subsequent refractive development dynamics internal model inferrer learns the temporal evolution, each time step's input simultaneously contains complete information on "how the eye looks" (visual features) and "what the eye power is" (refractive ground state quantization value).

[0081] Meanwhile, the internal model extrapolator of refractive development dynamics can also use simplified gated recurrent units to replace the current gated working memory structure, or adopt a sequence modeling architecture based on self-attention mechanism; or use temporal convolutional networks for time-dependent modeling.

[0082] S5. Obtain the fused feature vectors from multiple time points, form a fused feature vector sequence, input it into the refractive development dynamics internal model extrapolator, and generate a future multi-time point refractive state prediction sequence; convert the refractive state prediction sequence into an interpretable refractive state numerical sequence.

[0083] In one feasible implementation, the refractive development kinetics internal model inferrer of the present invention is defined as: an inferrer module that employs a gated working memory prediction loop structure to progressively generate a prediction sequence of future refractive states at multiple time points based on fused features. The sequence composed of fused feature vectors from multiple time points is passed as input to the refractive development kinetics internal model inferrer. This inferrer, employing a gated working memory prediction loop structure, can effectively capture long-term dependencies in the sequence data.

[0084] Specifically, the structural features of the inferr include: Forgetting Gating Unit: Used for selectively forgetting historical information that is irrelevant to the current prediction; Input gating unit: used to selectively integrate new information currently being input; Output gating unit: used to control the state information passed to the next time step; Each gating unit works in concert to achieve long-range dependency modeling of refractive development time series.

[0085] The inferrer internally constructs a dynamic model of refractive development, capable of capturing the intrinsic laws governing the evolution of refractive states over time, including nonlinear changes in developmental rates and transitional characteristics at different developmental stages. Through processing the entire input sequence, the inferrer can learn the inherent dynamic patterns and evolutionary laws of myopia development in children. It performs visual developmental dynamic trajectory extrapolation, a modeling process that predicts the refractive evolution trend at multiple consecutive time points based on the current state.

[0086] Furthermore, the developmental trajectory readout is the output module that converts the internal state sequence of the deducer into an interpretable numerical sequence of refractive states. The final step in the extrapolator's processing is generating predictions. The output of the last time step of the extrapolator within the refractive developmental dynamics model, after a transformation layer, ultimately generates a quantitative prediction of refractive status for one or more future years. This output is not a single risk score or classification label, but a specific sequence of future refractive power values. For example, if the input is current data, the model can output predicted refractive status values ​​for each of the next five years. This sequence constitutes a personalized, quantitative dynamic trajectory of visual development for the child, providing unprecedentedly accurate information for clinical decision-making.

[0087] The core technical solution of this invention uses the steady-state regulation, lateral inhibition, developmental plasticity and reentry loop of the visual system as biomimetic constraints, and decouples and connects "morphological cue extraction" (perceptual pathway) and "refractive dynamics deduction" (predictive pathway) in layers, avoiding the generalization instability and interpretation difficulties caused by relying solely on large-sample end-to-end fitting.

[0088] Central-peripheral receptive field: a receptive field structure that mimics the central-peripheral antagonism of retinal ganglion cells, used to provide the physiological basis and parameter design for supporting side inhibition enhancement.

[0089] Visual steady state and light adaptation: Simulate the light adaptation and contrast gain control process of the visual system to achieve a consistent steady-state baseline for input signals under different devices, exposure and imaging conditions.

[0090] Predictive coding error signal: The difference between the observed refractive state and the inference result of the internal model is used as a correction signal to constrain the temporal consistency of the developmental trajectory inference and improve robustness to data noise and domain shift.

[0091] Developmental plasticity parameter library: a set of pre-set plasticity window parameters for different age groups, used to drive receptive field tuning and staged adaptation of key hyperparameters of the model.

[0092] S6. Risk warning is given based on the numerical sequence of refractive state and the preset risk threshold.

[0093] In one feasible implementation, the threshold warning reflex of the present invention is defined as: a mechanism that automatically triggers a risk warning signal when the predicted refractive state reaches a preset risk threshold.

[0094] Specifically, based on the prediction results read from the developmental trajectory, a risk warning signal is automatically triggered when the predicted refractive state reaches or exceeds a preset risk threshold. This module sets up a multi-level threshold system, including: Prediction threshold for refractive errors (e.g., predicted refractive error of -0.50D); Moderate risk threshold (e.g., predicted refractive error of -3.00D); Severe risk threshold (e.g., predicted refractive error of -6.00D).

[0095] This invention employs a parameter adaptive tuning strategy, including: Data partitioning: Training set: 70% of the data is used for model parameter learning; Validation set: 15% of the data is used for hyperparameter tuning; Test set: 15% of the data is used for final performance evaluation.

[0096] Dual-objective optimization constraint: The system's parameter adaptive tuning process employs a weighted loss function design that simultaneously optimizes the threshold warning reflex (classification) and developmental trajectory readout (regression) tasks, including: The first optimization objective (classification optimization) is to optimize the classification accuracy of the threshold warning reflex module, using the cross-entropy loss function. The second optimization objective (regression optimization) is used to optimize the numerical prediction accuracy of the developmental trajectory readout module, using the mean squared error loss function. The two objectives are jointly optimized using a weighted summation method: The weighting coefficients are dynamically adjusted according to the prediction time span.

[0097] This invention employs attention retrospective mapping: a mechanism that automatically triggers a risk warning signal when the predicted refractive state reaches a preset risk threshold, providing a visual analysis method for clinically interpretable prediction results.

[0098] Furthermore, the data storage system of the present invention adopts a distributed storage architecture, which supports the storage and rapid retrieval of massive retinal optical imaging signals and corresponding refractive data, and complies with the HIPAA medical data privacy standard.

[0099] This invention also includes a user interaction terminal. The doctor's workstation is equipped with a visual interface that displays developmental trajectory readout curves, threshold alert reflex warning information, attention retrospective mapping heatmaps, refractive state change curves, etc., in the form of charts and graphs to support clinical decision-making.

[0100] This invention can be implemented using the programming language Python 3.12.0, employing the deep learning frameworks PyTorch or TensorFlow, with GPU support of NVIDIA A5000-24G or equivalent graphics cards, and a development platform of Linux / Windows. It complies with FDA medical device software guidance principles, follows the ISO 13485 medical device quality management system, meets HIPAA patient privacy protection requirements, and has a retinal optical imaging signal resolution of ≥512×512 pixels.

[0101] This invention uses the hierarchical processing and developmental plasticity of the visual system as biomimetic priors. It processes information according to the neural circuit links of “input stabilization—local contrast enhancement—developmental stage tuning—re-entry hierarchical coding—multi-sensory integration—refractive dynamics internal model deduction—threshold warning reflex”, so that the model can still achieve interpretable multi-year quantitative trajectory prediction under conditions of single medical visit and scarce data.

[0102] Stabilization and gain control: Corresponding to retinal light adaptation and contrast gain adjustment, the device / exposure difference is first eliminated at the input end, so that subsequent feature learning focuses more on the real structural differences rather than imaging conditions.

[0103] Center-periphery lateral inhibition: Drawing on the center-periphery antagonism and lateral inhibition mechanism of retinal ganglion cells, this method highlights local contrast cues related to myopia, such as blood vessels and optic disc boundaries, which is different from prior enhancement that relies solely on automatic fitting of convolutional kernels.

[0104] Developmental receptive field tuning: Using age group as a proxy variable for developmental stage, the spatial frequency response parameters are dynamically adjusted to simulate the pattern of changes in the receptive field scale of children's visual system with development, reducing the performance degradation caused by cross-age distribution shift. It is an adaptive parameter adjustment technology for children's fundus characteristics, including a complete preprocessing process of retinal stabilization preprocessing and retinal lateral inhibition enhancement.

[0105] Reentrant hierarchical coding: Introducing cross-layer bypass / reentrant connections, it obtains macroscopic structural representation while maintaining micro-texture sensitivity, avoiding the loss of fine-grained information caused by deep abstraction of a single path.

[0106] Internal model of refractive development dynamics: The internal model extrapolates refractive evolution using a gated working memory prediction circuit, outputting a continuous multi-year refractive sequence, which is different from static classification that only makes "whether it is myopia / whether it is high myopia".

[0107] Threshold alert reflex and attention backtracking: triggering early warnings through risk thresholds and backtracking the decision-making basis to the input space to form a heat map, making the prediction results closer to the clinical workflow and auditable.

[0108] A biomimetic prior-constrained retinal input mechanism: Based on visual homeostasis (light adaptation / gain control) and center-peripheral suppression as priors, a reproducible structural enhancement link is constructed, which is different from general image enhancement and end-to-end convolution fitting without mechanism constraints.

[0109] Temporal consistency constraints of the prediction loop: Long-term prediction is organized around internal model deduction, and can be corrected by combining prediction error signals to make the output more consistent with the continuity and auditability of the development process.

[0110] For the input standardization dimension, existing technologies often rely on data augmentation / normalization, which is difficult to cover differences in different devices and exposures. The bionic mechanism of this invention uses retinal stabilization preprocessing (light adaptation / gain control), which can reduce domain offset sensitivity and improve cross-device consistency.

[0111] For the dimension of local contrast cues, the common practice of existing technologies is to use end-to-end convolutional automatic learning, which lacks physiological constraints. The biomimetic mechanism of this invention uses center-periphery lateral inhibition enhancement (lateral inhibition), which can highlight myopia-related structural cues such as blood vessels, optic disc, and macula.

[0112] Regarding the adaptation dimension for developmental stages, existing technologies commonly use a unified model or simply use age as a feature. The bionic mechanism of this invention adopts receptive field tuning for developmental stages (plasticity window parameter library), which can better fit the developmental patterns of children and reduce cross-age generalization loss.

[0113] Regarding the representation path dimension, existing technologies commonly employ unidirectional deep abstraction, which easily smooths out fine-grained textures. The biomimetic mechanism of this invention uses reentrant layered encoding (cross-layer bypass / reentry), which can take into account both micro-texture and macro-structure, and improve the fidelity of effective information.

[0114] For the temporal prediction morphology dimension, existing technologies commonly employ multiple classifications or single-time-point regression. The biomimetic mechanism of this invention uses refractive dynamics internal model deduction (prediction loop) to output continuous trajectories for the next ten years, supporting long-term treatment planning.

[0115] This invention proposes a refractive development assessment method based on visual development prediction neural circuits. This method simulates the hierarchical processing mechanism of the visual nervous system, constructing a complete information processing pathway from retinal optical imaging signal acquisition, retinal stabilization preprocessing, reentrant hierarchical retinal choroidal coding, multisensory integration, refractive development dynamics internal model deduction to developmental trajectory readout.

[0116] This method employs a perception-prediction dual-pathway architecture and a visual developmental dynamic trajectory extrapolation algorithm based on single-visit data. It utilizes only retinal optical imaging signals and refractive ground state quantification values ​​from a single visit, achieving long-term prediction based on minimal data input. This enables accurate extrapolation of children's refractive developmental trajectories, including quantitative refractive power values ​​and risk classification prediction. The overall system architecture comprises nine functional modules: signal acquisition, retinal stabilization preprocessing, developmental receptive field tuning, retinal lateral inhibition enhancement, reentrant hierarchical coding, multisensory integration, dynamic extrapolation, developmental trajectory readout, and threshold warning reflex. This achieves full automation from data input to clinical decision support.

[0117] Unlike traditional end-to-end deep networks, this invention organizes the information processing chain according to the visual system in the order of "input stabilization - local contrast enhancement - developmental stage parameter tuning - multi-level reentrancy encoding - cross-modal fusion - internal model inference - threshold reflection". It explicitly embeds physiological mechanisms into the model structure and parameter adaptation strategy, thereby maintaining the stability and interpretability of predictions in a single medical visit scenario.

[0118] The method proposed in this invention exhibits significant advantages and superior performance compared to existing technologies in multiple dimensions, and these effects have been confirmed through rigorous internal and cross-group generalization verification.

[0119] High Prediction Accuracy: The visual development prediction neural circuit system of this invention has achieved industry-leading prediction accuracy: the model accuracy reaches 95% for predicting the risk of future myopia. For the more critical risk of high myopia, the accuracy reaches 99%; in the regression task of directly predicting future refractive state values, the overall average error of the model is 0.25D per year, showing extremely high quantitative prediction accuracy.

[0120] Data Efficiency and Scalability: One of the most prominent advantages of this invention is its extremely high data efficiency. Existing technologies use overly simplistic methods (direct splicing) to fuse visual features and refractive values, resulting in the following problems: significant differences in the feature spaces of the two modalities make effective information interaction difficult with simple splicing; the contribution weights of different modalities cannot be dynamically adjusted; and there is a lack of modeling of intermodal correlations. This invention proposes CMGFN (Cross-Modal Gated Fusion Network) to achieve deep fusion of visual and numerical features. This feature completely changes the previous reliance of prediction models on long-term, multi-dimensional data, making early and accurate predictions possible in initial screening or resource-limited scenarios.

[0121] Cross-group generalization validation capability: The true value of a medical AI model lies in its universality across different populations. This invention demonstrates its excellent generalization capability through rigorous cross-group generalization validation across multiple ethnic groups.

[0122] Attention Retrospective Mapping and Clinical Relevance: To open the "black box" of deep learning and understand the basis of model predictions, this invention integrates interpretability techniques such as attention retrospective mapping. By generating heatmaps, the regions the model focuses on when analyzing retinal optical imaging signals can be visualized. Analysis shows that the model's attention is highly concentrated in the following regions during prediction: the superior and inferior temporal retinal regions, the posterior pole, and optic disc features. These regions highly overlap with those that ophthalmologists focus on when diagnosing and assessing myopia-related complications, validating the model's effectiveness.

[0123] In this embodiment of the invention, compared to traditional methods that require data from multiple medical visits, the present invention can achieve accurate prediction using only retinal optical imaging signals and refractive ground state quantification values ​​from a single medical visit. This design significantly reduces data collection costs, enabling the technology to be widely applied in resource-constrained environments, and is particularly suitable for myopia screening needs in remote areas.

[0124] This invention simultaneously realizes threshold warning reflex (myopia risk assessment) and developmental trajectory reading (specific refractive error value), providing clinicians with more comprehensive decision support. Compared with existing methods that can only perform binary classification, the quantitative prediction function of this invention (average error 0.25D / year) can provide precise numerical guidance for the formulation of personalized treatment plans.

[0125] This invention achieves an accuracy rate of 95% in predicting myopia risk and 99% in predicting the risk of high myopia, significantly outperforming existing technologies. Furthermore, its ability to extrapolate the dynamic trajectory of visual development provides support for long-term treatment planning, a capability not found in existing technologies.

[0126] This invention constructs a database containing statistics on the prevalence and progression rate of myopia in different regions. After cross-group generalization validation across various geographical regions, climate types, urban-rural distributions, and education burden indices, its excellent generalization ability has been demonstrated. This cross-population applicability allows the technology to be promoted and applied globally, without geographical or racial limitations.

[0127] By explicitly embedding biomimetic mechanisms such as visual homeostasis regulation, lateral inhibition, developmental stage tuning, and reentrant hierarchical coding into the processing chain, this invention is less sensitive to domain shifts when faced with different imaging conditions of different devices, different age structures, and cross-regional population differences. At the same time, vigilance reflexes and attentional retrospection mapping directly align the prediction with clinical decision thresholds, making the model output easier for doctors to interpret, review, and adopt.

[0128] Figure 2 This is a block diagram of a refractive development assessment device based on visual developmental prediction neural circuits, according to an exemplary embodiment. The device is used in a refractive development assessment method based on visual developmental prediction neural circuits. (Refer to...) Figure 2 The device includes a retinal optical imaging signal acquisition module 310, a retinal stabilization preprocessing module 320, a re-entry layered retinal choroid encoder module 330, a multisensory integration core module 340, a refractive developmental dynamics internal model inferrer module 350, and a threshold warning reflex module 360. Among them: The retinal optical imaging signal acquisition module 310 is used to acquire color retinal optical imaging signals and refractive ground state quantization values ​​for each subject, specifically for adolescents and children.

[0129] The retinal stabilization preprocessing module 320 is used to perform retinal stabilization preprocessing on the acquired color retinal optical imaging signals to obtain preprocessed retinal optical imaging signals; wherein, the retinal stabilization preprocessing includes retinal stabilization processing, retinal lateral inhibition enhancement, and receptive field tuning during developmental stages.

[0130] The reentrant layered retinal choroid encoder module 330 is used to extract multi-level morphological features from the preprocessed retinal optical imaging signal through the reentrant layered retinal choroid encoder to obtain a feature map.

[0131] The multisensory integration kernel module 340 is used to perform spatial feature aggregation gate operation on the feature map to obtain feature vectors; the feature vectors are cascaded and integrated with the corresponding refractive ground state quantization values ​​to obtain fused feature vectors.

[0132] The refractive development dynamics internal model inferrer module 350 is used to acquire fused feature vectors from multiple time points, form a fused feature vector sequence, input it into the refractive development dynamics internal model inferrer, generate a future multi-time point refractive state prediction sequence, and convert the refractive state prediction sequence into an interpretable refractive state numerical sequence.

[0133] The threshold warning reflection module 360 ​​is used to provide risk warnings based on the refractive state numerical sequence and preset risk thresholds.

[0134] In this embodiment of the invention, compared to traditional methods that require data from multiple medical visits, the present invention can achieve accurate prediction using only retinal optical imaging signals and refractive ground state quantification values ​​from a single medical visit. This design significantly reduces data collection costs, enabling the technology to be widely applied in resource-constrained environments, and is particularly suitable for myopia screening needs in remote areas.

[0135] This invention simultaneously realizes threshold warning reflex (myopia risk assessment) and developmental trajectory reading (specific refractive error value), providing clinicians with more comprehensive decision support. Compared with existing methods that can only perform binary classification, the quantitative prediction function of this invention (average error 0.25D / year) can provide precise numerical guidance for the formulation of personalized treatment plans.

[0136] This invention achieves an accuracy rate of 95% in predicting myopia risk and 99% in predicting the risk of high myopia, significantly outperforming existing technologies. Furthermore, its ability to extrapolate the dynamic trajectory of visual development provides support for long-term treatment planning, a capability not found in existing technologies.

[0137] This invention constructs a database containing statistics on the prevalence and progression rate of myopia in different regions. After cross-group generalization validation across various geographical regions, climate types, urban-rural distributions, and education burden indices, its excellent generalization ability has been demonstrated. This cross-population applicability allows the technology to be promoted and applied globally, without geographical or racial limitations.

[0138] By explicitly embedding biomimetic mechanisms such as visual homeostasis regulation, lateral inhibition, developmental stage tuning, and reentrant hierarchical coding into the processing chain, this invention is less sensitive to domain shifts when faced with different imaging conditions of different devices, different age structures, and cross-regional population differences. At the same time, vigilance reflexes and attentional retrospection mapping directly align the prediction with clinical decision thresholds, making the model output easier for doctors to interpret, review, and adopt.

[0139] Figure 3 This is a schematic diagram of the structure of a child refractive development assessment device provided in an embodiment of the present invention, as shown below. Figure 3 As shown, the device for assessing children's refractive development may include the above-mentioned Figure 2 The illustrated device is a refractive development assessment device based on visual development prediction neural circuits. Optionally, the pediatric refractive development assessment device 410 may include a first processor 2001.

[0140] Optionally, the child refractive development assessment device 410 may also include a memory 2002 and a transceiver 2003.

[0141] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0142] The following is combined Figure 3 A detailed introduction to each component of the pediatric refractive development assessment device 410: The first processor 2001 is the control center of the pediatric refractive development assessment device 410. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0143] Optionally, the first processor 2001 can perform various functions of the child refractive development assessment device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0144] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 3 CPU0 and CPU1 are shown in the diagram.

[0145] In a specific implementation, as one example, the pediatric refractive development assessment device 410 may also include multiple processors, such as... Figure 3 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0146] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0147] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently, and may be connected via the interface circuit of the child refractive development assessment device 410. Figure 3 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0148] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0149] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 3 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0150] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently and be connected to the interface circuit of the child refractive development assessment device 410. Figure 3 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0151] It should be noted that, Figure 3 The structure of the child refractive development assessment device 410 shown in the diagram does not constitute a limitation on the router. Actual knowledge structure identification devices may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0152] Furthermore, the technical effect of the pediatric refractive development assessment device 410 can be referred to the technical effect of the refractive development assessment method based on visual development prediction neural circuit described in the above method embodiments, and will not be repeated here.

[0153] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0154] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0155] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), 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 programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention 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. 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 a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0156] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0157] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0158] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0159] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0160] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0161] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0162] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0163] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0164] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0165] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for assessing refractive development based on visual developmental prediction neural circuits, characterized in that, The method includes: S1. For adolescents and children, collect color retinal optical imaging signals and refractive ground state quantification values ​​for each subject; S2. Perform retinal stabilization preprocessing on the acquired color retinal optical imaging signals to obtain preprocessed retinal optical imaging signals; wherein, retinal stabilization preprocessing includes retinal stabilization processing, retinal lateral inhibition enhancement, and receptive field tuning during developmental stages. S3. For the preprocessed retinal optical imaging signal, multi-level morphological features are extracted by a reentry-type layered retinal choroid encoder to obtain a feature map. S4. Perform a spatial feature aggregation gate operation on the feature map to obtain a feature vector; concatenate and integrate the feature vector with the corresponding refractive ground state quantization value to obtain a fused feature vector; S5. Obtain the fused feature vectors from multiple time points, form a fused feature vector sequence, input it into the refractive development dynamics internal model inference tool, generate a future multi-time point refractive state prediction sequence; convert the refractive state prediction sequence into an interpretable refractive state numerical sequence. S6. Risk warning is given based on the numerical sequence of refractive state and the preset risk threshold.

2. The refractive development assessment method based on visual development prediction neural circuits according to claim 1, characterized in that, The retinal stabilization process in S2 includes: preserving and enhancing the anatomical structures in the retinal optical imaging signal, simulating optical distortion, and synthesizing pathological enhancement. Among them, the preservation and enhancement of anatomical structures in retinal optical imaging signals includes: rotational enhancement with the center of the optic disc in the anatomical structure as the origin, while maintaining the relative positional relationship between the optic disc and the macula; vertical flipping of the anatomical structure; and maintaining the overall morphology of the major vascular arches. The optical distortion simulation includes: adding radial gradient blur to simulate different degrees of refractive media turbidity; simulating flash illumination of different intensities and angles; and simulating the differences in color response curves of fundus cameras from different brands. The synthetic pathological enhancements include: adding controllable choroidal texture to retinal optical imaging signals; simulating different degrees of optic disc tilt through affine transformation; and using elastic deformation to simulate vascular morphological changes in high myopia.

3. The refractive development assessment method based on visual development prediction neural circuits according to claim 1, characterized in that, The enhanced retinal-side inhibition in S2 includes: For retinal vessels in anatomical structures, spatial frequency filters are used to enhance the vessel edge information; For the optic disc in the anatomical structure, the edge detection operator combined with the spatial frequency filter is used to enhance the pixel gradient of the optic disc boundary; For the macula in the anatomical structure, a spatial frequency filter is used to enhance high-frequency components and suppress low-frequency background noise.

4. The refractive development assessment method based on visual development prediction neural circuits according to claim 1, characterized in that, The developmental receptive field tuning in S2 includes: The spatial frequency response parameters of the spatial frequency filter during the retinal side suppression enhancement process are dynamically adjusted according to different age groups.

5. The refractive development assessment method based on visual development prediction neural circuits according to claim 1, characterized in that, The reentrant hierarchical retinal choroid encoder in S3 includes: multiple serially connected feature abstraction levels, with parallel bypass connections between adjacent feature abstraction levels, each feature abstraction level is equipped with an independent dual attention gating module, and a top-down feature propagation path is adopted.

6. The refractive development assessment method based on visual development prediction neural circuits according to claim 1, characterized in that, The spatial feature aggregation gate operation in S4 adopts a full spatial domain averaging strategy, averaging all pixel values ​​of each feature map to obtain a one-dimensional feature vector.

7. The refractive development assessment method based on visual development prediction neural circuits according to claim 1, characterized in that, The method further includes: The risk warning process is optimized using the cross-entropy loss function, and the process of converting the refractive state prediction sequence into an interpretable refractive state numerical sequence is optimized using the mean square error loss function. The two optimization processes are then jointly optimized using a weighted summation method.

8. A refractive development assessment device based on visual development prediction neural circuits, wherein the refractive development assessment device based on visual development prediction neural circuits is used to implement the refractive development assessment method based on visual development prediction neural circuits as described in any one of claims 1-7, characterized in that, The device includes: The retinal optical imaging signal acquisition module is used to acquire color retinal optical imaging signals and refractive ground state quantification values ​​for each subject, specifically for adolescents and children. The retinal stabilization preprocessing module is used to perform retinal stabilization preprocessing on the acquired color retinal optical imaging signals to obtain preprocessed retinal optical imaging signals; wherein, retinal stabilization preprocessing includes retinal stabilization processing, retinal lateral inhibition enhancement, and receptive field tuning during developmental stages. The reentrant layered retinal choroid encoder module is used to extract multi-level morphological features from the preprocessed retinal optical imaging signal to obtain a feature map. The multisensory integration kernel module is used to perform spatial feature aggregation gate operations on the feature map to obtain feature vectors; the feature vectors are then cascaded and integrated with the corresponding refractive ground state quantization values ​​to obtain fused feature vectors. The refractive development dynamics internal model inferrer module is used to obtain fused feature vectors from multiple time points, form a fused feature vector sequence, input it into the refractive development dynamics internal model inferrer, generate a future multi-time point refractive state prediction sequence; and convert the refractive state prediction sequence into an interpretable refractive state numerical sequence. The threshold warning reflection module is used to provide risk warnings based on the refractive state numerical sequence and preset risk thresholds.

9. A device for assessing refractive development in children, characterized in that, The pediatric refractive development assessment device includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 7.

Images