System for acquiring parameters related to visual function based on retinal cell types

Abstract

The application relates to a retinal cell type-based visual function related parameter acquisition system and belongs to the technical field of medical auxiliary systems. The application aims to accurately acquire a retinal cell type-based visual function related parameter acquisition system of L cone, M cone, S cone, ipRGCs and other different retinal cell type function related physiological parameters. The system of the application comprises a stimulation presentation module, a brightness calibration module, a biological signal acquisition module, a multi-dimensional parameter extraction and cell type specific parameter decoupling module, a parameter analysis and standardized output module, a user interaction module and a data storage module. The application can provide a portable, objective, non-invasive and cell type specific retinal function related parameter acquisition technology, provide comprehensive and reliable technical support for visual function evaluation, is suitable for large-scale parameter acquisition and home monitoring, and solves many limitations of the prior art.

Description

Technical Field

[0001] This invention belongs to the field of medical auxiliary systems technology. Background Technology

[0002] Traditional visual acuity, color vision, and visual field tests rely on the subject's subjective cooperation and verbal feedback. The results obtained are easily affected by factors such as the subject's comprehension ability, attention, and fatigue level. For children, the elderly, and patients with cognitive impairment, the reliability and repeatability of parameter acquisition are poor. Moreover, these methods can usually only obtain abnormal indicators when there is obvious impairment of visual function, and they are difficult to capture subtle parameter changes in the early stages of retinal cell function, thus failing to provide effective parameter support for the early assessment of visual function.

[0003] Early abnormal changes in retinal cell function are important precursors to visual impairment. The functional status of different retinal cell types, such as L-cones, M-cones, S-cones, and ipRGCs, is closely related to the health assessment of the visual system. L-cones and M-cones are mainly concentrated in the macula, and changes in their functional parameters can reflect the visual function status of the macula. S-cones are sensitive to ischemia and hypoxia, and ipRGCs, as a subtype of retinal ganglion cells, have functional parameters that reflect the functional status of retinal neural pathways. Therefore, accurately obtaining functional physiological parameters of different retinal cell types is of great significance for the objective assessment of visual function. Current techniques for detecting and assessing visual function parameters have many limitations, making it difficult to meet the needs for accurate, objective, and portable acquisition of retinal cell functional parameters.

[0004] Multifocal electroretinography (mfERG) and patterned electroretinography (PERG) can obtain electrophysiological parameters related to retinal function and are important technical means for assessing retinal function. However, these technologies have obvious limitations: they require the use of corneal contact electrodes or skin electrodes, making the operation complex and highly invasive, resulting in significant discomfort for patients during the examination; the examination time is long, and professional technicians are required to operate and interpret the parameters, resulting in a high professional threshold; the equipment cost is high, making it difficult to promote and apply in primary healthcare institutions, community health service centers, and large-scale parameter collection, resulting in poor clinical accessibility.

[0005] Colorimetric pupillography, an objective assessment technique that has emerged in recent years, uses pupillary responses induced by light stimuli of different wavelengths to obtain parameters related to retinal function. However, its technical design has several shortcomings, making it difficult to meet the needs for accurate parameter acquisition: First, the equipment has low integration, employing a separate design for the light source and camera, making it impossible to achieve portable and automated integrated parameter acquisition; second, the stimulus paradigm design is simple, often using single-wavelength flicker light, which cannot accurately distinguish different retinal cell types, and the acquired parameters lack cell type specificity, making it difficult to reflect the functional state of different cells; third, the parameter extraction dimension is singular, only measuring the basic parameter of pupillary contraction amplitude, failing to fully explore the dynamic information of pupillary responses, such as latency, contraction speed, and recovery time, resulting in low information utilization efficiency; fourth, it is not combined with eye movement responses, relying solely on a single pupillary index, making it difficult to comprehensively obtain parameters related to retinal function and failing to provide multi-dimensional support for visual function assessment.

[0006] Virtual reality (VR)-based ophthalmology systems have been developing gradually in recent years. These systems utilize VR devices to present visual stimuli and monitor the visual responses of subjects, enabling preliminary visual function assessments. However, these systems still have fundamental limitations and cannot meet the need for precise acquisition of retinal cell function parameters: First, they rely on the subject's subjective button responses, essentially constituting a behavioral response assessment, making it difficult to objectively and without interference acquire retinal cell function-related parameters. Second, they use conventional brightness / contrast visual stimuli, lacking cell type selectivity and failing to specifically activate different retinal cell types, making it difficult to obtain functional-specific parameters for different cells. Third, they do not utilize the objective physiological indicator of pupillary light reflex, failing to directly reflect the functional state of photoreceptor cells and ganglion cells, resulting in insufficient specificity in parameter acquisition. Fourth, the parameter acquisition dimensions are limited, only monitoring a limited number of visual response indicators, and failing to provide comprehensive parameter support for functional assessment of different regions and cell types of the retina.

[0007] Different retinal cell types correspond to different dimensions of visual system assessment. In clinical practice and research, there is an urgent need for objective technologies capable of selectively acquiring functional parameters related to different retinal cell types. This would enable precise and quantitative analysis of retinal cell function, providing comprehensive and reliable parameter support for visual function assessment. Such technologies must meet the following core requirements: first, portability—small and low-cost devices suitable for primary healthcare and large-scale data collection; second, non-invasiveness—no invasive procedures required, improving subject cooperation; third, objectivity—no subjective cooperation required, applicable to the entire population; and fourth, specificity—capable of accurately distinguishing different retinal cell types and acquiring cell-type-specific parameters. Current technologies have not yet simultaneously met all of these requirements, making filling this technological gap an important research direction in this field. Summary of the Invention

[0008] The purpose of this invention is to provide a system for acquiring visual function-related parameters based on retinal cell types, which can accurately acquire physiological parameters related to the functions of different retinal cell types such as L-cones, M-cones, S-cones, and ipRGCs.

[0009] The system of the present invention includes: S1, Stimulation Presentation Module: Generates a narrowband light stimulation sequence with cell type selectivity based on the spectral sensitivity function of the target retinal cell type; includes a high-resolution display integrated into the VR headset, with the core configuration being an OLED microdisplay; S2, Brightness Calibration Module: The perceived brightness of different narrowband light stimuli is calibrated by using the heterochromatic flicker photometric method, and an isoluminance lookup table is established; S3, Biosignal Acquisition Module: Synchronously acquires image sequences and timestamps through a high-speed camera unit integrated in the VR headset; the acquired raw signal data includes three types: pupil image sequence, corneal reflection image sequence and stimulus presentation timestamp; includes at least one infrared camera unit integrated in the VR headset, preferably a binocular infrared camera unit, with a core configuration of a global shutter CMOS sensor; S4. Multidimensional parameter extraction and cell type-specific parameter decoupling module: Extracts kinetic parameters and reaction feature parameters from image sequences, establishes a parameter decoupling model, and solves for cell type-specific functional parameters and parameter ratios; This module is the core processing unit of the system, integrating an embedded processor, an image processing unit, and a dedicated algorithm chip. S5, Parameter Analysis and Standardized Output Module: Inputs response parameters, specific parameters, and parameter ratios into a pre-trained multimodal data fusion analysis model, and outputs a standardized set of retinal functional parameters after analysis and verification; including support vector machines, random forests, gradient boosting trees, and deep neural networks; S6, User Interaction Module: Configured to receive user input, issue operation commands, and provide real-time feedback on the collection status; includes VR headset controllers, voice interaction interface, and dedicated interactive applications for external terminals; S7, Data Storage Module: Configured to store raw image data, intermediate parameters, and standardized parameter sets, supporting local data storage and remote transmission; the core configuration consists of a local storage unit and a cloud communication module.

[0010] The narrowband photostimulation sequence parameters of the present invention include: L cone cell stimulation center wavelength 560nm±15nm, M cone cell stimulation center wavelength 540nm±15nm, S cone cell stimulation center wavelength 460nm±15nm, ipRGCs stimulation center wavelength 480nm±15nm, and full width at half maximum (FWHM) ≤20nm.

[0011] The high-speed camera unit of this invention has a sampling frequency of not less than 120Hz and a spatial resolution of not less than 640×480 pixels; the high-speed camera unit synchronous acquisition adopts binocular synchronous acquisition, and acquires the reaction image sequences of the left eye and the right eye respectively.

[0012] The kinetic parameters of this invention include one or more of the following: initial contraction latency, contraction amplitude, contraction peak time, maximum contraction velocity, re-expansion half-life, scintillation stimulus gain, scintillation stimulus phase delay, and continuous illumination contraction maintenance index; the response characteristic parameters include one or more of the following: gaze stability represented by the area of ​​a bivariate contour ellipse, microscan video rate, and microscan visual amplitude.

[0013] The parameter decoupling model is a linear model, with the contributions of L-cones, M-cones, S-cones, and ipRGCs to the pupillary response as unknowns, and pupillary response parameters under multi-wavelength stimulation as observed values. The contribution weights of each cell type are solved by non-negative least squares method to obtain standardized cell type-specific function-related parameters, including L-cone cell function parameters, M-cone cell function parameters, S-cone cell function parameters, ipRGCs cell function parameters, and parameter ratios L-cone / S-cone parameter ratio and L-cone / ipRGCs parameter ratio. It also includes the calculation of the binocular parameter asymmetry index, where the binocular parameter asymmetry index is the left and right eye parameters divided by the binocular average.

[0014] The multimodal data fusion analysis model of this invention is constructed using machine learning algorithms, including one of support vector machines, random forests, gradient boosting trees, and deep neural networks. The model takes parameter effectiveness and stability as optimization objectives, outputs a comprehensive parameter effectiveness score, sets a score threshold to filter effective parameters to form a standardized parameter set, and presents the standardized parameter set in the form of numerical values, trend curves, and ratio comparison charts.

[0015] The narrowband light stimulation of this invention includes selective stimulation and parameter acquisition of rod cells, using narrowband light with a center wavelength of 500nm±15nm to extract functional parameters of rod cells.

[0016] The multi-dimensional parameter extraction and cell type-specific parameter decoupling module of this invention, namely the signal processing module, includes: a quality control system that monitors the signal acquisition quality in real time, automatically identifies and processes blinking and head movement artifacts, uses cubic spline interpolation to fill missing data and low-pass filtering to remove high-frequency noise, and automatically prompts for repeated acquisition when the data quality is substandard.

[0017] The data storage module of this invention is configured to receive parameter data, enabling multi-center data aggregation, model iterative updates, remote parameter analysis, and standardized report generation.

[0018] The VR headset of this invention features a lightweight ergonomic design, weighing ≤300g, making it comfortable to wear. It requires no professional technicians to operate and supports convenient parameter acquisition.

[0019] This invention provides a portable, objective, non-invasive, and cell-type-specific technique for acquiring retinal function-related parameters. It can accurately acquire functional physiological parameters of different retinal cell types, such as L-cones, M-cones, S-cones, and ipRGCs, providing comprehensive and reliable technical support for visual function assessment. It is also suitable for large-scale parameter acquisition and home monitoring, overcoming many limitations of existing technologies. Attached Figure Description

[0020] Figure 1 This is an overall structural block diagram of the visual function-related parameter acquisition system based on selective stimulation of retinal cell types in this embodiment of the invention, showing the connection relationship between the VR headset, signal processing unit, user terminal, and cloud data platform, as well as the core modules of each unit; Figure 2 This is a spectral distribution diagram of cell type selective stimulation in an embodiment of the present invention, showing the spectral sensitivity curves of four cell types: L-cones, M-cones, S-cones, and ipRGCs, as well as their corresponding narrowband stimulation wavelengths and full width at half maximum (FWHM). Figure 3 This is a schematic diagram illustrating the principle of isoluminance calibration using the heterochromatic flicker photometric measurement method in this embodiment of the invention, showing the presentation method of alternating flicker of two wavelengths of light, the subject adjustment interface, and the calibration process; Figure 4 This is a typical pupil response waveform diagram in an embodiment of the present invention, labeled with extracted pupil dynamics-related parameters such as initial contraction latency, contraction amplitude, contraction peak time, maximum contraction velocity, and re-dilation half-life; Figure 5 This is a statistical distribution chart of L-cone cell functional parameters (LFI) in normal subjects in this embodiment of the invention, showing the distribution range and mean of LFI in the normal population; Figure 6 This is a statistical distribution chart of ipRGCs cell function parameters (ipFI) in normal subjects in an embodiment of the present invention, showing the distribution range and mean of ipFI in the normal population; Figure 7 This is a scatter plot showing the distribution of the L / S ratio of normal subjects in this embodiment of the invention, illustrating the distribution characteristics of the L / S ratio in the normal population. Figure 8 This is a schematic diagram of the network structure of the multimodal data fusion analysis model based on the attention mechanism in this embodiment of the invention, showing the structure and parameter flow relationship of the model's input layer, modality coding layer, attention fusion layer, and analysis layer; Figure 9 This is a schematic diagram of the internal structure of the VR headset in an embodiment of the present invention, showing the layout of the display, infrared camera unit, infrared LED illumination, and controller; Figure 10 This is a schematic diagram of the user interaction and parameter output interface in an embodiment of the present invention, including a test guidance interface, a real-time data quality monitoring interface, a parameter visualization display interface, and a data export interface. Detailed Implementation

[0021] The present invention aims to overcome the shortcomings of the prior art and provide a system for obtaining visual function-related parameters based on selective stimulation of retinal cell types, so as to achieve accurate, objective and quantitative acquisition of physiological parameters related to the function of different retinal cell types.

[0022] 1. Provide a photostimulation system that can selectively activate different retinal cell types (L cones, M cones, S cones, ipRGCs), enabling independent acquisition and analysis of functional parameters of different cell types, and ensuring cell type specificity of the parameters; 2. Provides an objective parameter acquisition system that does not require subjective cooperation from the subject. By simultaneously collecting pupillary light reflex and eye movement response, it extracts multi-dimensional, cell type-specific visual function-related parameters to ensure the objectivity and reliability of the parameter results. 3. Provide a portable and low-cost parameter acquisition system that integrates all functions into a virtual reality device to achieve integrated and automated parameter acquisition, suitable for large-scale retinal functional parameter acquisition and home monitoring; 4. Provides a parameter analysis model based on multimodal data fusion to achieve accurate quantitative analysis of parameters related to retinal cell function, improve the reliability and effectiveness of parameters, and provide high-quality parameter support for visual function assessment; 5. Provide a standardized parameter output format to achieve parameter unification and comparability, providing a unified technical reference for dynamic evaluation of visual function and multi-center research, and improving the efficiency of parameter utilization; 6. It achieves technological scalability, easily extending to the acquisition of rod cell functional parameters, enriching the range of parameter acquisition, and adapting to diverse visual function assessment needs.

[0023] This invention is based on the following four key technological concepts to achieve accurate acquisition and quantitative analysis of parameters related to specific functions, thus overcoming the core limitations of existing technologies: First, different retinal cell types possess unique spectral sensitivity functions. The peak sensitivity of L-cone cells is around 560 nm, M-cone cells around 540 nm, S-cone cells around 460 nm, and the intrinsic photosensitivity peak of ipRGCs is around 480 nm. By selecting narrow-band light stimulation with a specific center wavelength and narrow full width at half maximum (FWHM), the target cell type can be preferentially activated. Combined with isoluminance calibration using heterochromatic scintillation photometry, interference from the luminance pathway on the pupillary response is eliminated, ensuring that the acquired response signal accurately reflects the functional state of the target cell type. This provides a reliable signal basis for parameter acquisition and achieves cell-type specificity of the parameters.

[0024] Secondly, the pupillary light reflex pathway is a low-level neural circuit consisting of the retina-anterior tectal nucleus-oculomotor nucleus-pupil sphincter, which does not involve the higher cortical areas of the brain. It can directly and objectively reflect the functional state of photoreceptor cells and ganglion cells, and physiological parameters free from subjective interference can be obtained based on this pathway. By acquiring the temporal changes in pupillary responses through a high-speed imaging system, rich dynamic parameters such as latency, amplitude, contraction velocity, and recovery time can be extracted. Simultaneously acquiring eye movement responses (fixation stability, microsaccades, etc.) can obtain supplementary parameters related to macular fixation function, achieving comprehensive acquisition of multi-dimensional parameters and providing richer technical references for visual function assessment.

[0025] Third, by integrating parameters from different sources such as pupillary dynamics parameters and ocular response parameters, a multimodal data fusion parameter analysis model is established to perform collaborative analysis and weighted fusion of different parameters, thereby improving the quantification accuracy and effectiveness of the parameters. Simultaneously, based on the stimulus response characteristics of different retinal cell types, a parameter decoupling model is established. Through mathematical algorithms, the contribution of different cell types to the response signal is separated, achieving effective decoupling of functionally related parameters for different cell types and obtaining standardized cell-type-specific parameter indicators.

[0026] Fourth, the entire process of cell type-selective stimulation, physiological signal acquisition, parameter extraction and analysis, and result output is integrated into a virtual reality (VR) headset. The VR headset's high-resolution display presents precise narrowband light stimulation, while a built-in eye-tracking camera simultaneously acquires pupil and eye movement data. An embedded processor performs real-time parameter extraction and analysis, outputting the results. The VR headset's enclosed optical design effectively controls ambient light interference, ensuring the accuracy of stimulus presentation and the stability of signal acquisition. Furthermore, the VR device is small, lightweight, and easy to operate, achieving portable and integrated parameter acquisition, making it suitable for primary healthcare and home monitoring.

[0027] The parameter acquisition system of this invention specifically includes the following modules: S1, Stimulus Presentation Module: Based on the spectral sensitivity function of the target retinal cell type, a narrowband light stimulation sequence with cell type selectivity is generated. The target retinal cell type includes one or more of L cone cells, M cone cells, S cone cells, and ipRGCs, ensuring that the stimulation can preferentially activate the target cell type, laying the foundation for obtaining specific parameters.

[0028] To achieve precise cell type selection, the wavelength and bandwidth parameters of narrowband photostimulation are set as follows: For selective stimulation of L-cone cells, narrow-band light with a center wavelength of 560nm±15nm and a full width at half maximum (FWHM) ≤20nm was selected. For selective stimulation of M cone cells, narrow-band light with a center wavelength of 540nm±15nm and a full width at half maximum (FWHM) ≤20nm was selected. For selective stimulation of S cone cells, narrow-band light with a center wavelength of 460nm±15nm and a full width at half maximum (FWHM) ≤20nm was selected. For selective stimulation of ipRGCs, narrowband light with a center wavelength of 480nm±15nm and a full width at half maximum (FWHM) ≤20nm is selected.

[0029] To comprehensively acquire functional parameters of different dimensions of the target cell type and reflect the complete functional state of the cell, the photostimulation sequence includes multiple time-domain modulation modes, with different modes adapted to acquire parameters of different functional dimensions: Single-pulse stimulation: Pulse width 100-1000ms, used to obtain parameters related to the transient response characteristics of cells; Dual-pulse stimulation: Two pulses spaced 1-4 seconds apart, used to obtain parameters related to cell function recovery characteristics; Scintillation stimulation: frequency 1-20Hz, used to obtain parameters related to the time-frequency response characteristics of cells; Continuous light stimulation: duration 1-10s, used to obtain parameters related to the continuous response characteristics of ipRGCs.

[0030] It includes a high-resolution display integrated into the VR headset, with a core configuration of an OLED microdisplay, a resolution ≥1920×1080, a refresh rate ≥120Hz, and a peak brightness ≥1000cd / m². This module is configured to generate a narrowband light stimulation sequence based on the spectral sensitivity function of the target retinal cell type, and has precise spectral control capabilities. Through a filter array, it can output narrowband light stimulation with an adjustable center wavelength and a full width at half maximum (FWHM) ≤20nm, ensuring cell type selectivity of the stimulation.

[0031] Preferably, the stimulus presentation module further includes noise masking technology to present dynamic brightness noise around the stimulus, eliminating the potential influence of edge perception on the pupil response signal and further improving the accuracy of parameter acquisition.

[0032] S2, Brightness Calibration Module: The perceived brightness of different narrowband light stimuli is calibrated using heterochromatic scintillation photometry, eliminating interference from the brightness pathway on pupillary responses. This ensures that the response signals acquired under different wavelength stimuli are only related to the functional state of the target cell type, rather than brightness perception, thus improving the specificity and accuracy of parameter acquisition. The system is configured to generate cell type-selective narrowband light stimulation sequences, possessing precise spectral control capabilities and capable of outputting narrowband light with an adjustable center wavelength and a full width at half maximum (FWHM) ≤ 20 nm.

[0033] The specific calibration process is as follows: 1. The two wavelengths of light A and B to be calibrated are alternately presented in the VR headset at a frequency of 20-30Hz; 2. Subjects adjust the brightness of one of the lights using a handle or button until the visual flickering is minimized. At this point, the perceived brightness of the two lights is determined to be equal, and this brightness value is recorded as the equal brightness point. 3. Perform pairwise calibration on all wavelengths used to establish a complete isoluminance lookup table; 4. During actual parameter acquisition, the brightness of each wavelength stimulus is automatically set according to the lookup table to ensure that all stimuli have the same perceived brightness and eliminate the interference of brightness factors on the parameters.

[0034] In conjunction with the stimulus presentation module, it is configured to achieve isoluminance calibration of stimuli of different wavelengths through heterochromatic scintillation photometry, and automatically establish, store and update a standardized isoluminance lookup table. In actual parameter acquisition, this module can automatically adjust the brightness of each wavelength stimulus according to the lookup table to ensure that all stimuli have the same perceived brightness, eliminate the interference of the brightness pathway on the response signal, and ensure the specificity of the parameters.

[0035] S3, Biosignal Acquisition Module: The cell type-selective photostimulation sequence is presented to the subject, and the subject's pupillary response signal and eye movement response signal are simultaneously acquired by a high-speed camera unit integrated in the VR headset. At the same time, the timestamp of the stimulus presentation is recorded to achieve precise time alignment between the stimulus signal and the response signal, ensuring the accuracy of subsequent parameter extraction.

[0036] To ensure the quality and accuracy of signal acquisition, the hardware parameters of the high-speed camera unit are set as follows: sampling frequency not less than 120Hz, spatial resolution not less than 640×480 pixels; preferably, a binocular synchronous acquisition method is adopted to acquire the response signals of the left and right eyes respectively, so as to realize the independent extraction and analysis of binocular parameters and provide parameter support for unilateral retinal function assessment.

[0037] The acquired raw signal data includes three categories, providing complete raw data for subsequent parameter extraction: Pupil image sequence: used to extract pupil diameter change curves and obtain dynamic parameters related to pupil response; Corneal reflection image sequence: Combined with pupil center position, it is used to calculate eye movement trajectory and extract eye movement response related parameters; Stimulus presentation timestamp: Used to achieve precise time alignment between stimulus and response signals, ensuring a one-to-one correspondence between parameter extraction and stimulus events.

[0038] It includes at least one infrared camera unit integrated into the VR headset, preferably a binocular infrared camera unit (one for the left eye and one for the right eye), with a core configuration of a global shutter CMOS sensor, a sampling frequency ≥120Hz (preferably 250Hz), and a spatial resolution ≥640×480 pixels; the module is equipped with 850nm infrared LED illumination, which can clearly image in dark environments, and is configured to synchronously acquire pupil response images and eye movement response images of the subject at a set sampling frequency, providing high-quality raw signals for subsequent parameter extraction.

[0039] The binocular synchronous acquisition method can acquire response signals from the left and right eyes separately, enabling independent extraction and analysis of binocular parameters, calculating the binocular parameter asymmetry index, and providing parameter support for unilateral retinal function assessment.

[0040] S4. Multidimensional parameter extraction and cell type-specific parameter decoupling module: i.e., signal processing module Based on the image sequence, dynamic parameters and response characteristic parameters are extracted, and a parameter decoupling model is established to solve for cell type-specific functional parameters and parameter ratios. Specifically, based on the acquired pupil response signal and eye movement response signal, quantitative parameters related to the function of the target cell type are extracted through image processing and algorithm analysis. A parameter decoupling model is then established to separate the contribution of different cell types to the response signal, obtain standardized cell type-specific function-related parameters, and achieve precise decoupling and quantification of parameters.

[0041] It includes three core components: (1) Pupil response related parameters Based on pupil image sequences, quantitative parameters reflecting the dynamic characteristics of pupillary response are extracted using algorithms such as image segmentation, ellipse fitting, and artifact removal. These parameters cover the entire pupillary contraction and recovery process, comprehensively reflecting the functional state of the target cell type, including but not limited to: Initial contraction latency: The time from the onset of stimulation to the onset of pupil constriction, measured in milliseconds (ms). Amplitude of pupil constriction: the percentage of the maximum pupil constriction diameter relative to the baseline, expressed as % Time to Peak: The time from the onset of stimulation to the attainment of maximum contraction, measured in milliseconds (ms). Maximum Velocity: The maximum rate at which the pupil constricts, measured in mm / s; Redilation half-time: The time required for the pupil to return to 50% of its baseline diameter, measured in seconds. Gain and phase delay under flickering stimulation: calculated by Fourier analysis, reflecting the pupil's response characteristics to stimuli of different frequencies; Sustained Construction Index (SCI): The ratio of the average contraction amplitude to the peak contraction amplitude in the last second of continuous stimulation, reflecting the pupil's sustained response characteristics.

[0042] (2) Eye movement response related parameters Based on corneal reflectance image sequences and changes in pupil center position, eye movement trajectories are accurately calculated, and quantitative parameters related to eye movement responses are extracted to reflect macular fixation function and the status of local retinal neural pathways, providing supplementary parameters for visual function assessment, including but not limited to: Fixation Stability: Represented by the area of ​​the bivariate contour ellipse (BCEA), in degrees²; Microsaccade Rate: The number of microsaccades per unit of time, measured in seconds (s). Microsaccade Amplitude: The average amplitude of microsaccades, measured in degrees.

[0043] (3) Decoupling of cell type-specific parameters To accurately separate the independent contributions of different cell types to pupillary response signals and eliminate signal interference between different cells, a parameter decoupling model was established to achieve effective extraction of cell type-specific parameters.

[0044] The specific decoupling process is as follows: 1. The contributions of four cell types—L cone, M cone, S cone, and ipRGCs—to pupillary response were treated as unknowns and normalized. 2. Using pupil response parameters under stimulation at multiple wavelengths as observed values, and based on the spectral sensitivity function of each cell type, calculate the relative activation weights of each cell type at different wavelengths; 3. The contribution weights of each cell type are solved by the least squares method to obtain standardized cell type-specific function-related parameters; 4. Based on cell type-specific parameters, calculate the cell type parameter ratio to provide a multi-dimensional reference for visual function assessment.

[0045] The final cell type-specific parameters and ratios obtained include: Core functional parameters: L cone cell functional parameters (LFI), M cone cell functional parameters (MFI), S cone cell functional parameters (SFI), and ipRGCs cell functional parameters (ipFI); Parameter ratios: L-cone / S-cone parameter ratio (L / S Ratio), L-cone / ipRGCs parameter ratio (L / ip Ratio); Unilateral assessment parameter: Binocular parameter asymmetry index, calculated as (left eye parameter - right eye parameter) / average value of both eyes.

[0046] As the core processing unit of the system, it integrates an embedded processor (ARM Cortex-A76 or higher), an image processing unit (GPU), and a dedicated algorithm chip, configured to perform the following core operations to convert raw signals into valid parameters: Based on pupil response image sequences, the pupil diameter change curve is extracted in real time using a deep learning segmentation algorithm (U-Net architecture). Abnormal signals such as blink artifacts and head movement interference are automatically identified and processed. Cubic spline interpolation is used to fill missing data and low-pass filtering is used to remove high-frequency noise. Based on corneal reflexes and changes in pupil center position, eye movement trajectories are accurately calculated and eye movement response parameters are extracted. By integrating pupillary response parameters and eye movement response parameters, a parameter decoupling model was established. The model was solved using non-negative least squares method to calculate cell type-specific function-related parameters, parameter ratios, and binocular parameter asymmetry index. All extracted parameters are input into a pre-trained multimodal data fusion analysis model to perform quantitative analysis and validity verification of the parameters, and output a comprehensive validity score of the parameters. Valid parameters are selected and integrated into a standardized parameter set.

[0047] Preferably, the signal processing module further includes a quality control system that monitors the signal acquisition quality and parameter extraction reliability in real time. The monitoring indicators include pupil tracking confidence, blink artifacts, head movement interference, etc. When the data quality does not meet the standards, it automatically prompts for repeated acquisition to ensure the validity and reliability of the output parameters.

[0048] S5, Parameter Analysis and Standardization Output Module: The extracted pupillary response parameters, eye movement response parameters, cell type-specific parameters, and parameter ratios are input into a pre-trained multimodal data fusion analysis model. The parameters are then quantitatively analyzed and their validity verified. Invalid parameters are removed, and a standardized set of retinal cell function-related parameters is finally output, providing a unified and comparable technical reference for visual function assessment.

[0049] The multimodal data fusion analysis model is constructed using machine learning algorithms, including but not limited to support vector machines, random forests, gradient boosting trees, and deep neural networks. The model is trained using retinal function-related parameter data from normal subjects, with the effectiveness and stability of the parameters as the core optimization objectives, and the extracted quantitative parameters as features, thereby improving the accuracy and reliability of the model's parameter analysis.

[0050] The core function of the model is: 1. Perform fusion analysis on multi-dimensional parameters, dynamically adjust the weight of each parameter, and improve the overall effectiveness of the parameters; 2. Output parameters and give a comprehensive validity score. Set a score threshold (e.g., ≥0.8) to filter out valid parameters. 3. Integrate effective parameters into a standardized parameter set and visualize them in the form of numerical values, trend curves, ratio comparison charts, heat maps, etc. At the same time, it supports exporting in standard formats such as Excel and PDF, so that the parameters can be stored, compared, and shared.

[0051] The module is configured to provide a standardized and visual representation of the effective retinal cell function parameters analyzed by the signal processing module, and supports the export of parameters in a standardized format. This module is integrated into the display interface of the VR headset and the display interface of the external terminal to adapt to different viewing needs.

[0052] Visualization formats include numerical displays, parameter trend charts, parameter ratio comparison charts, heat maps, etc., which intuitively present parameter results. Standardized export formats include Excel, PDF, etc., and the exported content includes raw data, processed parameters, validity scores, normal reference ranges, etc., providing storable, comparable, and analyzable parameter data for visual function assessment.

[0053] S6, User Interaction Module: Configured to receive user input, issue operation commands, and provide real-time feedback on the data collection status; including dedicated interactive applications for VR headset controllers, voice interaction interfaces, and external terminals (smartphones / tablets / computers), configured to receive user input and issue operation commands, while providing real-time feedback on the data collection status, thus improving the ease of operation of the system.

[0054] Core features include: Receive user input: basic information of the subject (age, gender, etc.) and test parameter settings (such as target cell type, stimulation mode, etc.); Issue operation instructions: start / pause / end test, start calibration, export data, etc. Real-time status feedback: test progress, data quality status, operation prompts, and alerts for parameter acquisition anomalies.

[0055] S7, Data Storage Module: It is configured to store the raw acquired signal data, extracted intermediate parameters, and the final output standardized parameter set. The core configuration consists of a local storage unit (memory ≥16G, storage ≥128G) and a cloud communication module (Wi-Fi / 5G), which supports local data storage and remote transmission.

[0056] This module enables long-term storage of parameter data, providing data support for dynamic monitoring and trend analysis of retinal functional parameters; at the same time, through the cloud communication module, the parameter data is uploaded to the cloud data platform to realize multi-center data aggregation, model iterative updates, and remote analysis.

[0057] Application of this invention in virtual reality devices for obtaining retinal function-related parameters: This invention can accurately obtain functional parameters of L-cone and M-cone cells related to visual function in the macular region, as well as functional parameters of S-cone and ipRGCs cells related to retinal neural pathway function, providing comprehensive and accurate parameter support for visual function assessment.

[0058] The applicable scenarios for this device include: 1. Screening in primary healthcare institutions: Deployed in community hospitals, township health centers, and physical examination centers to provide retinal function-related parameter collection services for large-scale populations and provide technical support for preliminary assessment of visual function; 2. Home-based dynamic monitoring: For use by subjects at home, to achieve long-term and regular monitoring of retinal functional parameters, obtain the trend of parameter changes, and provide data for dynamic assessment of visual function; 3. Scientific data support: Provides standardized, large-scale parameter data for basic and clinical research related to retinal function, promoting research progress in related fields; 4. Unilateral Functional Assessment: Based on the binocular parameter asymmetry index, it provides accurate parameters for the assessment of unilateral retinal function, filling the gap in the acquisition of unilateral retinal function parameters.

[0059] The device can be extended to acquire rod cell function-related parameters. Specifically, it uses narrowband light with a center wavelength of 500nm±15nm to present stimulation under dark adaptation conditions. Rod cell function-related parameters are extracted using the method of this invention, thereby realizing the acquisition and analysis of dark vision function-related parameters, enriching the parameter acquisition range, and adapting to more visual function assessment needs.

[0060] The system of this invention is an integrated virtual reality (VR) head-mounted display device, which integrates seven core functional modules: stimulus presentation, brightness calibration, biosignal acquisition, signal processing, parameter output, user interaction, and data storage. It can accurately acquire, analyze in real time, verify effectiveness, and output standardized parameters related to retinal cell function. The device is portable, easy to operate, requires no professional technicians, and is suitable for multi-scenario applications.

[0061] Preferred technical solution of the present invention: 1. The cell type-selective stimulation also includes selective stimulation of rod cells, using narrowband light with a center wavelength of 500nm±15nm and a full width at half maximum (FWHM) of ≤20nm, presented under dark adaptation conditions. The method of this invention extracts rod cell function-related parameters, realizes the collection and analysis of dark vision function-related parameters, and enriches the parameter acquisition range. 2. The biosignal acquisition module adopts a binocular synchronous acquisition method to calculate the pupillary response parameters and eye movement response parameters of the left and right eyes respectively, and accurately calculates the binocular parameter asymmetry index, providing accurate parameters for unilateral retinal function assessment and filling the gap in unilateral parameter acquisition; 3. The quality control system of the signal processing module can identify and mark abnormal signals such as blinking and head movement in real time, automatically correct minor abnormal signals, and prompt repeated acquisition for serious abnormal signals in an immediate manner to ensure the reliability and effectiveness of parameter acquisition and reduce the proportion of invalid data. 4. The multimodal data fusion analysis model adopts an attention-based multimodal fusion network, which can dynamically adjust the feature weights of pupil parameters, eye movement parameters, and cell type-specific parameters to achieve adaptive fusion analysis of parameters. Compared with traditional models, it improves the accuracy of parameter analysis by 10-15%. 5. The system is equipped with a cloud data platform that communicates with the data storage module, supporting the aggregation, standardized processing, model iterative updates, remote parameter analysis, and standardized report generation of multi-center parameter data. It is adapted to the needs of primary healthcare and large-scale parameter acquisition, and promotes the progress of retinal function research. 6. The VR headset adopts a lightweight and ergonomic design, weighing ≤300g. It is equipped with a comfortable headband and eye pads, making it comfortable to wear without pressure. It is easy to operate and supports a one-click testing mode, allowing parameter collection to be completed without professional technicians. It is suitable for different groups such as children, the elderly, and patients with cognitive impairment, thus improving the universality of the system. 7. The system is equipped with a battery life module with a battery life of ≥4 hours, supports wireless operation, eliminates the constraints of power cords, further improves the portability and flexibility of the device, and is suitable for mobile scenarios such as outdoor screening and on-site testing.

[0062] For the first time, precise acquisition of retinal cell type-specific parameters has been achieved, filling a technological gap. This invention is the first to deeply integrate the principle of cell type-selective stimulation with VR technology. Through precise narrowband light stimulation design (specific center wavelength, narrow half-width and full width at half-height) and isoluminance calibration, it achieves preferential activation of different retinal cell types such as L-cones, M-cones, S-cones, and ipRGCs, eliminating signal interference between different cells. By establishing a parameter decoupling model, it accurately separates the contribution of each cell type from the mixed pupillary response signal, obtains standardized cell type-specific function-related parameters, and realizes the "cell resolution" acquisition of retinal functional parameters. This fills the technical gap of existing technologies that cannot accurately obtain cell type-specific parameters, and provides a new technical dimension for the refined assessment of visual function.

[0063] Parameter acquisition is objective and free from subjective interference. This invention acquires parameters based on pupillary light reflex and eye movement response. The pupillary light reflex pathway is a low-level neural circuit that does not involve the higher cortical areas of the brain. It requires no subjective cooperation or verbal feedback from the subject, resulting in objective, quantifiable, and repeatable parameter results, completely eliminating reliance on the subject's subjective abilities. It is suitable for special populations that are difficult to cover using traditional methods, such as children, the elderly, and patients with cognitive impairment, solving the challenge of assessing retinal function parameters in special clinical populations and enabling parameter acquisition for the entire population. Furthermore, the temporal resolution of parameter extraction is at the millisecond level, capable of capturing subtle changes in retinal cell function parameters, improving the sensitivity of parameter acquisition, and allowing for the acquisition of effective parameters at the early stages of cell dysfunction.

[0064] Comprehensive acquisition of multi-dimensional parameters significantly enhances the value supported by these parameters. This invention simultaneously acquires two types of signals: pupillary response and eye movement response. It extracts a multi-dimensional parameter set, including pupillary dynamics parameters, eye movement characteristic parameters, cell type-specific parameters and parameter ratios, and binocular parameter asymmetry index. The parameters cover multiple aspects such as different retinal cell types, different functional dimensions, and binocular unilateral assessment. It can not only obtain functional parameters of different retinal cell types, but also obtain macular fixation function-related parameters, achieving comprehensive coverage of retinal function-related parameters. The multi-dimensional parameters complement each other and are analyzed synergistically, providing a more comprehensive and reliable technical reference for visual function assessment. Compared with the single-dimensional parameters of existing technologies, the supporting value is significantly improved.

[0065] VR integrated design enables portable, low-cost large-scale parameter acquisition. This invention integrates stimulus presentation, brightness calibration, signal acquisition, parameter extraction, analysis output, and data storage into a VR headset. The device features an integrated design, is compact, lightweight (≤300g), and easy to operate. Its cost is significantly lower than traditional devices such as OCT and multifocal ERG, drastically reducing the hardware cost of parameter acquisition. No professional technicians are required for operation; it supports a one-click testing mode and can be deployed in community hospitals, health check centers, pharmacies, and other primary healthcare institutions. It also enables long-term monitoring of retinal function parameters at home, providing a feasible technical means for large-scale retinal function parameter acquisition and filling a gap in retinal function parameter acquisition in primary healthcare.

[0066] Multimodal data fusion analysis significantly improves the reliability and effectiveness of parameters. This invention establishes a multimodal data fusion analysis model based on the attention mechanism, integrating parameters from different sources such as pupil and eye movement, dynamically adjusting the feature weights of each parameter to achieve adaptive fusion analysis of parameters, and simultaneously verifying the validity of parameters and eliminating invalid parameters. Compared with single-modal parameter analysis, multimodal fusion analysis improves the validity and reliability of parameters by 10-15%, effectively reducing errors in parameter acquisition and analysis, ensuring the accuracy and validity of output parameters, and providing high-quality parameter support for visual function assessment.

[0067] Standardized parameter output enables data to be compared, stored, and shared, thereby improving utilization efficiency. This invention standardizes and visualizes extracted retinal function-related parameters, while supporting the export and long-term storage of parameters in standard formats such as Excel and PDF. The standardized parameter format enables parameter comparisons across different times, devices, and institutions, providing a unified reference standard for dynamic monitoring of retinal function. Furthermore, the standardized parameters can be aggregated and analyzed through a cloud-based data platform, providing large-scale, standardized research data for basic and clinical research related to retinal function, significantly improving the efficiency of parameter utilization.

[0068] The system is highly scalable and adaptable to diverse parameter acquisition and application requirements. The technical solution of this invention has good scalability: the stimulation paradigm can be easily extended to the acquisition of rod cell functional parameters, realizing the collection and analysis of parameters related to dark vision function, thus enriching the scope of parameter acquisition; the multimodal data fusion analysis model adopts an updatable architecture, supporting iterative optimization of the model based on new parameter data, continuously improving the accuracy of parameter analysis; the cloud data platform supports multi-center data aggregation, remote analysis, and standardized report generation, adapting to my country's hierarchical medical system, while also supporting home monitoring and chronic disease management; the device also supports wireless operation and long battery life, adapting to diverse scenarios such as mobile screening and door-to-door testing, with broad application prospects.

[0069] Non-invasive collection, comfortable to wear, improves subject cooperation, and is suitable for long-term dynamic monitoring. This invention uses a VR headset to non-invasively collect pupil and eye movement response signals. There are no corneal contact electrodes or invasive operations, and the subject is comfortable to wear without any discomfort. The entire parameter acquisition process only takes 8-10 minutes, which is short and greatly improves the subject's cooperation during the examination. The device is portable and easy to operate, making it suitable for subjects to monitor retinal function parameters at home for a long time and regularly, and to obtain the trend of parameter changes, so as to provide timely and effective data support for the dynamic assessment and intervention of visual function.

[0070] Example 1: System Hardware Implementation like Figure 1 As shown, this embodiment provides a system for acquiring visual function-related parameters based on selective stimulation of retinal cell types. It is an integrated design for VR headsets, including a VR headset, a signal processing unit, a user terminal, and a cloud data platform.

[0071] The hardware configurations and connections are as follows: 1. VR Headset: Featuring a customized lightweight ergonomic design, weighing only 280g, it comes with a comfortable headband and silicone eye pads for a pressure-free fit. Key components include: Left eye display / Right eye display: OLED microdisplay, resolution 1920×1080, refresh rate 120Hz, peak brightness 1000cd / m², configured with a filter array to achieve narrowband light output with adjustable center wavelength and full width at half maximum (FWHM) of 15nm, which is the core hardware of the stimulus presentation module; Left eye infrared camera unit / right eye infrared camera unit: global shutter CMOS sensor, resolution 640×480, sampling frequency 250Hz, equipped with 850nm infrared LED illumination, is the core hardware of the biosignal acquisition module, which can clearly acquire pupil and eye movement images in dark environments; Optical system: Aspherical lens group, 100° field of view, 15mm exit pupil distance, to ensure clear presentation of stimuli and stability of eye tracking; Local controller: integrates an ARM Cortex-A76 processor, GPU, 16G memory and 128G local storage to realize real-time signal acquisition, preliminary processing and local data storage, and is the local hardware for the signal processing module and data storage module.

[0072] 2. Signal Processing Unit: Connects bidirectionally to the VR headset via Type-C wired / 5G wireless, and is the core processing unit of the system, including: Image processing module: Equipped with U-Net deep learning segmentation algorithm, it processes pupil and eye movement images in real time, extracting pupil diameter change curves and eye movement trajectories; Parameter extraction module: Equipped with parameter extraction algorithm and parameter decoupling model, it calculates pupil dynamic parameters and eye movement response parameters based on processed image data, and decouples them to obtain cell type specific parameters and ratios; Model Analysis Module: Loads a pre-trained multimodal data fusion analysis model, performs quantitative analysis and validity verification on the extracted parameters, and outputs a comprehensive validity score for the parameters; Data storage module: Enables long-term storage of raw image data, intermediate parameters, and standardized parameter sets, supporting local storage and cloud upload; Communication module: Equipped with a Wi-Fi / 5G communication chip, enabling wireless communication with VR headsets, user terminals, and cloud data platforms.

[0073] 3. User terminal: including smartphones, tablets, or personal computers, with a dedicated user interaction application (APP / web version) installed to achieve: Subject basic information entry and test parameter setting; Test operation instructions are issued (start / pause / end / calibrate); View test progress and data quality status in real time; Visualize parameter results and export standardized parameter files; View historical test data to perform parameter trend analysis.

[0074] 4. Cloud Data Platform: Built on cloud servers, equipped with a big data management system and model training engine, to achieve: Aggregation, standardization processing, and storage of multi-center parameter data; Iterative training and updating of the multimodal data fusion analysis model continuously improves model accuracy; Remote analysis of parameters and generation of standardized reports; Data sharing and scientific research data support provide data services for research related to retinal function.

[0075] The connections between the components are as follows: the VR headset communicates bidirectionally with the signal processing unit to transmit signals and issue commands; the signal processing unit communicates bidirectionally with the user terminal to enable interactive operation and result display; the signal processing unit communicates bidirectionally with the cloud data platform to upload data and update models; and the user terminal can directly access the cloud data platform to view historical data and remote analysis reports.

[0076] Example 2: Cell type-selective stimulation generation This embodiment details the generation of cell type-selective stimuli. Based on the L / M cone spectral sensitivity data of Stockman & Sharpe (2000) and the ipRGCs spectral sensitivity data of Dacey et al. (2005), spectral sensitivity functions for each cell type are established. A precise cell type-selective narrowband light stimulation sequence is generated through the stimulation presentation module. The specific steps are as follows: I. Determine the spectral sensitivity function for each cell type Based on existing research data, spectral sensitivity functions for four cell types were established to determine peak sensitivity and spectral range. L-cone cells: peak sensitivity 560nm, full width at half maximum (FWHM) approximately 100nm; M cone cells: peak sensitivity 540nm, full width at half maximum (FWHM) approximately 100nm; S-cone cells: peak sensitivity 460nm, full width at half maximum (FWHM) approximately 60nm; ipRGCs cells: peak sensitivity 480nm, full width at half maximum (FWHM) approximately 150nm.

[0077] II. Selection of narrowband stimulation wavelength and bandwidth Based on the spectral sensitivity function, the stimulation wavelength that minimizes activation of non-target cells was selected, while the full width at half maximum (FWHM) was set to ≤20 nm to ensure cell type selectivity. The final determination was as follows: L-cone cells: 560nm, full width at half maximum (FWHM) 15nm; M cone cells: 540nm, full width at half maximum (FWHM) 15nm; S-cone cells: 460nm, full width at half maximum (FWHM) 15nm; ipRGCs cells: 480nm, full width at half maximum (FWHM) 15nm.

[0078] III. Generating Multimodal Temporal Stimulus Sequences For each cell type, stimulation sequences containing four temporal modulation modes—single pulse, double pulse, flicker, and continuous illumination—are generated to comprehensively acquire functionally relevant parameters across different dimensions. Taking the L-cone cell assessment protocol as an example (total duration approximately 2 minutes): 1. Baseline period: 10 seconds, medium brightness white light (100 cd / m²), used to obtain pupil baseline diameter parameters; 2. Single-pulse stimulation: 560nm narrowband light, pulse width 500ms, repeated 5 times with a 5-second interval, used to obtain transient cellular response characteristics. 3. Dual-pulse stimulation: 560nm narrowband light, pulse width 500ms, two pulses with intervals of 1s, 2s, 3s, and 4s, repeated 3 times each, with an interval of 10 seconds, used to obtain cell function recovery characteristic parameters. 4. Scintillation stimulation: 560nm narrowband light, frequencies of 2Hz, 5Hz, 10Hz and 20Hz, each lasting 5 seconds, with a 5-second interval, used to obtain the cell's time-frequency response characteristics. 5. Continuous light stimulation: 560nm narrowband light for 5 seconds, used to obtain continuous cell response characteristics parameters; 6. Recovery period: 10 seconds, medium brightness white light, used to allow the pupil to return to baseline.

[0079] The evaluation protocol for M cone, S cone, and ipRGCs cells uses the same temporal modulation mode as that for L cone cells, only replacing the corresponding narrowband light wavelength. The complete parameter acquisition protocol includes the evaluation of the four cell types, with a total duration of approximately 8-10 minutes. The test order is randomized to eliminate the influence of the order effect on the parameters.

[0080] The stimulus sequence is generated by the built-in algorithm of the stimulus presentation module and accurately presented through the VR headset's display. During the presentation process, stimulus timestamps are recorded in real time to provide a basis for subsequent signal alignment.

[0081] Example 3: Isoluminance Calibration This embodiment details the isoluminance calibration method, implemented through a luminance calibration module. It employs a heterochromatic scintillation photometric method, using 560nm (L-cone peak) as the reference wavelength to calibrate isoluminance values ​​at 540nm, 480nm, and 460nm. A standardized isoluminance lookup table is established to eliminate interference from the luminance path on parameter acquisition. The specific steps are as follows: I. Calibration Preparation 1. Subjects wore VR headsets and underwent 2 minutes of dark adaptation to ensure that their visual systems were in a stable state; 2. The brightness calibration module calls the stimulus presentation module to present the calibration interface in the VR headset. The center of the interface has two alternately flashing light spots (2° in diameter), which are the reference light and the light to be calibrated, respectively.

[0082] II. Single Wavelength Calibration 1. Set the reference light to a 560nm narrowband light with a fixed brightness of 50cd / m², and the light to be calibrated to a 540nm narrowband light with an initial brightness of 50cd / m². 2. Two lights flash alternately at a frequency of 25Hz. The subject adjusts the brightness of the light to be calibrated using a VR controller until the visual flickering completely disappears. At this point, the perceived brightness of the two lights is determined to be equal. 3. Record the brightness value of the 540nm light at this time, repeat the calibration 3 times, and take the average value as the equal brightness value of the 540nm light relative to the 560nm light; 4. Following the above method, perform calibration of 480nm and 460nm light relative to 560nm light in sequence, and record the isoluminance values ​​for each wavelength.

[0083] III. Establishing an isoluminance lookup table The isoluminance values ​​for each wavelength are organized and a standardized isoluminance lookup table is established. The table contains information such as wavelength, reference luminance, isoluminance value, calibration count, and average value. The lookup table is stored in the luminance calibration module for use during actual parameter acquisition.

[0084] IV. Verification of Calibration Results Twenty normal subjects were selected, and their pupillary responses to white light flickering (without activating the color vision pathway) and colored light flickering at equal brightness at various wavelengths were tested under calibrated isoluminance conditions. The pupillary contraction amplitude parameters were compared. The results showed that the difference in pupillary response amplitude parameters induced by each wavelength after calibration was <5%, proving that the interference of the brightness pathway had been effectively eliminated and ensuring the specificity of the parameter acquisition.

[0085] During actual parameter acquisition, the brightness calibration module automatically calls the isoluminance lookup table and adjusts the actual brightness of each wavelength stimulus according to the reference brightness to ensure that the perceived brightness of all stimuli is consistent.

[0086] Example 4: Acquisition and Parameter Extraction of Pupil Response Signals I. Signal Acquisition Preparation 1. Subjects wear VR headsets, adjusting the headband and eye pads to ensure comfortable wear and no light leakage; 2. Perform dark adaptation for 2 minutes (ambient light < 1 lux) to ensure the retina is in a stable state; 3. The biosignal acquisition module completes camera calibration to ensure clear imaging of pupil and corneal reflections; 4. The stimulus presentation module begins to present cell type-selective stimulus sequences, and the biological signal acquisition module starts simultaneously, acquiring pupil image sequences at a frame rate of 250Hz and transmitting them to the signal processing unit in real time.

[0087] II. Pupil Diameter Extraction The image processing module of the signal processing unit completes the task, employing a U-Net-based deep learning pupil segmentation algorithm. The specific steps are as follows: 1. Image preprocessing: Histogram equalization is performed on each frame of pupil image to improve image contrast; Gaussian filtering (kernel size 3×3) is performed to remove image noise; 2. Pupil region segmentation: The preprocessed image is input into the pre-trained U-Net model, and the model outputs a binary mask of the pupil region to accurately segment the pupil region; 3. Ellipse Fitting: Perform ellipse fitting on the pupil mask, calculate the major and minor axes of the ellipse, and use the equivalent diameter as the pupil diameter. The formula is D=√(a×b), where a is the major axis and b is the minor axis. 4. Confidence assessment: Based on the error of ellipse fitting and the sharpness of the pupil edge, output the pupil tracking confidence score (0-1). A score ≥ 0.8 is considered a valid image.

[0088] The final result is the curve d(t) of pupil diameter changing over time, with a sampling rate of 250Hz, which provides a basis for subsequent parameter extraction.

[0089] III. Counterfeit Handling The image processing module of the signal processing unit automatically identifies and processes artifacts such as blinking and head movements to ensure the accuracy of parameter extraction. The specific method is as follows: 1. Blink artifact recognition and processing: Blinks are recognized based on the pupil diameter change rate. When the diameter suddenly drops to 0 within 100ms and then recovers quickly, it is determined to be a blink. The cubic spline interpolation method is used to fill the missing data during the blink period with the effective data before and after the blink. 2. Head movement artifact recognition and processing: Head movement is recognized based on the change in pupil position in the image. When the position change exceeds 5 pixels, it is determined to be head movement. An image registration algorithm is used to perform translation correction on the image to restore the normal position of the pupil. 3. High-frequency noise removal: Low-pass filtering is applied to the filled and corrected data with a cutoff frequency of 20Hz to remove high-frequency noise and smooth the pupil diameter curve.

[0090] IV. Extraction of Pupil Dynamics Parameters The parameter extraction module of the signal processing unit performs this task. For each stimulus event, based on the pupil diameter change curve d(t), the following core dynamic parameters are extracted (e.g., Figure 4 (as shown) 1. Baseline diameter: The average pupil diameter 500ms before stimulation, denoted as D_baseline; 2. Initial contraction latency: Using the velocity threshold method, when the pupil contraction velocity exceeds 0.5 mm / s, it is determined that the contraction has begun. The time from the start of the stimulus to this point in time is the latency. 3. Amplitude of contraction: The percentage change in pupil diameter D_min relative to the baseline during maximum contraction, calculated as: Amp = (D_baseline - D_min) / D_baseline × 100%; 4. Time to Peak: The time from the start of stimulation to the pupil reaching maximum contraction (D_min); 5. Maximum Velocity: The maximum rate of change of pupil diameter during the contraction phase, measured in mm / s; 6. Redilation Half-time: The time required for the pupil diameter to recover to (D_baseline - D_min) / 2 + D_min from the point of maximum contraction, in seconds; 7. Scintillation Stimulation Gain and Phase Delay: Fourier analysis is performed on the pupil diameter curve under scintillation stimulation. The ratio of output (pupil fluctuation amplitude) to input (stimulus modulation depth) is the gain, and the phase lag of the output relative to the input is the phase delay. 8. Sustained-light contraction maintenance index: The ratio of the average contraction amplitude in the last second of sustained stimulation to the peak contraction amplitude.

[0091] All extracted parameters are stored in the data storage module, providing data for subsequent parameter decoupling and analysis.

[0092] Example 5: Eye Movement Response Signal Acquisition and Parameter Extraction This embodiment details the method for acquiring and extracting eye movement response signals, which is completed collaboratively by a biosignal acquisition module and a signal processing module. This provides supplementary parameters for visual function assessment. The specific steps are as follows: I. Eye Tracking Calibration Before parameter acquisition, the biosignal acquisition module performs nine-point eye-tracking calibration, specifically as follows: 1. Nine calibration points are displayed sequentially in the VR headset (distributed in the center and around the field of vision), with each calibration point lasting for 2 seconds; 2. The subject fixates on each calibration point, and the biosignal acquisition module acquires the positions of the pupil center and corneal reflex point; 3. Based on the position data of 9 calibration points, an eye-tracking mapping model is established to ensure the accuracy of the gaze point coordinate calculation.

[0093] II. Eye Movement Signal Acquisition During cell type-selective stimulation presentation, the biosignal acquisition module is synchronized with pupil signal acquisition, acquiring corneal reflection image sequences and pupil image sequences at a frame rate of 250Hz, and transmitting them to the signal processing unit in real time to record the positional changes of the pupil center and corneal reflection point.

[0094] III. Calculation of gaze point coordinates The image processing module of the signal processing unit completes the calculation of the subject's fixation point coordinates (x(t), y(t)) based on the positional changes of the pupil center and corneal reflection point, using an eye-tracking mapping model, where t is time.

[0095] IV. Extraction of Eye Movement Response Parameters The parameter extraction module of the signal processing unit extracts three core parameters—gaze stability, microscan video rate, and microscan amplitude—based on the gaze point coordinate sequence. The specific method is as follows: (1) Fixation Stability Represented by the area of ​​the bivariate profile ellipse (BCEA), the calculation formula is as follows: BCEA = 2π × k × σ x × σᵧ × √(1 - ρ²) Where, σ x σᵧ is the standard deviation of the horizontal fixation position x(t), σᵧ is the standard deviation of the vertical fixation position y(t), ρ is the correlation coefficient between x(t) and y(t), and k is a constant (k=1.14 when the coverage probability is 68.2%). The smaller the BCEA value, the better the fixation stability.

[0096] (2) Micro-scan video rate and micro-scan amplitude The micro-scanning detection algorithm based on velocity threshold is adopted, and the specific steps are as follows: 1. Calculate the instantaneous velocity of the fixation point: Based on the 3-point difference method, calculate the fixation point velocity v(t) at each time point. The formula is v(t)=√[(x(t+1)-x(t-1))² + (y(t+1)-y(t-1))²] / (2Δt), where Δt is the sampling interval (4ms). 2. Microscan start determination: When the velocity v(t) exceeds 5 times the median velocity and the duration is ≥6ms, microscan is determined to have started. 3. Microscanning termination determination: When the velocity v(t) drops to less than 5 times the median velocity, the microscanning is determined to have ended; 4. Parameter calculation: Count the number of micro-saccades per unit time, which is the micro-saccade rate (times / s); calculate the distance the gaze point moves during each micro-saccade, and take the average value as the micro-saccade amplitude (degrees).

[0097] All extracted eye movement response parameters are stored in the data storage module and integrated with pupil dynamic parameters to provide data for subsequent multimodal fusion analysis.

[0098] Example 6: Decoupling Calculation of Cell Type-Specific Parameters This embodiment details the decoupling calculation method for cell type-specific parameters, which is completed by the parameter extraction module of the signal processing unit. By establishing a linear parameter decoupling model, specific functional parameters of four cell types—L cones, M cones, S cones, and ipRGCs—are decoupled from the pupillary response parameters of multi-wavelength stimulation. The specific steps are as follows: I. Establishing a linear parameter decoupling model Let L, M, S, and ip represent the contributions (normalized to 0-1) of four cell types (L cone, M cone, S cone, and ipRGCs) to the pupillary response, respectively. For a stimulus of any wavelength λ, the induced pupillary response parameter R(λ) can be expressed as a weighted sum of the contributions of each cell type, as shown in the model formula: R(λ) = w_L(λ) × L + w_M(λ) × M + w_S(λ) × S + w_ip(λ) × ip + ε Where w_L(λ), w_M(λ), w_S(λ), and w_ip(λ) are the relative activation weights of each cell type at wavelength λ, calculated based on the spectral sensitivity function of each cell type; ε is the error term, representing the influence of non-target factors (such as ambient light and equipment noise) on pupil response, ε≈0.

[0099] II. Constructing an overdetermined system of equations This invention employs narrowband light stimulation at five wavelengths (560nm, 540nm, 460nm, 480nm, and 500nm) to obtain pupil constriction amplitude parameters R(λ1), R(λ2), R(λ3), R(λ4), and R(λ5) at these five wavelengths. Based on the relative activation weights at each wavelength, an overdetermined set of equations is constructed: [R(λ1), R(λ2), R(λ3), R(λ4), R(λ5)]ᵀ = W × [L, M, S, ip]ᵀ Wherein, W is a 5×4 weight matrix, and the matrix elements are the relative activation weights of each cell type at each wavelength, which are calculated by the spectral sensitivity function. An example of the weight matrix is ​​shown below (for illustration only).

[0100] Table 1 Weight Matrix

[0101] III. Solving for the least squares solution The above overdetermined system of equations is solved using the nonnegative least squares (NNLS) method to obtain estimates of L, M, S, and ip, ensuring the nonnegativity of the solution (the cell contribution value cannot be negative).

[0102] IV. Calculation of standardized cell type-specific parameters The obtained L, M, S, and ip values ​​are normalized, and standardized cell type-specific function-related parameters are calculated using the following formula: LFI = L / (L+M+S+ip) MFI = M / (L+M+S+ip) SFI = S / (L+M+S+ip) ipFI = ip / (L+M+S+ip) Wherein, LFI is the functional parameter of L cone cells, MFI is the functional parameter of M cone cells, SFI is the functional parameter of S cone cells, and ipFI is the functional parameter of ipRGCs cells. All parameter values ​​are normalized to 0-1. The closer the parameter value is to 1, the better the functional status of the corresponding cell type.

[0103] V. Calculation of parameter ratios and binocular asymmetry index 1. Cell type parameter ratios: Based on standardized parameters, calculate L / S Ratio = LFI / SFI and L / ipRatio = LFI / ipFI to provide multi-dimensional references for visual function assessment; 2. Binocular asymmetry index: Calculate the LFI, MFI, SFI, and ipFI of the left and right eyes respectively, and calculate the binocular asymmetry index. The formula is: Binocular asymmetry index = (left eye parameter - right eye parameter) / average value of both eyes, which is used for unilateral retinal function assessment.

[0104] All decoupled parameters are stored in the data storage module and integrated with pupil dynamics parameters and eye movement response parameters to form a multi-dimensional parameter set, providing data for subsequent multimodal fusion analysis.

[0105] Example 7: Construction and Parameter Validation of Multimodal Data Fusion Analysis Model This embodiment details the construction method and parameter validity verification process of the multimodal data fusion analysis model, which is completed by the model analysis module of the signal processing unit. The model is built based on an attention mechanism to achieve multi-dimensional parameter fusion analysis and validity verification. The specific steps are as follows: I. Constructing a multi-dimensional feature set Forty-seven feature parameters were extracted from the parameter data of each subject and used as input features for the model. The feature set was divided into two categories: pupil feature parameters and eye movement feature parameters, specifically including: 1. Pupil characteristic parameters (32 dimensions): LFI, MFI, SFI, ipFI, L / S Ratio, L / ip Ratio (6 dimensions); pupil dynamic parameters under 5 wavelength stimulations (latency, amplitude, time to peak, maximum velocity, half-life, 5×5=25 dimensions); gain and phase delay of 4 frequency flicker stimulations (4×2=8 dimensions); continuous illumination contraction maintenance index (1 dimension). 2. Eye movement characteristic parameters (15 dimensions): fixation stability BCEA under 3 stimulus conditions (3 dimensions); microsaccade rate under 3 stimulus conditions (3 dimensions); microsaccade amplitude under 3 stimulus conditions (3 dimensions); fixation point distribution characteristics (horizontal / vertical range, standard deviation, skewness, kurtosis, 6 dimensions).

[0106] II. Constructing a Multimodal Fusion Network Model Based on Attention Mechanism Model architecture such as Figure 8 As shown, it consists of four layers: input layer, modality coding layer, attention fusion layer, and analysis layer. The specific structure is as follows: 1. Input layer: 47-dimensional multi-dimensional feature parameters are input, divided into pupil feature branch (32-dimensional), eye movement feature branch (15-dimensional), and cell type parameter branch (10-dimensional). 2. Modality Coding Layer: Three independent fully connected sub-networks encode the features of the three branches respectively, mapping pupil features from 32-dimensional to 64-dimensional, eye movement features from 15-dimensional to 32-dimensional, and cell type parameters from 10-dimensional to 32-dimensional. The activation function used in all layers is ReLU to achieve non-linear transformation of features. 3. Attention Fusion Layer: The attention weights of the three modal coding features are calculated using the formula: α_i = softmax(W_i × h_i + b_i), where i∈{pupil, eye movement, cell type}, W_i is the weight matrix, h_i is the coding feature, and b_i is the bias term. The coding features of the three modalities are weighted and fused based on the attention weights to obtain a 128-dimensional fused feature. 4. Analysis layer: The 128-dimensional fused features are input into the fully connected layer, first mapped to 64 dimensions, and a Dropout layer (dropout=0.5) is added to prevent overfitting. Then it is mapped to 1 dimension and the output parameter comprehensive effectiveness score (0-1) is given. The higher the score, the stronger the reliability and effectiveness of the parameters.

[0107] III. Model Training Using parameter data from 100 normal subjects as the training set, 5-fold cross-validation training was performed, with the training parameters set as follows: 1. Loss function: Mean squared error loss (MSE), with the optimization objective being the effectiveness of the parameters; 2. Optimizer: Adam optimizer, learning rate set to 0.001, weight decay to 1e-5; 3. Batch size: 16; 4. Number of training rounds: 100 rounds; 5. Early stopping mechanism: When the validation loss does not decrease for 10 consecutive rounds, stop training, save the optimal model, and avoid model overfitting.

[0108] IV. Parameter Validity Verification The multi-dimensional parameter set to be validated is input into the trained model, and the model outputs a comprehensive parameter effectiveness score, with a score threshold of 0.8. 1. If the score is ≥0.8, it is considered a valid parameter and included in the final standardized parameter set; 2. If the score is <0.8, it is determined to be an invalid parameter, and the quality control system will prompt to repeat the data collection to ensure the reliability of the output parameters.

[0109] This model enables the fusion analysis and effectiveness screening of multi-dimensional parameters, thereby improving the overall quality of the parameters.

[0110] Example 8: Parameter Standardization Output and Application This embodiment details the standardized output format and practical application scenarios of retinal function-related parameters. It is achieved collaboratively by a parameter output module, a user interaction module, and a cloud data platform, enabling the visualization, standardized export, and multi-scenario application of these parameters. The specific content includes: I. Standardized Parameter Output Format The parameter output module standardizes the validated parameters and provides both visual display and standardized export formats to suit different usage needs. (1) Visualization The display interface is integrated into the interactive applications of the VR headset and user terminal, and includes four forms: 1. Numerical Reports: The specific values ​​of all parameters are presented in tabular form, including pupillary dynamics parameters, oculomotor response parameters, cell type-specific parameters, parameter ratios, binocular asymmetry index, and reference ranges for normal populations and validity scores for the parameters. 2. Trend Curve: With time as the horizontal axis, it presents the curve of pupil diameter change and fixation point trajectory curve, intuitively showing the dynamic response characteristics of pupil and eye movement; 3. Comparison Charts: The bar charts show the comparison of specific parameters for the four cell types, and the scatter plots show the distribution of parameter ratios, providing a visual representation of the functional status of different cell types. 4. Heat map: Displays the distribution of fixation points of both eyes in the form of a heat map, reflecting the state of fixation function in the macular area.

[0111] (2) Standardized export It supports exporting parameter data in two standard formats: Excel and PDF. The exported content includes: 1. Basic information of the subject: name, gender, age, test time, device number, etc.; 2. Raw data: pupil diameter sequence, fixation point coordinate sequence, etc.; 3. Post-processing parameters: All extracted quantitative parameters and their values; 4. Effectiveness assessment: Parameter comprehensive effectiveness score, and determination of the effectiveness of each indicator; 5. Reference range: Parameter reference range for normal population.

Claims

1. A system for acquiring visual function-related parameters based on retinal cell type, characterized in that: S1, Stimulation Presentation Module: Generates a narrowband light stimulation sequence with cell type selectivity based on the spectral sensitivity function of the target retinal cell type; includes a high-resolution display integrated into the VR headset, with the core configuration being an OLED microdisplay; S2, Brightness Calibration Module: The perceived brightness of different narrowband light stimuli is calibrated by using the heterochromatic flicker photometric method, and an isoluminance lookup table is established; S3, Biosignal Acquisition Module: Synchronously acquires image sequences and timestamps through a high-speed camera unit integrated into the VR headset; the acquired raw signal data includes three types: pupil image sequence, corneal reflection image sequence, and stimulus presentation timestamp; It includes at least one infrared camera unit integrated into the VR headset, preferably a binocular infrared camera unit, with a core configuration of a global shutter CMOS sensor; S4. Multidimensional parameter extraction and cell type-specific parameter decoupling module: Extracts kinetic parameters and reaction feature parameters from image sequences, establishes a parameter decoupling model, and solves for cell type-specific functional parameters and parameter ratios; This module is the core processing unit of the system, integrating an embedded processor, an image processing unit, and a dedicated algorithm chip. S5, Parameter Analysis and Standardized Output Module: Inputs response parameters, specific parameters, and parameter ratios into a pre-trained multimodal data fusion analysis model, and outputs a standardized set of retinal functional parameters after analysis and verification; including support vector machines, random forests, gradient boosting trees, and deep neural networks; S6, User Interaction Module: Configured to receive user input, issue operation commands, and provide real-time feedback on the collection status; includes VR headset controllers, voice interaction interface, and dedicated interactive applications for external terminals; S7, Data Storage Module: Configured to store raw image data, intermediate parameters, and standardized parameter sets, supporting local data storage and remote transmission; the core configuration consists of a local storage unit and a cloud communication module.

2. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: The narrowband photostimulation sequence parameters include: L cone cell stimulation center wavelength 560nm±15nm, M cone cell stimulation center wavelength 540nm±15nm, S cone cell stimulation center wavelength 460nm±15nm, ipRGCs stimulation center wavelength 480nm±15nm, and full width at half maximum (FWHM) ≤20nm.

3. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: The sampling frequency of the high-speed camera unit is no less than 120Hz, and the spatial resolution is no less than 640×480 pixels. The high-speed camera unit synchronous acquisition adopts binocular synchronous acquisition, acquiring the reaction image sequences of the left eye and the right eye respectively.

4. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: The kinetic parameters include one or more of the following: initial contraction latency, contraction amplitude, contraction peak time, maximum contraction velocity, re-expansion half-life, scintillation stimulus gain, scintillation stimulus phase delay, and continuous illumination contraction maintenance index. The response characteristic parameters include one or more of the following: gaze stability, microscan video rate, and microscan visual amplitude, expressed as the area of ​​a bivariate contour ellipse.

5. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: The parameter decoupling model is a linear model, with the contributions of L-cones, M-cones, S-cones, and ipRGCs to the pupillary response as unknowns, and pupillary response parameters under multi-wavelength stimulation as observed values. The contribution weights of each cell type are solved by non-negative least squares method to obtain standardized cell type-specific function-related parameters, including L-cone cell function parameters, M-cone cell function parameters, S-cone cell function parameters, ipRGCs cell function parameters, and parameter ratios L-cone / S-cone parameter ratio and L-cone / ipRGCs parameter ratio. It also includes the calculation of the binocular parameter asymmetry index, where the binocular parameter asymmetry index is the left and right eye parameters divided by the binocular average.

6. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: The multimodal data fusion analysis model is constructed using machine learning algorithms, including one of support vector machines, random forests, gradient boosting trees, and deep neural networks. The model optimizes parameter effectiveness and stability, outputs a comprehensive parameter effectiveness score, sets a score threshold to filter effective parameters to form a standardized parameter set, which is presented in the form of numerical values, trend curves, and ratio comparison charts.

7. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: Narrow-band light stimulation includes selective stimulation and parameter acquisition of rod cells. Narrow-band light with a center wavelength of 500nm±15nm is used to extract functional parameters of rod cells.

8. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: The multi-dimensional parameter extraction and cell type-specific parameter decoupling module, i.e. the signal processing module, includes: a quality control system that monitors signal acquisition quality in real time, automatically identifies and processes blinking and head movement artifacts, uses cubic spline interpolation to fill missing data, low-pass filtering to remove high-frequency noise, and automatically prompts for repeated acquisition when data quality is substandard.

9. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: The data storage module is configured to receive parameter data, enabling multi-center data aggregation, model iterative updates, remote parameter analysis, and standardized report generation.

10. The system for acquiring visual function-related parameters based on retinal cell type according to claim 1, characterized in that: The VR headset features a lightweight, ergonomic design, weighing ≤300g, ensuring comfortable wear. It requires no professional technicians to operate and supports convenient parameter acquisition.

Images