Control system and control method for a corneal cross-linking device

Abstract

This invention relates to the field of medical device control technology, specifically to a control system and method for a corneal cross-linking device. The control system includes: a multimodal data fusion module configured to integrate preoperative corneal topography curvature matrix, OCT layered thickness data, and biomechanical parameters to generate a four-channel fusion tensor; a virtual corneal modeling module configured to use the four-channel fusion tensor as a condition to input a conditional generative adversarial network to generate a virtual corneal model; a dynamic energy planning module configured to predict the mechanical response based on the virtual corneal model and generate a heatmap of energy distribution; and an irradiation control module configured to perform patterned irradiation based on the heatmap. This application, by constructing a virtual corneal model and generating an energy distribution heatmap, can provide accurate patterned references for irradiation therapy, achieving better treatment results.

Description

Technical Field

[0001] This invention relates to the field of medical device control technology, specifically to a control system and control method for a corneal cross-linking device. Background Technology

[0002] With the increasing number of myopia patients, corneal cross-linking (CCL) has gradually gained attention as an effective treatment. CCL is a key technology for treating corneal ectasia diseases such as keratoconus. Its core principle is to activate riboflavin through ultraviolet (UVA) irradiation, inducing cross-linking of corneal collagen fibers to enhance mechanical strength. It is primarily performed using corneal cross-linking devices. Corneal cross-linking (CXL) devices are medical devices used to treat corneal diseases such as keratoconus, and their core function is to enhance the mechanical strength of corneal collagen fibers through photochemical effects. However, existing technologies have the following limitations: Insufficient parameter matching: Traditional cross-linking devices use linear parameters (such as a fixed UVA energy / time ratio), which cannot adapt to the asymmetric steep areas and uneven thickness distribution of keratoconus. Delayed risk control: Existing systems rely on intraoperative OCT monitoring, which cannot predict the biomechanical response under different energy distributions, resulting in the risk of postoperative corneal ectasia or perforation (clinical data shows a perforation rate >3% with traditional methods). Data utilization is limited: Preoperative examination data (such as Corvis ST biomechanical parameters) are only used for contraindication screening and are not dynamically correlated with irradiation parameters.

[0003] Therefore, existing technologies still need improvement. Summary of the Invention

[0004] In view of this, this application proposes a control system and control method for a corneal cross-linking device. The main purpose is to overcome the deficiencies of the above-mentioned background technology, and to generate a thermal map of energy distribution by constructing a virtual corneal model, so as to provide a precise patterned reference for irradiation treatment by the corneal cross-linking device and achieve better treatment results.

[0005] The first aspect of this invention provides a control system for a corneal crosslinking device, comprising a multimodal data fusion module, a virtual corneal modeling module, a dynamic energy planning module, and an irradiation control module. The multimodal data fusion module is configured to integrate preoperative corneal topography curvature matrix, OCT layered thickness data, and biomechanical parameters to generate a four-channel fusion tensor. The virtual corneal modeling module is configured to use the four-channel fusion tensor as a condition, inputting a conditional generative adversarial network to generate a virtual corneal model. The dynamic energy planning module is configured to generate a heatmap of energy distribution based on the mechanical response prediction of the virtual corneal model. The irradiation control module is configured to perform patterned irradiation based on the heatmap.

[0006] In some embodiments, the multimodal data fusion module includes a curvature processing unit, an OCT analysis unit, a biomechanical mapping unit, and a feature fusion unit. The curvature processing unit is configured to perform Gaussian filtering and normalization on the curvature matrix of the preoperative corneal topography map to generate a standardized curvature tensor. The OCT analysis unit is configured to segment the corneal epithelium, stroma, and endothelium in the OCT volumetric data using a pre-trained neural network, calculate the stroma thickness distribution map, and register it to the coordinate system of the curvature tensor. The biomechanical mapping unit converts the detected biomechanical parameters into a biomechanical constraint matrix, the biomechanical parameters including at least the amplitude of the first flattening deformation and the radius of curvature of the corneal concavity. The feature fusion unit is configured to concatenate the curvature tensor, the stroma thickness distribution map, and the biomechanical constraint matrix along the channel dimension to form a four-channel fusion tensor, and perform standardization processing on each channel.

[0007] In some embodiments, the feature fusion unit further includes a risk region enhancement submodule, which is configured to: identify pixel regions that meet the conditions based on a preset normalized thickness threshold, a normalized curvature threshold, and a first flattening deformation amplitude mapping threshold; and multiply the fusion tensor feature value corresponding to the pixel region by an enhancement coefficient.

[0008] In some embodiments, the dynamic energy planning module is configured to: apply an energy attenuation coefficient to regions that meet the stromal thickness threshold in the virtual corneal model; generate a spot pattern with a spot diameter of 50~100μm and a spot density in the central region that is 1.5~2 times that in the edge region.

[0009] In some embodiments, the control system of the corneal cross-linking device further includes a dynamic optimization module configured to: monitor riboflavin fluorescence intensity in real time; and update the irradiation strategy when the local riboflavin fluorescence intensity is lower than the average of the entire field by more than a preset ratio.

[0010] In some embodiments, the control system of the corneal crosslinking device further includes a motion compensation module, configured to: track pupil edge feature points and calculate eyeball deflection vector; establish an eyeball-robotic arm motion model based on the eyeball deflection vector; drive the lifting track and the robotic arm to complete synchronous offset within a preset time based on the eyeball-robotic arm motion model; and ensure that the projection position error of the light spot is ≤50μm through closed-loop control.

[0011] In some embodiments, the control system of the corneal cross-linking device further includes a three-level safety circuit breaker module, which comprises a primary protection unit, a secondary protection unit, and a tertiary protection unit. The primary protection unit is configured to detect UVA light density fluctuations in real time and trigger automatic calibration when the fluctuation amplitude exceeds a fluctuation threshold. The secondary protection unit is configured to cut off the UVA light source and lock the robotic arm travel when the distance sensor detects that the probe-cornea distance is less than a distance threshold. The tertiary protection unit is configured to automatically activate the energy attenuation mode in the local area when intraoperative OCT detects a local reduction in stromal thickness greater than a preset ratio, with an attenuation coefficient of 0.3 to 0.5.

[0012] In some embodiments, the control system of the corneal crosslinking device further includes an intraoperative adjustment module configured to: acquire the actual stromal layer thickness distribution detected by intraoperative OCT at preset time intervals, and adjust the applied energy when the deviation between the actual stromal layer thickness and the predicted stromal layer thickness of the virtual corneal model exceeds a threshold.

[0013] In some embodiments, the control system of the corneal crosslinking device further includes a postoperative assessment module configured to: compare changes in corneal biomechanical index before and after surgery; and generate a secondary treatment recommendation report when the corneal biomechanical index decreases less than an index threshold after a preset time postoperatively.

[0014] A second aspect of the present invention also provides a control method for a corneal crosslinking device, implemented through the control system described in any of the above embodiments, comprising the following steps: integrating preoperative corneal topographic curvature matrix, OCT layer thickness data, and biomechanical parameters to generate a four-channel fusion tensor; using the four-channel fusion tensor as a condition, inputting a conditional generative adversarial network to generate a virtual corneal model; generating a thermal map of energy distribution based on the mechanical response prediction of the virtual corneal model; and performing patterned irradiation based on the thermal map.

[0015] The beneficial effects of this invention are as follows: This application, through multimodal data fusion and virtual corneal modeling, can generate a highly accurate virtual corneal model, providing precise basis for irradiation treatment using corneal cross-linking devices. In particular, the enhanced modeling of the steep infratemporal quadrant, a high-incidence area of ​​keratoconus, can effectively improve the targeting and success rate of treatment. By generating a heat map of energy distribution, a precise reference is provided for irradiation, ensuring the accuracy of energy distribution and effectively avoiding over- or under-irradiation, thus improving the safety and effectiveness of treatment. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a control system module diagram of a corneal cross-linking device according to an embodiment of the present invention;

[0018] Figure 2 This is a flowchart of a multimodal data fusion process proposed in one embodiment of the present invention;

[0019] Figure 3 This is a flowchart illustrating the generation process of a virtual corneal model according to an embodiment of the present invention;

[0020] Figure 4 This is a block diagram of a robotic arm motion compensation control according to an embodiment of the present invention;

[0021] Figure 5 This is a flowchart of a control method for a corneal cross-linking device according to an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to specific examples and the accompanying drawings.

[0023] The first aspect of this invention provides a control system for a corneal cross-linking device. This control system employs a distributed architecture, consisting of a host computer system deployed on a Windows tablet and a slave computer system integrated into the treatment device control unit. The host computer system is responsible for multi-dimensional fusion analysis, constructing a virtual corneal model through a conditional generative adversarial network (GAN), and generating a UVA energy distribution heatmap. It also provides a touch-screen interface for setting treatment parameters, data visualization, and managing encrypted treatment records. The slave computer system carries an irradiation control module that receives commands from the host computer via a CAN bus, precisely controlling the irradiation intensity and duration of the UVA light source in real time, while simultaneously feeding back actual output data. The system employs a dual-redundancy safety mechanism, automatically maintaining safe power during communication interruptions to ensure the safety and reliability of the treatment process. This architecture fully leverages the advantages of the host computer in complex calculations and visualization, while utilizing the slave computer to achieve microsecond-level real-time control, meeting the stringent requirements for precision and stability in corneal cross-linking treatment.

[0024] Specifically, such as Figure 1As shown, the control system of the corneal cross-linking device includes a multimodal data fusion module, a virtual corneal modeling module, a dynamic energy planning module, and an irradiation control module. The multimodal data fusion module is configured to integrate preoperative corneal topography curvature matrix, OCT layered thickness data, and biomechanical parameters to generate a four-channel fusion tensor (channel dimensions: topography curvature, thickness gradient, biomechanical intensity, and lesion probability mask, where the lesion probability mask is derived data generated through the collaborative analysis of the first three channels and used to enhance the weight of the lesion region). The virtual corneal modeling module is configured to use a conditional generative adversarial network (cGAN) architecture, taking the four-channel fusion tensor as conditional input, to generate a virtual corneal model with enhancement features of the lesion region (especially the steep area of ​​the infratemporal quadrant). The dynamic energy planning module is configured to predict the mechanical response based on a virtual corneal model. It performs voxel-level prediction using a 3D CNN decoder to generate a heatmap of UVA energy distribution. The spatial resolution of the heatmap is no less than 50μm×50μm×10μm, prioritizing enhancement of the steep region in the infratemporal quadrant, which is a high-incidence area for keratoconus. The irradiation control module is configured to perform patterned irradiation based on the heatmap using a digital micromirror device (DMD). The control logic is as follows: the heatmap is converted into a DMD bitmap (using the Floyd-Steinberg dithering algorithm), and the irradiation pattern is updated every 400ms.

[0025] This application utilizes multimodal data fusion and virtual corneal modeling. The control system can generate highly accurate virtual corneal models, particularly for enhanced modeling of the steep infratemporal quadrant, a common site of keratoconus. This provides precise guidance for treatment, improving lesion localization accuracy by 40%, effectively enhancing treatment targeting and success rates. By generating high spatial resolution thermal maps, the accuracy of energy distribution is ensured, effectively avoiding over- or under-irradiation and improving treatment safety and effectiveness. Through patterned irradiation using a digital micromirror device (DMD), the control system can rapidly and accurately deliver UVA energy to various parts of the cornea, effectively shortening treatment time and improving treatment efficiency.

[0026] In some embodiments, the multimodal data fusion module is used to achieve accurate fusion of corneal multimodal data. The multimodal data fusion module includes a curvature processing unit, an OCT analysis unit, a biomechanical mapping unit, and a feature fusion unit. The curvature processing unit is configured to perform Gaussian filtering (σ=1.5) on the corneal topography curvature matrix to eliminate motion artifacts caused by eye movement or device noise, while preserving curvature abrupt changes (such as the steep gradient in the inferotemporal quadrant of keratoconus) and normalizing it to the [-1,1] interval, generating a 512×512 pixel curvature tensor. Patient-adaptive curvature normalization avoids population statistical bias, making the features of steep regions more prominent. The OCT analysis unit is configured to automatically segment the corneal epithelium, stroma, and endothelium in OCT volumetric data using a pre-trained U-Net neural network, outputting the boundary coordinates of each layer. Based on the segmentation results, it calculates the stromal thickness distribution map (unit: μm) and registers it to the curvature tensor coordinate system through affine transformation, ensuring spatial alignment with a registration error controlled within 50 μm (meeting the ISO 24294 ophthalmic image alignment standard). The biomechanical mapping unit is configured to convert the detected biomechanical parameters into a 512×512 pixel biomechanical constraint matrix, maintaining the same spatial resolution as the curvature data. Biomechanical parameters include, for example, the first planarization deformation amplitude (SP-A1, unit: mm) and the radius of curvature of the corneal concavity point (HC radius, unit: mm) obtained from Corvis ST detection. The feature fusion unit is configured to stitch together the curvature tensor, the matrix thickness distribution map, the biomechanical constraint matrix and the lesion probability mask along the channel dimension to form a four-channel fusion tensor (i.e., a 512×512×4-dimensional fusion tensor), and perform Z-score normalization on each channel.

[0027] This application combines corneal topography curvature information, OCT tomographic structural information, biomechanical parameters, and lesion probability masks to provide a more comprehensive assessment of corneal condition, avoiding the limitations of single-modality data. The generation process of the four-channel fusion tensor is as follows: Figure 2 As shown.

[0028] In some embodiments, the feature fusion unit further includes a risk region enhancement submodule for automatically identifying and enhancing high-risk corneal lesion areas, such as the infratemporal quadrant of keratoconus. The risk region enhancement submodule is configured to: identify eligible pixel regions based on preset normalized thickness thresholds, normalized curvature thresholds, and SP-A1 mapping thresholds; and multiply the fusion tensor feature value corresponding to the pixel region by an enhancement coefficient.

[0029] Specifically, for example, the preset normalized thickness threshold is -0.8. When the normalized thickness value is less than -0.8, the corresponding physical thickness is <400μm; the preset normalized curvature threshold is 0.6. When the normalized curvature value is greater than 0.6, it is significantly steep, which is a typical characteristic of keratoconus; the SP-A1 mapping threshold is 0.25. When the SP-A1 mapping value is greater than 0.25, the mechanical properties are weak and the risk of deformation is high.

[0030] The fusion tensor feature values ​​corresponding to pixel regions that meet the above conditions are multiplied by an enhancement factor of 1.2 to 1.8, giving them a higher weight in subsequent virtual modeling. The enhanced features significantly affect the attention distribution and gradient distribution of the energy heatmap in the cGAN generator, resulting in improved generation resolution in high-risk areas and an automatic increase in UVA energy density of 15 to 30% in these areas.

[0031] In some embodiments, the dynamic energy planning module is configured to: apply an energy attenuation coefficient to regions that meet the stromal thickness threshold in the virtual corneal model; generate a spot pattern with a spot diameter of 50-100 μm and a spot density in the central region that is 1.5-2 times that in the peripheral region. The spots can be, for example, honeycomb-shaped or gradient ring-shaped spots.

[0032] In this model, regions meeting the stromal thickness threshold (e.g., stromal thickness < 450 μm) are assigned an energy attenuation coefficient k using the formula: k = 0.9 - 1.3e^{-0.01d}, where d is the Euclidean distance from the current voxel to the thinnest point (in μm). For example, at the thinnest point (d = 0): k ≈ 0.3 (energy reduced to 30%, avoiding perforation); d = 100 μm: k ≈ 0.6 (energy restored to 60%); d ≥ 300 μm: k approaches 0.9 (retaining a 10% attenuation margin). The automatic response intraoperative dynamic energy planning module can adjust the attenuation coefficient in real time based on thickness changes detected by OCT.

[0033] Specifically, a flowchart for generating a virtual corneal model according to an embodiment of this application can be referred to. Figure 3 .

[0034] In some embodiments, the control system further includes a dynamic optimization module configured to: monitor riboflavin fluorescence intensity in real time; and update the irradiation strategy when the local riboflavin fluorescence intensity is lower than the average of the entire field area by more than a preset proportion. Optionally, the riboflavin fluorescence intensity can be monitored in real time using an embedded spectral sensor, converting the light intensity signal into a two-dimensional distribution map. The embedded spectral sensor has a detection wavelength of 365nm±5nm (riboflavin excitation peak), a signal-to-noise ratio >50dB, a spatial resolution of 100μm, a sampling rate of 10Hz, and covers an irradiation area with a diameter of 6mm. For example, when a local riboflavin fluorescence intensity is detected to be lower than the average of the entire field area by more than 30%, the local area is determined to be a low-response area, and the irradiation strategy can be updated using a Q-learning algorithm. For low-response areas, irradiation strategies can be updated by increasing energy density (up to +30%), extending irradiation time (up to +15s), or switching the spot pattern (e.g., from honeycomb pattern to spiral scan).

[0035] In some embodiments, the control system further includes a motion compensation module, which solves the core problem of spot drift caused by intraoperative micro-movements of the eyeball through a sub-millisecond dynamic tracking-compensation system, improving error control from the millimeter level to the micrometer level. For example... Figure 4 As shown, the motion compensation module is configured to: track pupil edge feature points and calculate eyeball deflection vector; establish an eyeball-robotic arm motion model based on the eyeball deflection vector; drive the lifting track of the corneal crosslinking device and the robotic arm to complete synchronous offset within a preset time based on the eyeball-robotic arm motion model; and ensure that the projection position error of the light spot is ≤50μm through closed-loop control.

[0036] Specifically, a PID control loop can be used for closed-loop control. In one embodiment, the motion compensation module includes a vision unit, such as a 200fps infrared camera (1280×1024 resolution), combined with an 850nm ring light source, to track 6-8 feature points at the edge of the pupil. The infrared camera acquires the coordinates of the pupil feature points every 5ms, and calculates the eyeball deflection vector (accuracy ±0.02°) through ellipse fitting; based on Δx eye Deviation from the target position (initial calibration point) e(t) Real-time update of PID output signal u(t) .Will u(t) Converted to joint torque command (via Jacobian matrix) J −1 The system maps the X / Y / θ axis motors to complete the compensation motion within 50ms. The CNC lifting track (Z-axis) compensates for longitudinal displacement, while the rigid robotic arm (X / Y / θ) compensates for lateral offset and rotation, adapting to complex eye movements (such as the Bell phenomenon). The next frame of camera data verifies the residual error after compensation; if it exceeds 50μm, a secondary correction (adaptive PID) is triggered.

[0037] In some embodiments, the control system further includes a three-level safety circuit breaker module, comprising a primary protection unit, a secondary protection unit, and a tertiary protection unit. The primary protection unit is configured to detect UVA light density fluctuations in real time and trigger automatic calibration when the fluctuation amplitude exceeds a fluctuation threshold (e.g., ±10%). For example, a UV light density sensor (wavelength 365nm) can be used to sample at a frequency of 1kHz to detect fluctuations; automatic calibration can employ closed-loop PID calibration, which can restore the light intensity to the set value within 200ms, avoiding insufficient energy or overexposure. The secondary protection unit is configured to cut off the UVA light source and lock the robotic arm stroke when the distance sensor detects that the distance between the probe and cornea of ​​the corneal cross-linking device is less than a distance threshold (e.g., <5mm). The probe-cornea distance can be detected, for example, using an infrared TOF sensor (accuracy ±0.1mm). When the probe-cornea distance is detected to be <5mm, a collision risk is identified, at which point the secondary protection unit cuts off the light source and locks the robotic arm stroke. The robotic arm's travel is controlled by an electromagnetic brake, which locks the arm's travel within 10ms upon receiving a signal. The Level 3 protection unit is configured to immediately trigger the digital micromirror device (DMD) to dynamically attenuate the UV energy in the abnormal area by 0.3–0.5 (reducing the 365nm UV energy from the standard 5.4 J / cm² to 1.6–2.7 J / cm²) when intraoperative OCT detects a local reduction in stromal thickness exceeding a preset percentage (e.g., 10%). Simultaneously, it activates the balancing fluid infusion system to promote corneal recovery to a safe threshold. The OCT can update the full corneal thickness map every 30 seconds to detect local dehydration or excessive thinning.

[0038] This three-level safety circuit breaker module upgrades passive alarms to active protection, improving clinical safety by 90%. Through a tiered and progressive protection mechanism, it addresses multi-dimensional safety risks during corneal cross-linking surgery.

[0039] In some embodiments, the control system further includes an intraoperative adjustment module, which addresses the planned-actual thickness deviation caused by corneal dehydration, uneven riboflavin penetration, etc., during surgery through real-time data-driven dynamic energy regulation. Specifically, the intraoperative adjustment module is configured to: acquire the actual thickness distribution T_real(t) detected by intraoperative OCT at preset time intervals (e.g., 60 seconds); and adjust the applied energy according to the following formula when the deviation between the actual thickness distribution T_real(t) and the thickness T_model predicted by the virtual corneal model exceeds a threshold:

[0040] ΔE=0.5×|T_{real}(t)-T_{model}| / T_{model}\quad (J / cm²)

[0041] When ΔE > 0.3 J / cm², the energy distribution is recalculated.

[0042] The coefficient 0.5 in the formula was determined through finite element optimization to balance response speed and stability.

[0043] Through the settings of the above-mentioned intraoperative adjustment module, the energy can be automatically increased by 7.5% (ΔE=0.375 J / cm²) for thickness reduction caused by dehydration (e.g., 400μm→370μm); and the energy can be intelligently reduced to avoid over-irradiation for abnormally thickened areas (e.g., edema).

[0044] In some embodiments, the control system further includes a postoperative assessment module configured to: compare changes in corneal biomechanical index before and after surgery; and generate a secondary treatment recommendation report when the corneal biomechanical index decreases by less than an index threshold after a preset postoperative time. Specifically, taking the Corvis ST system as an example, the stress-strain relationship analysis of the Corvis ST system identifies whether the corneal resistance to deformation is maintained at a safe threshold (e.g., a critical value of ≥30% reduction in biomechanical index). When the corneal biomechanical index decreases by <30% after a preset postoperative time (e.g., 6 months), a secondary treatment recommendation report is generated. This application can provide early warning of corneal ectasia risk and trigger an intervention mechanism when biomechanical preservation is insufficient, detecting subclinical abnormalities 3-6 months earlier than traditional topographic screening.

[0045] A second aspect of the present invention provides a control method for a corneal cross-linking device, implemented through the control system described in any of the above embodiments, such as... Figure 5 As shown, the procedure includes the following steps: integrating preoperative corneal topography curvature matrix, OCT layered thickness data, and biomechanical parameters to generate a four-channel fusion tensor; using the four-channel fusion tensor as a condition, inputting a conditional generative adversarial network to generate a virtual corneal model; generating a heat map of energy distribution based on the mechanical response prediction of the virtual corneal model; and performing patterned irradiation based on the heat map.

[0046] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0047] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A control system for a corneal cross-linking device, characterized in that, include: The multimodal data fusion module is configured to integrate preoperative corneal topography curvature matrix, OCT layered thickness data and biomechanical parameters to generate a four-channel fusion tensor. The virtual corneal modeling module is configured to use the four-channel fusion tensor as a condition, input a conditional generative adversarial network, and generate a virtual corneal model. The dynamic energy planning module is configured to generate a heat map of energy distribution based on the mechanical response prediction of the virtual corneal model. The irradiation control module is configured to perform patterned irradiation based on the heat map.

2. The control system according to claim 1, characterized in that, The multimodal data fusion module includes: The curvature processing unit is configured to perform Gaussian filtering and normalization on the curvature matrix of the preoperative corneal topography to generate a standardized curvature tensor. The OCT analysis unit is configured to segment the corneal epithelium, stroma, and endothelium in OCT volume data using a pre-trained neural network, calculate the stroma thickness distribution map, and register it to the coordinate system of the curvature tensor. The biomechanical mapping unit converts the detected biomechanical parameters into a biomechanical constraint matrix, wherein the biomechanical parameters include at least the amplitude of the first flattening deformation and the radius of curvature of the corneal concavity point. The feature fusion unit is configured to stitch the curvature tensor, matrix layer thickness distribution map and biomechanical constraint matrix along the channel dimension to form a four-channel fusion tensor, and to standardize each channel.

3. The control system according to claim 2, characterized in that, The feature fusion unit further includes a risk region enhancement submodule, which is configured as follows: Pixel regions that meet the conditions are identified based on preset normalized thickness threshold, normalized curvature threshold and first flattening deformation amplitude mapping threshold. Multiply the fusion tensor feature value corresponding to the pixel region by the enhancement factor.

4. The control system according to claim 2, characterized in that, The dynamic energy planning module is configured as follows: An energy attenuation coefficient is applied to regions that meet the stromal thickness threshold in the virtual corneal model. A light spot arrangement pattern is generated, wherein the diameter of the light spot is 50~100μm, and the light spot density in the central region is 1.5~2 times that in the edge region.

5. The control system according to claim 1, characterized in that, It also includes a dynamic optimization module, which is configured as follows: Real-time monitoring of riboflavin fluorescence intensity; When the local riboflavin fluorescence intensity is lower than the average of the entire field by more than a preset ratio, the irradiation strategy is updated.

6. The control system according to claim 4, characterized in that, It also includes a motion compensation module, which is configured as follows: Track pupil edge feature points and calculate eyeball deflection vector; An eye-robotic arm motion model is established based on the eyeball deflection vector; Based on the eyeball-robotic arm motion model, the lifting track and the robotic arm complete synchronous offset within a preset time. Closed-loop control ensures that the projection position error of the light spot is ≤50μm.

7. The control system according to claim 2, characterized in that, It also includes a three-level safety fuse module, which includes: The primary protection unit is configured to detect UVA optical density fluctuations in real time and trigger automatic calibration when the fluctuation amplitude exceeds the fluctuation threshold. The secondary protection unit is configured to cut off the UVA light source and lock the robotic arm's stroke when the distance sensor detects that the probe-cornea distance is less than a distance threshold. The level 3 protection unit is configured to automatically activate the energy attenuation mode in the local area when intraoperative OCT detects a local reduction in the thickness of the stromal layer greater than a preset ratio, with an attenuation coefficient of 0.3~0.

5.

8. The control system according to claim 2, characterized in that, It also includes an intraoperative adjustment module, which is configured as follows: The actual stromal layer thickness distribution detected by intraoperative OCT is obtained at preset time intervals. When the deviation between the actual stromal layer thickness and the predicted stromal layer thickness of the virtual corneal model exceeds a threshold, the applied energy is adjusted.

9. The control system according to claim 1, characterized in that, It also includes a postoperative assessment module, which is configured as follows: Compare changes in corneal biomechanical index before and after surgery; When the corneal biomechanical index decreases to less than the index threshold after a preset time post-surgery, a secondary treatment recommendation report is generated.

10. A control method for a corneal cross-linking device, characterized in that, The control system implemented according to any one of claims 1-9 includes the following steps: Integrate preoperative corneal topography curvature matrix, OCT layered thickness data and biomechanical parameters to generate a four-channel fusion tensor; Using the four-channel fusion tensor as a condition, a conditional generative adversarial network is input to generate a virtual corneal model; Based on the mechanical response prediction of the virtual corneal model, a heat map of energy distribution is generated; Patterned illumination is performed based on the heatmap.

Images