Intelligent corneal transplantation operation planning and postoperative management system

Abstract

The invention provides an intelligent corneal transplantation operation planning and postoperative management system, and the system comprises a model construction module which is used for collecting the preoperative multi-modal eye image data of a patient, and constructing a three-dimensional digital twinborn body; the preoperative planning module is used for generating a corresponding operation scheme through analogue simulation based on the three-dimensional digital twin and generating a recovery trajectory prediction model; the intra-operation correction module is used for collecting multi-mode eye image data in the operation in real time, dynamically updating the three-dimensional digital twin body, correcting an operation scheme and recovering a trajectory prediction model; and the postoperative monitoring module is used for periodically collecting self-examination data of the patient, inputting the self-examination data into the recovery trajectory prediction model to calculate the deviation probability of the self-examination data and the prediction data, and automatically generating a risk level assessment report. According to the method, the eye digital twins are constructed and updated in real time, so that accurate planning and intelligent monitoring of the whole process of the operation are realized, and the operation safety and the postoperative rehabilitation quality are improved.

Description

Technical Field

[0001] This invention relates to the field of digital medical technology, and in particular to an intelligent corneal transplantation surgery planning and postoperative management system. Background Technology

[0002] Corneal disease is one of the leading causes of blindness worldwide. For patients with end-stage corneal disease, corneal transplantation is the only effective way to restore vision. Traditional corneal transplantation surgery has significant limitations in planning, execution, and postoperative management. Preoperative planning relies on two-dimensional static images and the doctor's experience, lacking precise three-dimensional quantitative analysis, making it difficult to achieve personalized plans and predict postoperative results. During the surgery, the plan cannot be adjusted in a timely manner based on real-time dynamic changes in eye tissues, increasing surgical risks. Postoperative management relies on regular hospital check-ups, and there are blind spots in monitoring patients at home, making it difficult to detect complications such as rejection and infection in a timely manner, and easily missing the best intervention opportunity. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent corneal transplantation surgery planning and postoperative management system. By constructing and updating a digital twin of the eye in real time, it can achieve precise planning and intelligent monitoring of the entire surgical process, thereby improving surgical safety and postoperative recovery quality.

[0004] To achieve the above objectives, the present invention provides an intelligent corneal transplantation surgery planning and postoperative management system, the system comprising: The model building module is used to collect multimodal ocular image data of patients before surgery and to build a three-dimensional digital twin based on the multimodal ocular image data; The preoperative planning module is used to generate corresponding surgical plans based on a 3D digital twin through simulation, and to generate a recovery trajectory prediction model based on the surgical plan. The intraoperative correction module is used to collect multimodal ocular image data in real time during surgery, dynamically update the three-dimensional digital twin, and correct the surgical plan and recovery trajectory prediction model. The postoperative monitoring module is used to periodically collect patients' self-examination data, input the self-examination data into the recovery trajectory prediction model to calculate the probability of deviation between the self-examination data and the predicted data, and automatically generate a risk level assessment report.

[0005] Furthermore, a three-dimensional digital twin is constructed based on multimodal eye image data, specifically including: S11. Acquire multimodal ocular imaging data of the patient before surgery, wherein the multimodal ocular imaging data includes at least first morphological data providing ocular tomographic volume information and second morphological data providing ocular surface curvature information; S12. Register and fuse the first morphological data and the second morphological data into the same three-dimensional coordinate system to obtain a fused image; S13. Based on the fused images, the patient's ocular anatomy is segmented using a deep learning segmentation network to obtain the segmentation results; S14. Based on the segmentation results, perform three-dimensional geometric reconstruction to generate a three-dimensional mesh model. Based on prior medical knowledge, assign biomechanical material properties to the three-dimensional mesh model to generate a three-dimensional digital twin.

[0006] Furthermore, the generation of the surgical plan specifically includes: S21. Assign a set of surgical variables to a three-dimensional digital twin of the cornea, wherein the set of surgical variables includes at least graft parameters and suture parameters; S22. Using postoperative biomechanical safety indicators and postoperative optical quality indicators as joint optimization objectives, a quantifiable objective function is constructed. In the solution space of the surgical variable set, at least two sets of candidate surgical plans are generated through optimization algorithms.

[0007] Furthermore, for each group of candidate surgical plans, the following procedures are performed sequentially: S31. Based on the finite element analysis method, simulate the surgical operation of the surgical plan and calculate the mechanical safety parameters of the cornea after surgery. The mechanical safety parameters include at least the deformation state and the internal stress distribution. S32. Based on the ray tracing method, calculate the optical quality parameters of the cornea after the simulated postoperative corneal morphology. The optical quality parameters include at least the modulation transfer function and wavefront aberration. S33. Substitute the mechanical safety parameters and optical quality parameters of each candidate surgical plan into the quantifiable objective function for evaluation, and output at least one optimal solution as the recommended surgical plan.

[0008] Furthermore, the generation of the trajectory prediction model specifically includes: S41. Obtain a historical patient postoperative follow-up time series dataset, which includes preoperative static features, surgical plan features, biomechanical and optical features generated based on the surgical plan, and corneal index measurements at different time points. The biomechanical and optical features include at least the maximum stress value, average stress value, and predicted astigmatism value of the cornea after surgery. S42. Using preoperative static features, surgical plan features, and biomechanical and optical features as inputs, and the changes in corneal indices over time as outputs, the model is trained using a machine learning framework that outputs prediction uncertainty to obtain an initial recovery trajectory prediction model. S43. The recovery trajectory prediction model is used to output the predicted values ​​of corneal indicators at future time points and their confidence intervals.

[0009] Furthermore, the 3D digital twin is dynamically updated, specifically including: S51. Real-time acquisition of intraoperative multimodal ocular image data, wherein the multimodal ocular image data provides at least the current surface geometry information of the eye; S52. Perform real-time segmentation of multimodal eye image data based on machine learning models to obtain current eye surface point cloud data; S53. Spatial registration is performed between the current ocular surface point cloud data and the preoperative three-dimensional digital twin to calculate the dynamic transformation field that compensates for intraoperative ocular movement and tissue deformation. S54. Deformation update of the geometric shape of a three-dimensional digital twin based on a dynamic transformation field; S55. Based on the updated geometry, the force distribution within the eye is estimated in real time and the mechanical state of the three-dimensional digital twin is updated synchronously using the inverse mechanics inversion algorithm.

[0010] Further revisions to the surgical plan include: S61. Based on the real-time updated three-dimensional digital twin of the eye during surgery, calculate the deviation between the current biomechanical-optical parameters and the preoperative surgical plan target. S62. When the deviation value exceeds the preset threshold, the already performed and irreversible surgical operation is used as a fixed constraint and locked to the real-time updated three-dimensional digital twin of the eye. S63. Using the remaining adjustable surgical parameters as optimization variables and minimizing the deviation as the optimization objective, output the correction parameters through a local optimization algorithm; S64. Convert the modified parameters into intraoperative guidance information and modify the surgical plan in real time.

[0011] Further refinements to the trajectory prediction model include: S71. Based on the intraoperative real-time updated three-dimensional digital twin of the eye, extract the current mechanical-optical coupling features to form an intraoperative dynamic feature set; S72. Combine the intraoperative dynamic feature set with the patient's static features and the surgical plan features to obtain a specific input vector; S73. The recovery trajectory prediction model is updated by taking the patient-specific input vector and the postoperative initial state measurement value obtained during the operation as input and using a machine learning framework that outputs prediction uncertainty.

[0012] Further calculation of the probability of deviation includes: S81. Obtain the patient's self-check measurement value at a certain time point after surgery, and extract the prediction distribution information of the corresponding time point in the recovery trajectory prediction model; S82. Based on the predicted distribution information, the self-checked measurement value is converted into a standardized deviation and mapped to a preset probability metric to obtain the deviation probability.

[0013] Compared with the prior art, the beneficial effects of the present invention are: This invention provides an intelligent corneal transplant surgery planning and postoperative management system. The model building module constructs a three-dimensional digital twin by collecting multimodal ocular image data, reconstructing the complex structure and lesion details of the cornea, overcoming the limitations of subjective interpretation of traditional two-dimensional images. The preoperative planning module generates personalized surgical plans, reducing the risk of postoperative complications such as astigmatism and graft rejection due to plan deviations. The intraoperative correction module collects image data in real time and dynamically updates the three-dimensional digital twin, promptly correcting the surgical plan and recovery trajectory prediction model, compensating for the shortcomings of traditional surgery relying on physician experience and improving surgical precision. The postoperative monitoring module periodically collects patient self-assessment data, calculates the probability of deviation based on the recovery trajectory prediction model, automatically generates a risk level assessment report, automatically identifies abnormal trends, and achieves early warning of complications. This invention, by constructing a digital twin that is synchronously updated with the real eye and performing preoperative simulation planning and intelligent monitoring, improves the accuracy, safety, and predictability of postoperative management, maximizing personalized treatment, early warning of complication risks, and long-term patient recovery. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application 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 embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the structure of an intelligent corneal transplantation surgery planning and postoperative management system provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the process of constructing a three-dimensional digital twin based on multimodal eye image data, as provided in an embodiment of the present invention. Figure 3 A schematic diagram illustrating the process of generating a surgical plan according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the operation process for each group of candidate surgical plans provided in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the generation process of the recovery trajectory prediction model provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the dynamic updating process of a three-dimensional digital twin provided in an embodiment of the present invention; Figure 7 A schematic diagram illustrating the modification process of the surgical plan provided in an embodiment of the present invention; Figure 8 A schematic diagram illustrating the correction process of the recovery trajectory prediction model provided in this embodiment of the invention; Figure 9 This is a schematic diagram illustrating the calculation process of the deviation probability provided in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0017] Reference Figure 1 This embodiment provides an intelligent corneal transplantation surgery planning and postoperative management system, the system comprising: The model building module is used to collect multimodal ocular image data of patients before surgery and to build a three-dimensional digital twin based on the multimodal ocular image data.

[0018] The preoperative planning module is used to generate corresponding surgical plans based on a 3D digital twin through simulation, and to generate a recovery trajectory prediction model based on the surgical plan.

[0019] The intraoperative correction module is used to acquire multimodal ocular image data in real time during surgery, dynamically update the three-dimensional digital twin, and correct the surgical plan and recovery trajectory prediction model.

[0020] The postoperative monitoring module is used to periodically collect patients' self-examination data, input the self-examination data into the recovery trajectory prediction model to calculate the probability of deviation between the self-examination data and the predicted data, and automatically generate a risk level assessment report.

[0021] In this embodiment, a model building module collects multimodal ocular image data of the patient before surgery and constructs a three-dimensional digital twin. This overcomes the limitations of traditional two-dimensional images, comprehensively presenting the complex structure and lesion details of the cornea, providing a precise digital foundation for subsequent surgical planning, and allowing doctors to more intuitively and comprehensively understand the patient's ocular condition. The preoperative planning module, relying on the three-dimensional digital twin, uses simulation to develop a personalized surgical plan and generates a recovery trajectory prediction model based on the surgical plan. In other words, the surgical plan combines precise digital models and simulation results, improving the scientific rigor and specificity of the surgical plan, while providing a reference predictive standard for postoperative recovery assessment.

[0022] The intraoperative correction module collects multimodal ocular image data in real time during surgery, dynamically updates the 3D digital twin, and then corrects the surgical plan and recovery trajectory prediction model. This solves the problem of traditional surgery's inability to respond promptly to dynamic changes in the cornea during surgery, allowing the surgical plan to be flexibly adjusted according to the actual situation and improving the precision of the surgical operation. The postoperative monitoring module periodically collects the patient's self-assessment data and calculates the probability of deviation between the self-assessment data and the predicted data through the recovery trajectory prediction model. It automatically generates a risk level assessment report, changing the traditional model of relying on patients to come to the hospital for follow-up examinations, improving the convenience and compliance of patient follow-up examinations, and thus timely identifying potential postoperative risks, providing data support for doctors to formulate intervention plans.

[0023] As a preferred embodiment, a three-dimensional digital twin is constructed based on multimodal eye image data, specifically including: S11. Acquire preoperative multimodal ocular imaging data of the patient, wherein the multimodal ocular imaging data includes at least first morphological data providing ocular tomographic volume information and second morphological data providing ocular surface curvature information.

[0024] S12. Register and fuse the first morphological data and the second morphological data into the same three-dimensional coordinate system to obtain a fused image.

[0025] S13. Based on the fused images, the patient's ocular anatomy is segmented using a deep learning segmentation network to obtain the segmentation results.

[0026] S14. Based on the segmentation results, perform three-dimensional geometric reconstruction to generate a three-dimensional mesh model. Based on prior medical knowledge, assign biomechanical material properties to the three-dimensional mesh model to generate a three-dimensional digital twin.

[0027] In this embodiment, firstly, by integrating first-modal data providing tomographic volume information and second-modal data providing surface curvature information, the information limitations of a single modality image are broken, revealing the deep lesion details of each layer of corneal tissue, reflecting the geometric features of the eye surface, and achieving complete coverage of structural information. Secondly, different modal data are registered to the same three-dimensional coordinate system to eliminate spatial deviations during data acquisition, ensuring that the positional relationships of each anatomical structure in the fused image are completely consistent with the real eye, providing a spatial benchmark for subsequent analysis. Thirdly, deep learning segmentation networks are used to automatically segment the anatomical structures of the eye to capture the boundaries and abnormal areas of fine structures such as the epithelium, stroma, and endothelium. Combined with prior medical knowledge, the three-dimensional mesh model is endowed with biomechanical material properties, enabling the digital twin to simulate the elasticity, hardness, and other mechanical characteristics of real eye tissue. This provides a physical basis close to clinical reality for surgical simulation, stress analysis, and other scenarios, and provides core support for preoperative planning, intraoperative navigation, and postoperative prediction of corneal transplantation surgery.

[0028] Specifically, the core pathological features of keratoconus patients are central corneal thinning, a cone-shaped protrusion on the anterior surface, and are often accompanied by rupture of Descemet's membrane and stromal edema. When constructing a digital twin, we first acquire OCT data (first morphological data) that clearly shows the degree of thinning of the corneal stroma, the location of the Descemet's membrane rupture, and the extent of edema, and corneal topography data (second morphological data) that accurately presents the curvature gradient and irregular areas of the anterior surface from the periphery to the center. After registering the first and second morphological data to the same three-dimensional coordinate system, the images are fused to completely reconstruct the three-dimensional morphology of the corneal lesion.

[0029] Then, deep learning segmentation networks (such as U-Net and nnUNet) are used to automatically segment the boundaries of the epithelial, stromal, and endothelial layers, clarifying the extent of lesions in each layer. Finally, based on prior medical knowledge, different tissue layers are assigned corresponding biomechanical properties, such as decreased elastic modulus in thinned stromal areas and decreased hardness in edematous areas. Ultimately, a three-dimensional digital twin is generated to help doctors visually observe the three-dimensional distribution of lesions, accurately measure the thickness of thinned corneal areas, and the height and curvature of corneal cones, avoiding misjudgments in surgical plans due to missing deep lesions caused by single image information.

[0030] As a preferred embodiment, the generation of the surgical plan specifically includes: S21. Assign a set of surgical variables to the three-dimensional digital twin of the cornea, wherein the set of surgical variables includes at least the graft parameters and the suture parameters.

[0031] S22. Using postoperative biomechanical safety indicators and postoperative optical quality indicators as joint optimization objectives, a quantifiable objective function is constructed. In the solution space of the surgical variable set, at least two sets of candidate surgical plans are generated through optimization algorithms.

[0032] In this embodiment, key variables such as graft parameters (including diameter, thickness, and curvature) and suture parameters (including suture method, number of stitches, and tension intensity) are explicitly incorporated into the surgical plan design. This transforms the surgical operation into a quantifiable and controllable set of parameters, providing a foundation for generating the surgical plan. Postoperative biomechanical safety, such as structural stability and good wound healing, and optical quality, including low astigmatism and clear vision, are often mutually restrictive. Based on a joint optimization objective, finite element analysis simulation can automatically find the optimal balance point, rather than sacrificing one of them. For example, overly tight sutures may increase astigmatism, while overly loose sutures may lead to mechanical instability. Optimization algorithms, such as genetic algorithms, generate multiple candidate plans in the solution space. This preserves the diversity of plans to accommodate individual differences while ensuring the scientific validity of the plans through a quantified objective function. This allows doctors to select the most suitable plan for the patient with data support, significantly reducing surgical risks caused by experience bias and improving the controllability of postoperative outcomes.

[0033] As a preferred embodiment, for each group of candidate surgical plans, the following procedures are performed sequentially: S31. Based on the finite element analysis method, simulate the surgical operation of the surgical plan and calculate the mechanical safety parameters of the cornea after surgery. The mechanical safety parameters include at least the deformation state and the internal stress distribution.

[0034] S32. Based on ray tracing, calculate the optical quality parameters of the cornea after surgery according to the simulated postoperative corneal morphology. The optical quality parameters include at least the modulation transfer function and wavefront aberration.

[0035] S33. Substitute the mechanical safety parameters and optical quality parameters of each candidate surgical plan into the quantifiable objective function for evaluation, and output at least one optimal solution as the recommended surgical plan.

[0036] In this embodiment, the finite element method simulates the mechanical impact of surgical procedures on the cornea, quantifies postoperative deformation and stress distribution, and identifies potential risks such as graft loosening and tissue damage caused by excessive suture tension in advance, thus overcoming the limitations of traditional methods that rely solely on static morphological assessment. The ray tracing method calculates optical quality parameters based on simulated postoperative corneal morphology, objectively predicting the patient's postoperative visual outcome and avoiding the subjectivity of relying solely on experience to judge optical prognosis. By using a quantifiable objective function to comprehensively evaluate mechanical and optical parameters and output the optimal solution, the method ensures both biomechanical safety and patient needs for visual quality, allowing the recommended plan to simultaneously meet the dual goals of surgical safety and functional recovery. This provides doctors with data-driven and visualized scientific evidence for decision-making, reducing the risk of postoperative complications and the possibility of unsatisfactory results.

[0037] As a preferred embodiment, the generation of the trajectory prediction model specifically includes: S41. Obtain a historical patient postoperative follow-up time series dataset, which includes preoperative static features, surgical plan features, biomechanical and optical features generated based on the surgical plan, and corneal index measurements at different time points. The biomechanical and optical features include at least the maximum stress value, average stress value, and predicted astigmatism value of the cornea after surgery. S42. Using preoperative static features, surgical plan features, and biomechanical and optical features as inputs, and the changes in corneal indices over time as outputs, the model is trained using a machine learning framework that outputs prediction uncertainty to obtain an initial recovery trajectory prediction model. S43. The recovery trajectory prediction model is used to output the predicted values ​​of corneal indicators at future time points and their confidence intervals.

[0038] In this embodiment, patients vary greatly in age, health status, and healing ability. Even if they undergo the exact same surgical procedure, the recovery process and outcome may be completely different. Therefore, the input to the recovery trajectory prediction model not only includes the patient's preoperative basic information, such as age and corneal thickness, and surgical procedure parameters, such as graft size and suture method, but also incorporates the biomechanical and optical features generated by the surgical simulation, including postoperative stress distribution and predicted astigmatism. This achieves full-chain feature coverage of preoperative condition, surgical operation, and immediate postoperative effect, avoiding prediction bias caused by incomplete information and allowing the recovery trajectory prediction model to capture key factors affecting recovery from multiple perspectives.

[0039] Therefore, by training based on time-series data, the dynamic recovery pattern is captured. Using historical postoperative follow-up time-series data of patients for training, the recovery trajectory prediction model learns the inherent patterns of corneal parameters changing over time, such as the rate of edema resolution and astigmatism trends. It outputs predicted values ​​for multiple future time points, providing doctors with a complete prediction of the recovery process. By quantifying prediction uncertainty, clinical applicability is enhanced. The predicted value and confidence interval are output, with the position of the predicted value within the confidence interval clearly informing the doctor of the reliability of the prediction. When the predicted value exceeds the confidence interval and the deviation is greater than a preset threshold, this prediction may indicate that the patient's condition is special or there is a high risk, requiring extra attention from the doctor. For example, when the postoperative corneal curvature reaches 44.0D and significantly exceeds the upper limit of the model's prediction interval [40.5D, 42.0D], the system will immediately mark it as a high-risk deviation, indicating potential problems such as corneal bulging or overly tight sutures, requiring timely intervention.

[0040] As a preferred embodiment, dynamically updating the three-dimensional digital twin specifically includes: S51. Real-time acquisition of intraoperative multimodal ocular image data, wherein the multimodal ocular image data provides at least the current surface geometry information of the eye.

[0041] S52. Based on the machine learning model, multimodal eye image data is segmented in real time to obtain the current eye surface point cloud data.

[0042] S53. Spatial registration is performed between the current ocular surface point cloud data and the preoperative three-dimensional digital twin to calculate the dynamic transformation field that compensates for intraoperative ocular movement and tissue deformation.

[0043] S54. The geometric shape of the three-dimensional digital twin is updated based on the dynamic transformation field.

[0044] S55. Based on the updated geometry, the force distribution within the eye is estimated in real time and the mechanical state of the three-dimensional digital twin is updated synchronously using the inverse mechanics inversion algorithm.

[0045] In this embodiment, during surgery, the eyeball may move, and changes in intraocular pressure may cause tissue deformation. Furthermore, the surgical procedure itself fundamentally alters the eye's geometry and biomechanics. The pre-operative 3D digital twin will quickly become distorted after the surgery begins, and without dynamic updates, navigation based on it will become meaningless and even misleading. By keeping the digital twin synchronized with the real eye, the surgeon can be guided to perform the next step on the altered tissue.

[0046] First, multimodal data is acquired in real time using intraoperative imaging equipment, focusing on capturing the current geometric information of the eye surface to provide the latest realistic basis for model updates. Next, a machine learning model is used to quickly segment the real-time images and extract the point cloud data of the current eye surface to accurately reflect the morphological changes of the cornea caused by intraocular pressure fluctuations and instrument operations during surgery. This is more efficient and closer to the real-time surgical rhythm than manual recognition. Subsequently, the real-time point cloud data is spatially registered with the preoperatively constructed three-dimensional digital twin. The algorithm calculates the dynamic transformation field to compensate for intraoperative eye movement and tissue deformation, eliminating the problem of spatial misalignment between the virtual model and the actual eye, and laying a spatial benchmark for subsequent morphological updates.

[0047] Based on this, the geometry of the digital twin is adjusted in real time using a dynamic transformation field, so that the corneal surface contour and tissue position of the virtual model are completely matched with the actual state during the operation. Finally, the force distribution inside the eye is inferred from the updated geometry using an inverse mechanics algorithm, such as the pressure in the contact area between the graft and the recipient and the tension changes at the suture point, and the mechanical state of the model is updated synchronously.

[0048] As a preferred embodiment, the modification of the surgical plan specifically includes: S61. Based on the intraoperative real-time updated three-dimensional digital twin of the eye, calculate the deviation between the current biomechanical-optical parameters and the preoperative surgical plan target.

[0049] S62. When the deviation value exceeds the preset threshold, the already performed and irreversible surgical operation is used as a fixed constraint and locked to the real-time updated three-dimensional digital twin of the eye.

[0050] S63. Using the remaining adjustable surgical parameters as optimization variables and minimizing the deviation as the optimization objective, the correction parameters are output through a local optimization algorithm.

[0051] S64. Convert the modified parameters into intraoperative guidance information and modify the surgical plan in real time.

[0052] The modified parameters are converted into intraoperative guidance information to adjust the surgical plan in real time.

[0053] In this embodiment, the operations performed during the surgery are irreversible, and subsequent strategies must be adjusted immediately upon discovering problems. Based on the dynamically updated three-dimensional digital twin of the eye during the operation, the deviation between the current mechanical-optical parameters of the cornea and the target values ​​set in the preoperative plan is calculated in real time. For example, the preoperative target for uniform distribution of suture tension is ≤5kPa, but real-time calculation during the operation finds that the tension in a certain area reaches 12kPa, and the deviation value exceeds the acceptable range. When the deviation value exceeds the preset threshold, the system will first lock the already performed and irreversible surgical operations as fixed constraints into the digital twin to avoid secondary damage to the completed operations due to the correction of subsequent steps, thus ensuring the safety of the correction process.

[0054] Subsequently, the remaining adjustable surgical parameters are used as optimization variables. With minimizing the current deviation and approaching the preoperative target as the core optimization objectives, the optimal correction parameters are quickly calculated through a local optimization algorithm. For example, for areas with excessive tension, specific parameters are output to reduce the tension of the remaining sutures by 20% and to add one suture on the opposite side to balance the tension. Finally, the system automatically converts the abstract correction parameters into intraoperative guidance information that doctors can directly execute, such as setting the tension of the 8th suture to 8 kPa and the suture position offset by 0.5 mm. This information is updated in real time and pushed to the surgical navigation interface to guide doctors to correct subsequent operations and ensure that the final surgical outcome meets the preoperative planning expectations.

[0055] As a preferred embodiment, the correction of the trajectory prediction model specifically includes: S71. Based on the intraoperative real-time updated three-dimensional digital twin of the eye, extract the current mechanical-optical coupling features to form an intraoperative dynamic feature set.

[0056] S72. Combine the intraoperative dynamic feature set with the patient's static features and the surgical plan features to obtain a specific input vector.

[0057] S73. The recovery trajectory prediction model is updated by taking the patient-specific input vector and the postoperative initial state measurement value obtained during the operation as input and using a machine learning framework that outputs prediction uncertainty.

[0058] In this embodiment, the correction process is based on the real-time updated three-dimensional digital twin of the eye during the operation. It extracts mechanical-optical coupling features that reflect the actual effect of the current surgery, such as the correlation features between the actual fit of the implant during the operation, the real-time suture tension distribution and the immediate postoperative astigmatism value, to form a dynamic and accurate intraoperative dynamic feature set, which is used to make up for the shortcomings of preoperative prediction relying solely on simulation data. The dynamic feature set is deeply integrated with the patient's preoperative static features, including age and the degree of underlying disease, and the surgical plan features that are finally determined, including actual implant parameters and suture method, to construct a patient-specific input vector, ensuring that the input data can accurately match the actual situation of the individual during the operation.

[0059] The specific input vector and the initial postoperative state measurements obtained during the operation, such as immediate postoperative corneal thickness and endothelial cell density, are input into a machine learning framework with the function of predicting uncertainty output. The model parameters are adjusted through real-time iterative training, and the prediction logic of the recovery trajectory prediction model is updated. For example, if the graft fit is found to be slightly lower than the preoperative simulation during the operation, the recovery trajectory prediction model will combine this dynamic feature to correct the prediction results such as the postoperative edema reduction speed and astigmatism recovery trend, and update the confidence interval of the prediction value at the same time to ensure that the prediction can reflect the impact of the actual situation during the operation on the postoperative recovery.

[0060] As a preferred embodiment, the calculation of the deviation probability specifically includes: S81. Obtain the patient's self-check measurement value at a certain time point after surgery, and extract the prediction distribution information of the corresponding time point in the recovery trajectory prediction model; S82. Based on the predicted distribution information, the self-checked measurement value is converted into a standardized deviation and mapped to a preset probability metric to obtain the deviation probability.

[0061] In this embodiment, self-measured values ​​such as corneal curvature, visual acuity, and intraocular pressure are collected from patients one month post-surgery. Simultaneously, predicted distribution information for these indicators at each time point is extracted from the recovery trajectory prediction model, including predicted values, confidence intervals, and probability density functions. This ensures accurate matching between actual and predicted data in both the time and indicator dimensions, laying the foundation for subsequent comparative analysis. Next, based on the extracted predicted distribution information, the self-measured values ​​are converted into standardized deviations using algorithms such as Z-score standardization. For example, if a patient's self-measured corneal curvature one month post-surgery is 43.5D... The model predicts a mean of 42.0D and a standard deviation of 0.8D at this time point. The standardized deviation is calculated as (43.5-42.0) / 0.8=1.875, which directly reflects the degree of deviation between the actual value and the predicted mean. The standardized deviation is mapped to a preset probability measurement system to obtain the deviation probability, such as probability calculation based on normal distribution or probability matrix based on historical risk data. For example, the deviation probability corresponding to the standardized deviation of 1.875 is 96.9%, which means that there is a high probability that the patient's current recovery status deviates from the normal trajectory predicted by the model.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent corneal transplant surgery planning and postoperative management system, characterized in that, The system includes: The model building module is used to collect multimodal ocular image data of patients before surgery and to build a three-dimensional digital twin based on the multimodal ocular image data; The preoperative planning module is used to generate corresponding surgical plans based on a 3D digital twin through simulation, and to generate a recovery trajectory prediction model based on the surgical plan. The intraoperative correction module is used to collect multimodal ocular image data in real time during surgery, dynamically update the three-dimensional digital twin, and correct the surgical plan and recovery trajectory prediction model. The postoperative monitoring module is used to periodically collect patients' self-examination data, input the self-examination data into the recovery trajectory prediction model to calculate the probability of deviation between the self-examination data and the predicted data, and automatically generate a risk level assessment report.

2. The intelligent corneal transplant surgery planning and postoperative management system according to claim 1, characterized in that, Constructing a 3D digital twin based on multimodal eye image data, specifically including: S11. Acquire multimodal ocular imaging data of the patient before surgery, wherein the multimodal ocular imaging data includes at least first morphological data providing ocular tomographic volume information and second morphological data providing ocular surface curvature information; S12. Register and fuse the first morphological data and the second morphological data into the same three-dimensional coordinate system to obtain a fused image; S13. Based on the fused images, the patient's ocular anatomy is segmented using a deep learning segmentation network to obtain the segmentation results; S14. Based on the segmentation results, perform three-dimensional geometric reconstruction to generate a three-dimensional mesh model. Based on prior medical knowledge, assign biomechanical material properties to the three-dimensional mesh model to generate a three-dimensional digital twin.

3. The intelligent corneal transplant surgery planning and postoperative management system according to claim 1, characterized in that, The generation of a surgical plan specifically includes: S21. Assign a set of surgical variables to a three-dimensional digital twin of the cornea, wherein the set of surgical variables includes at least graft parameters and suture parameters; S22. Using postoperative biomechanical safety indicators and postoperative optical quality indicators as joint optimization objectives, a quantifiable objective function is constructed. In the solution space of the surgical variable set, at least two sets of candidate surgical plans are generated through optimization algorithms.

4. The intelligent corneal transplant surgery planning and postoperative management system according to claim 3, characterized in that, For each group of candidate surgical plans, execute them sequentially: S31. Based on the finite element analysis method, simulate the surgical operation of the surgical plan and calculate the mechanical safety parameters of the cornea after surgery. The mechanical safety parameters include at least the deformation state and the internal stress distribution. S32. Based on the ray tracing method, calculate the optical quality parameters of the cornea after the simulated postoperative corneal morphology. The optical quality parameters include at least the modulation transfer function and wavefront aberration. S33. Substitute the mechanical safety parameters and optical quality parameters of each candidate surgical plan into the quantifiable objective function for evaluation, and output at least one optimal solution as the recommended surgical plan.

5. The intelligent corneal transplant surgery planning and postoperative management system according to claim 1, characterized in that, The generation of the trajectory prediction model specifically includes: S41. Obtain a historical patient postoperative follow-up time series dataset, which includes preoperative static features, surgical plan features, biomechanical and optical features generated based on the surgical plan, and corneal index measurements at different time points. The biomechanical and optical features include at least the maximum stress value, average stress value, and predicted astigmatism value of the cornea after surgery. S42. Using preoperative static features, surgical plan features, and biomechanical and optical features as inputs, and the changes in corneal indices over time as outputs, the model is trained using a machine learning framework that outputs prediction uncertainty to obtain an initial recovery trajectory prediction model. S43. The recovery trajectory prediction model is used to output the predicted values ​​of corneal indicators at future time points and their confidence intervals.

6. The intelligent corneal transplant surgery planning and postoperative management system according to claim 1, characterized in that, Dynamically updating the 3D digital twin includes: S51. Real-time acquisition of intraoperative multimodal ocular image data, wherein the multimodal ocular image data provides at least the current surface geometry information of the eye; S52. Perform real-time segmentation of multimodal eye image data based on machine learning models to obtain current eye surface point cloud data; S53. Spatial registration is performed between the current ocular surface point cloud data and the preoperative three-dimensional digital twin to calculate the dynamic transformation field that compensates for intraoperative ocular movement and tissue deformation. S54. Deformation update of the geometric shape of a three-dimensional digital twin based on a dynamic transformation field; S55. Based on the updated geometry, the force distribution within the eye is estimated in real time and the mechanical state of the three-dimensional digital twin is updated synchronously using the inverse mechanics inversion algorithm.

7. The intelligent corneal transplant surgery planning and postoperative management system according to claim 1, characterized in that, The revision of the surgical plan specifically includes: S61. Based on the real-time updated three-dimensional digital twin of the eye during surgery, calculate the deviation between the current biomechanical-optical parameters and the preoperative surgical plan target. S62. When the deviation value exceeds the preset threshold, the already performed and irreversible surgical operation is used as a fixed constraint and locked to the real-time updated three-dimensional digital twin of the eye. S63. Using the remaining adjustable surgical parameters as optimization variables and minimizing the deviation as the optimization objective, output the correction parameters through a local optimization algorithm; S64. Convert the modified parameters into intraoperative guidance information and modify the surgical plan in real time.

8. The intelligent corneal transplant surgery planning and postoperative management system according to claim 1, characterized in that, The correction of the trajectory prediction model includes: S71. Based on the real-time updated three-dimensional digital twin of the eye during surgery, extract the current mechanical-optical coupling features to form an intraoperative dynamic feature set; S72. Combine the intraoperative dynamic feature set with the patient's static features and the surgical plan features to obtain a specific input vector; S73. The recovery trajectory prediction model is updated by taking the patient-specific input vector and the postoperative initial state measurement value obtained during the operation as input and using a machine learning framework that outputs prediction uncertainty.

9. The intelligent corneal transplant surgery planning and postoperative management system according to claim 1, characterized in that, The calculation of the probability of deviation specifically includes: S81. Obtain the patient's self-check measurement value at a certain time point after surgery, and extract the prediction distribution information of the corresponding time point in the recovery trajectory prediction model; S82. Based on the predicted distribution information, the self-checked measurement value is converted into a standardized deviation and mapped to a preset probability metric to obtain the deviation probability.

Images