An eye surgery planning method and system based on OCT data modeling

Abstract

The application discloses an eye surgery planning method and system based on OCT data modeling, relates to the technical field of surgical plan planning, and the method acquires OCT blood vessel imaging data and structural OCT data of a target eye, performs pixel-level semantic segmentation on pretreated multi-modal OCT data based on deep learning, labels eye structures, associates OCT image features with biomechanical properties, and generates a dynamic eye tissue model; multi-dimensional quantitative features are extracted from the model, input into a hybrid recommendation engine, combined with collaborative filtering based on historical case similarity and sequence decision based on reinforcement learning and model interaction, and a surgical planning scheme is generated. The application solves the problem that existing surgical planning methods highly depend on subjective experience of doctors and cannot extract complete eye lesion features from single static data, improves the modeling accuracy of OCT data, realizes dynamic capture of the eye state, and improves the precision and automation level of surgical planning.

Description

Technical Field

[0001] This application relates to the field of surgical planning technology, and in particular to a method and system for planning ocular surgery based on OCT data modeling. Background Technology

[0002] Optical coherence tomography (OCT) has become a core standard for the diagnosis and preoperative evaluation of ophthalmic diseases. It can provide high-resolution images of the microscopic structure of the eye, providing an irreplaceable data foundation for accurately measuring tissue thickness and identifying subtle pathological changes. However, current ophthalmic surgical planning relies heavily on the personal experience of doctors, and the use of OCT data is mostly limited to static observation and simple measurement. It lacks in-depth quantitative analysis and simulation of three-dimensional structure, dynamic changes, and the impact of surgical interactions. Existing surgical navigation systems also lack the intelligent planning capabilities that are deeply integrated with preoperative multimodal data and can be dynamically adjusted.

[0003] In recent years, deep learning-based OCT image analysis has demonstrated high precision in the automatic detection of ocular lesions and tissue layer segmentation. However, these AI models are positioned as diagnostic aids, and there is a gap between their output and surgical decisions. How to automatically generate and optimize specific surgical parameters by combining the features identified by the model with biomechanical principles and surgical goals is a challenge currently faced by AI in ocular surgery planning. There is an urgent need for a method that can deeply analyze OCT data, integrate multi-dimensional information, and introduce intelligent algorithms such as reinforcement learning for automated ocular surgery planning. Summary of the Invention

[0004] To overcome the problems of over-reliance on doctors' subjective experience, insufficient OCT data mining, and disconnect between preoperative planning and intraoperative execution in existing ophthalmic surgical planning, this application provides an ophthalmic surgical planning method and system based on OCT data modeling. It collects multimodal OCT data of the target eye to construct a dynamic three-dimensional model and applies a hybrid intelligent recommendation engine to automatically generate surgical planning schemes to assist doctors in making better decisions.

[0005] This application provides a method for planning ocular surgery based on OCT data modeling, including:

[0006] Step S10: Acquire multimodal OCT data of the target eye, including OCT vascular imaging data and structural OCT data, and preprocess and fuse the multimodal data. Step S20: Perform pixel-level semantic segmentation based on deep learning on the preprocessed OCT data, annotate the eye structure, and associate the OCT image features with biomechanical properties to generate a dynamic eye tissue model. Step S30: Extract multidimensional quantitative features from the eye tissue model, input them into the hybrid recommendation engine, combine collaborative filtering based on historical case similarity and sequential decision-making based on reinforcement learning and model interaction, fuse the results of the two methods to generate a surgical planning scheme, and optimize the scheme details through a multi-objective evaluation algorithm.

[0007] Furthermore, during the multimodal OCT data acquisition process of the target eye, multiple scanning modes are employed according to the needs of the target disease and surgical planning: Parallel B-scans were performed in the fovea region of the macula using a high-definition linear scanning method to obtain three-dimensional images of each retinal layer with the highest signal-to-noise ratio, which were used to detect minute tears and epiretinal membranes. Using the optic disc as the origin, 24 radial B-scans were performed to obtain a three-dimensional image of the optic disc, which was used to assess the optic disc morphology, cup-to-disc ratio, and thickness of the nerve fiber layer along the disc margin, in order to plan glaucoma surgery. Based on eye-tracking technology, a wide-field three-dimensional retinal image with a range of 12mm×12mm is stitched together to observe lesions in the peripheral area and the spatial relationship between the lesion and the central area. Based on the above three scanning modes, structural OCT data of the target eye are obtained. At the same scanning position of the eye, repeated B-scans are performed on the blood flow signals of the superficial and deep retinal vascular plexus and choroidal capillary layer to obtain blood flow information of the retina and choroidal capillary layer and generate OCT vascular imaging data.

[0008] Furthermore, the preprocessing steps for structural OCT data include: A three-dimensional filtering algorithm is used for processing. Based on the non-local similarity in the three-dimensional image of eye tissue, Wiener filtering is combined with hard thresholding to suppress speckle noise while preserving tissue edges. Inter- and intra-layer corrections are applied to remove motion artifacts in structural OCT data. Mutual information between adjacent B-scan results is calculated. Alignment is achieved through a non-rigid transformation model to correct translation and deformation caused by patient breathing and heartbeat. An image gradient-based algorithm is used to correct intra-B-scan distortion caused by patient eye saccades. Signal attenuation compensation is performed based on the depth information of the structural OCT three-dimensional image to highlight the deep tissue details of the retinal pigment epithelium and choroid.

[0009] For OCT angiography data, blood flow signals are extracted and differential, variance and amplitude decorrelation analyses are performed to generate blood flow angiography images. Based on image layer subtraction, projection shadow artifacts formed by large blood vessels on the surface of the eye are reduced on the image. The processed OCT angiography data is fused with structural OCT data at the corresponding location to reconstruct three-dimensional OCTA volume data, reflecting the three-dimensional spatial distribution of blood flow.

[0010] Furthermore, the pixel-level semantic segmentation based on deep learning employs a pre-trained 3D U-net backbone network. An attention gating mechanism is used in the skip connections of the encoder and decoder of the network to focus on key areas of ocular tissue and suppress irrelevant background areas. The network input is fused and registered multimodal OCT data of the eye, and the output is semantic segmentation results. The network identifies and labels the various layers of retinal structure, posterior vitreous cortex, lesion areas and vascular network. Based on the labeled data, the surface of each layer of ocular tissue is reconstructed into a three-dimensional mesh, generating a triangular mesh surface model representing the boundary of each tissue to describe the geometric morphology of ocular tissue. A constitutive model was established based on biomechanical principles, defining the retina and cornea as hyperelastic materials and the vitreous body as a viscoelastic material, and setting parameters such as elastic modulus, Poisson's ratio, and density. A biomechanical prior knowledge base was established, which included the range of material parameters for different ocular tissues and under different physiological and pathological conditions obtained through in vitro experiments, microscopic measurements, and literature reports. Image features related to tissue mechanical properties, including local intensity statistics, texture features, and structural features, are extracted from multimodal OCT data of the eye. A mapping relationship between image features and estimated elastic modulus is established based on a prior knowledge base. An elastic modulus value is predicted for each location in the multimodal OCT 3D image according to the mapping relationship. Based on the semantic segmentation results, a typical Poisson's ratio is assigned to different types of ocular tissues, generating a biomechanical parameter field corresponding to the OCT image space, reflecting the abnormal elastic modulus of the ocular lesion area.

[0011] Furthermore, the quantitative features extracted from the ocular tissue model include the geometric features and location information of ocular lesions, as well as the structural and vascular features of adjacent ocular tissues; The historical surgical case database contains patients' ocular features, surgical procedures, and postoperative efficacy assessments. A feature similarity metric is used to calculate the weighted Euclidean distance. The database is then used to search for the K most similar historical cases to the current patient. Through a matrix factorization-based collaborative filtering method, one or more recommended surgical procedures A based on group experience are generated. Secondly, the ophthalmic surgery planning is modeled as a sequential decision-making process. The state space consists of the current lesion state and tissue stress of the ophthalmic tissue model, and the action space consists of discretized surgical operation steps, involving operation positions and surgical equipment parameters. The reward function is defined as the expected effect of the surgical simulation, expressed as the degree of ophthalmic lesion elimination minus the degree of tissue damage. Deep reinforcement learning is used to train in a virtual environment that interacts with a patient-specific ophthalmic tissue model, learning the optimal surgical strategy for the current patient's unique anatomical structure, and outputting an optimized surgical plan B based on model exploration.

[0012] This application also provides an ocular surgery planning system based on OCT data modeling, including: OCT data acquisition and processing module: used to acquire multimodal OCT data of the target eye, including OCT vascular imaging data and structural OCT data, and to preprocess and fuse the multimodal data. Eye tissue modeling module: used to perform pixel-level semantic segmentation based on deep learning on preprocessed OCT data, annotate eye structures, and associate OCT image features with biomechanical properties to generate dynamic eye tissue models; Surgical planning generation module: This module extracts multidimensional quantitative features from the ocular tissue model, inputs them into a hybrid recommendation engine, combines collaborative filtering based on historical case similarity with sequential decision-making based on reinforcement learning and model interaction, merges the results of the two methods to generate a surgical planning scheme, and optimizes the scheme details through a multi-objective evaluation algorithm.

[0013] This application also proposes an eye surgery planning device based on OCT data modeling. The device includes: a memory, a processor, and a program such as an eye surgery planning algorithm based on OCT data modeling stored in the memory and executable on the processor. The program such as the eye surgery planning algorithm based on OCT data modeling is the step to implement the eye surgery planning method based on OCT data modeling as described above.

[0014] This application also provides a computer program product, which includes programs such as an eye surgery planning algorithm based on OCT data modeling. When the program such as the eye surgery planning algorithm based on OCT data modeling is executed by a processor, it implements the eye surgery planning method based on OCT data modeling as described above.

[0015] This application discloses the following technical effects: This application provides a method and system for ocular surgery planning based on OCT data modeling. By performing in-depth processing and modeling of the patient's ocular OCT data, a dynamic ocular tissue model integrating geometric features and biomechanical properties is constructed. This transforms static image data into an object that can be physically simulated and quantitatively analyzed, providing a reliable three-dimensional model foundation for precise surgical planning. The method integrates collaborative filtering based on historical big data statistics and reinforcement learning based on interactive exploration of the three-dimensional model to automatically generate and optimize the initial surgical plan for a specific patient. This overcomes the limitation of existing ocular surgery planning, which can only passively guide patients, and provides doctors with better surgical plan decision-making suggestions. The method proposed in this application combines expert experience, biological principles, and artificial intelligence, reducing the over-reliance on personal experience in ocular surgery planning and improving the precision of complex ocular surgical operations. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating an ophthalmic surgery planning method based on OCT data modeling, provided as an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of an eye surgery planning system based on OCT data modeling, provided in an embodiment of this application. Detailed Implementation

[0019] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In the following description, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0022] Example 1: This application provides a method for planning eye surgery based on OCT data modeling, such as... Figure 1 As shown, the method includes: Step S10: Acquire multimodal OCT data of the target eye, including OCT vascular imaging data and structural OCT data, and preprocess and fuse the multimodal data.

[0023] In this embodiment, multiple scanning modes are used according to the target disease and surgical planning requirements: Parallel B-scans were performed in the fovea region of the macula using a high-definition linear scanning method to obtain three-dimensional images of each retinal layer with the highest signal-to-noise ratio, which were used to detect minute tears and epiretinal membranes. Using the optic disc as the origin, 24 radial B-scans were performed to obtain a three-dimensional image of the optic disc, which was used to assess the optic disc morphology, cup-to-disc ratio, and thickness of the nerve fiber layer along the disc margin, in order to plan glaucoma surgery. Based on eye-tracking technology, a wide-field three-dimensional retinal image with a range of 12mm×12mm is stitched together to observe lesions in the peripheral area and the spatial relationship between the lesion and the central area. Based on the above three scanning modes, structural OCT data of the target eye are obtained. At the same time, at the same scanning position of the eye, repeated B-scans are performed on the blood flow signals of the superficial and deep retinal vascular plexus and choroidal capillary layer to obtain blood flow information of the retina and choroidal capillary layer and generate OCT vascular imaging data. The preprocessing steps for OCT data of eye structures include: A three-dimensional filtering algorithm is used for processing. Based on the non-local similarity in the three-dimensional image of eye tissue, Wiener filtering is combined with hard thresholding to suppress speckle noise while preserving tissue edges. Inter- and intra-layer corrections are applied to remove motion artifacts in structural OCT data. Mutual information between adjacent B-scan results is calculated. Alignment is achieved through a non-rigid transformation model to correct translation and deformation caused by patient breathing and heartbeat. An image gradient-based algorithm is used to correct intra-B-scan distortion caused by patient eye saccades. Signal attenuation compensation is performed based on the depth information of the structural OCT three-dimensional image to highlight the deep tissue details of the retinal pigment epithelium and choroid. Furthermore, for the OCT angiography data of the target eye, blood flow signals are extracted and subjected to differential, variance, and amplitude decorrelation analysis to generate blood flow angiography images. Based on image layer subtraction, projection shadow artifacts formed by large blood vessels on the surface of the eye are reduced on the image. The processed OCT angiography data is then fused with the structural OCT data of the corresponding location to reconstruct three-dimensional OCTA volume data, reflecting the three-dimensional spatial distribution of blood flow. For structural OCT data collected sequentially from different scanning modes that may be shifted due to slight patient movements, a feature-point-based elastic registration algorithm is used for spatial alignment, unifying them into a common coordinate system to form a basic three-dimensional model of the fundus structure. A three-dimensional non-rigid registration algorithm based on maximizing mutual information is adopted to calculate the transformation matrix that achieves the best spatial alignment between OCTA data and structural OCT data. The transformation matrix is ​​then applied to the original OCTA data to generate OCTA data that is voxel-level aligned with the structural OCT data. After preprocessing and registration, the OCTA data and structural OCT data of the target eye are fused to generate multimodal three-dimensional volume data. At each three-dimensional spatial coordinate of the eye, structural information and blood flow information are associated to reflect the reflectivity of eye tissues, characterize the anatomical boundaries of the retina, cornea and each layer of structure, and indicate the blood flow signal intensity, showing the location of blood vessels and perfusion status.

[0024] Step S20: Perform pixel-level semantic segmentation based on deep learning on the preprocessed OCT data, annotate the eye structure, and associate the OCT image features with biomechanical properties to generate a dynamic eye tissue model.

[0025] In this embodiment, the pixel-level semantic segmentation based on deep learning adopts a pre-trained 3D U-net backbone network. An attention gating mechanism is used in the skip connections of the encoder and decoder of the network to focus on the key areas of the ocular tissue and suppress irrelevant background areas. The network input is the fused and registered multimodal OCT data of the eye, and the output is the semantic segmentation result. The network identifies and labels the various layers of the retina, the posterior vitreous cortex, lesion areas and vascular networks. Based on the labeled data, the surface of each layer of ocular tissue is reconstructed into a three-dimensional mesh to generate a triangular mesh surface model representing the boundary of each tissue, which describes the geometric morphology of the ocular tissue. A constitutive model was established based on biomechanical principles, defining the retina and cornea as hyperelastic materials and the vitreous humor as a viscoelastic material, and setting parameters such as elastic modulus, Poisson's ratio, and density. A biomechanical prior knowledge base was established, including the range of material parameters for different ocular tissues under different physiological and pathological conditions obtained through in vitro experiments, microscopic measurements, and literature reports. For example, the elastic modulus of a normal retina is about 0.1-1 MPa, while the elastic modulus of an edematous retina is reduced, and the elastic modulus of a fibrotic proliferative membrane can be as high as several MPa.

[0026] Image features related to tissue mechanical properties, including local intensity statistics, texture features, and structural features, are extracted from multimodal OCT data of the eye. A mapping relationship between image features and estimated elastic modulus is established based on a prior knowledge base. An elastic modulus value is predicted for each location in the multimodal OCT 3D image according to the mapping relationship. Based on the semantic segmentation results, Poisson's ratio is assigned to different types of ocular tissues to generate a biomechanical parameter field corresponding to the OCT image space, reflecting the abnormal elastic modulus of the ocular lesion area. The generated triangular mesh surface of the key eye interface is imported into finite element processing software to generate a three-dimensional volume mesh. Based on specific eye surgery requirements, local refinement is performed in the expected surgical area. Personalized biomechanical parameter fields generated according to the patient's eye condition are mapped onto each element of the finite element model. Find the position of the center point of each cell in the original OCT image coordinate system, query the elastic modulus value predicted by the parameter mapping model at that position, and assign it to the cell; at the same time, assign the corresponding Poisson's ratio and density according to the tissue type of the cell (determined by the segmentation result); The final result is a finite element model containing personalized geometric and material properties, which serves as an ocular tissue model for surgical simulation and planning of precise surgical procedures.

[0027] Step S30: Extract multidimensional quantitative features from the eye tissue model, input them into the hybrid recommendation engine, combine collaborative filtering based on historical case similarity and sequential decision-making based on reinforcement learning and model interaction, fuse the results of the two methods to generate a surgical planning scheme, and optimize the scheme details through a multi-objective evaluation algorithm.

[0028] In this embodiment, the multidimensional quantitative characteristics of ocular tissue consist of geometric morphological features, biomechanical features, functional imaging features, and case semantic features: Geometric features: axial length, vitreous cavity volume, and foveal curvature; taking glaucoma as an example, calculate the cup-to-disc ratio, disc rim area, and distribution of nerve fiber layer thickness; Biomechanical characteristics: mean elastic modulus and standard deviation of modulus of the lesion area and surrounding healthy tissue; Functional imaging features: vascular density and blood flow index of superficial and deep capillary plexuses in the macula; vascular tortuosity and fractal dimension; Pathological semantic features: based on the semantic segmentation results, the posterior detachment state of the top vitreous body, the direction of traction on the epiretinal membrane, and the type of retinal edema.

[0029] One approach to hybrid recommendation engines is collaborative filtering recommendation based on the similarity of historical cases, including: A historical surgical case database was established, including patient ocular characteristics, surgical procedures (structured operation sequence: surgical type, surgical path coordinates, glass extent, and instruments used), and postoperative efficacy assessments (anatomical success rate, postoperative BCVA improvement value, and presence or absence of complications). A feature similarity metric was used to calculate the weighted Euclidean distance between ocular features. The database was then searched for the K most similar historical cases to the current patient. A collaborative filtering method based on matrix factorization was used, with the success rate of similar cases serving as a confidence reference, to generate one or more recommended surgical plans A based on group experience. The second approach of the hybrid recommendation engine is sequential decision-making based on reinforcement learning and model interaction, including: The ophthalmic surgery planning is modeled as a sequential decision-making process. The state space consists of the current lesion state and tissue stress of the ophthalmic tissue model, while the action space consists of discretized surgical operation steps, involving operation location and surgical equipment parameters. The reward function is defined as the expected effect of the surgical simulation, expressed as the degree of ophthalmic lesion elimination minus the degree of tissue damage. Deep reinforcement learning is used to train in a virtual environment that interacts with a patient-specific ophthalmic tissue model, learning the optimal surgical strategy for the current patient's unique anatomical structure, and outputting an optimized surgical plan B based on model exploration.

[0030] The surgical plans generated by the two approaches are input into the fusion decision network to evaluate which approach or hybrid approach is more likely to succeed for patients with given characteristics. The network's training data comes from a historical case database. The network finally outputs 1-3 preliminary surgical plans, including surgical operation sequences, spatial parameters, physical parameters, and expected results. The initial surgical plan is imported into the virtual surgical environment, and multi-dimensional evaluation indicators for each surgical plan are automatically calculated. Considering effectiveness, safety, efficiency, and stability, the surgical plan parameters are fine-tuned through multi-objective optimization, and the final optimal surgical plan is generated based on the expert experience provided by the doctor. The effectiveness indicators consist of lesion clearance rate and anatomical repositioning. The lesion clearance rate is calculated by dividing the volume of the lesion after virtual surgery simulation by the original volume of the lesion. The anatomical repositioning is expressed as retinal flatness and macular foveal morphology recovery score. Safety indicators include maximum tissue stress, cumulative stress exposure in functional areas, and volume of damage to healthy tissue. Maximum resistive stress represents the peak stress that occurs in all ocular tissues, reflecting the risk of acute injury. Cumulative stress exposure in functional areas represents the stress-time integral of key areas such as the macula and optic nerve fiber layer during the operation. Volume of damage to healthy tissue represents the volume of healthy tissue that has been subjected to stress exceeding the damage threshold. Efficiency indicators are quantified as estimated surgical time and complexity of operation steps. Based on the complexity of virtual surgical simulation steps and virtual operation time, the number of surgical steps and the number of instrument switching times are recorded. The stability index is quantified as the scheme's fault tolerance rate. By introducing small parameter perturbations during the virtual surgical simulation, the simulation is re-enhanced and the degree of decline in the target achievement rate is assessed.

[0031] Example 2: The ophthalmic surgery planning system based on OCT data modeling provided in this embodiment of the invention can execute the ophthalmic surgery planning method based on OCT data modeling provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method, such as... Figure 2 As shown, it includes the following modules: OCT data acquisition and processing module: used to acquire multimodal OCT data of the target eye, including OCT vascular imaging data and structural OCT data, and to preprocess and fuse the multimodal data. Eye tissue modeling module: used to perform pixel-level semantic segmentation based on deep learning on preprocessed OCT data, annotate eye structures, and associate OCT image features with biomechanical properties to generate dynamic eye tissue models; Surgical planning generation module: This module extracts multidimensional quantitative features from the ocular tissue model, inputs them into a hybrid recommendation engine, combines collaborative filtering based on historical case similarity with sequential decision-making based on reinforcement learning and model interaction, merges the results of the two methods to generate a surgical planning scheme, and optimizes the scheme details through a multi-objective evaluation algorithm.

[0032] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0033] Example 3: This application provides an eye surgery planning device based on OCT data modeling. The eye surgery planning device based on OCT data modeling includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute an eye surgery planning method based on OCT data modeling in the above embodiment.

[0034] Example 4: This application provides a computer program product including a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication system, or installed from a storage system. When the computer program is executed by a processing system, it performs the functions defined in the method of Example 1 of this application.

[0035] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for planning eye surgery based on OCT data modeling, characterized in that, The method includes: Step S10: Acquire multimodal OCT data of the target eye, including OCT vascular imaging data and structural OCT data, and preprocess and fuse the multimodal data. Step S20: Perform pixel-level semantic segmentation based on deep learning on the preprocessed OCT data, annotate the eye structure, and associate the OCT image features with biomechanical properties to generate a dynamic eye tissue model. Step S30: Extract multidimensional quantitative features from the eye tissue model, input them into the hybrid recommendation engine, combine collaborative filtering based on historical case similarity and sequential decision-making based on reinforcement learning and model interaction, fuse the results of the two methods to generate a surgical planning scheme, and optimize the scheme details through a multi-objective evaluation algorithm.

2. The ophthalmic surgery planning method based on OCT data modeling as described in claim 1, characterized in that, In step S10, during the multimodal OCT data acquisition of the target eye, multiple scanning modes are used according to the target disease and surgical planning requirements: Parallel B-scans were performed in the fovea region of the macula using a high-definition linear scanning method to obtain three-dimensional images of each retinal layer with the highest signal-to-noise ratio, which were used to detect minute tears and epiretinal membranes. Using the optic disc as the origin, 24 radial B-scans were performed to obtain a three-dimensional image of the optic disc, which was used to assess the optic disc morphology, cup-to-disc ratio, and thickness of the nerve fiber layer along the disc margin, in order to plan glaucoma surgery. Based on eye-tracking technology, a wide-field three-dimensional retinal image with a range of 12mm×12mm is stitched together to observe lesions in the peripheral area and the spatial relationship between the lesion and the central area. Based on the above three scanning modes, structural OCT data of the target eye are obtained. At the same scanning position of the eye, repeated B-scans are performed on the blood flow signals of the superficial and deep retinal vascular plexus and choroidal capillary layer to obtain blood flow information of the retina and choroidal capillary layer and generate OCT vascular imaging data.

3. The ophthalmic surgery planning method based on OCT data modeling as described in claim 2, characterized in that, The preprocessing steps for the structured OCT data include: A three-dimensional filtering algorithm is adopted, which is based on the non-local similarity in the three-dimensional image of eye tissue and combined with Wiener filtering. The hard thresholding method is used to suppress speckle noise while preserving tissue edges. Inter- and intra-layer corrections are applied to remove motion artifacts in structural OCT data. Mutual information between adjacent B-scan results is calculated. Alignment is achieved through a non-rigid transformation model to correct translation and deformation caused by patient breathing and heartbeat. An image gradient-based algorithm is used to correct intra-B-scan distortion caused by patient eye saccades. Signal attenuation compensation is performed based on the depth information of the structural OCT three-dimensional image to highlight the deep tissue details of the retinal pigment epithelium and choroid.

4. The ophthalmic surgery planning method based on OCT data modeling as described in claim 2, characterized in that, The preprocessing steps for the OCT angiography data include: Blood flow signals are extracted from OCT angiography data and subjected to differential, variance, and amplitude decorrelation analysis to generate blood flow angiography images. Based on image layer subtraction, projection shadow artifacts formed by large blood vessels on the surface of the eye are reduced. The processed OCT angiography data is then fused with structural OCT data at the corresponding locations to reconstruct three-dimensional OCTA volume data, reflecting the three-dimensional spatial distribution of blood flow.

5. The ophthalmic surgery planning method based on OCT data modeling as described in claim 1, characterized in that, In step S20, the pixel-level semantic segmentation based on deep learning adopts a pre-trained 3D U-net backbone network. An attention gating mechanism is used in the skip connections of the encoder and decoder of the network to focus on the key areas of the eye tissue and suppress irrelevant background areas. Using fused and registered multimodal OCT data of the eye as network input, the system outputs semantic segmentation results, identifies and labels the various layers of retinal structure, posterior vitreous cortex, lesion areas and vascular networks, and performs three-dimensional mesh reconstruction on the surface of each layer of ocular tissue based on the labeled data, generating triangular mesh surface models representing the boundaries of each tissue to describe the geometric morphology of ocular tissue.

6. The ophthalmic surgery planning method based on OCT data modeling as described in claim 1, characterized in that, In step S20, the process of obtaining the biomechanical properties includes: A constitutive model was established based on biomechanical principles, defining the retina and cornea as hyperelastic materials and the vitreous body as a viscoelastic material, and setting parameters such as elastic modulus, Poisson's ratio, and density. A biomechanical prior knowledge base was established, which included the range of material parameters for different ocular tissues and under different physiological and pathological conditions obtained through in vitro experiments, microscopic measurements, and literature reports. Image features related to tissue mechanical properties are extracted from multimodal OCT data of the eye. A mapping relationship between image features and estimated elastic modulus is established based on a prior knowledge base. An elastic modulus value is predicted for each location in the multimodal OCT three-dimensional image according to the mapping relationship. Based on the semantic segmentation results, Poisson's ratio is assigned to different types of ocular tissues to generate a biomechanical parameter field corresponding to the OCT image space, reflecting the abnormal elastic modulus of the ocular lesion area.

7. The ophthalmic surgery planning method based on OCT data modeling as described in claim 1, characterized in that, In step S30, the collaborative filtering recommendation based on historical case similarity includes: A historical surgical case database was established, including patients' ocular features, surgical procedures, and postoperative efficacy evaluations. A feature similarity metric was used to calculate the weighted Euclidean distance between ocular features. The database was then searched for the K most similar historical cases to the current patient. A collaborative filtering method based on matrix factorization was used, with the success rate of similar cases serving as a confidence reference, to generate a recommended surgical plan based on group experience.

8. The ophthalmic surgery planning method based on OCT data modeling as described in claim 1, characterized in that, The sequential decision-making based on reinforcement learning and model interaction described in step S30 includes: The planning of ocular surgery is modeled as a sequential decision-making process. The state space consists of the current lesion state and tissue stress of the ocular tissue model, while the action space consists of discretized surgical operation steps, involving operation location and surgical equipment parameters. The reward function is defined as the expected effect of the surgical simulation, expressed as the degree of ocular lesion elimination minus the degree of tissue damage. Using the proximal policy optimization algorithm in deep reinforcement learning, the algorithm is trained in a virtual environment that interacts with a patient-specific ocular tissue model to learn the optimal surgical strategy for the current patient's unique anatomical structure and output an optimized surgical plan based on model exploration.

9. The ocular surgery planning method based on OCT data modeling as described in claim 1, characterized in that, In step S30, the surgical planning scheme generated by the hybrid recommendation engine is imported into the virtual surgical environment. Multi-dimensional evaluation indicators for each surgical planning scheme are automatically calculated. Considering effectiveness, safety, efficiency and stability, the surgical plan parameters are optimized through a multi-objective optimization evaluation algorithm. Based on the expert experience provided by the doctor, the final optimal surgical planning scheme is generated.

10. An ocular surgery planning system based on OCT data modeling, characterized in that, The system is used to implement the ocular surgery planning method based on OCT data modeling as described in any one of claims 1-9, and the system comprises: OCT data acquisition and processing module: used to acquire multimodal OCT data of the target eye, including OCT vascular imaging data and structural OCT data, and to preprocess and fuse the multimodal data. Eye tissue modeling module: used to perform pixel-level semantic segmentation based on deep learning on preprocessed OCT data, annotate eye structures, and associate OCT image features with biomechanical properties to generate dynamic eye tissue models; Surgical planning generation module: This module extracts multidimensional quantitative features from the ocular tissue model, inputs them into a hybrid recommendation engine, combines collaborative filtering based on historical case similarity with sequential decision-making based on reinforcement learning and model interaction, merges the results of the two methods to generate a surgical planning scheme, and optimizes the scheme details through a multi-objective evaluation algorithm.

Images