Adaptive optical calibration intelligent intraocular lens implantation navigation system and method thereof

Abstract

The present application relates to the field of medical devices, in particular to an intelligent intraocular lens implant navigation system and method with adaptive optical calibration, comprising: a microscope module for collecting intraocular real-time image data; an adaptive optical system module connected with the microscope module for real-time evaluation of intraocular tissue optical aberration distribution and correction; a three-dimensional deep learning motion compensation module for predicting eye movement and compensating for micro-motion interference; a registration algorithm module for establishing spatial registration between OCT and intraocular imaging systems; an augmented reality projection module for superimposing registration guide lines on surgical microscope imaging; an intelligent intraocular imaging system for identifying intraocular tissue structure and instrument position, through multi-modal fusion perception, deep learning intelligent analysis, adaptive optical calibration and augmented reality guidance, precise implantation of intraocular lenses is achieved, and practical application shows that the lens implantation position accuracy can be improved to ±0.1mm.

Description

Technical Field

[0001] This invention relates to the field of medical devices, and in particular to an intelligent intraocular lens implantation navigation system and method with adaptive optics calibration. Background Technology

[0002] With the accelerating aging of society, cataracts have become one of the leading causes of blindness worldwide. In cataract surgery, the precise implantation of the intraocular lens (IOL) has a decisive impact on postoperative visual quality. Traditional cataract surgery relies heavily on the surgeon's experience for IOL implantation, lacking precise navigation and struggling to cope with intraoperative micro-movements and optical aberrations. This leads to inaccurate IOL positioning and reduced postoperative visual quality, especially for patients with corneal astigmatism, where satisfactory results are even more difficult to achieve.

[0003] Existing surgical navigation systems generally suffer from the following problems: First, they lack real-time optical calibration capabilities and cannot cope with changes in optical aberrations during surgery; second, they lack sufficient compensation for micro-movements of the eyeball, affecting navigation accuracy; third, multimodal data fusion is imperfect, resulting in limited spatial registration accuracy; and fourth, they lack intuitive augmented reality guidance, leading to high operational complexity.

[0004] Therefore, there is an urgent need for an intelligent navigation system that can calibrate optical aberrations in real time, accurately compensate for micro-movements of the eyeball, achieve precise spatial registration, and provide intuitive surgical guidance in order to improve the accuracy of intraocular lens implantation and postoperative visual quality. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent intraocular lens implantation navigation system and method with adaptive optics calibration. This system solves the problems existing in the prior art by using multimodal fusion perception, deep learning intelligent analysis, adaptive optics calibration and augmented reality guidance, thereby improving the accuracy of intraocular lens implantation and postoperative visual quality.

[0006] This invention proposes an intelligent intraocular lens implantation navigation system adapted to optical calibration, comprising:

[0007] Microscope module, used to acquire real-time image data inside the eye;

[0008] An adaptive optics system module, which is signal-connected to the microscope module, is used to evaluate and correct the optical aberration distribution of intraocular tissues in real time.

[0009] A 3D deep learning motion compensation module is used to predict eye movements and compensate for micro-motion disturbances;

[0010] The registration algorithm module is used to establish spatial registration between OCT and intraocular imaging systems;

[0011] Augmented reality projection module for overlaying registration guide lines onto surgical microscope images;

[0012] Intelligent intraocular imaging system for identifying intraocular structures and instrument locations.

[0013] Preferably, the adaptive optics system module includes:

[0014] The optical aberration assessment unit is used to train a virtual optical imaging module for intraocular images based on image data, and to evaluate the distribution of optical aberrations in intraocular tissues in real time.

[0015] A lens control unit is used to generate a control strategy for the contour of the intraocular lens based on the optical aberration assessment.

[0016] The imaging quality adjustment unit is used to adjust the imaging quality and the shape of the artificial lens in real time.

[0017] Preferably, the three-dimensional deep learning motion compensation module includes:

[0018] Feature extraction network unit, used to train feature extraction networks for cornea / lens / iris / pupil;

[0019] The motion feature parameter unit is used to acquire eye movement feature parameters;

[0020] A depth particle filter unit is used to estimate the displacement and attitude angle of intraocular surgical instruments;

[0021] The compensation matrix generation unit is used to generate a compensation matrix based on the virtual projection correction module.

[0022] Preferably, the registration algorithm module includes:

[0023] The spatial mapping unit is used to establish the spatial image mapping relationship between intraocular OCT and intraocular imaging systems.

[0024] The guideline calculation unit is used to calculate the guideline based on the spatial position, three-dimensional orientation, and image mapping relationship between the intraocular instruments / lens.

[0025] Preferably, the augmented reality projection module includes:

[0026] A guideline generation unit is used to generate guidelines based on OCT and intraocular imaging.

[0027] An attitude recognition unit is used to identify the spatial attitude of intraocular surgical instruments;

[0028] An image fusion unit is used to overlay registration guide lines onto surgical microscope images;

[0029] The visual tracking unit is used to perform visual tracking using the optical characteristics of intraocular instruments to obtain the registration relationship between the registration guide line and the surgical instruments.

[0030] Preferably, the intelligent intraocular imaging system includes:

[0031] Convolutional neural network units are used to design feature extraction algorithms to identify intraocular surgical instruments, lenses, corneas, lenticels, pupils, irises, and aqueous humor;

[0032] The contour generation unit is used to segment the three-dimensional contour of intraocular tissues based on OCT images;

[0033] Deep neural network model unit, used to generate and train deep neural network models;

[0034] The verification unit is used to match key points of the contour image segmented from OCT images and intraocular images to verify the accuracy of the model.

[0035] As a preferred option, it also includes:

[0036] The eye movement prediction module is used to construct a central coordinate model of the intraocular lens based on the optical imaging principles and spatial registration of intraocular surgery.

[0037] A virtual projection correction module is used to track the spatial position between intraocular surgical instruments and artificial lenses in real time, and a coordinate transformation mechanism based on optical projection is designed.

[0038] Preferably, the eye movement prediction module includes:

[0039] Central coordinate model unit, used to construct the central coordinate model of the intraocular lens;

[0040] Deep learning model unit, used to obtain eye movement feature parameters with the help of deep learning network model;

[0041] The estimation unit is used to estimate the displacement and attitude angle of intraocular surgical instruments using a deep particle filter algorithm.

[0042] The tracking unit is used to track OCT data and estimate the spatial position of intraocular surgical instruments.

[0043] Preferably, the virtual projection correction module includes:

[0044] The spatial positioning tracking unit is used to track the spatial position between intraocular surgical instruments and the intraocular lens in real time;

[0045] The image acquisition unit is used to acquire real-time intraocular images based on the position of the intraocular instruments.

[0046] Coordinate transformation unit, used to design coordinate transformation mechanisms based on optical projection;

[0047] The accuracy correction unit is used to perform tracking accuracy correction through the eye movement prediction module and spatial registration.

[0048] An adaptive optics-calibrated intelligent intraocular lens implantation navigation method, comprising the following steps:

[0049] S100, acquires real-time intraocular image data;

[0050] S200 uses an adaptive optics system to assess and correct the distribution of optical aberrations in intraocular tissues in real time.

[0051] S300 uses a 3D deep learning motion compensation module to predict eye movements and compensate for micro-motion interference.

[0052] S400 establishes spatial registration between OCT and intraocular imaging systems through a registration algorithm module;

[0053] S500 generates guide lines based on OCT and intraocular imaging;

[0054] S600 identifies the spatial orientation of intraocular surgical instruments;

[0055] S700, superimposes registration guide lines onto surgical microscope imaging;

[0056] The S800 uses the optical characteristics of intraocular instruments for visual tracking to obtain the registration relationship between the registration guide line and the surgical instruments.

[0057] The S900 uses an intelligent intraocular imaging system to identify intraocular tissue structures and instrument locations.

[0058] The S1000 adjusts the position of the surgical instruments in real time according to the registration relationship between the registration guide line and the surgical instruments, thus completing the precise implantation of the intraocular lens.

[0059] Compared with existing technologies, the present invention has the following significant advantages:

[0060] The intelligent intraocular lens implantation navigation system and method with adaptive optics calibration of this invention achieves precise implantation of the intraocular lens through multimodal fusion perception, deep learning intelligent analysis, adaptive optics calibration, and augmented reality guidance. This solves problems inherent in traditional techniques such as insufficient optical aberration correction, insufficient compensation for micro-movements of the eye, limited spatial registration accuracy, and unintuitive surgical guidance. Practical applications show that this system can improve the positional accuracy of lens implantation to ±0.1mm, the angular accuracy to ±2°, reduce surgical time by 30%, and significantly improve postoperative visual quality, providing better treatment outcomes for cataract patients. Attached Figure Description

[0061] Figure 1This is an overall architecture diagram of the intelligent intraocular lens implantation navigation system with adaptive optics calibration according to the present invention;

[0062] Figure 2 This is a structural diagram of the adaptive optics system module;

[0063] Figure 3 This is a flowchart of the workflow for a 3D deep learning motion compensation module.

[0064] Figure 4 A schematic diagram illustrating the implementation principle of the registration algorithm module;

[0065] Figure 5 A schematic diagram of an augmented reality projection module;

[0066] Figure 6 A diagram illustrating the recognition performance of an intelligent intraocular imaging system;

[0067] Figure 7 This is a structural diagram of the eye movement prediction module;

[0068] Figure 8 This is a schematic diagram illustrating the working principle of the virtual projection correction module.

[0069] Figure 9 This is a flowchart of the navigation method of the present invention;

[0070] Figure 10 This is a comparison chart of clinical application effects. Detailed Implementation

[0071] Please refer to the attached document. Figure 1-10 The intelligent intraocular lens implantation navigation system with adaptive optics calibration provided by this invention mainly comprises six core modules: a microscope module 1, an adaptive optics system module 2, a three-dimensional deep learning motion compensation module 3, a registration algorithm module 4, an augmented reality projection module 5, and an intelligent intraocular imaging system 6. These six modules are connected by signals to form a closed-loop control system, achieving precise navigation for intraocular lens implantation.

[0072] Microscope module 1 is the fundamental sensing device of the entire system. It employs a high-resolution digital microscope with a field of view of 12mm × 9mm, a resolution of 4K, and a frame rate of no less than 60fps to ensure the acquisition of clear real-time intraocular images. This module connects to adaptive optics system module 2 via a dedicated interface to transmit the acquired raw image data in real time.

[0073] The adaptive optics system module 2 is one of the core innovative modules of this invention. It receives image data from the microscope module 1 and uses advanced optical processing algorithms to evaluate and correct the distribution of optical aberrations in intraocular tissues in real time. This module can cope with optical aberrations caused by changes in the intraoperative environment, ensuring image quality and accuracy, which is especially important for complex cases (such as patients with high astigmatism).

[0074] The 3D deep learning motion compensation module 3 is responsible for predicting eye movements and compensating for micro-motion interference. Micro-motion is a key factor affecting surgical precision, especially during the delicate implantation of an intraocular lens. This module uses deep learning algorithms to predict the trajectory of eye micro-motion, generate compensation parameters, and coordinate with surgical instrument control to counteract the effects of micro-motion.

[0075] Registration algorithm module 4 establishes spatial registration between OCT and intraocular imaging systems. During surgery, it is often necessary to fuse data from two different modalities: OCT and microscopic images. This module uses a precise spatial registration algorithm to establish an accurate correspondence between the two imaging modalities, providing three-dimensional spatial coordinates for surgical navigation.

[0076] Augmented Reality Projection Module 5 overlays registration guide lines onto the surgical microscope image. This module superimposes the calculated optimal implantation path and position onto the real-time microscope image in an augmented reality manner, providing doctors with intuitive visual guidance, simplifying the operation process, and improving surgical precision.

[0077] The intelligent intraocular imaging system 6 is responsible for identifying intraocular structures and instrument locations. This system employs deep learning methods to identify and track various intraocular structures and surgical instruments in real time, providing fundamental data support for other modules.

[0078] The modules are connected via a high-speed data bus to ensure real-time data transmission with a latency of less than 10ms, in order to meet the stringent requirements of real-time surgical navigation.

[0079] The adaptive optics system module 2 is one of the key innovations of this invention. Its core function is to evaluate and correct intraocular optical aberrations in real time, ensuring image quality and navigation accuracy. This module mainly includes three functional units: an optical aberration evaluation unit 21, a crystal control unit 22, and an imaging quality adjustment unit 23.

[0080] The optical aberration assessment unit 21 trains a virtual optical imaging module based on image data of intraocular images to evaluate the distribution of optical aberrations in intraocular tissues in real time. This unit uses a deep convolutional neural network to construct an optical aberration distribution prediction model, and its core algorithm can be expressed as follows:

[0081] W(x,y)=f CNN (I(x,y),θ),

[0082] Here, W(x,y) represents the predicted aberration distribution function, I(x,y) is the input intraocular image, and θ is the network parameter. The network uses ResNet-50 as its backbone and enhances aberration feature extraction capabilities by adding an adaptive feature fusion layer. Training data comes from intraocular images of 1000 patients with varying degrees of cataracts, and labeled data is provided by wavefront aberrometer measurements. This unit can complete aberration assessment of a single frame within 5ms, significantly faster than the 50ms response time of traditional wavefront sensors.

[0083] Based on optical aberration evaluation results, the lens control unit 22 generates a control strategy for the contour of the intraocular lens. This unit optimizes the lens shape adjustment parameters using a deep reinforcement learning algorithm, the framework of which is as follows:

[0084] A t =π(S) t |θ π ),

[0085] Among them, A t S represents the control action at time t. t The current state (including the current aberration distribution and crystal position), π is the strategy function, and θ is the current state. π The parameters are those of the policy network. The reward function of this algorithm is designed as a weighted combination of aberration improvement and operational complexity. Practical application shows that this method can improve crystal position control accuracy to ±0.1mm and control angular error within ±2 degrees, significantly outperforming traditional manual adjustment methods.

[0086] The imaging quality adjustment unit 23 is responsible for real-time adjustment of imaging quality and intraocular lens shape. This unit receives control commands from the lens control unit 22 and uses micro-actuators to precisely adjust the imaging system and lens shape. The typical adjustment cycle is 20ms, effectively responding to intraoperative environmental changes and maintaining stable image quality.

[0087] The 3D deep learning motion compensation module 3 is another key innovation of this invention, solving the problem that traditional systems struggle to handle subtle eye movements. This module comprises four functional units: a feature extraction network unit 31, a motion feature parameter unit 32, a depth particle filter unit 33, and a compensation matrix generation unit 34.

[0088] Feature extraction network unit 31 is responsible for training the feature extraction network for the cornea, lens, iris, and pupil. This unit adopts a multi-branch cascaded convolutional network architecture, with dedicated feature extraction sub-networks designed for different intraocular tissues. The network training employs a transfer learning strategy, first pre-training on a large-scale general ophthalmic image dataset, and then fine-tuning on a surgical-specific dataset. This network maintains a feature extraction accuracy of over 95% under different lighting conditions and tissue states.

[0089] The motion feature parameter unit 32 acquires eye movement feature parameters based on the feature extraction results. This unit extracts the frequency, amplitude, and pattern features of eye movements through time series analysis. The main parameters include: microtremors (frequency 40-100Hz, amplitude 0.1-5.0μm), drift (frequency 0.1-0.5Hz, amplitude 20-100μm), and microsaccades (frequency 1-5Hz, amplitude 10-20μm). These parameters constitute the eye movement feature vector, providing a foundation for subsequent compensation.

[0090] The deep particle filtering unit 33 innovatively integrates deep learning with particle filtering algorithms to estimate the displacement and attitude angles of intraocular surgical instruments. Its core algorithm is as follows:

[0091]

[0092] Where, x t The state vector at time t (containing position and attitude information), y 1:t For the observation sequence, and Here, represents the weight and state of the i-th particle, N is the number of particles (set to 1000 in this system), and δ is the Dirac function. This unit optimizes the particle generation and weight update strategy through a deep neural network, solving the degradation problem of traditional particle filtering in eye movement estimation, improving tracking accuracy by 40% and computational efficiency by 2 times.

[0093] The compensation matrix generation unit 34 generates the compensation matrix for the virtual projection correction module based on the depth particle filtering results. The compensation matrix is ​​calculated using the following formula:

[0094] C = R·T·S

[0095] Where C is the compensation matrix, R is the rotation matrix, T is the translation matrix, and S is the scaling matrix. This matrix is ​​used to adjust the position and orientation of the guide line to counteract the effects of micro-movements of the eyeball. In practical applications, this compensation technology can control the navigation error within 0.1mm, meeting the requirements of high-precision surgery.

[0096] The registration algorithm module 4 is responsible for establishing spatial registration between the OCT and intraocular imaging systems, which is the foundation for achieving precise navigation. This module includes two functional units: a spatial mapping unit 41 and a guide line calculation unit 42.

[0097] Spatial mapping unit 41 establishes the spatial image mapping relationship between intraocular OCT and the intraocular imaging system. This unit adopts a multi-point matching initial registration method based on anatomical features and optimizes the registration accuracy through an iterative nearest-point algorithm. The registration transformation can be expressed as:

[0098]

[0099] Where T is the transformation matrix, p i and q i These represent the corresponding feature points in the OCT image and intraocular image, respectively, where n is the number of feature points. To address tissue deformation, this unit further introduces a non-rigid transformation model, using a B-spline function to describe local deformation. This method improves spatial registration accuracy to ±0.1 mm, significantly surpassing existing technologies.

[0100] The guideline calculation unit 42 calculates the guideline based on the spatial registration results, taking into account the spatial position, three-dimensional orientation, and image mapping relationship between the intraocular instruments / lens. Guideline generation involves the following steps: First, determining the ideal implantation position and orientation of the lens; second, calculating the optimal path based on the current instrument position and the target position; and finally, projecting the path into the microscope's field of view to generate the guideline. To accommodate different surgical stages, the system provides three guideline modes: position guidance (accuracy ±0.1mm), direction guidance (accuracy ±2°), and depth guidance (accuracy ±0.2mm), which surgeons can switch flexibly as needed.

[0101] Augmented reality projection module 5 visually overlays the registration guideline onto the surgical microscope's field of view, providing doctors with real-time navigation. This module includes four functional units: guideline generation unit 51, pose recognition unit 52, image fusion unit 53, and visual tracking unit 54.

[0102] The guide line generation unit 51 generates a guide line based on OCT and intraocular imaging data. This unit generates a visually clear and intuitive guide line according to the optimal implantation path calculated by the registration algorithm module 4. The guide line employs an adaptive display strategy, dynamically adjusting the line thickness (0.1-0.5mm), color (adjustable in RGB mode), and transparency (adjustable from 20% to 80%) based on the surgical stage and instrument position, minimizing interference with the surgical field while ensuring visibility.

[0103] The pose recognition unit 52 is responsible for identifying the spatial pose of intraocular surgical instruments. This unit employs a 3D convolutional neural network to extract instrument pose information from multi-angle images. The network input is a sequence of consecutive multi-frame images, and the output is the six-degree-of-freedom pose parameters of the instrument (position coordinates x, y, z and Euler angles α, β, γ). This method achieves a pose recognition accuracy of 98.5% on a standard test set, with an average processing time of 8 ms / frame.

[0104] Image fusion unit 53 overlays the guide line onto the surgical microscope image. This unit employs a semi-transparent overlay technique to clearly display the guide information while preserving the details of the original image. The fusion algorithm can be represented as:

[0105] I fusion (x,y)=(1-α)·Imicroscope (x,y)+α·I guideline (x,y),

[0106] Among them, I f usion is the fused image, I m icroscope is the original image from a microscope, I g uideline is the guide line image, and α is the transparency parameter (usually set to 0.6). This unit also implements edge enhancement and adaptive contrast adjustment to ensure clear visibility of the guide line under different lighting conditions.

[0107] The visual tracking unit 54 utilizes the optical characteristics of intraocular instruments for visual tracking to obtain the registration relationship between the registration guide line and the surgical instruments. This unit achieves real-time instrument tracking based on a combined algorithm of Kalman filtering and template matching. The tracking accuracy can reach ±0.05mm under normal surgical conditions, with a tracking frequency of 60Hz, effectively handling complex situations such as rapid instrument movement and partial occlusion.

[0108] Through the coordinated operation of the above five modules, the adaptive optics-calibrated intelligent intraocular lens implantation navigation system of this invention can achieve precise and stable surgical navigation, significantly improving the accuracy of intraocular lens implantation and postoperative visual quality. This system is particularly suitable for complex cataract surgeries (such as those involving high astigmatism, small pupils, iris damage, etc.) and multifocal lens implantation surgeries requiring ultra-high precision.

[0109] The intelligent intraocular imaging system 6 is the basic support system of this invention, responsible for identifying various intraocular structures and the location of surgical instruments. This system includes four functional units: a convolutional neural network unit 61, a contour generation unit 62, a deep neural network model unit 63, and a verification unit 64.

[0110] The convolutional neural network unit 61 is designed with a feature extraction algorithm to identify intraocular surgical instruments, lenses, corneas, irises, aqueous humor, and other tissue structures. This unit employs an improved U-Net network architecture, using an encoder-decoder structure to achieve accurate segmentation of intraocular tissues. The network's loss function is a weighted combination of Dice loss and cross-entropy loss.

[0111] L=α·L dice +(1-α)·L CE ,

[0112] Among them, L dice For Dice's loss, L CECross-entropy loss is used, and α is the weighting coefficient (set to 0.7). This network was trained on a dataset of 3000 diverse intraocular images, achieving an average segmentation accuracy of 97.8% and a processing speed of 15ms / frame, meeting real-time recognition requirements. Even in complex cases (such as late-stage cataract patients), the algorithm maintains an accuracy rate of over 92%, significantly outperforming traditional image processing methods.

[0113] The contour generation unit 62 segments the three-dimensional contours of intraocular tissues based on OCT images. This unit first performs denoising and enhancement preprocessing on the raw OCT data, then uses a 3D convolutional neural network for voxel-level segmentation, and finally reconstructs the three-dimensional surface model using the Marching Cubes algorithm. The contour generation accuracy can reach the voxel level (typical resolution of 0.01 mm). 3 This unit meets the requirements for high-precision surgical navigation. It also enables real-time contour updates (update frequency 5Hz), allowing it to adapt to dynamic changes in the morphology of intraocular tissues during surgery.

[0114] The Deep Neural Network Model Unit 63 is responsible for generating and training deep neural network models. This unit employs a modular design, dynamically combining and adjusting the network structure according to different tissue types and surgical stages. Core technologies include: few-shot learning (requiring only 20-30 samples to adapt to new instruments or special cases), active learning (prioritizing samples with high information content to improve training efficiency), and model distillation (transferring complex model knowledge to a lightweight network to improve running speed). This unit enables semi-automatic model updates; hospitals only need to provide a small amount of new data for the system to adaptively optimize, significantly reducing maintenance costs.

[0115] The validation unit 64 performs key point matching on contour images segmented from OCT images and intraocular images to verify the model's accuracy. This unit employs a robust feature matching method based on the RANSAC algorithm to extract and match SIFT feature points from both modalities. During validation, the system sets strict accuracy thresholds (positional error <0.15mm, angular error <3°), and only models that pass validation are applied to actual surgical navigation. The validation unit also has a self-diagnostic function, capable of detecting and reporting potential recognition errors, ensuring the system adopts a conservative strategy under uncertain conditions to guarantee surgical safety.

[0116] The system of the present invention also includes an eye movement prediction module 7 and a virtual projection correction module 8, which further enhance the adaptability and accuracy of the system.

[0117] The eye movement prediction module 7 constructs a central coordinate model of the intraocular lens based on the optical imaging principles and spatial registration of intraocular surgery. This module applies deep learning technology to establish a mapping relationship between eye movement characteristics and lens position changes, achieving accurate prediction of eye movements. Preferably, this module uses a Long Short-Term Memory (LSTM) network combined with an attention mechanism to model eye movement sequences. The algorithm can be expressed as:

[0118] h t =LSTM(x t ,h t-1 ,c t-1 ),

[0119]

[0120] Where, x t Let h be the eye position feature at time t. t and c t These are the hidden states and cell states of the LSTM, respectively. This module predicts the future position. It can predict the eye movement trajectory within 200ms, with a prediction error controlled within ±0.08mm, providing sufficient reaction time for micro-motion compensation.

[0121] The virtual projection correction module 8 tracks the spatial position between intraocular surgical instruments and the intraocular lens in real time and designs a coordinate transformation mechanism based on optical projection. This module constructs a complete optical projection model, accurately mapping three-dimensional spatial coordinates onto the two-dimensional microscope field of view. Preferably, this module employs binocular stereo vision technology, obtaining depth information through parallax calculation and combining it with the optical projection equation to achieve accurate coordinate system transformation. This technology enables the system to adapt to different models of surgical microscopes and optical systems, significantly improving the product's versatility and adaptability.

[0122] This invention establishes a complete spatial coordinate transformation chain through the collaborative work of the eye movement prediction module 7 and the virtual projection correction module 8, from the three-dimensional space inside the eye to the two-dimensional field of view of the microscope, and then to the augmented reality guide line, realizing seamless visual navigation and greatly improving surgical accuracy and the doctor's operating experience.

[0123] The eye movement prediction module 7 includes four functional units: a central coordinate model unit 71, a deep learning model unit 72, an estimation unit 73, and a tracking unit 74.

[0124] The central coordinate model unit 71 constructs the central coordinate model of the intraocular lens. Based on the anatomy of the eye, this unit establishes a three-dimensional coordinate system with the center of the lens capsule as the origin. Preferably, this coordinate system is determined by three key anatomical landmarks: the corneal apex, the pupillary center, and the center of the anterior surface of the lens. These three points form a reference plane, and combined with biological parameters such as axial length and corneal curvature, the ideal implantation position of the lens is accurately calculated. This model also considers individual differences, allowing for personalized adjustments based on preoperative examination data (such as corneal topography and axial length measurement) to adapt to the ocular characteristics of different patients.

[0125] Deep learning model unit 72 obtains eye movement feature parameters using a deep learning network model. This unit employs a multimodal fusion strategy, simultaneously processing image data from an eye-tracking camera and a microscope. Preferably, this unit uses a two-stream network architecture, processing spatial features in one stream and temporal features in the other, finally combining the two streams of information through a feature fusion layer. Trained on an eye movement dataset (containing eye movement data from 500 patients with different types of eye movements), this network accurately identifies and predicts three main types of micro-movements: microtremors, drift, and microsaccades. The network output includes key parameters such as frequency, amplitude, and direction, providing accurate data for subsequent compensation.

[0126] The estimation unit 73 uses a deep particle filter algorithm to estimate the displacement and attitude angle of the intraocular surgical instruments. The particle filter algorithm first generates a set of particles based on the motion model, with each particle representing a possible state hypothesis; then it updates the particle weights based on observation data; finally, it updates the particle distribution through resampling. Preferably, the deep particle filter of this invention introduces a neural network to optimize particle generation and weight calculation, and the algorithm is expressed as follows:

[0127]

[0128] Where, q θ The proposed distribution, parameterized by a neural network, can adaptively generate particles based on historical states and current observations. This algorithm significantly improves the stability and accuracy of tracking, maintaining reliable tracking performance even under conditions of rapid eye movement or partial occlusion of the field of vision.

[0129] The tracking unit 74 tracks OCT data to estimate the spatial position of intraocular surgical instruments. This unit reconstructs the three-dimensional contour and positional information of the instruments by analyzing OCT scan image sequences in real time. Preferably, this unit employs a sliding window strategy, processing multiple consecutive frames of OCT data simultaneously to enhance the stability of position estimation. The tracking results are updated at a frequency of 10Hz, providing key information such as instrument position, attitude, and movement speed, supporting surgical navigation and micro-motion compensation. This unit also implements automatic instrument type recognition, enabling it to adapt to different types of surgical instruments and improving the system's versatility.

[0130] The virtual projection correction module 8 includes four functional units: spatial position tracking unit 81, image acquisition unit 82, coordinate transformation unit 83, and accuracy correction unit 84.

[0131] The spatial positioning tracking unit 81 tracks the spatial position between intraocular surgical instruments and the intraocular lens in real time. This unit employs a multi-sensor fusion strategy, comprehensively utilizing microscope images, OCT data, and special marker information to achieve high-precision positioning tracking. Preferably, this unit uses an extended Kalman filter algorithm to fuse multi-source data; the state update and observation equations are as follows:

[0132]

[0133] In this unit, the extended Kalman filter algorithm is used to fuse multi-source data. The state update and observation equations are as follows:

[0134]

[0135] P t|t =(IK t H t )P t|t-1 ,

[0136] in, For state estimation, P is the covariance matrix, K is the Kalman gain, z is the observation value, f and h are the state transition function and observation function, respectively, F and H are the corresponding Jacobian matrices, and Q and R are the process noise and observation noise covariances. The algorithm can control the position tracking accuracy to ±0.05mm and the attitude angle accuracy to ±1°, meeting the requirements of precision surgery.

[0137] The image acquisition unit 82 acquires real-time intraocular images based on the position of the intraocular instruments. This unit is tightly integrated with the microscope module 1 and automatically adjusts imaging parameters (such as depth of focus, field of view, and exposure settings) according to the current surgical stage and instrument position. Preferably, this unit implements intelligent focusing, automatically adjusting the focal plane position by analyzing the image sharpness distribution to ensure clear imaging of key areas. This technology significantly reduces the surgeon's workload, allowing them to focus on the surgery itself without frequently adjusting microscope settings.

[0138] The coordinate transformation unit 83 is designed based on an optical projection-based coordinate transformation mechanism. This unit is responsible for converting three-dimensional spatial coordinates into two-dimensional microscope field-of-view coordinates, which is crucial for achieving precise guided projection. Preferably, this unit adopts a perspective projection model, combined with the characteristics of the optical system, to establish accurate coordinate transformation relationships:

[0139]

[0140] Where (u,v) are pixel coordinates, (X,Y,Z) are world coordinates, K is the camera intrinsic parameter matrix, and R and t are the rotation matrix and translation vector, respectively. This unit obtains precise parameter values ​​through a calibration procedure and considers the nonlinear distortion of the microscope optical system to ensure projection accuracy. Experiments show that this coordinate transformation mechanism can control the projection error within 1 pixel, equivalent to a physical size of 0.02 mm at 4K resolution, meeting the requirements of ultra-precise surgery.

[0141] The accuracy correction unit 84 performs tracking accuracy correction through the eye movement prediction module 7 and spatial registration. This unit continuously monitors and optimizes tracking accuracy via a closed-loop feedback mechanism. Preferably, this unit is designed with a self-calibration program that assesses the current tracking accuracy based on the reprojection error of specific marker points and automatically adjusts the tracking parameters. The calibration cycle is 500ms, enabling timely responses to accuracy reductions caused by factors such as changes in illumination and tissue deformation. This unit also implements an anomaly detection function; when the tracking accuracy falls below a preset threshold (typically 0.2mm), the system issues a warning and provides auxiliary positioning information to ensure surgical safety.

[0142] The present invention also provides an intelligent intraocular lens implantation navigation method with adaptive optics calibration, including a complete navigation process of steps S100 to S1000.

[0143] Step S100: Acquire real-time intraocular image data. This step uses microscope module 1 to acquire high-resolution intraocular images, providing basic data for subsequent processing. Preferably, timestamps and microscope parameters (such as magnification, working distance, etc.) are recorded simultaneously during acquisition to facilitate subsequent processing and analysis. The image acquisition frequency is no less than 60fps to ensure the capture of rapid micro-movements of the eyeball and the movement of surgical instruments.

[0144] Step S200 involves real-time evaluation and correction of intraocular tissue optical aberration distribution using the adaptive optics system module 2. This step first analyzes the intraocular image quality, identifies aberration types and distributions, then calculates the optimal correction parameters, and finally performs real-time correction via the optical system. Preferably, the correction process employs a closed-loop control strategy, continuously evaluating and adjusting to progressively optimize image quality until preset standards are met (e.g., contrast > 0.8, sharpness > 0.85, etc.). This step significantly improves intraocular image quality, laying the foundation for subsequent analysis and recognition.

[0145] Step S300 involves using the 3D deep learning motion compensation module 3 to predict eye movements and compensate for micro-motion interference. This step first extracts eye movement features from the image sequence, then predicts short-term movement trends using a deep learning model, and finally generates compensation parameters to counteract the effects of micro-motion. Preferably, the compensation employs a feedforward-feedback combined strategy, both predictively compensating for regular movements and correcting prediction errors in real time, achieving comprehensive and effective micro-motion suppression. Experiments show that this step can reduce the impact of micro-motion by more than 85%, significantly improving surgical accuracy.

[0146] Step S400 involves establishing spatial registration between the OCT and intraocular imaging systems using registration algorithm module 4. This step achieves accurate fusion of multimodal data, providing a unified reference coordinate system for 3D navigation. Preferably, the registration process employs a two-stage strategy of coarse registration followed by fine registration. First, the initial transformation is quickly determined through feature point matching. Then, the transformation parameters are precisely adjusted through an iterative optimization algorithm, ultimately achieving sub-millimeter level registration accuracy. This step solves the problem of inaccurate multimodal data fusion in traditional systems, providing a solid foundation for precise navigation.

[0147] Steps S500 to S800 implement the complete augmented reality guidance process. First, a guideline is generated based on OCT and intraocular imaging. Then, the surgical instrument posture is identified, and the guideline is superimposed onto the microscope's field of view. Finally, visual tracking ensures the guideline and instrument maintain correct registration. Preferably, the guideline display employs a "stepped" strategy, dynamically adjusting the display method according to the distance between the instrument and the target location. More precise guidance is provided at close range, while a more macroscopic directional indication is provided at long range. This innovative display strategy makes the navigation process more intuitive and smooth, reducing the cognitive burden on doctors.

[0148] Step S900 involves using the intelligent intraocular imaging system 6 to identify intraocular structures and instrument locations. This step utilizes deep learning technology to analyze intraocular images in real time, identifying key structures (such as the cornea, iris, and lens capsule) and instrument locations. Preferably, the identification results are displayed intuitively in an augmented reality view using different colors and markings, helping doctors better understand the surgical scenario. This function is particularly important for patients with atypical anatomy or diseased anatomical structures.

[0149] Step S1000 involves adjusting the position of the surgical instruments in real time based on the registration relationship between the registration guide line and the surgical instruments to achieve precise implantation of the intraocular lens (IOL). This step is the final goal of the entire navigation process, guiding the surgeon to accurately implant the IOL at the predetermined location through the synergistic effect of the preceding steps. Preferably, the system provides a real-time feedback mechanism, indicating the deviation between the current position and the target position through visual and audible cues. When a preset accuracy threshold is reached (typically a center offset <0.1mm and a tilt angle <2°), the system provides a confirmation signal, indicating that the implantation position meets the requirements. This step significantly improves the accuracy of IOL implantation, especially for patients with high astigmatism and in multifocal lens implantation surgeries.

[0150] Example 1

[0151] The system of this invention was applied in a cataract surgery patient with high corneal astigmatism (4.5D). Preoperatively, the patient's axial length was measured to be 24.8 mm using a wavefront aberrometer, and corneal curvature K1 = 42.5D, K2 = 47.0D. The system automatically calculated the optimal intraocular lens power as +21.5D and the astigmatism correction axis as 85°. During the surgery, the system corrected optical aberrations in real time, compensated for micro-movements of the eyeball (amplitude up to 120 μm), and indicated the optimal implantation position and direction using augmented reality guide lines. The final lens implantation position deviation was 0.08 mm, and the axis deviation was 1.8°, both superior to the traditional surgical standards (0.3 mm and 5°, respectively). At a 3-month follow-up, the patient's uncorrected visual acuity reached 1.0, astigmatism decreased to 0.75D, and the visual quality satisfaction score was 9.2 (out of 10).

[0152] Example 2

[0153] The system of this invention was applied in the surgery of a complex cataract patient with a small pupil (diameter <4mm) and iris injury. Due to the small and irregular pupil, traditional surgery struggles to ensure precise lens implantation. This system accurately identifies the edge of the iris defect and the location of the lens capsule using an intelligent intraocular imaging system 6, assesses and corrects complex optical aberrations in real time, and uses augmented reality guide lines to indicate a safe operating path and ideal implantation position. The entire surgery took only 17 minutes, 35% shorter than the average time for similar surgeries in hospitals. The lens was successfully implanted in the center, with no postoperative complications, and the patient's vision recovered well.

[0154] The above embodiments demonstrate that the adaptive optics-calibrated intelligent intraocular lens implantation navigation system and method of the present invention perform excellently in both complex and routine cataract surgeries, significantly improving surgical accuracy, shortening surgical time, and enhancing postoperative visual quality, and has broad clinical application prospects.

[0155] It should be noted that 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 principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent intraocular lens implantation navigation system with adaptive optics calibration, characterized in that, include: Microscope module, used to acquire real-time image data inside the eye; An adaptive optics system module, which is signal-connected to the microscope module, is used to evaluate and correct the optical aberration distribution of intraocular tissues in real time. A 3D deep learning motion compensation module is used to predict eye movements and compensate for micro-motion disturbances; The registration algorithm module is used to establish spatial registration between OCT and intraocular imaging systems; Augmented reality projection module for overlaying registration guide lines onto surgical microscope images; Intelligent intraocular imaging system for identifying intraocular structures and instrument locations; The eye movement prediction module is used to construct a central coordinate model of the artificial lens based on the optical imaging principle and spatial registration of intraocular surgery. It applies deep learning technology to establish a mapping relationship between eye movement characteristics and lens position changes, thereby enabling the prediction of eye movement. A virtual projection correction module is used to track the spatial position between intraocular surgical instruments and artificial lenses in real time, and a coordinate transformation mechanism based on optical projection is designed. The adaptive optics system module includes: The optical aberration assessment unit is used to train a virtual optical imaging module for intraocular images based on image data, and to evaluate the distribution of optical aberrations in intraocular tissues in real time. A crystal control unit is used to generate a control strategy for the contour of the artificial lens based on the optical aberration assessment. The crystal control unit optimizes the crystal shape adjustment parameters through a deep reinforcement learning algorithm, the framework of which is as follows: ， in, This represents the control action at time t. This represents the current state, which includes the current aberration distribution and crystal position. For the policy function, For the policy network parameters, the reward function of the deep reinforcement learning algorithm is designed as a weighted combination of aberration improvement degree and operational complexity; The imaging quality adjustment unit is used to adjust the imaging quality and the shape of the intraocular lens in real time. The imaging quality adjustment unit receives control commands from the lens control unit and adjusts the imaging system and the shape of the lens through micro-actuators. The three-dimensional deep learning motion compensation module includes: Feature extraction network unit, used to train feature extraction networks for cornea, lens, iris, and pupil; The motion feature parameter unit is used to acquire eye movement feature parameters; A depth particle filter unit is used to estimate the displacement and attitude angle of intraocular surgical instruments; The compensation matrix generation unit is used to generate a compensation matrix based on the virtual projection correction module.

2. The intelligent artificial lens implantation navigation system with adaptive optics calibration according to claim 1, characterized in that, The registration algorithm module includes: The spatial mapping unit is used to establish the spatial image mapping relationship between intraocular OCT and intraocular imaging systems. The guideline calculation unit is used to calculate the guideline based on the spatial position, three-dimensional orientation, and image mapping relationship between the intraocular instruments, the lens, and the optical channels.

3. The intelligent intraocular lens implantation navigation system with adaptive optics calibration according to claim 1, characterized in that, The augmented reality projection module includes: A guideline generation unit is used to generate guidelines based on OCT and intraocular imaging. An attitude recognition unit is used to identify the spatial attitude of intraocular surgical instruments; An image fusion unit is used to overlay registration guide lines onto surgical microscope images; The visual tracking unit is used to perform visual tracking using the optical characteristics of intraocular instruments to obtain the registration relationship between the registration guide line and the surgical instruments.

4. The intelligent intraocular lens implantation navigation system with adaptive optics calibration according to claim 1, characterized in that, The intelligent intraocular imaging system includes: Convolutional neural network units are used to design feature extraction algorithms to identify intraocular surgical instruments, lenses, corneas, lenticels, pupils, irises, and aqueous humor; The contour generation unit is used to segment the three-dimensional contour of intraocular tissues based on OCT images; Deep neural network model unit, used to generate and train deep neural network models; The verification unit is used to match key points of the contour image segmented from OCT images and intraocular images to verify the accuracy of the model.

5. The intelligent intraocular lens implantation navigation system with adaptive optics calibration according to claim 1, characterized in that, The eye movement prediction module includes: Central coordinate model unit, used to construct the central coordinate model of artificial crystals; Deep learning model unit, used to obtain eye movement feature parameters with the help of deep learning network model; The estimation unit is used to estimate the displacement and attitude angle of intraocular surgical instruments using a deep particle filter algorithm. The tracking unit is used to track OCT data and estimate the spatial position of intraocular surgical instruments.

6. The intelligent intraocular lens implantation navigation system with adaptive optics calibration according to claim 1, characterized in that, The virtual projection correction module includes: The spatial positioning tracking unit is used to track the spatial position between intraocular surgical instruments and the intraocular lens in real time; The image acquisition unit is used to acquire real-time intraocular images based on the position of the intraocular instruments. Coordinate transformation unit, used to design coordinate transformation mechanisms based on optical projection; The accuracy correction unit is used to perform tracking accuracy correction through the eye movement prediction module and spatial registration.

Images