Optimization method for positioning measurement of implanted crystalline lens after ophthalmologic operation

Abstract

The invention relates to an optimization method for positioning measurement of an implanted crystalline lens after an ophthalmic operation, which comprises the following steps of: scanning an implanted astigmatism correction type artificial crystalline lens by using frequency sweep source optical coherence tomography (SS-OCT) to obtain image data containing a positioning mark, processing the image data by using a deep learning framework, and obtaining a positioning mark of the implanted astigmatism correction type artificial crystalline lens. Automatically identifying the astigmatism mark on the intraocular lens, and calculating the actual implantation axial position of the intraocular lens according to the identified astigmatism mark. The method has the advantages that axial position measurement can be completed without mydriasis, discomfort of a patient is reduced, dependence on operators is reduced, and measurement accuracy and efficiency are improved. The system comprises SS-OCT equipment and a deep learning processing module, and realizes automatic and non-invasive measurement of the implantation axial position of the artificial lens.

Description

Technical Field

[0001] The invention relates to the technical field of ophthalmic surgery, and in particular to an optimization method for positioning measurement of an implanted lens after ophthalmic surgery. Background Art

[0002] Astigmatism-correcting intraocular lenses play an important role in correcting refractive errors, and the accuracy of their implantation is crucial to the correction effect. Intraocular lenses that can correct human eye astigmatism include phakic posterior chamber intraocular lenses and intraocular lenses. Doctors decide whether to correct astigmatism based on the patient's astigmatism and select the appropriate intraocular lens based on the actual situation. Astigmatism intraocular lenses have astigmatism marks to help doctors align them to the appropriate position during surgery, thereby achieving effective astigmatism correction. Among them, ICL is an intraocular lens implanted in the posterior chamber of the eye, mainly used to correct refractive errors such as high myopia and astigmatism. Its characteristics include astigmatism marks, which facilitate accurate positioning of the actual axis of the lens implantation. There is a central hole with a diameter of 360 microns in the center of the ICL to facilitate the circulation of aqueous humor. IOL is an intraocular lens used after cataract surgery. It is implanted in the eye to replace the natural lens and can also correct astigmatism. The astigmatism correction effect of these two lenses depends on the precise positioning during surgery and the accuracy of the actual axis of implantation.

[0003] At present, the actual axial position of the implanted astigmatism-correcting intraocular lens is traditionally measured by indirect measurement of the axial position of the TICL through refractive change analysis before and after surgery or wavefront aberration instruments, but these methods lack precise imaging verification and are difficult to provide accurate axial position information. In addition, the method used more often in clinical practice is anterior segment photography measurement after mydriasis. This method not only requires a long time for mydriasis preparation, but also bears the risk of increased intraocular pressure and angle closure due to mydriasis; during the examination, the imaging of the anterior segment photography depends on strong visible light, and the patient is uncomfortable during the examination due to strong light. After the examination, the patient has blurred vision and photophobia for 4-6 hours due to the mydriatic agent. And it is heavily dependent on the operator's experience and manual measurement, and there are subjective errors.

[0004] Therefore, an optimization method for the positioning and measurement of implanted lenses after ophthalmic surgery is proposed. In order to locate the implantation axis of the postoperative astigmatism-correcting intraocular lens in the anterior segment of the patient's eye, there is a great demand for designing an imaging and quantitative evaluation of the implantation axis of the intraocular lens without mydriatic agents. Among them, swept-source optical coherence tomography (SS-OCT) is an advanced ophthalmic imaging technology. By using near-infrared light to perform high-resolution imaging of the anterior and posterior segments of the eyeball, it can provide three-dimensional structure and position information of the postoperative intraocular lens in the anterior segment of the eye. The non-visible light imaging characteristics of SS-OCT make it have significant advantages in reducing patient discomfort and improving patient cooperation, and it is particularly suitable for postoperative evaluations that require precise imaging. Therefore, designing an optimization method for quantitative evaluation of the implantation axis of the postoperative astigmatism-correcting intraocular lens based on SS-OCT and artificial intelligence is conducive to more intuitive follow-up and observation of the position of the astigmatism-correcting intraocular lens by clinicians, and significantly improves the efficiency of postoperative evaluation during clinical diagnosis and treatment. Summary of the invention

[0005] The technical problem to be solved by the present invention is that positioning measurement imaging verification is inaccurate and difficult to provide accurate axial position information, requires long-term mydriasis preparation, relies on strong light during the examination, and makes the patient uncomfortable before and after the examination.

[0006] In order to solve the above technical problems, the technical solution provided by the present invention is: an optimization method for positioning measurement of implanted lens after ophthalmic surgery, which comprises the following steps:

[0007] The implanted astigmatism-correcting intraocular lens is scanned using swept-source optical coherence tomography (SS-OCT) to obtain image data containing positioning markers;

[0008] Processing the image data using a deep learning framework to automatically identify astigmatism markers on the intraocular lens;

[0009] Based on the identified astigmatism markers, the actual implantation axis of the intraocular lens is calculated.

[0010] Furthermore, the deep learning framework uses a convolutional neural network (CNN) model to accurately identify astigmatism markers and central holes in SS-OCT images.

[0011] Furthermore, the SS-OCT scan is performed without using mydriatic agents to reduce patient discomfort.

[0012] Furthermore, the image data is B-scan image data, and the multi-layer B-scan image data is used to generate a three-dimensional point cloud matrix, and a feature map for a deep learning model is obtained through dimensionality reduction processing.

[0013] Furthermore, the processing module of the deep learning framework includes a first deep learning model for detecting the depth of the artificial lens and a second deep learning model for identifying positioning marks.

[0014] The present invention has the following advantages:

[0015] 1. Improve patient comfort: SS-OCT's non-visible light scanning avoids the mydriasis process, reduces patient discomfort, and improves patient cooperation.

[0016] 2. Automated measurement: Combined with deep learning technology, automated measurement of the axial position of intraocular lens implantation is achieved, reducing dependence on operators and reducing human errors.

[0017] 3. High precision and reliability: The deep learning model has high-precision image recognition capabilities, ensuring the accuracy and reliability of the measurement results.

[0018] 4. Improve work efficiency: Measurement can be completed without pupil dilation, which shortens the detection time and improves clinical work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the overall process of an optimization method for positioning measurement of a lens implanted after ophthalmic surgery according to the present invention.

[0020] Figure 2 It is a schematic diagram of a B-scan image of an artificial lens with astigmatism correction obtained by using SS-OCT in an optimization method for positioning measurement of a lens implanted after ophthalmic surgery according to the present invention.

[0021] Figure 3 The present invention is an optimization method for positioning measurement of implanted lenses after ophthalmic surgery, which uses a B-scan image of an astigmatism-correcting intraocular lens acquired by SS-OCT to obtain a recognition result through a first deep learning model.

[0022] Figure 4 The present invention is a method for optimizing the positioning measurement of a lens implanted after ophthalmic surgery, based on a feature map of 384×384 pixels containing astigmatism markers in a deep learning framework.

[0023] Figure 5 The present invention is a method for optimizing the positioning measurement of a lens implanted after ophthalmic surgery, based on the recognition result of a feature map containing astigmatism markers of 384×384 pixels in a deep learning framework through a second deep learning model.

[0024] Figure 6 It is a schematic diagram of implantation axis calculated based on the recognition result of the second deep learning model in the deep learning framework of an optimization method for positioning measurement of an implanted lens after ophthalmic surgery of the present invention.

[0025] Figure 7 The invention discloses an optimization method for positioning measurement of a lens implanted after ophthalmic surgery and an implant axis measurement result imaged by the optimization method in a bright environment. DETAILED DESCRIPTION

[0026] The present invention is further described in detail below in conjunction with the embodiments.

[0027] Example 1

[0028] An optimization method for positioning measurement of implanted lens after ophthalmic surgery, comprising the following modules:

[0029] SS-OCT scanning module: used to obtain high-resolution three-dimensional image data of the patient's eyes.

[0030] Data preprocessing module: processes the original image data and generates a format suitable for deep learning model input.

[0031] Deep learning recognition module (see Figure 1 ): Mainly includes two deep learning models:

[0032] Model 1: Used to detect the depth position of the intraocular lens.

[0033] Model 2: Used to identify positioning markers on the intraocular lens.

[0034] Axis calculation module: Calculate the actual implantation axis of the intraocular lens based on the recognition results.

[0035] Result display module: presents the calculation results in a visual form for reference by clinicians.

[0036] The following steps are involved:

[0037] Step 1: SS-OCT scan to obtain image data

[0038] The patient did not use mydriasis. In a dark room, the SS-OCT device was used to scan the eye implanted with an astigmatism-correcting intraocular lens to obtain high-resolution three-dimensional image data containing positioning markers.

[0039] Patient preparation:

[0040] Ambient lighting: Performed in a dark room. Wait 60 seconds after turning off the ambient light in the examination room to allow the patient's pupil to adapt to the dark so as to increase the pupil diameter and improve the visibility of the astigmatism mark.

[0041] Gaze fixation: Instruct the patient to look at the internal fixation light and keep the eyes steady without blinking or moving the eyes.

[0042] Scan parameter settings:

[0043] Mode: front section mode.

[0044] Scanning range: 9mm×9mm.

[0045] Number of B-scans: 384 layers, each layer has an image resolution of 384 × 3296 pixels. An example of a B-scan of one patient is shown below: Figure 2 .

[0046] Step 2: Data Preprocessing

[0047] The acquired raw image data is preprocessed, mainly including:

[0048] Image denoising: Use Gaussian filtering and other methods to reduce image noise and enhance image quality.

[0049] Image normalization: Normalize the image pixel values ​​to the range of [0, 1] to facilitate model processing.

[0050] Step 3: Intraocular lens depth detection

[0051] Use model 1 (such as a three-dimensional convolutional neural network, 3D-CNN) to detect the depth position range of the intraocular lens zICL H, zICL L in the intermediate B-scan image

[0052] Model structure:

[0053] For example, in the structure of Mask-RCNN, in order to save computing resources, only the middle B-scan is input to detect the depth of the intraocular lens. The model outputs the mask and target detection frame of the ICL. Assuming the coordinates of the upper left corner of the target detection frame are (XH, YH) and the coordinates of the lower right corner are (XL, YL), the depth position range of the intraocular lens is zICL H=YH; zICL L=YL

[0054] Model training:

[0055] Collect 240 images with intraocular lens masks. Training set loss: 0.4594; validation set loss: 0.4518; test set loss: 0.5207. On the test set, the Dice similarity coefficients are 92.15%, and the average Intersection over Union (mIOU) scores are 83.59%.

[0056] 3D point cloud generation: Multi-layer B-scan image data I (x, y, z) is cropped according to the depth of the intraocular lens to extract, and then a 3D point cloud matrix V (x, y, z) is generated.

[0057] Where x, y are the two-dimensional space coordinates, and z is the depth layer.

[0058] Feature map generation: Use 1×1 convolution to convolve the 3D point cloud matrix along the height direction and compress it into a clear feature map containing the astigmatism markers of the artificial lens, such as Figure 4 .

[0059] Step 4: Positioning Mark Identification

[0060] Model 2 (such as Mask R-CNN) is used to identify the positioning marks on the intraocular lens on the feature map. The recognition results are as follows: Figure 5 .

[0061] Model structure:

[0062] For example, the structure of Mask RCNN inputs the feature map and outputs the located astigmatism marks and center holes.

[0063] Model training:

[0064] Collect 400 images with intraocular lens masks. Training set loss: 0.4867; validation set loss: 0.4826; test set loss: 0.4872. On the test set, the Dice similarity coefficients are 93.75%, and the average Intersection over Union (mIOU) scores are 86.42%.

[0065] The model outputs the results of identifying the astigmatism markers and the central hole (if any), such as Figure 5 .Position information of astigmatism marks (xi, yi) where i=1, 2 represents two marks.

[0066] Step 5: Axis calculation

[0067] According to the identified position of the positioning mark, the actual implantation axis angle θ of the intraocular lens is calculated.

[0068] Double-mark identification:

[0069] Calculate the slope of the line between two anchor marks:

[0070]

[0071] Calculate the axis angle:

[0072]

[0073] Normalize the angle to the range [0°,180°].

[0074] Single-marker recognition:

[0075] The central hole position (x c ,y c ) for reference.

[0076] Calculate the slope of the line connecting the marked point and the center hole:

[0077]

[0078] Calculate the axis angle:

[0079]

[0080] Step 6: Results display and analysis

[0081] Results:

[0082] The calculated axial angle θ is displayed on the user interface.

[0083] like Figure 6 , shows the double astigmatism marker measurements of one patient measured by the deep learning-based framework using SS-OCT imaging in a darkroom.

[0084] like Figure 7 , shows the single astigmatism marker measurements of the same patient measured by the deep learning-based framework using SS-OCT imaging in a bright environment.

[0085] Result analysis:

[0086] Compare the actual implant axis with the surgically planned axis and calculate the deviation angle

[0087] Δθ.

[0088] If the deviation exceeds the allowable range, the doctor can be prompted to take appropriate measures.

[0089] Through the above-mentioned implementation mode, the present invention achieves the following optimizations: small measurement error and axial angle error: compared with the traditional anterior segment photography method, the average measurement error is within ±0.5°, which meets the clinical accuracy requirements; high efficiency: no mydriasis is required, the examination time is short, and the measurement time for each patient is shortened to about 3 minutes, which is suitable for large-scale clinical applications; automation: the whole process is automated, and operators can use it proficiently after simple training without manual intervention, which reduces the difficulty of operation and reduces human errors; high patient comfort: all patients are examined without mydriasis, avoiding the discomfort caused by mydriasis and improving patient acceptance.

[0090] Therefore, the automated and high-precision measurement of the axial position of the astigmatism-correcting intraocular lens implanted after ophthalmic surgery is achieved, which has important clinical application value.

[0091] Although the present invention has been described in detail above with general descriptions and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements may be made thereto based on the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all fall within the scope of protection claimed by the present invention.

Claims

1. A method for optimizing the positioning measurement of a lens implanted after ophthalmic surgery, characterized in that: It includes the following steps: The implanted astigmatism-correcting intraocular lens is scanned using swept-source optical coherence tomography (SS-OCT) to obtain image data containing positioning markers; Processing the image data using a deep learning framework to automatically identify astigmatism markers on the intraocular lens; Based on the identified astigmatism markers, the actual implantation axis of the intraocular lens is calculated.

2. The method for optimizing the positioning measurement of a lens implanted after ophthalmic surgery according to claim 1, characterized in that: The deep learning framework uses a convolutional neural network (CNN) model to accurately identify astigmatism markers and central holes in SS-OCT images.

3. The method for optimizing the positioning measurement of a lens implanted after ophthalmic surgery according to claim 1, characterized in that: The SS-OCT scans were performed without the use of mydriatic agents to reduce patient discomfort.

4. The method for optimizing the positioning measurement of a lens implanted after ophthalmic surgery according to claim 1, characterized in that: The image data is B-scan image data. Multi-layer B-scan image data is used to generate a three-dimensional point cloud matrix, and a feature map for a deep learning model is obtained through dimensionality reduction processing.

5. The method for optimizing the positioning measurement of a lens implanted after ophthalmic surgery according to claim 1, characterized in that: The processing module of the deep learning framework includes a first deep learning model for detecting the depth of the intraocular lens and a second deep learning model for identifying positioning marks.

Images