An eyeball a-ultrasonic biological measurement method and system assisted by b-ultrasonic image

Abstract

The application discloses an eyeball A-ultrasonic biological measurement method and system under the assistance of B-ultrasonic images, and relates to the field of ophthalmic image processing. The method comprises the following steps: standard scanning of eyeballs is performed by using a B-ultrasonic scanning technology to obtain standard eyeball section images; an automatic judgment model is constructed to mark the standard eyeball section images to obtain three surface distances; distance parameter measurement is performed on the standard eyeball section images based on the three surface distances to obtain a first measurement result; standard scanning of eyeballs is performed by using an A-ultrasonic scanning technology to obtain a standard A-ultrasonic scanning waveform image; the standard A-ultrasonic scanning waveform image is marked based on the first measurement result to obtain four surface waves; and the second measurement result is obtained based on the four surface waves to complete A-ultrasonic waveform biological parameter measurement. The application can effectively improve the accuracy and efficiency of A-ultrasonic waveform identification in the case that existing A-ultrasonic technology is prone to inaccurate eyeball waveform identification in some cases.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of ophthalmic image processing, in particular to an eyeball A-mode ultrasonic biological measurement method and system assisted by B-mode ultrasonic images. BACKGROUND

[0002] Eyeball biological measurement refers to measuring various parameters of eyeball structure by various examination methods to provide reference for diagnosis and treatment of eye diseases. A-mode ultrasonic is also called A-mode ultrasonic. It detects the echo of sound waves according to the relationship between time and amplitude of sound waves. Ophthalmic ultrasonic imaging is to measure biological parameters of eyes by A-mode ultrasonic. When the probe is placed in front of the eyeball, the sound waves propagate forward. According to the propagation characteristics of sound waves, each interface with density difference reflects once, forming a wave peak. The returning echo will be arranged according to the returning time, and the height of the wave peak represents the intensity of the echo. The higher the wave peak, the stronger the echo. A-mode ultrasonic detects the position of the wave peak and calculates the corresponding biological parameters of the eyeball (such as corneal thickness, anterior chamber depth, lens thickness, vitreous length and axial length of the eyeball) through sound velocity matching. Accurate measurement of these parameters is crucial for calculating the degree of intraocular artificial lens (1.0mm of axial measurement error will cause an average of 3.0D of refractive error).

[0003] Although the existing A-mode ultrasonic biological measurement technology can judge the interfaces of different structures in the eyeball for most people, and thus accurately measure various biological parameters, it is difficult to accurately identify the A-mode ultrasonic waveform for special types of cataract patients, retinal diseases and vitreous opacity disease patients, which requires the operator to have rich experience to adjust the waveform identification, thereby affecting the clinical work efficiency and easily leading to measurement error of eyeball biological parameters.

[0004] The existing B-mode ultrasonic imaging technology can easily judge the interfaces of different structures in the eye (such as posterior corneal surface, anterior lens surface, posterior lens surface and retinal surface), but due to the insufficient resolution of B-mode ultrasonic imaging, it is difficult to accurately define and measure the distance between each structure interface in the eyeball, thereby affecting the accurate measurement of anterior chamber depth, lens thickness, vitreous length and axial length of the eyeball and other eyeball biological parameters. SUMMARY

[0005] To solve the above technical problems, the present application provides an accurate eyeball A-mode ultrasonic waveform automatic identification biological measurement method and system combined with B-mode ultrasonic image measurement technology, which is beneficial to efficiently and accurately complete the eyeball A-mode ultrasonic biological parameter measurement.

[0006] To achieve the above purpose, the present application provides an eyeball A-mode ultrasonic biological measurement method assisted by B-mode ultrasonic images, comprising the following steps:

[0007] adopting B ultrasonic scanning technology to perform standard scanning on the eyeball to obtain standard eyeball section images;

[0008] constructing an automatic judgment model to mark the standard eyeball section images to obtain: corneal anterior surface, denoted as S1; lens anterior surface, denoted as S2; lens posterior surface, denoted as S3 and retinal surface, denoted as S4, three surface distances between the four surfaces, the automatic judgment model adopts an image segmentation positioning algorithm based on a YOLO model to realize automatic position judgment, and the steps include:

[0009] 1. Image annotation

[0010] using Labelme annotation software to perform instance segmentation annotation on the images, marking the positions of the corneal posterior surface, the lens anterior surface, the lens posterior surface and the retinal surface respectively, and then generating a label file containing object categories and bounding boxes;

[0011] 2. Model training

[0012] selecting a YOLOv11 model for training, adjusting the input resolution of the picture to 640*640, setting the batch size to 16, using an Adam optimizer, setting the initial learning rate to 0.001, setting the learning rate decay factor to 0.01, setting the random horizontal flip to 0.5, setting the random scaling to 0.5, and enabling the Mosaic data enhancement strategy;

[0013] 3. Model inference

[0014] setting the confidence threshold of the inference process to 0.25, the model outputs the bounding box coordinates of the corneal anterior surface, the lens anterior surface, the lens posterior surface and the retinal surface, calculates the center point coordinates of each bounding box according to the bounding box coordinates, and then calculates the three surface distances through the value of the X-axis of the center point coordinates; based on the three surface distances, distance parameter measurement is performed on the standard eyeball section images to obtain a first measurement result; the distance parameters include: corneal anterior surface to lens anterior surface distance, corneal anterior surface to lens posterior surface distance and corneal anterior surface to retinal surface distance;

[0015] adopting A ultrasonic scanning technology to perform standard scanning on the eyeball to obtain standard A ultrasonic scanning waveform images;

[0016] based on the first measurement result, marking the standard A ultrasonic scanning waveform images to obtain four surface waves, including: corneal anterior surface wave, lens anterior surface wave, lens posterior surface wave and retinal anterior surface wave;

[0017] based on the four surface waves, obtaining a second measurement result to complete A ultrasonic waveform biological parameter measurement.

[0018] Preferably, the corneal anterior surface to the lens anterior surface distance is denoted as D1; the corneal anterior surface to the lens posterior surface distance is denoted as D2; and the corneal anterior surface to the retina surface distance is denoted as D3.

[0019] Preferably, in the standard A-mode scanning waveform image, from left to right, the first single high wave marked as S1 is a surface wave, and D1 is taken as a reference to select a surface wave marked as S2 with the smallest distance adjacent to D1; D2 is taken as a reference to select a surface wave marked as S3 with the smallest distance adjacent to D2; and D3 is taken as a reference to select a surface wave marked as S4 with the smallest distance adjacent to D3.

[0020] Preferably, based on the surface waves of S1, S2, S3 and S4, the distance between S1 and S2 is measured as the anterior chamber depth, the distance between S2 and S3 is measured as the lens thickness, the distance between S3 and S4 is measured as the vitreous cavity depth, and the distance between S1 and S4 is measured as the axial length.

[0021] The application also provides an eye A-mode biological measurement system assisted by a B-mode image, which is used to implement the above method and comprises a first acquisition module, a first marking module, a second acquisition module, a second marking module and a measurement module.

[0022] The first acquisition module is used to perform standard scanning on an eye by using a B-mode scanning technology to obtain a standard eye section image.

[0023] The first marking module is used to mark the standard eye section image by using an automatic judgment model to obtain a corneal anterior surface denoted as S1, a lens anterior surface denoted as S2, a lens posterior surface denoted as S3 and a retina surface denoted as S4, and three surface distances between the four surfaces, wherein the automatic judgment model adopts an image segmentation positioning algorithm based on a YOLO model to realize automatic position judgment, and the steps include:

[0024] 1. Image annotation

[0025] An image is annotated by using a Labelme annotation software to annotate the positions of the edge boxes of four categories of the corneal anterior surface, the lens anterior surface, the lens posterior surface and the retina surface, respectively, and then a label file containing object categories and boundary boxes is generated.

[0026] 2. Model training

[0027] The YOLOv11 model is selected for training, the input resolution of the picture is adjusted to 640*640, the batch size is set to 16, the Adam optimizer is used, the initial learning rate is set to 0.001, the learning rate decay factor is set to 0.01, the random horizontal flip is set to 0.5, the random scaling is set to 0.5, and the Mosaic data enhancement strategy is enabled;

[0028] 3、Model inference

[0029] The confidence threshold of the inference process is set to 0.25, the model outputs the bounding box coordinates of the four categories of corneal anterior surface, lens anterior surface, lens posterior surface and retina surface, the center point coordinates of each bounding box are calculated according to the bounding box coordinates, and then the three surface distances are calculated through the value of the X-axis of the center point coordinates; based on the three surface distances, the distance parameter measurement is performed on the standard eye section image to obtain a first measurement result; the distance parameters include: the distance from the corneal anterior surface to the lens anterior surface, the distance from the corneal anterior surface to the lens posterior surface and the distance from the corneal anterior surface to the retina surface;

[0030] The second acquisition module is configured to use A-scan technology to perform standard scanning on the eyeball to obtain a standard A-scan waveform image;

[0031] The second marking module is configured to mark the standard A-scan waveform based on the first measurement result to obtain four surface waves, including: a corneal anterior surface wave, a lens anterior surface wave, a lens posterior surface wave and a retina anterior surface wave.

[0032] The measurement module is configured to complete A-scan waveform biological parameter measurement based on the four surface waves.

[0033] Compared with the prior art, the beneficial effects of the present application are as follows:

[0034] The present application aims at the situation that the existing A-scan technology in clinical practice is prone to inaccurate identification of eyeball waveform in some cases, and adds B-scan image measurement information to assist in identifying A-scan waveform, which can effectively improve the accuracy and efficiency of A-scan waveform identification, thereby improving the accuracy of eyeball biological parameter measurement. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0036] Figure 1 The first measurement result acquisition flowchart of the embodiment of the present application is shown in the following table:

[0037] Figure 2 A second measurement result acquisition flowchart of an embodiment of the present application;

[0038] Figure 3 An A-mode scanning waveform image of an embodiment of the present application;

[0039] Figure 4 A scanning cross-section image marking schematic diagram of an embodiment of the present application;

[0040] Figure 5 An A-mode scanning waveform image based on B-mode auxiliary of an embodiment of the present application. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0042] In order to make the above objectives, characteristics and advantages of the present application more apparent, comprehensible and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0043] Embodiment one

[0044] The embodiment provides an eyeball A-mode biological measurement method under B-mode image auxiliary, and steps include:

[0045] S1. Adopting B-mode scanning technology to perform standard scanning on the eyeball to acquire a standard eyeball cross-section image.

[0046] S2. Constructing an automatic judgment model to mark the standard eyeball cross-section image.

[0047] Through marking the standard eyeball cross-section image, the following is obtained: a cornea front surface, marked as S1; a lens front surface, marked as S2; a lens back surface, marked as S3; and a retina surface, marked as S4; and three surface distances between the four surfaces, the automatic judgment model adopts an image segmentation positioning algorithm based on a YOLO model to realize automatic position judgment, and steps include:

[0048] 1. Image annotation

[0049] The image is instance segmentation labeled using Labelme labeling software, and the bounding box positions of the corneal anterior surface, lens anterior surface, lens posterior surface and retina surface are labeled respectively, and then a label file containing object categories and bounding boxes is generated. Since the B-ultrasound resolution cannot accurately measure the cornea, the average thickness of the cornea is taken as a constant, 0.5 mm, and the posterior surface interface of the cornea is marked as the anterior surface of the cornea by shifting left by 0.5 mm. A total of 1000 B-ultrasound images are labeled.

[0050] 2. Model training

[0051] The YOLOv11 (COCO pre-training) model is selected for training, the input resolution of the picture is adjusted to 640*640, the batch size is set to 16, the Adam optimizer is used, the initial learning rate is set to 0.001, the learning rate decay factor is set to 0.01, the random horizontal flip (fliplr) is set to 0.5, the random scaling (scale) is set to 0.5, and the Mosaic data augmentation strategy is enabled. The operating system is Ubuntu 20.04, the GPU is 1 NVIDIA GeForce RTX 3090 (24G video memory), the programming framework is PyTorch, and a total of 100 rounds of training are performed to realize model training for segmentation and positioning of the corneal anterior surface, lens anterior surface, lens posterior surface and retina surface on B-ultrasound images.

[0052] 3. Model inference

[0053] The confidence threshold of the inference process is set to 0.25, the model outputs the bounding box coordinates of the corneal anterior surface, lens anterior surface, lens posterior surface and retina surface, the center point coordinates of each bounding box are calculated according to these bounding box coordinates, and then the three surface distances between the four surfaces are calculated through the value of the X-axis of the center point coordinates. Finally, based on the three surface distances, the distance parameters of the standard eyeball section image are measured to obtain the first measurement result, including: the distance from the corneal anterior surface to the lens anterior surface, denoted as D1, the distance from the corneal anterior surface to the lens posterior surface, denoted as D2, and the distance from the corneal anterior surface to the retina surface, denoted as D3. Among them, D1 is the anterior chamber depth, D3 is the axial length, D2-D1 is the lens thickness, and D3-D2 is the vitreous cavity depth. The overall process is shown in Figure 1

[0054] S3. Adopt A-scan technology to scan the eyeball to obtain a standard A-scan waveform image.

[0055] S4. Based on the first measurement result, mark the standard A-scan waveform image to obtain four surface waves.

[0056] ​The four surface waves include: corneal anterior surface wave, lens anterior surface wave, lens posterior surface wave and retina anterior surface wave. The method for identifying the A-mode wave peaks of different structure interfaces in the eye includes: in a standard A-mode scanning waveform image, from left to right, marking the first single high wave as a surface wave S1, taking D1 as a reference, selecting a single high wave with the minimum distance adjacent to D1 as a surface wave S2; taking D2 as a reference, selecting a single high wave with the minimum distance adjacent to D2 as a surface wave S3; taking D3 as a reference, selecting a single high wave with the minimum distance adjacent to D3 as a surface wave S4.

[0057] S5. Based on the four surface waves, a second measurement result is obtained, and the A-mode waveform biological parameter measurement is completed.

[0058] Based on the surface waves of S1, S2, S3 and S4, the distance between S1 and S2 is measured as the anterior chamber depth, the distance between S2 and S3 is measured as the lens thickness, the distance between S3 and S4 is measured as the vitreous cavity depth, and the distance between S1 and S4 is measured as the axial length of the eye, a second measurement result is obtained, and the A-mode waveform biological parameter measurement is completed, and the process is as shown in Figure 2

[0059] Example Two

[0060] This example is illustrated by taking retinal detachment as an example.

[0061] The prior art is used to perform A-mode scanning on the eyeball to be measured, and a standard A-mode scanning waveform (as shown in Figure 3 ) is obtained. The single high waves presented from left to right in the figure are respectively the corneal anterior surface S1, the lens anterior surface S2, the lens posterior surface S3 and the retina surface S4, and the measured axial length of the eye (the distance between the S1 interface and the S4 interface) is 23.45 mm.

[0062] B-mode scanning is performed on the eyeball to be measured, and the required standard scanning section is obtained. The corneal anterior surface, the lens anterior surface, the lens posterior surface and the retina surface of the scanning section image are judged and marked, and the measurement of the anterior chamber depth, the lens thickness, the vitreous cavity depth and the axial length of the eye is completed (as shown in Figure 4 ), wherein the axial length of the eye is 23.97 mm.

[0063] The waveform of the A-mode is re-marked, from left to right, the first single high wave is marked as the corneal anterior surface wave (S1 B ), the anterior chamber depth measured by B-mode is taken as a reference, the single high wave with the minimum distance adjacent to the anterior chamber depth is marked as the lens anterior surface wave (S2 B ), and the sum of the anterior chamber depth and the lens thickness measured by B-mode is taken as a reference, the single high wave with the minimum distance adjacent to the measurement value is marked as the lens posterior surface wave (S3​B ); take the axial length measured by B-ultrasound as a reference, select the single high wave adjacent to the minimum distance as the retina pre-surface wave (S4 B ), see Figure 5 .

[0064] Measure the distance between S1 B and S2 B as the anterior chamber depth, the distance between S2 B and S3 B as the lens thickness, the distance between S3 B and S4 B as the vitreous cavity depth, and the distance between S1 B and S4 B as the axial length. As shown in Figure 5 , the corneal pre-surface wave, the lens pre-surface wave, and the lens post-surface wave identified by the present technical solution are consistent with the prior art ( Figure 3 ), while the retina surface wave S4 B is inconsistent with S4 identified by the prior art. From the B-ultrasound scanning section of the eyeball, it can be judged that the retina is detached in the present embodiment, while the prior art identifies the raised retina as the eyeball wall surface, thus failing to accurately reflect the true axial length. In the present embodiment, the axial length measured by the present technical solution is 23.95 mm, compared with 23.45 measured by the prior art. The difference of 0.50 mm will result in a refractive error of about 1.5 D, which will have a significant impact on the calculation of the patient's intraocular lens degree.

[0065] The above-described embodiments are only descriptions of the preferred modes of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those of ordinary skill in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. A method for measuring eyeballs using A-ultrasound biometry assisted by B-ultrasound images, characterized in that the steps include: Use B-ultrasound scanning technology to perform standard scans on the eyeball and obtain standard cross-sectional images of the eyeball; An automatic judgment model is constructed to mark the standard eyeball section image to obtain: the anterior surface of the cornea, denoted as S1; The three surface distances between the front surface of the lens, denoted as S2; the back surface of the lens, denoted as S3; and the retinal surface, denoted as S4, are automatically determined using an image segmentation and positioning algorithm based on the YOLO model. The steps include: Use Labelme annotation software to perform instance segmentation and annotation on the image, annotating the border positions of four categories: the anterior surface of the cornea, the anterior surface of the lens, the posterior surface of the lens, and the retinal surface. Then generate a label file containing the object category and bounding box; The YOLOv11 model was selected for training. The input image resolution was adjusted to 640*640, the batch size was set to 16, the Adam optimizer was used, the initial learning rate was set to 0.001, the learning rate decay factor was set to 0.01, the random horizontal flip was set to 0.5, the random scale was set to 0.5, and the Mosaic data augmentation strategy was enabled. The confidence threshold of the inference process is set to 0.

25. The model outputs frame coordinates of four categories: the anterior surface of the cornea, the anterior surface of the lens, the posterior surface of the lens, and the retinal surface. The center point coordinates of each frame are calculated based on these frame coordinates, and then the three surface distances are calculated using the X-axis value of the center point coordinates. Based on the three surface distances, the distance parameters of the standard eye section image are measured to obtain a first measurement result. The distance parameters include: the distance from the anterior surface of the cornea to the anterior surface of the lens, the distance from the anterior surface of the cornea to the posterior surface of the lens, and the distance from the anterior surface of the cornea to the retinal surface. A-scan technology is used to perform a standard scan of the eyeball to obtain a standard A-scan waveform image; Based on the first measurement result, the standard A-scan waveform image is marked to obtain four surface waves, including: anterior corneal surface wave, anterior lens surface wave, posterior lens surface wave and anterior retinal surface wave; Based on the four surface waves, a second measurement result is obtained, and the A-ultrasound waveform biological parameter measurement is completed.

2. The eyeball A-ultrasound biological measurement method assisted by B-ultrasound images according to claim 1, characterized in that: The distance from the anterior surface of the cornea to the anterior surface of the lens is recorded as D1; ​​the distance from the anterior surface of the cornea to the posterior surface of the lens is recorded as D2; and the distance from the anterior surface of the cornea to the surface of the retina is recorded as D3.

3. The eyeball A-ultrasound biological measurement method assisted by B-ultrasound images according to claim 2, characterized in that: In the standard A-scan waveform image, from left to right, the first single high wave is marked as the surface wave S1, and with D1 as a reference, the surface wave with the single high wave at the minimum distance adjacent to D1 is selected as S2; with D2 as a reference, the surface wave with the single high wave at the minimum distance adjacent to D2 is selected as S3; with D3 as a reference, the surface wave with the single high wave at the minimum distance adjacent to D3 is selected as S4.

4. The eyeball A-ultrasound biological measurement method assisted by B-ultrasound images according to claim 3, characterized in that: Based on the surface waves of the four surfaces S1, S2, S3 and S4, the distance from S1 to S2 was measured as the anterior chamber depth, the distance from S2 to S3 was measured as the lens thickness, the distance between S3 and S4 was measured as the vitreous cavity depth, and the distance between S1 and S4 was measured as the axial length.

5. A B-ultrasound image-assisted eyeball A-ultrasound biological measurement system, the system being used to implement the method according to any one of claims 1 to 4, characterized in that: include: a first acquisition module, a first marking module, a second acquisition module, a second marking module and a measurement module; The first acquisition module is used to perform a standard scan of the eyeball using B-ultrasound scanning technology to obtain a standard eyeball cross-sectional image; The first marking module is used to construct an automatic judgment model to mark the standard eyeball section image to obtain: the anterior surface of the cornea, denoted as S1; The three surface distances between the front surface of the lens, denoted as S2; the back surface of the lens, denoted as S3; and the retinal surface, denoted as S4, are automatically determined using an image segmentation and positioning algorithm based on the YOLO model. The steps include: Use Labelme annotation software to perform instance segmentation and annotation on the image, annotating the border positions of four categories: the anterior surface of the cornea, the anterior surface of the lens, the posterior surface of the lens, and the retinal surface. Then generate a label file containing the object category and bounding box; The YOLOv11 model was selected for training. The input image resolution was adjusted to 640*640, the batch size was set to 16, the Adam optimizer was used, the initial learning rate was set to 0.001, the learning rate decay factor was set to 0.01, the random horizontal flip was set to 0.5, the random scale was set to 0.5, and the Mosaic data augmentation strategy was enabled. The confidence threshold of the inference process is set to 0.

25. The model outputs frame coordinates of four categories: the anterior surface of the cornea, the anterior surface of the lens, the posterior surface of the lens, and the retinal surface. The center point coordinates of each frame are calculated based on these frame coordinates, and then the three surface distances are calculated using the X-axis value of the center point coordinates. Based on the three surface distances, the distance parameters of the standard eye section image are measured to obtain a first measurement result. The distance parameters include: the distance from the anterior surface of the cornea to the anterior surface of the lens, the distance from the anterior surface of the cornea to the posterior surface of the lens, and the distance from the anterior surface of the cornea to the retinal surface. The second acquisition module is used to perform a standard scan of the eyeball using A-scan technology to obtain a standard A-scan waveform image; The second marking module is used to mark the standard A-scan waveform based on the first measurement result to obtain four surface waves, including: anterior corneal surface wave, anterior lens surface wave, posterior lens surface wave and anterior retinal surface wave; The measurement module is used to complete the measurement of A-ultrasound waveform biological parameters based on the four surface waves.

Images