Fundus OCT imaging diopter adaptive matching method and device

Abstract

The invention provides a fundus OCT imaging diopter adaptive matching method and device, and relates to the technical field of retina imaging, and the method comprises the following steps: detecting pupil positions in real time, automatically tracking pupils, and controlling a reference arm motor and a sample arm motor to move to initial positions; a reference arm motor is controlled to move, meanwhile, a retina structure in an OCT image is detected through a deep learning algorithm, and the eye axis length of the testee is automatically matched; the reference arm motor and the sample arm motor are controlled to move synchronously, the retina focusing position is found by calculating the composite information entropy of the OCT image, and the diopter of the testee is automatically compensated; and detecting and adjusting the position of the retina in the image by using a deep learning algorithm to cover retina imaging with a large view. According to the method, the OCT imaging device is automatically controlled, an automatic solution is provided for eye ground screening of people of multiple age groups and various ametropia groups, the overall image quality collected by OCT is improved, and the operation simplicity and the imaging speed are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of retinal imaging technology, and specifically to a method and device for adaptive matching of fundus OCT imaging diopter. Background Art

[0002] Optical Coherence Tomography (OCT) is a non-invasive three-dimensional real-time imaging method with micron-level resolution, which is widely used in the biomedical field, especially for high-precision imaging of the fundus retina structure in ophthalmic diagnosis. OCT constructs cross-sectional or volume images of the internal structure of a sample by measuring the reflection characteristics of low-coherence light in tissues at different depths. Its imaging conditions include coherence gating and confocal gating. Coherence gating determines whether a retinal image can be obtained within the imaging range, and confocal gating determines whether the retinal image is clear and bright.

[0003] However, in traditional OCT systems, in order to ensure the acquisition of high-quality fundus images, the operator needs to manually align the OCT device with the eye and manually adjust the distance between the device and the eye, and then perform complex refractive compensation settings to adapt to the eye characteristics of different patients. This process is not only time-consuming and laborious, but also prone to inaccurate adjustment due to human factors, affecting the final imaging effect. In addition, manual adjustment limits the efficiency and accuracy of OCT devices in clinical applications. Especially when facing a large number of patients, it is difficult to achieve rapid and accurate examinations.

[0004] The invention patent "A Method for Judging Fundus Refractive Compensation and Imaging Optimization Using OCT Signals" with the patent number CN112168132A proposes to calculate the concavity and convexity of the retinal structure to control the optical path and adjust the vertical position of the retinal image according to the image intensity. However, when the retina is relatively flat during a small-range scan of the retina, and a concave-shaped retinal virtual image will appear under large optical path difference adjustment, this method of calculating concavity and convexity will fail; and adjusting the vertical position according to the retinal intensity will also have problems of unstable detection. Summary of the Invention

[0005] The purpose of the present invention is to provide a device and method for adaptive matching of fundus OCT imaging diopter, which can automatically track the patient's pupil in real time, automatically match the axial length of different patients' eyes, perform refractive compensation and re-focus quickly and accurately, significantly improve the operation simplicity and imaging speed, and at the same time reduce the uncertainty brought by manual intervention, thereby providing a more reliable and efficient diagnostic tool for doctors.

[0006] A method for adaptive matching of fundus OCT imaging diopter, the method includes:

[0007] Step S1: Set the initial position, end position, and terminal position of the reference arm motor and the sample arm motor. Set the initial position, end position, and terminal position of the sample arm motor. Control the reference arm motor to move to the initial position of the reference arm motor, and control the sample arm motor to move to the initial position of the sample arm motor;

[0008] Step S2: Control the reference arm motor to move to the end position. When the retinal structure appears in the OCT image, record the optical path of the reference arm and the sample arm of the OCT system at this time. Then, control the reference arm motor to move to the motor position corresponding to the reference arm optical path to match the axial length of the subject;

[0009] Step S3: Control the reference arm motor and the sample arm motor to move synchronously to the terminal position of any one of the motors to find the focusing position corresponding to the maximum value of the composite information entropy of the OCT image. Control the reference arm motor and the sample arm motor to move to the focusing position to achieve diopter compensation; wherein, the composite information entropy represents the weighted sum of the information entropy, brightness, and contrast of the image;

[0010] Step S4: Detect the position of the retinal structure in the OCT image, and control the reference arm motor to move to move the retinal structure up and down to a height position between one-half and three-quarters of the height of the OCT image.

[0011] Further, the specific steps of controlling the reference arm motor to move to the end position, recording the optical path of the reference arm and the sample arm of the OCT system when the retinal structure appears in the OCT image, and then controlling the reference arm motor to move to the motor position corresponding to the reference arm optical path to match the axial length of the subject include:

[0012] Step S2-1: Control the reference arm motor to move to the specified end position. During the movement, every time a frame of OCT image is acquired, detect and record the reliability of the retinal structure in the OCT image, and record the position of the reference arm motor at this time;

[0013] Step S2-2: Find the motor position corresponding to the maximum value of the retinal detection reliability, and control the reference arm motor to move to this position;

[0014] Step S2-3: Continuously acquire OCT images within 1 second and detect whether there is a retinal structure. If there is, enter Step S2-4; otherwise, control the reference arm motor to move to the initial position and enter Step S2-1 again;

[0015] Step S2-4: Calculate the difference between the height position of the detected retinal structure in the OCT image and the specified height;

[0016] Step S2-5: Control the reference arm motor to move forward or backward according to the positive or negative of the height difference, so that the retinal structure is located between one-half and three-quarters of the specified OCT image height.

[0017] Further, the specific steps of controlling the reference arm motor and the sample arm motor to synchronously move to the end position of any one motor, finding the focusing position corresponding to the maximum composite information entropy of the OCT image, and controlling the reference arm motor and the sample arm motor to move to the focusing position to achieve diopter compensation include:

[0018] Step S3-1: Control the reference arm motor and the sample arm motor to synchronously move towards the end position;

[0019] Step S3-2: During the synchronous movement of the reference arm motor and the sample arm motor, whenever an OCT image is acquired, calculate and record the composite information entropy of the OCT image, and at the same time record the positions of the reference arm and the sample arm motors at this time;

[0020] Step S3-3: Determine whether the end of the reference arm motor or the end of the sample arm motor is reached. If the end of one of the motors is reached, execute Step S3-4; otherwise, return to Step S3-1;

[0021] Step S3-4: Find the motor position corresponding to the maximum composite information entropy, and control the reference arm motor and the sample arm motor to move to this position;

[0022] Step S3-5: Continuously acquire OCT images within 1 second and detect whether there is a retinal structure. If there is, enter Step S5-4; otherwise, end the motor control;

[0023] Step S3-6: Calculate the difference between the height position of the detected retinal structure in the OCT image and the specified height;

[0024] Step S3-7: Control the reference arm motor to move forward or backward according to the positive or negative of the height difference, so that the retinal structure is located between one-half and three-quarters of the specified OCT image height.

[0025] Further, in the said Step S2-1, the detection and recording of the credibility of the retinal structure in the OCT image is realized by a pre-constructed YOLOv8 neural network detection model. Specifically, it includes the following steps:

[0026] Step S4-1: Collect an OCT image dataset, manually annotate the retinal structure in the OCT image with a rectangular box, and the annotation categories include the fovea centralis, the optic nerve bundle, and other retinal layers. Divide all the annotated OCT images and their annotations into a training set and a validation set;

[0027] Step S4-2: Construct a YOLOv8 neural network detection model, set the number of classes, learning rate, number of training epochs, and image resolution of the YOLOv8 neural network detection model, train the YOLOv8 neural network detection model on the training set and validate it on the validation set, and export the weight file;

[0028] Step S4-3: Deploy the YOLOv8 neural network detection model, including the neural network and weights, on a device with a graphics rendering processor, input the OCT image into the detection model, and the output is the detection rectangle box and confidence of the retinal structure in the OCT image.

[0029] Further, the calculation of the composite information entropy specifically includes:

[0030] S3-41: Calculate the grayscale histogram of the OCT image, and calculate the probability distribution of each grayscale level according to the grayscale histogram. The formula is p(i) = n i / N, where i represents a certain grayscale level, n i represents the number of times i appears in the image, and N represents the total number of pixels in the image;

[0031] S3-42: Calculate the information entropy of the OCT image. The formula is H = -∑ i p(i)·log2(p(i));

[0032] S3-43: Calculate the mean and variance of the OCT image and use them as brightness and contrast respectively. The mean formula is The variance formula is where I i represents a pixel value of the OCT image, M represents the mean, and C represents the variance;

[0033] S3-44: Calculate the composite information entropy. The formula is αH + βM + γC, where α, β, and γ are weights.

[0034] Meanwhile, the present invention discloses a fundus OCT imaging diopter adaptive matching device, and the device includes: a high-speed swept-source light, a high-speed balanced detector, an optical fiber coupler, a robotic arm pod, and a reference arm optical path;

[0035] The high-speed swept-source light and the high-speed balanced detector are connected to one end of the optical fiber coupler, and the robotic arm pod and the reference arm optical path are connected to the other end of the optical fiber coupler;

[0036] The reference arm optical path includes a first optical fiber collimator and a reference arm motor module arranged in sequence;

[0037] The robotic arm pod includes a sample arm motor module and an intermediate optical path; the intermediate optical path includes an infrared camera, a second lens, a second dichroic mirror, a third lens, a first dichroic mirror, a fundus eyepiece, and a circular display screen; the infrared camera, the second lens, the second dichroic mirror, the third lens, the first dichroic mirror, and the fundus eyepiece are arranged in sequence, the circular display screen is arranged below the second dichroic mirror, and the sample arm motor module is arranged below the first dichroic mirror.

[0038] Specifically, the reference arm motor module includes a first lens and a mirror arranged in sequence; the motor drives the first lens and the mirror to move back and forth to change the optical path of the reference arm.

[0039] Specifically, the sample arm motor module includes a second fiber collimator, a two-dimensional scanning galvanometer, and a fourth lens; the motor drives the second fiber collimator, the two-dimensional scanning galvanometer, and the fourth lens to move up and down to change the focusing position of the light after passing through the fourth lens, so as to change the focal position after passing through the first dichroic mirror and the fundus eyepiece, and realize refocusing on different fundus positions.

[0040] Specifically, the robotic arm pod is installed on a six-axis robotic arm. The pupil coordinates are obtained by positioning the pupil through the infrared camera, and the coordinates are sent to the robotic arm, and then the robotic arm pod is driven to automatically align with the pupil.

[0041] After adopting the above scheme, the beneficial effects of the present invention are as follows: A method and device for adaptive matching of refractive power in fundus OCT imaging proposed in this application can automatically track the patient's pupil in real time, automatically align the patient's gaze direction throughout the process, can automatically match subjects with different eye axis lengths, and can stably identify the retinal structure within the range of 14 mm to 35 mm of the eye axis length, accurately set the optical path, basically covering the age range from infants to adults, and providing a solution for fundus screening for multiple age groups; by controlling the movement of the motor to achieve focus movement and calculating the composite information entropy to control the motor at an appropriate position, it can solve the defocusing problems caused by various refractive errors, covering hyperopia and myopia populations with refractive powers from 10 D (1000 degrees of hyperopia) to -20 D (2000 degrees of myopia), and improving the brightness and clarity of fundus OCT imaging; this method and device can achieve real-time and non-destructive fundus imaging without directly contacting the patient, significantly improving the operation simplicity and imaging speed, while reducing the uncertainty brought by manual intervention, thus providing a more reliable and efficient diagnostic tool for doctors. This invention is particularly suitable for scenarios where the detection object needs to be frequently changed, such as large-scale screening or outpatient settings, and is of great significance for improving the level of medical services. Description of the Drawings

[0042] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0043] Figure 1 Flow chart of the fundus OCT imaging diopter adaptive matching method according to an embodiment of the present invention;

[0044] Figure 2 Schematic diagram of a device for optimizing fundus imaging according to an embodiment of the present invention;

[0045] Figure 3 Flow chart of the automatic eye axis length matching method according to an embodiment of the present invention;

[0046] Figure 4 Result diagram of the automatic eye axis length matching method according to an embodiment of the present invention;

[0047] Figure 5 Flow chart of the automatic diopter compensation method according to an embodiment of the present invention;

[0048] Figure 6 Result diagram of the automatic diopter compensation method according to an embodiment of the present invention;

[0049] Reference numerals: 1 - first fiber collimator, 2 - first lens, 3 - mirror, 4 - second fiber collimator, 5 - two-dimensional scanning galvanometer, 6 - fourth lens, 7 - first dichroic mirror, 8 - fundus eyepiece, 9 - infrared camera, 10 - second lens, 11 - second dichroic mirror, 12 - third lens, 13 - circular display screen. Specific embodiments

[0050] The following will clearly and completely describe the technical solutions of the present invention in combination with the embodiments. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0051] The device of this embodiment is detailed in the attached Figure 2 , and the device includes: a high-speed swept-source light, a high-speed balanced detector, an optical fiber coupler, a robotic arm pod, and a reference arm optical path. The light output by the swept-source light is divided into two paths by the optical fiber coupler. One path enters the reference arm optical path, and the other path enters the robotic arm pod. Then, the light returned from the reference arm optical path and the robotic arm pod passes through the optical fiber coupler again and enters the balanced detector for detection.

[0052] The reference arm optical path includes a first fiber collimator 1 and a reference arm motor module arranged in sequence. The reference arm motor module is controlled to drive the first lens 2 and the mirror 3 to move, and the moving direction is close to or away from the first fiber collimator 1, so as to change the optical path of the reference arm and reach the equal optical path position that matches the optical path of the sample arm. People of different ages, those with myopia or hyperopia have different eye axis lengths. For example, the eye axis length of infants is about 14 mm, the eye axis length of people with high myopia is about 35 mm, and the imaging range of OCT is generally 6 mm. If the optical path of the reference arm is changed, it will be impossible to image the retina in the range of 14 mm - 35 mm. In this embodiment, changing the optical path of the reference arm can automatically match the eye axis lengths of 14 mm - 35 mm at the equal optical path position of the reference arm and the sample arm, so as to cope with subjects with a wide range of different eye axis lengths.

[0053] The robotic arm pod is installed on a six-axis robotic arm. The robotic arm pod includes a sample arm motor module, an infrared camera 9, a second lens 10, a second dichroic mirror 11, a third lens 12, a first dichroic mirror 7, a fundus eyepiece 8, and a circular display screen 13. The infrared camera 9, the second lens 10, the second dichroic mirror 11, the third lens 12, the first dichroic mirror 7, and the fundus eyepiece 8 are arranged in sequence. The circular display screen 13 is arranged below the second dichroic mirror 11, and the sample arm motor module is arranged below the first dichroic mirror 7.

[0054] The infrared camera 9 collects the eye image of the subject through the second lens 10, the second dichroic mirror 11, the third lens 12, the first dichroic mirror 7, and the fundus eyepiece 8, and measures the position of the pupil of the human eye. This position is sent to the six-axis robotic arm, and the robotic arm drives the pod to align with the pupil so that the OCT beam can enter the pupil and reach the retina. The sample arm motor module includes a second fiber collimator 4, a two-dimensional scanning galvanometer 5, and a fourth lens 6. The motor drives the second fiber collimator 4, the two-dimensional scanning galvanometer 5, and the fourth lens 6 to move up and down, changing the distance between the OCT beam and the first dichroic mirror. Since the focal lengths of the fourth lens 6 and the fundus eyepiece 8 are fixed, the up and down movement of the motor will change the focal position of the OCT beam entering the human eye on the retina. A normal human eye will accurately focus the parallel OCT beam on the retina. However, when the OCT beam enters the eye of a person with refractive error, the focus will be in front of or behind the retina, and it cannot be focused on the retina, resulting in problems such as low OCT imaging brightness, poor contrast, and blurred detail imaging. In this embodiment, the focus of the outgoing OCT beam is changed by the movement of the motor to cope with subjects with different symptoms of refractive error, and it can cover hyperopia and myopia populations with refractive powers ranging from 10 D (1000 degrees of hyperopia) to -20 D (2000 degrees of myopia).

[0055] For a method of adaptive matching of refractive power for fundus OCT imaging in this embodiment, please refer to the flowchart in Figure 1, the method includes the following steps:

[0056] Step S1: Set the initial position, end position, and terminal position of the reference arm motor, and set the initial position, end position, and terminal position of the sample arm motor. Control the reference arm motor to move to the initial position of the reference arm motor, and control the sample arm motor to move to the initial position of the sample arm motor;

[0057] In the above steps, the initial position, end position, and terminal position of the reference arm motor and the sample arm motor. The initial position and end position of the reference arm motor refer to the positions where the reference arm motor is located when the optical path is equal to the eye axis lengths of 14 mm and 35 mm; the initial position and end position of the sample arm motor are the positions where the sample arm motor is located when the focus is in front of and behind the normal retina; the terminal position of the reference arm motor and the sample arm motor refers to the farthest position that the motor can reach, and this position is determined by the size of the motor.

[0058] Step S2: Control the reference arm motor to move to the end position. When the retinal structure appears in the OCT image, record the optical paths of the reference arm and the sample arm of the OCT system at this time. Then, control the reference arm motor to move to the motor position corresponding to the reference arm optical path to match the eye axis length of the subject.

[0059] In specific implementation, the reference arm motor does not stop midway and moves uniformly towards the end position. Here, the specific moving speed can be set conventionally by those skilled in the art. During the process, wait for one frame of OCT image for detection, and then loop to wait for the next frame of OCT image for detection until the reference arm motor has moved to the end position, then pause the detection of the OCT image.

[0060] Step S3: Control the reference arm motor and the sample arm motor to move synchronously to the terminal position of any one of the motors, find the focusing position corresponding to the maximum value of the composite information entropy of the OCT image, and control the reference arm motor and the sample arm motor to move to the focusing position to achieve diopter compensation; among them, the composite information entropy represents the weighted sum of the information entropy, brightness, and contrast of the image.

[0061] In specific implementation, the reference arm motor and the sample arm motor also do not stop midway and move uniformly to the terminal position. During the process, calculate the composite information entropy for each frame of OCT image collected. Note that the above terminal position should be the terminal position corresponding to the motor.

[0062] Step S4: Detect the position of the retinal structure in the OCT image, and control the reference arm motor to move to move the retinal structure up and down so that it is between one-half and three-quarters of the height of the OCT image.

[0063] In the above step, moving the retinal structure up and down to the three - quarter height position of the OCT image is because the large - field retinal OCT image presents a concave shape, and there is a relatively long distance from the bottom to the top of the concave retina. Placing the bottom between one - half and three - quarters of the OCT image height can better cover the retina within the entire field of view. In specific implementation, if the OCT field of view is small, the height position can be adjusted to between one - quarter and three - quarters, etc.

[0064] Specifically, for the flowchart of step S2, please refer to Figure 3 , and it specifically includes:

[0065] Step S2 - 1: Control the reference - arm motor to move to the specified end position, that is, the equal - optical - path position corresponding to an eye - axis length of 35 mm. During the movement, whenever a frame of OCT image is acquired, detect the retinal structure in the OCT image and record the confidence level, and at the same time record the motor position at this time. Detecting the retina and having a high confidence level means that the motor position at this time is the equal - optical - path position corresponding to this eye - axis length.

[0066] Step S2 - 2: Find the motor position corresponding to the maximum value of the retinal - detection confidence level. This step is to sort all the recorded confidence levels to find the maximum value, and record the motor position when this confidence level was recorded, and then control the reference - arm motor to move back to this position.

[0067] Step S2 - 3: Continuously acquire OCT images within 1 second and detect whether there is a retinal structure. If there is, enter step S2 - 4; otherwise, control the reference - arm motor to move to the initial position and execute steps S2 - 1 to S2 - 3 again. This step is to re - confirm whether the current reference - arm motor position is the equal - optical - path position. If the confirmation is successful, proceed to the next step; if the confirmation fails, it means that there is a mistake in the adjustment process and readjustment is needed, which improves the stability and success rate of the method execution.

[0068] Step S2 - 4: Calculate the difference between the height position of the detected retinal structure in the OCT image and the specified height. In specific implementation, the specified height is set between one - half and three - quarters of the OCT image height to cover the bottom and top of the retina.

[0069] Step S2-5: According to the positive or negative of the height difference, control the reference arm motor to move forward or backward so that the retinal structure is located between one-half and three-quarters of the specified OCT image height. Specifically, when implemented, if the retinal structure is above half of the OCT image height, it means that the reference arm optical path is relatively large, and the reference arm motor needs to move closer to the first fiber collimator to reduce the reference arm optical path. If the retinal structure is below three-quarters of the OCT image height, the reference arm motor moves away from the first fiber collimator to increase the optical path. The distance the motor moves is proportional to the height difference of the OCT image. Specifically, when implemented, by fixing the sample and moving the reference arm motor in a fixed step, the pixel height of the sample offset in the OCT image is measured to obtain this proportion. Specifically, when implemented, if the total height of the OCT image is 1000, the position of the detected retinal structure in the 500-750th row of the OCT image is considered reasonable.

[0070] Please refer to Figure 4 , during the movement of the reference arm motor, the OCT images collected will show the retina at different OCT image height positions. Before Figure (a), due to the large difference between the reference arm optical path and the sample arm optical path, there is no retinal structure. The reference arm motor continues to move, and the retina moves from the position in Figure (a) to the position in Figure (c). The detection algorithm detects the retinal structure and gives a confidence level. At the position in Figure (b), there is a maximum confidence level of 0.87. The reference arm motor continues to move, and at the position in Figure (d), a retinal inverted image appears, which is not the correct optical path position, and the algorithm does not detect the retinal structure. After the reference arm motor moves to the end position, find the position corresponding to the maximum confidence level and control the reference arm to move to this position to obtain the correct position in Figure (e), corresponding to equal optical paths of the reference arm and the sample arm.

[0071] Specifically, the flowchart of step S3 is shown in detail in Figure 5 , specifically including:

[0072] Step S3-1: Control the reference arm motor and the sample arm motor to move synchronously to the end position; the end position is determined by the physical dimensions of the reference arm motor and the sample arm motor, referring to the farthest distance that the motor can reach;

[0073] Step S3-2: During the movement of the above motors, whenever an OCT image is obtained, calculate and record the composite information entropy of the OCT image, and record the positions of the reference arm and the sample arm motors at this time;

[0074] Step S3-3: Determine whether the end of the reference arm motor or the end of the sample arm motor is reached. If either motor reaches the end, it means that the two motors cannot move synchronously. Therefore, when one of the motors reaches the end, execute step S3-4 downward, otherwise loop through steps S3-1 to S3-3;

[0075] Step S3-4, sort all recorded composite information entropies, find the motor position corresponding to the maximum composite information entropy, at which the OCT image brightness is the highest and the detail imaging is the clearest, and control the reference arm motor and the sample arm motor to move to this position;

[0076] Step S3-5, continuously acquire OCT images within 1 second and detect whether there is a retinal structure. If so, proceed to step S5-4, otherwise terminate the motor control; this step is to reconfirm whether the reference arm motor position is at the equal optical path position at this time to prevent the optical path difference from changing due to motor asynchrony during the adjustment process;

[0077] Step S3-6, calculating the difference between the height position of the detected retinal structure in the OCT image and the specified height; in specific implementation, the specified height is set between one half and three quarters of the height of the OCT image to cover the bottom and top of the retina;

[0078] Step S3-7, according to the positive or negative height difference, control the reference arm motor to move forward or backward to make the retinal structure located at the specified OCT image position; in specific implementation, if the retinal structure is located above one-half the height of the OCT image, it means that the reference arm optical path is too large, and it is necessary to move the reference arm motor toward the first fiber optic collimator to reduce the reference arm optical path; if the retinal structure is located below three-quarters of the height of the OCT image, the reference arm motor moves away from the first fiber optic collimator to increase the optical path; in specific implementation, if the total height of the OCT image is 1000, the positions of the detected retinal structures in the 500-750th lines of the OCT image are considered reasonable.

[0079] See also Figure 6 , the reference arm motor and the sample arm motor move synchronously, changing the focus position, the OCT image changes from low-brightness (a) and (b) to high-brightness, clear (c), the composite information entropy increases, the focus changes from defocus to refocus, the reference arm motor and the sample arm motor continue to move, the composite information entropy decreases, the focus continues to move and becomes defocused again to obtain (d), reaching the end of the reference arm motor or the sample arm motor, finding the motor position corresponding to the maximum value of the composite information entropy and moving to that position to obtain the high-brightness, clear (e), completing the refocusing.

[0080] Specifically, the detection of retinal structure and reliability described in steps S2 and S3 specifically includes:

[0081] Step S4-1: First, use the device to collect a large number of OCT images, including those with clear retinas, unclear retinas, complete retinas, incomplete retinas, no retinas, large OCT fields of view, small OCT fields of view, positive-image retinas, and inverted-image retinas; manually annotate the correct retinal structures in the OCT images using a rectangular box, and the annotation categories include the fovea centralis, optic nerve bundle, and other retinal layers. The total number of all annotated images is 5000. Divide all annotated images and their annotations into a training set and a validation set, and the ratio of the training set to the validation set is 7:3;

[0082] Step S4-2: Build a YOLO v8 neural network, set the number of categories of the neural network to 4, including the fovea centralis 0, optic nerve bundle 1, other retinal layers 2, and inverted-image retina 3, the learning rate to 0.0001, the number of training epochs to 200, and the image resolution to 640×640. Train the neural network on the training set and validate it on the validation set, and export the weight file;

[0083] Step S4-3: Deploy the detection model on a device with a graphics rendering processor. The model includes the neural network and the weights. Sample the collected OCT images to a size of 640×640, input them into the detection model, and the output is the detection rectangular box and confidence of the retinal structure in the OCT image.

[0084] Specifically, the calculation of the composite information entropy in Step S3 specifically includes:

[0085] Calculate the grayscale histogram of the OCT image. According to the histogram data, calculate the probability distribution of each grayscale level. If a certain grayscale level i appears n i times in the image, and the total number of pixels in the image is N, then the probability of this grayscale level is p(i) = n i / N; The information entropy of the entire image can be obtained by calculating H = -∑ i p(i)·log2(p(i)); In addition to calculating the information entropy, it is also necessary to calculate the mean and variance of the image as the values of brightness and contrast. The mean of the image is calculated as where I i represents the pixel value of the image, and the variance of the image is calculated as The composite information entropy is obtained by calculating αH + βM + γC. Specifically, for an OCT image with a height of 1000 and a width of 800, the number of pixels N is 800000, and the parameters of the composite information entropy are α = 1, β = 0.01, and γ = 0.1.

[0086] The foregoing description is for illustration and not for limitation. Many embodiments and many applications other than the example provided will be apparent to those skilled in the art upon reading the above description. Accordingly, the scope of the present teachings should not be determined with reference to the above description, but should be determined with reference to the appended claims and the full scope of equivalents to which those claims are entitled. For the sake of completeness, all articles and references, including patent applications and publications, are incorporated herein by reference. Omission of any aspect of the subject matter disclosed herein from the foregoing claims is not a waiver of that subject matter nor should it be considered that the applicant has not considered that subject matter to be part of the disclosed subject matter of the application.

Claims

1. An adaptive matching method for refractive power in fundus OCT imaging, characterized in that, The method includes: Step S1: Set the initial position, end position, and terminal position of the reference arm motor, set the initial position, end position, and terminal position of the sample arm motor, control the reference arm motor to move to the initial position of the reference arm motor, and control the sample arm motor to move to the initial position of the sample arm motor; Step S2: Control the reference arm motor to move to the end position. When the retinal structure appears in the OCT image, record the optical paths of the reference arm and the sample arm of the OCT system at this time. Then, control the reference arm motor to move to the motor position corresponding to the optical path of the reference arm to match the axial length of the subject; Step S3: Control the reference arm motor and the sample arm motor to synchronously move to the terminal position of any one of the motors, find the focusing position corresponding to the maximum value of the composite information entropy of the OCT image, and control the reference arm motor and the sample arm motor to move to the focusing position to achieve diopter compensation; where the composite information entropy represents the weighted sum of the information entropy, brightness, and contrast of the image; Step S4: Detect the position of the retinal structure in the OCT image, and control the reference arm motor to move to move the retinal structure up and down so that it is between one-half and three-fourths of the height of the OCT image.

2. The method for adaptive matching of refractive power in fundus OCT imaging according to claim 1, wherein The specific steps of controlling the reference arm motor to move to the end position, when the retinal structure appears in the OCT image, recording the optical paths of the reference arm and the sample arm of the OCT system at this time, and then controlling the reference arm motor to move to the motor position corresponding to the optical path of the reference arm to match the axial length of the subject include: Step S2-1: Control the reference arm motor to move to the specified end position. During the movement, whenever an OCT image is obtained, detect and record the reliability of the retinal structure in the OCT image, and at the same time record the position of the reference arm motor at this time; Step S2-2: Find the position of the reference arm motor corresponding to the maximum value of the retinal detection reliability, and control the reference arm motor to move to this position; Step S2-3: Continuously obtain OCT images within 1 second and detect whether there is a retinal structure. If there is, enter Step S2-4; otherwise, control the reference arm motor to move to the initial position and enter Step S2-1 again; Step S2-4: Calculate the difference between the height position of the detected retinal structure in the OCT image and the specified height; Step S2-5: According to the positive or negative of the height difference, control the reference arm motor to move forward or backward so that the retinal structure is between one-half and three-fourths of the height of the OCT image.

3. The method for adaptively matching the refractive power in fundus OCT imaging according to claim 1, wherein The specific steps of controlling the reference arm motor and the sample arm motor to synchronously move to the terminal position of any one of the motors, find the focusing position corresponding to the maximum value of the composite information entropy of the OCT image, and control the reference arm motor and the sample arm motor to move to the focusing position to achieve diopter compensation include: Step S3-1: Control the reference arm motor and the sample arm motor to synchronously move to the terminal position; Step S3-2: During the synchronous movement of the reference arm motor and the sample arm motor, whenever an OCT image is obtained, calculate and record the composite information entropy of the OCT image, and at the same time record the positions of the reference arm and the sample arm motors at this time; Step S3-3: Determine whether the end of the reference arm motor or the sample arm motor is reached. If the end of one of the motors is reached, execute Step S3-4; otherwise, return to Step S3-1. Step S3-4: Find the motor position corresponding to the maximum value of the composite information entropy, and control the reference arm motor and the sample arm motor to move to this position. Step S3-5: Continuously acquire OCT images within 1 second and detect whether there is a retinal structure. If there is, enter Step S5-4; otherwise, end the motor control. Step S3-6: Calculate the difference between the height position of the detected retinal structure in the OCT image and the specified height. Step S3-7: According to the positive or negative of the height difference, control the reference arm motor to move forward or backward so that the retinal structure is located at a lower position in the specified OCT image.

4. A fundus OCT imaging diopter adaptive matching method according to claim 2, characterized in that In the above Step S2-1, the detection and recording of the credibility of the retinal structure in the OCT image are realized by a pre-constructed YOLOv8 neural network detection model.

5. A method for adaptively matching the refractive power in fundus OCT imaging according to claim 3, characterized in that, In Step S3-4, the calculation of the composite information entropy specifically includes: S3-41. Calculate the grayscale histogram of the OCT image, and calculate the probability distribution of each gray level according to the grayscale histogram. The formula is p(i) = n i / N, where i represents a certain gray level, and n i represents the number of times i appears in the OCT image, and N represents the total number of pixels in the OCT image; S3-42. Calculate the information entropy of the OCT image. The formula is H = -∑ i p(i)·log2(p(i)), where H represents the information entropy; S3-43. Calculate the mean and variance of the OCT image and use them as brightness and contrast respectively. The mean formula is The variance formula is where I i represents a pixel value of the OCT image, M represents the mean, and C represents the variance; S3-44: Calculate the composite information entropy, and the formula is Com_H = αH + βM + γC, where Com_H represents the composite information entropy, and α, β, and γ are weights.

6. An apparatus for adaptively matching the diopter in fundus OCT imaging, characterized in that, The device includes: a high-speed swept-source light, a high-speed balanced detector, an optical fiber coupler, a robotic pod, and a reference arm optical path. The high-speed swept-source light and the high-speed balanced detector are connected to one end of the optical fiber coupler, and the robotic arm pod and the reference arm optical path are connected to the other end of the optical fiber coupler. The reference arm optical path includes a first fiber collimator and a reference arm motor module arranged in sequence. The robotic arm pod includes a sample arm motor module and an intermediate optical path. The intermediate optical path includes an infrared camera, a second lens, a second dichroic mirror, a third lens, a first dichroic mirror, a fundus eyepiece, and a circular display screen. The infrared camera, the second lens, the second dichroic mirror, the third lens, the first dichroic mirror, and the fundus eyepiece are arranged in sequence. The circular display screen is arranged below the second dichroic mirror, and the sample arm motor module is arranged below the first dichroic mirror.

7. An ophthalmic OCT imaging refractive power adaptive matching device according to claim 6, wherein, The reference arm motor module includes a first lens and a mirror arranged in sequence. The motor drives the first lens and the mirror to move forward and backward to change the optical path of the reference arm.

8. An adaptive matching device for fundus OCT imaging diopter according to claim 6, characterized in that, The sample arm motor module includes a second fiber collimator, a two-dimensional scanning galvanometer, and a fourth lens. The motor drives the second fiber collimator, the two-dimensional scanning galvanometer, and the fourth lens to move up and down to change the focusing position of the light after passing through the fourth lens, so as to change the focal position after passing through the first dichroic mirror and the fundus eyepiece, and realize re-focusing on different fundus positions.

9. The fundus OCT imaging diopter adaptive matching device according to claim 6, wherein The robotic arm pod is installed on a six-axis robotic arm. The pupil coordinates are obtained by positioning the pupil through the infrared camera, and the coordinates are sent to the robotic arm, and then the robotic arm pod is driven to automatically align with the pupil. The robotic arm tracks the pupil in real time during use to ensure that the OCT imaging beam enters the pupil unobstructed.

Images