Optical biological measurement device and method for automatically judging eyeball state

Abstract

The invention discloses an optical biological measurement device and method for automatically judging an eyeball state, and relates to the field of medical image processing, and the device comprises a pre-scanning image acquisition module, an eyeball state judgment module and a measurement module; the pre-scanning image acquisition module is used for acquiring a standard scanning image; the eyeball state judgment module is used for judging based on the scanning image to obtain an eyeball state; the measurement module is used for completing biological measurement based on the eyeball state. According to the method, feature extraction is carried out on the position graph of the crystalline lens by adopting a deep learning technology according to the scanned image, the feature is compared with a preset database, meanwhile, the crystalline lens state is accurately judged according to a judgment criterion, and biological parameter scanning and (or) measurement are further carried out according to the correct eyeball state; and the accuracy and efficiency of biological parameter measurement can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to an optical biological measurement device and method for automatically judging the state of an eyeball. Background Art

[0002] Ocular biometry involves the use of various examination methods to measure various parameters of the ocular structure, providing a reference for the diagnosis and treatment of ocular diseases. Accurate measurement of these parameters is particularly crucial for assessing myopia progression and calculating the power of intraocular lenses (IOLs) in eyes undergoing cataract surgery. Generally speaking, the eye undergoing ocular biometry has various conditions, depending on the condition of the intraocular lens, which can affect the measurement of ocular biometric parameters. These conditions include aphakia, pseudophakia, completely dislocated lenses, and pseudophakia with an intraocular lens. Because ultrasound and light waves propagate at different speeds in different media (lens, IOL, silicone oil, and aqueous humor), varying ocular conditions can significantly affect the accuracy of ocular biometric measurements.

[0003] Existing optical ocular biometry technologies require the operator to inquire about the patient's medical history (such as whether the patient has undergone cataract surgery and intraocular lens implantation in the past, whether the patient has received silicone oil implantation, etc.) and manually judge the different ocular conditions according to the specific circumstances of the measurement during the examination, so as to adjust the ocular condition for correct biometric measurement. However, in clinical practice, some patients provide unclear or incorrect medical histories, which makes it impossible to accurately judge the ocular condition, affecting the measurement of ocular biological parameters. In addition, because axial length measurement is performed on the visual axis, some patients have conditions such as subluxation or complete dislocation of the lens, which means that the light wave or ultrasound wave does not pass through the lens medium on the visual axis (in this case, the ocular condition should be selected as aphakia), and clinical operators usually measure according to the phakic eye mode, which leads to errors in ocular biometric measurement. Summary of the Invention

[0004] To solve the problems in the above background, the present invention aims to provide an eye ultrasound biological measurement device and method that can automatically and accurately judge different eye conditions, which can be used to efficiently and accurately complete the measurement of eye biological parameters in clinical practice.

[0005] To achieve the above-mentioned object, the present invention provides an optical biological measurement device for automatically judging the state of an eyeball, comprising: a pre-scan image acquisition module, an eyeball state discrimination module, and a measurement module;

[0006] The pre-scan image acquisition module is used to acquire a standard scan image;

[0007] The eyeball state determination module is used to determine the eyeball state based on the scanned image; the eyeball state refers to the lens state; the eyeball state determination module uses different determination criteria for different eyes to determine the eyeball state, and the determination criteria are as follows:

[0008] When the lens / pupil area ratio = 1, it is judged that there is no lens dislocation;

[0009] When the lens / pupil area ratio = 0, it is judged as complete lens dislocation;

[0010] When the lens / pupil area ratio is >0 but <1, it is judged as subluxation of the lens;

[0011] The measurement module is used to complete biological measurement based on the eyeball state.

[0012] Preferably, the lens states include: aphakia, phakia, intraocular lens, ICL and subluxated lens.

[0013] Preferably, the eyeball state discrimination module includes: a pupil area automatic recognition unit, a pupil area image preprocessing unit, and a pupil area image classification unit;

[0014] The pupil area automatic recognition unit is used to recognize the pupil area based on the scanned image and obtain a circular outline of the pupil area;

[0015] The pupil area image preprocessing unit is used to preprocess the circular outline to obtain a processing result;

[0016] The pupil area image classification unit is used to perform classification and discrimination based on the processing result to obtain the eyeball state.

[0017] Preferably, the workflow of the pupil area automatic identification unit includes:

[0018] The acquired color image is converted into a grayscale image and Gaussian blurred using a 5x5 Gaussian kernel;

[0019] Canny edge detection is performed with a low threshold of 50 and a high threshold of 150 to capture edge features in the blurred image.

[0020] On the basis of edge detection, the Hough transform method based on image gradient information is used to locate the circular contour and complete the recognition.

[0021] Preferably, the workflow of the pupil area image preprocessing unit includes:

[0022] Segment the identified circular contour to obtain the pupil area image;

[0023] The resolution of the pupil area image is adjusted to 112×112 and normalized to complete the preprocessing.

[0024] Preferably, the workflow of the pupil area image classification unit includes: using a convolutional neural network with a ResNet50 structure to train an image classification model. During the training process, the initial value of the learning rate is set to 1×10 -3 , and adopt the learning rate decay strategy for dynamic adjustment; select Adam as the optimizer to update the model weights; use cross entropy loss as the loss function; set the batch size to 64;

[0025] The image classification model uses the trained model weights to forward propagate the pupil area image and outputs the probability distribution of each category through the softmax layer, thereby completing the automatic classification and recognition of the eye state.

[0026] Preferably, when the result of the determination is that the lens is partially dislocated, the determination is made based on whether the lens position is detected at the pupil center, and the criteria are as follows:

[0027] The lens is detected at the center of the pupil, and the phakic eye measurement mode is selected;

[0028] The lens is not detected at the pupil center, so the aphakic eye measurement mode is selected.

[0029] The present invention also provides an optical biological measurement method for automatically determining the state of the eyeball, which is applied to the above-mentioned system and comprises the following steps:

[0030] Acquire standard scan images;

[0031] The eyeball state is determined based on the scanned image; the eyeball state refers to the lens state; the eyeball state determination module uses different determination criteria for different eyeballs to determine the eyeball state, and the determination criteria are as follows:

[0032] When the lens / pupil area ratio = 1, it is judged that there is no lens dislocation;

[0033] When the lens / pupil area ratio = 0, it is judged as complete lens dislocation;

[0034] When the lens / pupil area ratio is >0 but <1, it is judged as subluxation of the lens;

[0035] Based on the eyeball status, biological measurements are performed.

[0036] Preferably, the lens states include: aphakia, phakia, intraocular lens, ICL and subluxated lens.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The present invention uses deep learning technology to extract features of the lens position graphic based on the scanned image and compares it with a preset database. At the same time, according to the discrimination criteria, the lens state is accurately judged, and biological parameters are scanned and (or) measured based on the correct eye state, which can effectively improve the accuracy and efficiency of biological parameter measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 Schematic diagram of the determination process of an embodiment of the present invention;

[0041] Figure 2 Schematic diagram of the system structure of an embodiment of the present invention;

[0042] Figure 3 Schematic diagram of the lens state according to an embodiment of the present invention; (a) indicates that the lens is not completely dislocated and the visual axis area does not pass through the lens medium; (b) indicates that the lens is not completely dislocated, but the visual axis area can pass through the lens medium. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Example 1:

[0046] This embodiment provides an optical biometric device for automatically determining the state of an eyeball, comprising: a pre-scan image acquisition module, an eyeball state determination module, and a measurement module; the pre-scan image acquisition module is used to acquire a standard scanning waveform and a scanning image; the eyeball state determination module is used to determine the state of the eyeball based on the scanning waveform and the scanning image; and the measurement module is used to perform biometric measurements based on the eyeball state.

[0047] The following will describe in detail how the present invention solves technical problems in real life in conjunction with this embodiment.

[0048] First, the pre-scan image acquisition module uses existing eye biometric scanning technology to perform a standard pre-scan on the eye according to the default eye state (crystalline eye) to obtain a standard scan image and (or) scan waveform.

[0049] The eye state identification module then uses a convolutional neural network from deep learning to extract features from the scanned waveform and image, specifically the lens image. Based on a pre-set database and established criteria, the module then determines the lens state. Different criteria are used for different ocular biometric techniques. The main process involves database construction and pupil area identification.

[0050] Database construction: Biometric scan images of 50 eyes, including aphakic eyes, phakic eyes, phakic eyes after intraocular lens implantation (ICL eyes), and eyes with subluxated lens, were obtained to construct the database.

[0051] Identify the pupil area: Extract features from the pupil area image and use a deep learning algorithm to identify the eye state, including aphakia, phakia, intraocular lens (ICL), and subluxated lens.

[0052] The specific process is as follows:

[0053] First, the automatic pupil region recognition unit implements an image processing pipeline using the OpenCV library to accurately detect the circular outline of the pupil region from a given image. The processing begins by converting the original color image to grayscale using the cv2.cvtColor function. The image is then Gaussian blurred using a 5x5 Gaussian kernel using the cv2.GaussianBlur function, effectively smoothing the image and reducing noise. Next, Canny edge detection is performed using the cv2.Canny function, with a low threshold of 50 and a high threshold of 150 to capture edge features in the image. Based on edge detection, the module further calls the cv2.HoughCircles function to identify circles, using the cv2.HOUGH_GRADIENT method based on image gradient information to locate the circles. To improve detection accuracy, the minimum distance minDist between circle centers is set to 100 to reduce redundant detections. param1 is set to 100 as the high threshold for Canny edge detection, while param2 is set to 20 to determine the accumulator threshold for circle center detection. Furthermore, by setting minRadius=35 and maxRadius=40, the module limits the radius of the detection circle, ensuring that only circles of a specific size are recognized, thereby accurately locating the pupil area. The circle with the largest radius ultimately identified is the circular outline of the pupil area, and the center of this circle is the center of the pupil.

[0054] After identifying the circular outline of the pupil area, the pupil image preprocessing unit segments the image to create the smallest square that covers the circular outline. The upper left corner of this square is set to [(center x coordinate - circle radius * 1.10), (center y coordinate - circle radius * 1.10)], and the lower right corner is set to [(center x coordinate + circle radius * 1.10), (center y coordinate + circle radius * 1.10)]. The radius is multiplied by 1.10 to ensure that the pupil area is fully covered. After segmenting the square image of the pupil area, the image resolution is resized to 112×112 and normalized to accommodate the input of the subsequent model.

[0055] Finally, the pupil area image classification unit is used for automatic classification and recognition. This is a multi-classification task, the goal is to automatically classify pupil area images into 7 categories. First, 1500 original images were collected and pupil area images were extracted. Then, each image was annotated into 7 categories, with approximately 200 samples in each category. Then, a convolutional neural network with a ResNet50 structure was used to train an image classification model. During the training process, the initial value of the learning rate was set to 1×10 -3, while adopting a learning rate decay strategy for dynamic adjustment; Adam was selected as the optimizer to update model weights; cross-entropy loss was used as the loss function; and the batch size was set to 64. The model was trained using the PyTorch programming framework and 100 epochs in a Ubuntu environment. During the prediction phase, the model uses the trained model weights to perform forward propagation on new pupil area images. The softmax layer outputs the probability distribution for each category, thus completing the task of automatically classifying and recognizing eye states.

[0056] like Figure 1 As shown in the figure, the eyeball status is judged by the judgment criteria for lens subluxation. The judgment criteria are as follows:

[0057] The lens / pupil area ratio is analyzed. When the lens / pupil area ratio = 1, it is automatically determined that there is no lens dislocation and the phakic eye measurement mode is automatically selected.

[0058] When the lens / pupil area ratio = 0, it is automatically judged as complete lens dislocation and the aphakic eye measurement mode is automatically selected.

[0059] When the lens / pupil area ratio is greater than 0 but less than 1, it is considered a subluxated lens. This determination is made automatically based on whether the lens is detected at the pupil center. If the lens is detected at the pupil center, the phakic measurement mode is automatically selected; if the lens is not detected at the pupil center, the aphakic measurement mode is automatically selected.

[0060] The above-mentioned different eye states, among which the lens state includes aphakia, pseudophakia, complete dislocation of the lens, and pseudophakia with an intraocular lens.

[0061] Finally, the measurement module performs correct biological measurement based on the judgment result of the eyeball state judgment module or re-scans with the correct eyeball state to perform correct biological parameter measurement. Figure 2 shown.

[0062] Example 2:

[0063] This embodiment also provides an optical biometric method for automatically determining the state of an eyeball, comprising the following steps:

[0064] S1. Acquire a standard scan image.

[0065] The existing eye biometric scanning technology is used to perform a standard pre-scan of the eye according to a default eye state (phakic eye) to obtain a standard scanning waveform and (or) scanning image.

[0066] S2. Perform judgment based on the scanned image to obtain the state of the eyeball.

[0067] A convolutional neural network, a deep learning framework, extracts features from the scanned waveform and image at the lens image location. The lens state is then determined based on a pre-set database and established criteria. Different criteria are used for different ocular biometric techniques. The main steps are database construction and pupil area identification.

[0068] Database construction: Biometric scan images of 50 eyes, including aphakic eyes, phakic eyes, phakic eyes after intraocular lens implantation (ICL eyes), and eyes with subluxated lens, were obtained to construct the database.

[0069] Identify the pupil area: Extract features from the pupil area image and use a deep learning algorithm to identify the eye state, including aphakia, phakia, intraocular lens (ICL), and subluxated lens.

[0070] The deep learning algorithm described above consists of three parts. The automatic pupil region recognition component utilizes the OpenCV library to implement an image processing pipeline designed to accurately detect the circular outline of the pupil region from a given image. The processing first converts the original color image to grayscale using the cv2.cvtColor function. The image is then Gaussian blurred using a 5x5 Gaussian kernel using the cv2.GaussianBlur function, effectively smoothing the image and reducing noise. Next, Canny edge detection is performed using the cv2.Canny function, with a low threshold of 50 and a high threshold of 150 to capture edge features in the image. Based on edge detection, the module further calls the cv2.HoughCircles function to identify circles, using the cv2.HOUGH_GRADIENT method based on image gradient information to locate the circles. To improve detection accuracy, the minimum distance minDist between circle centers is set to 100 to reduce redundant detections. param1 is set to 100 as the high threshold for Canny edge detection, while param2 is set to 20 to determine the accumulator threshold for circle center detection. Furthermore, by setting minRadius=35 and maxRadius=40, the module limits the radius of the detection circle, ensuring that only circles of a specific size are recognized, thereby accurately locating the pupil area. The circle with the largest radius ultimately identified is the circular outline of the pupil area, and the center of this circle is the center of the pupil.

[0071] After identifying the circular outline of the pupil area, the pupil image preprocessing process segments the image to create a minimum square that covers the circular outline. The upper left corner of this square is set to [(center x coordinate - radius * 1.10), (center y coordinate - radius * 1.10)], and the lower right corner is set to [(center x coordinate + radius * 1.10), (center y coordinate + radius * 1.10)]. The radius is multiplied by 1.10 to ensure that the pupil area is fully covered. After segmenting the square image of the pupil area, the image resolution is resized to 112×112 and normalized to accommodate the input of the subsequent model.

[0072] Finally, the pupil area image classification part is used for automatic classification and recognition. This is a multi-classification task, the goal is to automatically classify pupil area images into 7 categories. First, 1500 original images are collected and pupil area images are extracted. Then, each image is annotated into 7 categories, with about 200 samples in each category. Then, a convolutional neural network with a ResNet50 structure is used to train an image classification model. During the training process, the initial value of the learning rate is set to 1×10 -3 , while adopting a learning rate decay strategy for dynamic adjustment; Adam was selected as the optimizer to update model weights; cross-entropy loss was used as the loss function; and the batch size was set to 64. The model was trained using the PyTorch programming framework and 100 epochs in a Ubuntu environment. During the prediction phase, the model uses the trained model weights to perform forward propagation on new pupil area images. The softmax layer outputs the probability distribution for each category, thus completing the task of automatically classifying and recognizing eye states.

[0073] like Figure 1 As shown in the figure, the eyeball status is judged by the judgment criteria for lens subluxation. The judgment criteria are as follows:

[0074] The lens / pupil area ratio is analyzed. When the lens / pupil area ratio = 1, it is automatically determined that there is no lens dislocation and the phakic eye measurement mode is automatically selected.

[0075] When the lens / pupil area ratio = 0, it is automatically judged as complete lens dislocation and the aphakic eye measurement mode is automatically selected.

[0076] When the lens / pupil area ratio is greater than 0 but less than 1, it is considered a subluxated lens. This determination is made automatically based on whether the lens is detected at the pupil center. If the lens is detected at the pupil center, the phakic measurement mode is automatically selected; if the lens is not detected at the pupil center, the aphakic measurement mode is automatically selected.

[0077] The above-mentioned different eye states, among which the lens state includes aphakia, pseudophakia, complete dislocation of the lens, and pseudophakia with an intraocular lens.

[0078] S3. Based on the eye state, complete biological measurement.

[0079] According to the determination result of the eyeball state determination module, correct biological measurement is performed or correct biological parameter measurement is performed after rescanning with the correct eyeball state.

[0080] Example 3:

[0081] In order to illustrate the advancement of the present invention compared to the prior art, this embodiment is specially set up for verification.

[0082] An analysis of data from 21 clinical patients (23 eyes) revealed significant differences in ocular biological parameters measured using different eye status measurement modes (especially in axial length, with an average difference of 0.21 mm). This can lead to significant differences in the reserved power of intraocular lenses in clinical practice (with an average difference of up to 0.55 D). The axial length measured in the aphakic mode is larger than that measured in the phakic mode; there are also significant differences in the measured power of intraocular lenses. As shown in Table 1:

[0083] Table 1

[0084]

[0085] In the table, P < 0.05 indicates that the difference is statistically significant, and the intraocular lens power is counted based on the reserved power close to 0D.

[0086] like Figure 3 As shown, the degree and extent of lens subluxation vary. Specifically, Figure 3 (a) Indicates that the lens is not completely dislocated, and the visual axis does not pass through the lens medium; Figure 3 (b) indicates that the lens is not completely dislocated, but the visual axis passes through the lens medium. Incorrectly selecting the measurement mode can lead to measurement errors and increase the patient's postoperative refractive error. Therefore, the choice of optical ocular biometry mode should be determined based on whether the visual axis passes through the lens medium.

[0087] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. An optical biological measurement device for automatically judging the state of the eyeball, characterized in that: include: Pre-scan image acquisition module, eye state judgment module and measurement module; The pre-scan image acquisition module is used to acquire a standard scan image; The eyeball state determination module is used to determine the eyeball state based on the scanned image; the eyeball state refers to the lens state; the eyeball state determination module uses different determination criteria for different eyes to determine the eyeball state, and the determination criteria are as follows: When the lens / pupil area ratio = 1, it is judged that there is no lens dislocation; When the lens / pupil area ratio = 0, it is judged as complete lens dislocation; When the lens / pupil area ratio is >0 but <1, it is judged as subluxation of the lens; The measurement module is used to complete biological measurement based on the eyeball state.

2. The optical biological measurement device for automatically judging the state of the eyeball according to claim 1, characterized in that: The lens states include: aphakia, phakia, intraocular lens, ICL and subluxated lens.

3. The optical biological measurement device for automatically judging the state of the eyeball according to claim 1, characterized in that: The eyeball state discrimination module includes: a pupil area automatic recognition unit, a pupil area image preprocessing unit and a pupil area image classification unit; The pupil area automatic recognition unit is used to recognize the pupil area based on the scanned image and obtain a circular outline of the pupil area; The pupil area image preprocessing unit is used to preprocess the circular outline to obtain a processing result; The pupil area image classification unit is used to perform classification and discrimination based on the processing result to obtain the eyeball state.

4. The optical biological measurement device for automatically judging the state of the eyeball according to claim 3, characterized in that: The workflow of the pupil area automatic recognition unit includes: The acquired color image is converted into a grayscale image and Gaussian blurred using a 5x5 Gaussian kernel; Canny edge detection is performed with a low threshold of 50 and a high threshold of 150 to capture edge features in the blurred image. On the basis of edge detection, the Hough transform method based on image gradient information is used to locate the circular contour and complete the recognition.

5. The optical biological measurement device for automatically judging the state of the eyeball according to claim 3, characterized in that: The workflow of the pupil area image preprocessing unit includes: Segment the identified circular contour to obtain the pupil area image; The resolution of the pupil area image is adjusted to 112×112 and normalized to complete the preprocessing.

6. The optical biological measurement device for automatically judging the state of the eyeball according to claim 3, characterized in that: The workflow of the pupil area image classification unit includes: using a convolutional neural network with a ResNet50 structure to train an image classification model. During the training process, the initial value of the learning rate is set to 1×10 -3 , and adopt the learning rate decay strategy for dynamic adjustment; select Adam as the optimizer to update the model weights; use cross entropy loss as the loss function; set the batch size to 64; The image classification model uses the trained model weights to forward propagate the pupil area image and outputs the probability distribution of each category through the softmax layer, thereby completing the automatic classification and recognition of the eye state.

7. The optical biological measurement device for automatically judging the state of the eyeball according to claim 1, characterized in that: When the result of the judgment is that the lens is partially dislocated, the judgment is made based on whether the lens position is detected at the pupil center. The criteria are as follows: The lens is detected at the center of the pupil, and the phakic eye measurement mode is selected; The lens is not detected at the pupil center, so the aphakic eye measurement mode is selected.

8. An optical biological measurement method for automatically judging the state of the eyeball, the method being applied to the device according to any one of claims 1 to 7, characterized in that the steps include: Acquire standard scan images; Performing a judgment based on the scanned image to obtain an eyeball state; the eyeball state refers to a lens state; The eyeball state discrimination module uses different discrimination criteria for different eyeballs to discriminate the eyeball state. The discrimination criteria are as follows: When the lens / pupil area ratio = 1, it is judged that there is no lens dislocation; When the lens / pupil area ratio = 0, it is judged as complete lens dislocation; When the lens / pupil area ratio is >0 but <1, it is judged as subluxation of the lens; Based on the eyeball status, biological measurements are performed.

9. The optical biological measurement method for automatically judging the state of the eyeball according to claim 8, characterized in that: The lens states include: aphakia, phakia, intraocular lens, ICL and subluxated lens.

Images