Method for acquiring cornea biomechanical parameters

Abstract

The invention relates to a method for acquiring cornea biomechanical parameters, which comprises the following steps of: 1, applying excitation to a cornea through an air pulse to deform the cornea, and acquiring an image in the cornea deformation process through a high-speed cornea section imaging means in the deformation process; step 2, constructing an Attention U-Net neural network deep learning model, segmenting the cornea image by using the constructed Attention U-Net neural network deep learning model, and accurately positioning the cornea boundary; 3, based on the processing result of the cornea image in the step 2, 35 cornea biomechanical parameters are calculated.

Description

Technical Field

[0001] The invention belongs to the technical field of corneal detection, and in particular relates to a method for measuring corneal biomechanical parameters. Background Art

[0002] Measuring corneal biomechanical parameters can not only objectively evaluate corneal morphology and biomechanical properties, but also provide a research basis for measuring mechanical material parameters and properties of other soft tissues such as the cornea. On the other hand, as the number of ophthalmic patients in my country continues to increase, ophthalmic medical devices are in great demand. Therefore, the independent research and development of domestic corneal biomechanical measurement equipment will help reduce medical costs and improve the quality and accessibility of ophthalmic medical services. Accurately measuring corneal biomechanical parameters will help promote the research and development of domestic corneal biomechanical equipment, and can provide important theoretical support for demand-oriented design, functional optimization, introduction of new technologies, and product innovation of equipment.

[0003] The existing methods for measuring corneal biomechanical parameters have certain defects, resulting in unsatisfactory parameter measurement results. Therefore, it is necessary to construct a method for accurately measuring corneal biomechanical parameters.

[0004] SUMMARY OF THE INVENTION

[0005] In view of the deficiencies of the prior art, the present invention provides a method for obtaining corneal biomechanical parameters with high measurement accuracy.

[0006] The above-mentioned object of the present invention is achieved by the following technical solutions:

[0007] A method for obtaining corneal biomechanical parameters comprises the following steps:

[0008] Step 1: Excite the cornea through air pulses to cause the cornea to deform. During the deformation process, images of the corneal deformation process are collected through high-speed corneal cross-sectional imaging.

[0009] Step 2: construct an Attention U-Net neural network deep learning model, and use the constructed Attention U-Net neural network deep learning model to segment the corneal image and accurately locate the corneal boundary;

[0010] Step 3: Based on the processing results of the corneal image in step 2, 35 corneal biomechanical parameters are calculated.

[0011] Furthermore, in step 1, the high-speed corneal cross-sectional imaging method may be a Scheimpflug method or an optical coherence tomography method.

[0012] Moreover, in step 2, building the Attention U-Net neural network deep learning model includes:

[0013] Based on the deep learning model of U-Net architecture, the Attention Gate module is introduced to form the Attention U-Net neural network deep learning model, which consists of an encoder, a decoder and an Attention Gate module. Establishing the Attention U-Net neural network deep learning model includes the following three steps: data processing, model training and model verification.

[0014] Data processing: Use the data annotation tool LabelMe to complete the data set annotation of the collected images and identify the areas of interest in the images. The annotation results will be used as labels for training the model so that they can be used for subsequent training of the Attention U-Net network.

[0015] Model training: First, define an Attention Gate layer in the Python Keras model, which will perform pixel-by-pixel addition, ReLU activation, 1x1 convolution, and Sigmoid operations to generate attention weights; second, perform element-by-element product operations on the obtained attention weight coefficients and the feature maps in the encoding phase to optimize the feature maps; then, select part of the data set as the training data set to start the training process, and the model will learn how to predict the corresponding segmentation results based on the input image. Finally, save the trained model.

[0016] Model verification: In the verification data set, the Dice Similarity Coefficient (DSC) was used as the evaluation index of the segmentation process. The DSC value of each image in the test set was calculated, and the overall average DSC value was statistically calculated to reflect the overall performance of the model. According to the verification results, the model parameters were adjusted and the trained Attention U-Net neural network deep learning model was saved.

[0017] Moreover, in step 2, the corneal image is segmented using the constructed Attention U-Net neural network deep learning model, including:

[0018] Based on the trained Attention U-Net neural network deep learning model, the overall algorithm architecture is constructed. According to the algorithm architecture, the corneal image is segmented and the specific steps of accurately locating the corneal boundary include:

[0019] Input data: The corneal image data collected based on high-speed corneal cross-sectional imaging technology is used as the input data of this system;

[0020] Input model: Input image data into the Attention U-Net model for image segmentation;

[0021] Output result: The corneal image data after segmentation is used as the output result of this system.

[0022] Furthermore, step 3 includes:

[0023] In step 3, the resolution of the corneal image after segmentation is read as M*N pixels, the number of frames is I, and the upper and lower boundary positions of the cornea in the image are recorded as y i，m，n , where i represents the image frame number, m represents the image row data, and n represents the image column data, and subsequent biomechanical parameter calculations are performed, including:

[0024] 3.1 According to the position of the upper corneal boundary y i，m，n , take the upper border of the cornea to determine the corneal vertex position y Apex (n = Apex), calculate the upper boundary deformation displacement DA of the corneal vertex and the eye movement displacement WEMA in each frame image, and the corneal vertex deviation displacement DLA is calculated by the following formula:

[0025] DLA==DA--WEMA

[0026] 3.2 Calculate the maximum eye movement displacement WEMA MAX and time WEMT: The maximum eye movement displacement is the maximum value of WEMA. Record the maximum eye movement displacement frame number i. The maximum eye movement displacement time is calculated by the following formula:

[0027] WEMT=(i-1) / f

[0028] Wherein, f is the frame rate of high-speed corneal cross-sectional imaging technology.

[0029] 3.3 Calculate the maximum concave moment HCT, the deformation amplitude HCDA and the deviation amplitude HCDLA at the maximum concave moment: record the maximum deformation displacement frame number imax of the corneal vertex, and the maximum concave moment is calculated by the following formula:

[0030] HCT=(imax-1) / f

[0031] HCDA takes the maximum value of the corneal vertex deformation displacement, and HCDLA takes the deviation displacement of the corneal vertex at the HCT moment.

[0032] 3.4 Calculate the first applanation moment A1T: In the first half of the frame images, take the average value of the coordinate heights of the adjacent columns a at the vertex of the upper surface of the cornea in each frame image. The formula is as follows:

[0033] y1 i =(y i，Apex-a +…y i，Apex +…+y i，Apex+a ) / (2a+1)

[0034] Record y1i =y i，Apex The corresponding frame number il is used to calculate the first flattening moment:

[0035] A1T=(i1-1) / f

[0036] 3.5 Calculate the second applanation moment A2T: In the second half of the frame images, take the average value of the coordinate heights of the adjacent columns a at the vertex on the upper surface of the cornea in each frame image. The formula is as follows:

[0037] y2 i =(y i，Apex-a +…y i，Apex +…+y i，Apex+a ) / (2a+1)

[0038] Record y2 i =y i，Apex The corresponding frame number i2 is used to calculate the second flattening moment:

[0039] A2T=(i2-1) / f

[0040] 3.6 Calculate the deformation amplitude A1DA and deviation amplitude A1DLA at the first applanation moment: A1DA is the deformation displacement value of the corneal vertex at the moment A1T, and A1DLA is the offset displacement value of the corneal vertex at the moment A1T.

[0041] 3.7 Calculate the deformation amplitude A2DA and deviation amplitude A2DLA at the second applanation moment: A2DA is the deformation displacement value of the corneal vertex at the moment A2T, and A2DLA is the offset displacement value of the corneal vertex at the moment A2T.

[0042] 3.8 Calculate the maximum deviation amplitude DLAM and the maximum deviation amplitude moment DLAMT: The maximum deviation amplitude is the maximum deviation displacement value of the corneal vertex, and the frame number imax1 of the maximum deviation displacement is recorded. The maximum deviation amplitude moment is calculated by the following formula:

[0043] DLAMT=(imax1-1) / f

[0044] 3.9 Calculate the central corneal thickness CCT: Take the average height difference of the upper and lower boundary position data of each column a adjacent to the left and right of the corneal vertex in each frame image, the formula is as follows:

[0045] CCT i =(Δy i，Apex-a +…+Δy i，Apex +…+Δy i，Apex+a ) / (2a+1)

[0046] CCT is the average value of CCTi data in each group.

[0047] 3.10 Calculate the distance PD between the two peaks at the moment of maximum concavity: record the position data y of the upper border of the cornea at the moment of HCT n , record the average of the first maximum value column and the last maximum value column in the first Apex column data as the left end peak value n 左 ; In the subsequent columns of data in the Apex column, record the average value of the column where the first maximum value is located and the column where the last maximum value is located as the right end peak value n 右 The distance PD between the two peaks at the maximum depression moment is calculated by the following formula:

[0048] PD=n 右 -n 左

[0049] 3.11 Calculate the 1mm maximum deformation amplitude ratio DA Ratio Max[1mm], that is, the maximum ratio of the deformation amplitude at the corneal vertex to the deformation amplitude 1mm away from the corneal vertex: take the average deformation displacement of the two points 1mm away from the corneal vertex in each frame image, the formula is as follows:

[0050] DA 1mm =(DA 1mm左 +DA 1mm右 ) / 2

[0051] According to the deformation displacement DA of the upper boundary of the corneal vertex in each frame image, DA Ratio [1mm] is calculated by the following formula:

[0052] DA Ratio[1mm]=DA / DA 1mm

[0053] The 1mm maximum deformation amplitude ratio is the maximum value of DA Ratio[1mm] in each group.

[0054] 3.12 Calculate the 2mm maximum deformation amplitude ratio DA Ratio Max[2mm], that is, the maximum ratio of the deformation amplitude at the corneal vertex to the deformation amplitude 2mm away from the corneal vertex: take the average deformation displacement of the two points 2mm away from the corneal vertex in each frame image, the formula is as follows:

[0055] DA 2mm =(DA 2mm左 +DA 2mm右 ) / 2

[0056] According to the deformation displacement DA of the upper boundary of the corneal vertex in each frame image, DA Ratio [2mm] is calculated by the following formula:

[0057] DA Ratio[2mm]=DA / DA 2mm

[0058] The 2mm maximum deformation amplitude ratio is the maximum value of DA Ratio[2mm] in each group.

[0059] 3.13 Calculate the 1mm maximum deviation amplitude ratio DLA Ratio Max[1mm], that is, the maximum ratio of the deviation amplitude at the corneal vertex to the deviation amplitude at 1mm away from the corneal vertex: take the average deviation displacement of the two points 1mm away from the corneal vertex in each frame image, the formula is as follows:

[0060] DLA 1mm =(DLA 1mm左 +DLA 1mm右 ) / 2

[0061] According to the deviation displacement DLA of the upper boundary of the corneal vertex in each frame image, DLA Ratio [1mm] is calculated by the following formula:

[0062] DLA Ratio[1mm]=DLA / DLA 1mm

[0063] The 1mm maximum deviation amplitude ratio is the maximum value of DLA Ratio[1mm] in each group.

[0064] 3.14 Calculate the 2mm maximum deviation amplitude ratio DLA Ratio Max[2mm], that is, the maximum ratio of the deviation amplitude at the corneal vertex to the deviation amplitude 2mm away from the corneal vertex: take the average deviation displacement of the two points 2mm away from the corneal vertex in each frame image, the formula is as follows:

[0065] DLA 2mm =(DLA 2mm左 +DLA 2mm右 ) / 2

[0066] According to the deviation displacement DLA of the upper boundary of the corneal vertex in each frame image, DLA Ratio [2mm] is calculated by the following formula:

[0067] DLA Ratio[2mm]=DLA / DLA 2mm

[0068] The 2mm maximum deviation amplitude ratio is the maximum value of DLA Ratio[2mm] in each group.

[0069] 3.15 Calculate the first applanation velocity A1V and the second applanation velocity A2V: record the deformation amplitude of two adjacent groups of corneal vertices, and the velocity V is calculated by the following formula:

[0070] V=(DA i+1 -DA i-1 ) / 2

[0071] At the first flattening moment A1T and the second flattening moment A2T, the flattening speed is calculated according to A1DA and A2DA respectively.

[0072] 3.16 Calculate the first flattening length A1L and the second flattening length A2L: According to the corneal upper boundary position data y in each frame image i，m，n , record the first maximum value column n1 and the last maximum value column n2, and the flattening length is calculated by the following formula:

[0073] L=n2-n1

[0074] At the moments A1T and A2T respectively, the flattening length is calculated according to the above formula.

[0075] 3.17 Calculate the first applanation deviation length A1DLL, the second applanation deviation length A2DLL and the deviation length HCDLL at the maximum concave moment: the line connecting the two farthest points where the cornea deviates is the deviation length. Determine the position where the upper boundary deviates in each frame image, record the column n1 where the leftmost end of the deviation position is located and the column n2 where the rightmost end is located. The deviation length DLL can be calculated by the following formula:

[0076] DLL=n2-n1

[0077] At A1T, A2T and HCT respectively, A1DLL, A2DLL and HCDLL are calculated according to the above formulas.

[0078] 3.18 Calculate the first applanation arc length change A1dArcL, the second applanation arc length change A2dArcL, the maximum depression arc length change HCdArcL and the maximum arc length change dArcLM: Calculate the arc length ArcL in the 7mm area near the center of the corneal vertex in each frame image, and calculate it by the following formula:

[0079]

[0080] Record the arc length ArcL0 of the cornea in the initial state, and calculate the arc length change dArcL of the cornea. The formula is as follows:

[0081] dArcL=ArcL--ArcL0

[0082] At A1T, A2T and HCT respectively, A1dArcL, A2dArcL and HCdArcL are calculated according to the above formula. The maximum arc length change is the maximum value of dArcL.

[0083] 3.19 Calculate the first applanation deviation area A1DLAr, the second applanation deviation area A2DLAr and the maximum concave deviation area HCDLAr: Calculate the corneal deviation area according to the deviation amplitude DLA of each point on the upper border of the cornea in each frame image, and the formula is as follows:

[0084]

[0085] At A1T, A2T and HCT respectively, A1DLAr, A2DLAr and HCDLAr are calculated according to the above formulas.

[0086] 3.20 Calculate the corneal curvature radius HCT Radius at maximum indentation and the maximum inverse concave radius MaxInverseRadius, IRM: Calculate the corneal vertex curvature radius Radius in each frame image, and calculate HCTRadius at the HCT moment. The formula for the inverse concave radius is as follows:

[0087] Inverse Radius=1 / Radius

[0088] The maximum inverse concave radius is the maximum value of Inverse Radius.

[0089] 3.21 The above 35 parameter calculation methods are all performed at the pixel size level and need to be converted into actual sizes based on the physical size of the image resolution or pixel density.

[0090] The advantages and positive effects of the present invention are:

[0091] The present invention provides a method for accurately measuring corneal biomechanical parameters. It makes up for the shortcomings of previous measurement methods through high-speed corneal cross-section imaging, accurate positioning of corneal boundaries, and precise calculation of parameters according to definitions, providing important theoretical support for the research and development of domestic corneal biomechanical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 is a flow chart of a method for obtaining corneal biomechanical parameters of the present invention;

[0093] Figure 2 It is the Attention U-Net neural network deep learning model diagram constructed by the present invention;

[0094] Figure 3 It is the overall algorithm flow chart constructed by the present invention.

[0095] Figure 4 The embodiment of the present invention obtains an image of the cornea in its initial form;

[0096] Figure 5This is the image after segmentation by the deep learning model in the embodiment of the present invention. DETAILED DESCRIPTION

[0097] The structure of the present invention is further described below with reference to the accompanying drawings and by way of examples. It should be noted that the present examples are descriptive rather than restrictive.

[0098] A method for obtaining corneal biomechanical parameters, see Figure 1-Figure 3 , the invention comprises the following steps:

[0099] Step 1: Excite the cornea through air pulses to cause the cornea to deform, and collect images of the corneal deformation process through high-speed corneal cross-sectional imaging; wherein, the high-speed corneal cross-sectional imaging method can adopt the Scheimpflug method or the optical coherence tomography (OCT) method.

[0100] Step 2: construct an Attention U-Net neural network deep learning model, and use the constructed Attention U-Net neural network deep learning model to segment the corneal image and accurately locate the corneal boundary;

[0101] Step 3: Based on the processing results of the corneal image in step 2, 35 corneal biomechanical parameters are calculated.

[0102] The image segmentation in step 2 is as follows:

[0103] Based on the U-Net architecture, the Attention Gate module is introduced to form the Attention U-Net neural network deep learning model. Figure 2 As shown in the figure. The model mainly consists of encoder, decoder and Attention Gate modules. Building the Attention U-Net neural network deep learning model includes the following three steps: data processing, model training and model verification.

[0104] Data processing: Use the data annotation tool LabelMe to complete the data set annotation of the collected images and identify the areas of interest in the images. The annotation results will be used as labels for training the model so that they can be used for subsequent training of the Attention U-Net network.

[0105] Model training: First, define an Attention Gate layer in the Python Keras model, which will perform pixel-by-pixel addition, ReLU activation, 1x1 convolution, and Sigmoid operations to generate attention weights; second, perform element-by-element product operations on the obtained attention weight coefficients and the feature maps in the encoding phase to optimize the feature maps; then, select part of the data set as the training data set to start the training process, and the model will learn how to predict the corresponding segmentation results based on the input image. Finally, save the trained model.

[0106] Model verification: In the verification data set, the Dice Similarity Coefficient (DSC) was used as the evaluation index of the segmentation process. The DSC value of each image in the test set was calculated, and the overall average DSC value was statistically calculated to reflect the overall performance of the model. According to the verification results, the model parameters were adjusted and the trained Attention U-Net neural network deep learning model was saved.

[0107] Based on the trained Attention U-Net neural network deep learning model, the overall algorithm architecture is constructed. According to the algorithm architecture, the specific steps of segmenting the corneal image and accurately locating the corneal boundary are as follows: Figure 3 As shown:

[0108] Input data: The corneal image data collected based on high-speed corneal cross-sectional imaging technology is used as the input data of this system.

[0109] Input model: Input image data into the Attention U-Net model for image segmentation.

[0110] Output result: The corneal image data after segmentation is used as the output result of this system.

[0111] In step 3, the resolution of the corneal image after segmentation is read as M*N pixels, the number of frames is I, and the upper and lower boundary positions of the cornea in the image are recorded as y i，m，n (where i represents the number of image frames, m represents the image row data, and n represents the image column data) to perform subsequent biomechanical parameter calculations, including:

[0112] 3.1 According to the position of the upper corneal boundary y i，m，n , take the upper border of the cornea to determine the corneal vertex position y Apex (n = Apex), calculate the corneal vertex upper boundary deformation displacement (DA) and eye movement displacement (WEMA) in each frame image, and the corneal vertex deviation displacement (DLA) is calculated by the following formula:

[0113] DLA=DA--WEMA

[0114] 3.2 Calculate the maximum eye movement displacement (WEMA MAX) and time (WEMT): The maximum eye movement displacement is the WEMA maximum value. Record the maximum eye movement displacement frame number i. The maximum eye movement displacement time is calculated by the following formula:

[0115] WEMT=(i-1) / f

[0116] Wherein, f is the frame rate of high-speed corneal cross-sectional imaging technology.

[0117] 3.3 Calculate the maximum concave moment (HCT), the maximum concave moment deformation amplitude (HCDA) and the maximum concave moment deviation amplitude (HCDLA): record the maximum deformation displacement frame number imax of the corneal vertex, and the maximum concave moment is calculated by the following formula:

[0118] HCT=(imax-1) / f

[0119] HCDA takes the maximum value of the corneal vertex deformation displacement, and HCDLA takes the deviation displacement of the corneal vertex at the HCT moment.

[0120] 3.4 Calculate the first applanation time (AlT): In the first half of the frame images, take the average value of the coordinate heights of the adjacent columns a at the vertex of the upper surface of the cornea in each frame image. The formula is as follows:

[0121] y1 i =(y i，Apex-a +…y i，Apex +…+y i，Apex+a ) / (2a+1)

[0122] Record y1 i =y i，Apex The corresponding frame number il is used to calculate the first flattening moment:

[0123] A1T=(i1-1) / f

[0124] 3.5 Calculate the second applanation moment (A2T): In the second half of the frame number images, take the average value of the coordinate height of each a column adjacent to the vertex on the upper surface of the cornea in each frame image, the formula is as follows:

[0125] y2 i =(y i，Apex-a +…y i，Apex +…+y i，Apex+a ) / (2a+1)

[0126] Record y2 i =y i，Apex The corresponding frame number i2 is used to calculate the second flattening moment:

[0127] A2T=(i2-1) / f

[0128] 3.6 Calculate the deformation amplitude (A1DA) and deviation amplitude (A1DLA) at the first applanation moment: A1DA is the deformation displacement value of the corneal vertex at the A1T moment, and A1DLA is the deviation displacement value of the corneal vertex at the A1T moment.

[0129] 3.7 Calculate the deformation amplitude (A2DA) and deviation amplitude (A2DLA) at the second applanation moment: A2DA is the deformation displacement value of the corneal vertex at the A2T moment, and A2DLA is the offset displacement value of the corneal vertex at the A2T moment.

[0130] 3.8 Calculate the maximum deviation amplitude (DLAML) and the maximum deviation amplitude moment (DLAMT): The maximum deviation amplitude is the maximum deviation displacement value of the corneal vertex, and the frame number imax1 of the maximum deviation displacement is recorded. The maximum deviation amplitude moment is calculated by the following formula:

[0131] DLAMT=(imax1-1) / f

[0132] 3.9 Calculate the central corneal thickness (CCT): Take the average height difference of the upper and lower boundary position data of each column a adjacent to the left and right of the corneal vertex in each frame image. The formula is as follows:

[0133] CCT i =(Δy i，Apex-a +…+Δy i，Apex +…+Δy iApex+a ) / (2a+1)

[0134] CCT is the CCT of each group i The mean of the data.

[0135] 3.10 Calculate the distance between the two peaks (PD) at the moment of maximum concavity: record the position data of the upper border of the cornea at the moment of HCT n , record the average of the first maximum value column and the last maximum value column in the first Apex column data as the left end peak value n 左 ; In the subsequent columns of data in the Apex column, record the average value of the column where the first maximum value is located and the column where the last maximum value is located as the right end peak value n 右 The distance PD between the two peaks at the maximum depression moment is calculated by the following formula:

[0136] PD=n 右 -n 左

[0137] 3.11 Calculate the 1mm maximum deformation amplitude ratio (DA Ratio Max[1mm]), that is, the maximum ratio of the deformation amplitude at the corneal vertex to the deformation amplitude 1mm away from the corneal vertex: take the average deformation displacement of the two points 1mm away from the corneal vertex in each frame image, the formula is as follows:

[0138] DA 1mm =(DA 1mm左 +DA 1mm右 ) / 2

[0139] According to the deformation displacement DA of the upper boundary of the corneal vertex in each frame image, DA Ratio [1mm] is calculated by the following formula:

[0140] DA Ratio[1mm]=DA / DA 1mm

[0141] The 1mm maximum deformation amplitude ratio is the maximum value of DA Ratio[1mm] in each group.

[0142] 3.12 Calculate the 2mm maximum deformation amplitude ratio (DA Ratio Max[2mm]), that is, the maximum ratio of the deformation amplitude at the corneal vertex to the deformation amplitude 2mm away from the corneal vertex: take the average deformation displacement of the two points 2mm away from the corneal vertex in each frame image, the formula is as follows:

[0143] DA2mm=(DA2mm left+DA2mm right) / 2

[0144] According to the deformation displacement DA of the upper boundary of the corneal vertex in each frame image, DA Ratio [2mm] is calculated by the following formula:

[0145] DA Ratio[2mm]=DA / DA 2mm

[0146] The 2mm maximum deformation amplitude ratio is the maximum value of DA Ratio[2mm] in each group.

[0147] 3.13 Calculate the 1mm maximum deviation amplitude ratio (DLA Ratio Max [1mm]), that is, the maximum ratio of the deviation amplitude at the corneal vertex to the deviation amplitude at 1mm from the corneal vertex: take the average deviation displacement of the two points 1mm to the left and right of the corneal vertex in each frame image, the formula is as follows:

[0148] DLA 1mm =(DLA 1mm左 +DLA 1mm右 ) / 2

[0149] According to the deviation displacement DLA of the upper boundary of the corneal vertex in each frame image, DLA Ratio [1mm] is calculated by the following formula:

[0150] DLA Ratio[1mm]=DLA / DLA 1mm

[0151] The 1mm maximum deviation amplitude ratio is the maximum value of DLA Ratio[1mm] in each group.

[0152] 3.14 Calculate the 2mm maximum deviation amplitude ratio (DLA Ratio Max[2mm]), that is, the maximum ratio of the deviation amplitude at the corneal vertex to the deviation amplitude 2mm away from the corneal vertex: take the average deviation displacement of the two points 2mm away from the corneal vertex in each frame image, and the formula is as follows:

[0153] DLA 2mm =(DLA 2mm左 +DLA 2mm右 ) / 2

[0154] According to the deviation displacement DLA of the upper boundary of the corneal vertex in each frame image, DLA Ratio [2mm] is calculated by the following formula:

[0155] DLA Ratio[2mm]=DLA / DLA 2mm

[0156] The 2mm maximum deviation amplitude ratio is the maximum value of DLA Ratio[2mm] in each group.

[0157] 3.15 Calculate the first applanation velocity (A1V) and the second applanation velocity (A2V): Record the amplitude of the deformation of two adjacent groups of corneal vertices, and the velocity V is calculated by the following formula:

[0158] V=(DA i+1 -DA i-1 ) / 2

[0159] At the first flattening moment A1T (i=i1) and the second flattening moment A2T (i=i2), the flattening speed is calculated according to A1DA and A2DA respectively.

[0160] 3.16 Calculate the first flattening length (A1L) and the second flattening length (A2L): According to the corneal upper boundary position data y in each frame image i，m，n , record the first maximum value column n1 and the last maximum value column n2, and the flattening length is calculated by the following formula:

[0161] L=n2-n1

[0162] At the moments A1T and A2T respectively, the flattening length is calculated according to the above formula.

[0163] 3.17 Calculate the first applanation deviation length (A1DLL), the second applanation deviation length (A2DLL) and the maximum concave deviation length (HCDLL): The line connecting the two farthest points where the cornea deviates is the deviation length. Determine the position where the upper boundary deviates in each frame image, record the column n1 where the leftmost end of the deviation position is located and the column n2 where the rightmost end is located. The deviation length DLL can be calculated by the following formula:

[0164] DLL=n2-n1

[0165] At A1T, A2T and HCT respectively, A1DLL, A2DLL and HCDLL are calculated according to the above formulas.

[0166] 3.18 Calculate the first applanation arc length change (A1dArcL), the second applanation arc length change (A2dArcL), the maximum depression arc length change (HCdArcL) and the maximum arc length change (dArcLM): Calculate the arc length ArcL in the 7mm area (n=a~n=b) near the center of the corneal vertex in each frame image, and calculate it by the following formula:

[0167]

[0168] Record the arc length ArcL0 of the cornea in the initial state, and calculate the arc length change of the cornea (dArcL). The formula is as follows:

[0169] dArcL=ArcL--ArcL0

[0170] At A1T, A2T and HCT respectively, A1dArcL, A2dArcL and HCdArcL are calculated according to the above formula. The maximum arc length change is the maximum value of dArcL.

[0171] 3.19 Calculation of the first applanation deviation area (A1DLAr), the second applanation deviation area (A2DLAr) and the maximum concave deviation area (HCDLAr): The corneal deviation area is calculated according to the deviation amplitude DLA of each point on the upper border of the cornea in each frame image, and the formula is as follows:

[0172]

[0173] At A1T, A2T and HCT respectively, A1DLAr, A2DLAr and HCDLAr are calculated according to the above formulas.

[0174] 3.20 Calculate the corneal curvature radius at maximum indentation (HCT Radius) and the maximum inverse concave radius (MaxInverse Radius, IRM): Calculate the corneal vertex curvature radius Radius in each frame image, and calculate HCTRadius at the HCT moment. The formula for the inverse concave radius is as follows:

[0175] Inverse Radius = 1 / Radius

[0176] The maximum inverse concave radius is the maximum value of Inverse Radius.

[0177] 3.21 The above 35 parameter calculation methods are all performed at the pixel size level and need to be converted into actual sizes based on the physical size of the image resolution or pixel density.

[0178] Example:

[0179] Get the image of the cornea in its initial shape. Figure 4 ;

[0180] The image is segmented by the deep learning model. Figure 5 :

[0181] Through the parameter calculation method described in the present invention, the corneal biomechanical parameter results are as follows:

[0182]

[0183]

[0184] Although the embodiments and drawings of the present invention are disclosed for illustrative purposes, those skilled in the art will appreciate that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for obtaining corneal biomechanical parameters, characterized in that: The steps include: Step 1: Excite the cornea by air pulses to cause the cornea to deform, and during the deformation process, collect images of the corneal deformation process by high-speed corneal cross-sectional imaging; Step 2: construct an Attention U-Net neural network deep learning model, and use the constructed Attention U-Net neural network deep learning model to segment the corneal image and accurately locate the corneal boundary; Step 3: Based on the processing results of the corneal image in step 2, 35 corneal biomechanical parameters are calculated.

2. The method for obtaining corneal biomechanical parameters according to claim 1, characterized in that: In step 1, the high-speed corneal cross-sectional imaging method may be a Scheimpflug method or an optical coherence tomography method.

3. The method for obtaining corneal biomechanical parameters according to claim 1, characterized in that: In step 2, building the Attention U-Net neural network deep learning model includes: Based on the deep learning model of U-Net architecture, the Attention Gate module is introduced to form the Attention U-Net neural network deep learning model, which consists of an encoder, a decoder and an Attention Gate module. The establishment of the Attention U-Net neural network deep learning model includes the following three steps: data processing, model training and model verification; Data processing: Use the data annotation tool LabelMe to complete the data set annotation of the collected images and identify the areas of interest in the images. The annotation results will be used as labels for training the model; Model training: First, define an Attention Gate layer in the Keras model of Python. This layer will perform pixel-by-pixel addition, ReLU activation, 1x1 convolution, and Sigmoid operations to generate attention weights. Secondly, perform element-by-element product operations on the obtained attention weight coefficients and the feature maps in the encoding stage to optimize the feature maps. Next, select part of the data set as the training data set to start the training process. The model will learn how to predict the corresponding segmentation results based on the input image. Finally, save the trained model. Model verification: In the verification data set, the Dice similarity coefficient is used as the evaluation index of the segmentation process. The DSC value of each image in the test set is calculated, and the overall average DSC value is counted to reflect the overall performance of the model. According to the verification results, the model parameters are adjusted and the trained Attention U-Net neural network deep learning model is saved.

4. The method for obtaining corneal biomechanical parameters according to claim 1, characterized in that: In step 2, the corneal image is segmented using the constructed Attention U-Net neural network deep learning model, including: Based on the trained Attention U-Net neural network deep learning model, the overall algorithm architecture is constructed. According to the algorithm architecture, the corneal image is segmented and the specific steps of accurately locating the corneal boundary include: Input data: The corneal image data collected based on high-speed corneal cross-sectional imaging technology is used as the input data of this system; Input model: Input image data into the Attention U-Net model for image segmentation; Output result: The corneal image data after segmentation is used as the output result of this system.

5. The method for acquiring corneal biomechanical parameters according to claim 1, characterized in that: Step 3 includes: In step 3, the resolution of the corneal image after segmentation is read as M*N pixels, the number of frames is I, and the upper and lower boundary positions of the cornea in the image are recorded as y i，m，n , where i represents the image frame number, m represents the image row data, and n represents the image column data, and subsequent biomechanical parameter calculations are performed, including: 3.1 According to the upper and lower border positions of the cornea i，m，n , take the upper border of the cornea to determine the corneal vertex position y Apex (n = Apex), calculate the upper boundary deformation displacement DA of the corneal vertex and the eye movement displacement WEMA in each frame image, and the corneal vertex deviation displacement DLA is calculated by the following formula: DLA=DA-WEMA 3.2 Calculate the maximum eye movement displacement WEMA MAX and time WEMT: The maximum eye movement displacement is the maximum value of WEMA. Record the maximum eye movement displacement frame number i. The maximum eye movement displacement time is calculated by the following formula: WEMT=(i-1) / f Wherein, f is the frame rate of high-speed corneal cross-sectional imaging technology; 3.3 Calculate the maximum concave moment HCT, the deformation amplitude HCDA and the deviation amplitude HCDLA at the maximum concave moment: record the maximum deformation displacement frame number imax of the corneal vertex, and the maximum concave moment is calculated by the following formula: HCT=(imax-1) / f HCDA takes the maximum value of the corneal vertex deformation displacement, and HCDLA takes the deviation displacement of the corneal vertex at the HCT moment; 3.4 Calculate the first applanation moment A1T: In the first half of the frame images, take the average value of the coordinate heights of the adjacent columns a at the vertex of the upper surface of the cornea in each frame image. The formula is as follows: <h2 style=";text-align:left;direction:ltr">y1<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (y)<h2 style=";text-align:left;direction:ltr"> i，Apex-a <h2 style=";text-align:left;direction:ltr"> +…y<h2 style=";text-align:left;direction:ltr"> i，Apex <h2 style=";text-align:left;direction:ltr"> +…+y<h2 style=";text-align:left;direction:ltr"> i，Apex+a <h2 style=";text-align:left;direction:ltr"> ) / (2a+1) Record y1 i =y i，Apex The corresponding frame number i1 is used to calculate the first flattening moment: A1T=(i1-1) / f 3.5 Calculate the second applanation time A2T: In the second half of the frame number image, take the average value of the coordinate height of each adjacent column a at the vertex of the upper surface of the cornea in each frame image, the formula is as follows: <h2 style=";text-align:left;direction:ltr">y2<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (y)<h2 style=";text-align:left;direction:ltr"> i，Apex-a <h2 style=";text-align:left;direction:ltr"> +…y<h2 style=";text-align:left;direction:ltr"> i，Apex <h2 style=";text-align:left;direction:ltr"> +…+y<h2 style=";text-align:left;direction:ltr"> i，Apex+a <h2 style=";text-align:left;direction:ltr"> ) / (2a+1) Record y2 i =y i，Apex The corresponding frame number i2 is used to calculate the second flattening moment: A2T=(i2-1) / f 3.6 Calculate the deformation amplitude A1DA and deviation amplitude A1DLA at the first applanation moment: A1DA is the deformation displacement value of the corneal vertex at the moment A1T, and A1DLA is the deviation displacement value of the corneal vertex at the moment A1T; 3.7 Calculate the deformation amplitude A2DA and deviation amplitude A2DLA at the second applanation moment: A2DA is the deformation displacement value of the corneal vertex at the A2T moment, and A2DLA is the deviation displacement value of the corneal vertex at the A2T moment; 3.8 Calculate the maximum deviation amplitude DLAM and the maximum deviation amplitude moment DLAMT: The maximum deviation amplitude is the maximum deviation displacement value of the corneal vertex, and the frame number imax1 of the maximum deviation displacement is recorded. The maximum deviation amplitude moment is calculated by the following formula: DLAMT=(imax1-1) / f 3.9 Calculate the central corneal thickness CCT: Take the average height difference of the upper and lower boundary position data of each column a adjacent to the left and right of the corneal vertex in each frame image, the formula is as follows: CCT i =(Δy iApex-a +…+Δy i，Apex +…+Δy i，Apex+a ) / (2a+1) CCT is the CCT of each group i The average value of the data; 3.10 Calculate the distance PD between the two peaks at the moment of maximum concavity: record the position data y of the upper border of the cornea at the moment of HCT n , record the average of the first maximum value column and the last maximum value column in the first Apex column data as the left end peak value n 左 ; In the subsequent columns of data in the Apex column, record the average value of the column where the first maximum value is located and the column where the last maximum value is located as the right end peak value n 右 The distance PD between the two peaks at the maximum depression moment is calculated by the following formula: PD=n 右 -n 左 3.11 Calculate the 1mm maximum deformation amplitude ratio DA Ratio Max[1mm], that is, the maximum ratio of the deformation amplitude at the corneal vertex to the deformation amplitude 1mm away from the corneal vertex: take the average deformation displacement of the two points 1mm away from the corneal vertex in each frame image, the formula is as follows: YES 1mm =(YES 1mm左 +YES 1mm右 ) / 2 According to the deformation displacement DA of the upper boundary of the corneal vertex in each frame image, DA Ratio [1mm] is calculated by the following formula: DA Ratio[1mm]=DA / DA 1mm The 1mm maximum deformation amplitude ratio is the maximum value of DA Ratio[1mm] in each group; 3.12 Calculate the 2mm maximum deformation amplitude ratio DA Ratio Max[2mm], that is, the maximum ratio of the deformation amplitude at the corneal vertex to the deformation amplitude 2mm away from the corneal vertex: take the average deformation displacement of the two points 2mm away from the corneal vertex in each frame image, the formula is as follows: YES 2mm =(YES 2mm左 +YES 2mm右 ) / 2 According to the deformation displacement DA of the upper boundary of the corneal vertex in each frame image, DA Ratio [2mm] is calculated by the following formula: DA Ratio[2mm]=DA / DA 2mm The 2mm maximum deformation amplitude ratio is the maximum value of DA Ratio[2mm] in each group; 3.13 Calculate the 1mm maximum deviation amplitude ratio DLA Ratio Max[1mm], that is, the maximum ratio of the deviation amplitude at the corneal vertex to the deviation amplitude at 1mm away from the corneal vertex: take the average deviation displacement of the two points 1mm away from the corneal vertex in each frame image, the formula is as follows: FOR 1mm =(FOR 1mm左 +FOR 1mm右 ) / 2 According to the deviation displacement DLA of the upper boundary of the corneal vertex in each frame image, DLA Ratio [1mm] is calculated by the following formula: FOR Ratio[1mm]=FOR / FOR 1mm The 1mm maximum deviation amplitude ratio is the maximum value of DLA Ratio[1mm] in each group; 3.14 Calculate the 2mm maximum deviation amplitude ratio DLA Ratio Max[2mm], that is, the maximum ratio of the deviation amplitude at the corneal vertex to the deviation amplitude 2mm away from the corneal vertex: take the average deviation displacement of the two points 2mm away from the corneal vertex in each frame image, the formula is as follows: FOR 2mm =(FOR 2mm左 +FOR 2mm右 ) / 2 According to the deviation displacement DLA of the upper boundary of the corneal vertex in each frame image, DLA Ratio [2mm] is calculated by the following formula: FOR Ratio[2mm]=FOR / FOR 2mm The 2mm maximum deviation amplitude ratio is the maximum value of DLA Ratio[2mm] in each group; 3.15 Calculate the first applanation velocity A1V and the second applanation velocity A2V: Record the deformation amplitudes of two adjacent groups of corneal vertices, and the velocity V is calculated by the following formula: V=(DA i+1 -Yes i-1 ) / 2 At the first flattening moment A1T and the second flattening moment A2T, the flattening speed is calculated according to A1DA and A2DA; 3.16 Calculate the first flattening length A1L and the second flattening length A2L: According to the corneal upper boundary position data y in each frame image i，m，n , record the first maximum value column n1 and the last maximum value column n2, and the flattening length is calculated by the following formula: L=n2-n1 At the time A1T and A2T respectively, the flattening length is calculated according to the above formula; 3.17 Calculate the first applanation deviation length A1DLL, the second applanation deviation length A2DLL and the deviation length HCDLL at the maximum concavity: the line connecting the two farthest points where the cornea deviates is the deviation length; determine the position where the upper boundary deviates in each frame image, record the column n1 where the leftmost end of the deviation position is located and the column n2 where the rightmost end is located. The deviation length DLL can be calculated by the following formula: DLL=n2-n1 At A1T, A2T and HCT respectively, calculate A1DLL, A2DLL and HCDLL according to the above formula; 3.18 Calculate the first applanation arc length change A1dArcL, the second applanation arc length change A2dArcL, the maximum depression arc length change HCdArcL and the maximum arc length change dArcLM: Calculate the arc length ArcL in the 7mm area near the center of the corneal vertex in each frame image, and calculate it by the following formula: Record the arc length ArcL0 of the cornea in the initial state, and calculate the arc length change dArcL of the cornea. The formula is as follows: dArcL=ArcL-ArcL0 At A1T, A2T and HCT respectively, A1dArcL, A2dArcL and HCdArcL are calculated according to the above formulas, and the maximum arc length change is the maximum value of dArcL; 3.19 Calculate the first applanation deviation area A1DLAr, the second applanation deviation area A2DLAr and the maximum concave deviation area HCDLAr: Calculate the corneal deviation area according to the deviation amplitude DLA of each point on the upper border of the cornea in each frame image, and the formula is as follows: At A1T, A2T and HCT respectively, calculate A1DLAr, A2DLAr and HCDLAr according to the above formula; 3.20 Calculate the corneal curvature radius HCT Radius and the maximum inverse concave radius Max InverseRadius, IRM: Calculate the corneal vertex curvature radius Radius in each frame image, and calculate the HCT Radius at the HCT moment; the formula for the inverse concave radius is as follows: Inverse Radius = 1 / Radius The maximum inverse concave radius is the maximum value of Inverse Radius; 3.21 The above 35 parameter calculation methods are all performed at the pixel size level and need to be converted into actual sizes based on the physical size or pixel density of the image resolution provided by the imaging technology or equipment.

Images