A method, apparatus, and medium for corneal image segmentation

Abstract

According to an example embodiment of this disclosure, a method, apparatus, and medium for corneal image segmentation are provided. The method includes: acquiring a first OCT image including the cornea; obtaining a first corneal mask image based on the first OCT image via an image segmentation neural network model; determining the category of non-missing points in each column based on the thickness of non-missing points in each column, the average thickness of non-missing points, and the adjacency relationship between non-missing points and missing points, the categories including normal points, missing points, and abnormal points; fitting normal points on the upper and lower surfaces of the cornea in the first corneal mask image to obtain a fitting curve for the upper corneal surface and a fitting curve for the lower corneal surface; and reconstructing the upper and lower corneal surfaces based on the categories of non-missing points on the upper and lower corneal surfaces, respectively, to obtain a corneal segmentation result. Thus, accurate segmentation of corneal images is possible even when the cornea is incomplete due to foreign bodies or congenital corneal damage.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure generally relate to the field of image processing, and in particular, to a method for corneal image segmentation, an electronic device and a computer readable medium. BACKGROUND

[0002] Optical coherence tomography (OCT) was proposed by Huang et al. of MIT in 1991, and has been greatly developed and applied in the past few decades. This technology uses the principle of Michelson interferometer, changes one of the reflecting mirrors to a sample, and obtains the depth information of the sample by coherent imaging of the reflected light of the sample arm and the reflected light of the reference arm. OCT technology has the characteristics of fast imaging, deep imaging depth, non-invasive, non-contact, low price, etc., and has been widely developed and applied in the field of medical detection. Compared with other imaging methods in terms of resolution and imaging depth, OCT imaging technology fills the gap between confocal microscopy and ultrasonic technology.

[0003] According to the specific imaging principle and system design, the OCT system can be divided into time domain OCT (TD-OCT) and frequency domain OCT (FD-OCT). According to the difference of the light source and receiver used, the frequency domain OCT can be divided into spectral domain OCT (SD-OCT) and swept source OCT (SS-OCT). At present, the frequency domain OCT has basically replaced the time domain OCT as the main equipment for ophthalmic detection because it can directly obtain the depth information of the sample, save the movement of the reference arm, and greatly improve the imaging speed.

[0004] As the outermost layer of the anterior segment of the eye, the cornea plays an important role in eye vision, and the light gathering ability of the cornea makes objects visible to the human eye. Any morphological change of the cornea will cause ophthalmic problems. Therefore, quantitative analysis of the cornea and morphological characteristics of the cornea play an important role in various ophthalmic diseases. In actual clinical practice, OCT is used to visualize corneal tissue. Automatic and accurate segmentation of corneal tissue in OCT images is a quite critical technology.

[0005] There are many classic solutions for corneal segmentation in OCT images. For example, using graph theory and dynamic programming, using semi-classical methods and morphological operations, using threshold processing and fitting methods, etc. These methods have considerable limitations, first, they are quite sensitive to image quality, second, many methods are affected by parameters. In recent years, with the development of deep learning, UNET network based on convolutional network has been widely used in medical image segmentation. In addition to ordinary U-Net segmentation, there are Attention U-Net, Asymmetric U-Net, etc.

[0006] After learning and training with the U-Net network, for most OCT images, the method can quickly and effectively perform pixel-level segmentation. However, due to the existence of other interference factors, the picture segmented by the U-Net network may not be a complete cornea, and other non-corneal regions may be introduced. These additional defects are difficult to be perfectly solved by improving the network.

[0007] First, the imaging principle of OCT belongs to coherent imaging, and there is a large amount of speckle noise, which can cover small features and make the tissue boundary blurred, which will affect the segmentation effect of U-Net to some extent. Second, foreign matters such as eye lashes can also cause the imaging of the target object to be incomplete. Finally, additional cross lines may be generated in the scanning process. SUMMARY

[0008] Embodiments of the present disclosure provide a method, an electronic device and a computer readable medium for corneal image segmentation, so that the corneal image can be accurately segmented in the case that the cornea is incomplete due to foreign matters and the cornea is congenitally damaged.

[0009] In a first aspect of the present disclosure, a method for corneal image segmentation is provided. The method comprises: obtaining a first OCT image comprising a cornea; obtaining, based on the first OCT image, a first corneal mask image via an image segmentation neural network model, the image segmentation neural network model being trained by taking a set of OCT images comprising a cornea as input and a set of corresponding corneal mask images as labels, the first corneal mask image comprising missing points indicating non-cornea and non-missing points indicating cornea; determining the non-missing point thickness of each column and the average non-missing point thickness in the first corneal mask image; determining the category of the non-missing points in each column based on the non-missing point thickness of each column, the average non-missing point thickness and the adjacent relationship between the non-missing points and the missing points, the category including normal points, missing points and abnormal points; fitting the normal points on the upper surface and the lower surface of the cornea in the first corneal mask image respectively to obtain a corneal upper surface fitting curve and a corneal lower surface fitting curve; and reconstructing the upper surface and the lower surface of the cornea based on the category of the non-missing points on the upper surface and the lower surface of the cornea respectively, using the corneal upper surface fitting curve and the corneal lower surface fitting curve, to obtain a corneal segmentation result.

[0010] In a second aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the first aspect of the present disclosure.

[0011] In a third aspect of the present disclosure, a non-transitory computer readable storage medium storing computer instructions is provided, and the computer instructions are configured to cause a computer to execute the method according to the first aspect of the present disclosure.

[0012] The summary is provided to introduce a selection of concepts, in a simplified form, that are further described below in the detailed description. The summary is not intended to identify key or essential features of the disclosure, nor is it intended to limit the scope of the disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0013] The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which like reference characters refer to like parts throughout the figures, and in which:

[0014] Figure 1 A schematic diagram of an example environment 100 according to an embodiment of the present disclosure is shown;

[0015] Figure 2 A schematic diagram of a method 200 for corneal image segmentation according to an embodiment of the present disclosure is shown;

[0016] Figure 3 A schematic diagram of a first OCT image 300 according to an embodiment of the present disclosure is shown;

[0017] Figure 4 A schematic diagram of a first corneal mask image 400 according to an embodiment of the present disclosure is shown;

[0018] Figure 5 A schematic diagram of a second OCT image 500 according to an embodiment of the present disclosure is shown;

[0019] Figure 6 A schematic diagram of a normal point, an abnormal point, a missing point and an interference point according to an embodiment of the present disclosure is shown;

[0020] Figure 7 A schematic diagram of a corneal segmentation result 700 according to an embodiment of the present disclosure is shown;

[0021] Figure 8 A schematic diagram of a comparison 800 of a pure deep learning reconstructed corneal thickness topography (left) and a corrected reconstructed corneal thickness topography (right) according to an embodiment of the present disclosure is shown;

[0022] Figure 9 A block diagram of an electronic device 900 suitable for implementing an embodiment of the present disclosure is shown schematically.

[0023] In the various drawings, like or corresponding reference numbers refer to like or corresponding parts throughout the figures. DETAILED DESCRIPTION

[0024] Preferred embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure is more thoroughly and completely conveyed to those skilled in the art, and the scope of the present disclosure is fully conveyed to those skilled in the art.

[0025] As described above, due to the presence of other interference factors, the picture segmented by the U-Net network may not be a complete cornea, and other non-corneal regions may be introduced. These additional defects are difficult to be perfectly solved by improving the network.

[0026] Firstly, the imaging principle of OCT belongs to coherent imaging, and there is a large amount of speckle noise, which can cover small features and make the tissue boundary blurred, which will affect the segmentation effect of U-Net to some extent. Secondly, foreign matters such as eye lashes can also cause the imaging of the target object to be incomplete. Finally, additional cross lines can also be generated during the scanning process.

[0027] In order to solve the above problems or other problems not mentioned, the present disclosure provides a scheme for corneal image segmentation. In the scheme, a computing device can be used to obtain a first OCT image including a cornea. Subsequently, the computing device can obtain a first corneal mask image via an image segmentation neural network model based on the first OCT image, the image segmentation neural network model being trained by taking a set of OCT images including a cornea as input and a set of corresponding corneal mask images as labels, the first corneal mask image including missing points indicating non-cornea and non-missing points indicating cornea. Then, the computing device can determine the non-missing point thickness of each column and the average non-missing point thickness in the first corneal mask image; and determine the category of the non-missing points in each column based on the non-missing point thickness of each column, the average non-missing point thickness and the adjacent relationship between the non-missing points and the missing points, the category including normal points, missing points and abnormal points. The computing device can also respectively fit the normal points on the upper surface and the lower surface of the cornea in the first corneal mask image to obtain a corneal upper surface fitting curve and a corneal lower surface fitting curve. The computing device can then reconstruct the upper surface and the lower surface of the cornea based on the category of the non-missing points on the upper surface and the lower surface of the cornea, respectively, using the corneal upper surface fitting curve and the corneal lower surface fitting curve, to obtain a corneal segmentation result.

[0028] Thus, the corneal image can be accurately segmented in the case that the cornea is incomplete due to foreign matters or the cornea is congenitally damaged.

[0029] The preferred embodiments of the present disclosure will be described in more detail with reference to the drawings.

[0030] Figure 1 A schematic diagram illustrating an example environment 100 is shown. As Figure 1 shown, the example environment 100 includes a computing device 110, a first OCT image 120 including a cornea, an image segmentation neural network model 130, and a cornea segmentation result 140.

[0031] The computing device 110 can include, but is not limited to, an ophthalmic medical device, a personal computer, a personal digital assistant, a wearable device, a tablet computer, a smartphone, etc. In some embodiments, the computing device 110 can have or be coupled to an image acquisition apparatus for acquiring the first OCT image 120.

[0032] The computing device 110 can be configured to obtain the first OCT image 120 including the cornea. Subsequently, the computing device 110 can obtain, based on the first OCT image 120, a first cornea mask image via the image segmentation neural network model 130, the image segmentation neural network model 130 being trained by an OCT image set including the cornea as input and a corresponding cornea mask image set as label, the first cornea mask image including missing points indicating non-cornea and non-missing points indicating cornea. Then, the computing device 110 can determine non-missing point thicknesses of columns in the first cornea mask image and a non-missing point thickness mean value; and determine, based on the non-missing point thicknesses of the columns, the non-missing point thickness mean value, and a neighboring relationship between the non-missing points and the missing points, a class of the non-missing points in the columns, the class including normal points, missing points, and abnormal points. The computing device 110 can further fit the normal points on a superior corneal surface and an inferior corneal surface of the first cornea mask image respectively to obtain a superior corneal surface fitting curve and an inferior corneal surface fitting curve. The computing device 110 can then reconstruct the superior corneal surface and the inferior corneal surface based on the class of the non-missing points of the superior corneal surface and the inferior corneal surface respectively, using the superior corneal surface fitting curve and the inferior corneal surface fitting curve, to obtain the cornea segmentation result 140.

[0033] Thus, by setting the labels of different types of points and combining the superior and inferior corneal surface fitting curves to reconstruct the superior and inferior corneal surfaces, the cornea image can be accurately segmented in the case that the cornea is not complete due to foreign matter or the cornea is congenitally damaged.

[0034] Figure 2 A schematic diagram illustrating an example of a method 200 for cornea image segmentation according to embodiments of the present disclosure is shown. In Figure 2 various embodiments, the acts performed by a computing device, such as the computing device 110, are illustrated. It should be understood that the method 200 can include additional acts not shown and / or can omit some of the acts shown, and the scope of the disclosure is not limited in this regard. Figure 1

[0035] ​At frame 202, a first OCT image including the cornea is acquired. The first OCT image can be as follows: Figure 3 As shown.

[0036] At box 204, a first corneal mask image is obtained based on the first OCT image via an image segmentation neural network model. The image segmentation neural network model is trained using a set of OCT images including the cornea as input and a corresponding set of corneal mask images as labels. The first corneal mask image includes indicators of missing points outside the cornea and indicators of non-missing points inside the cornea.

[0037] like Figure 4 As shown, the first corneal mask image 400 includes a corneal region 410 and a non-corneal region 420. The corneal region 410 includes non-missing points indicating the cornea, which are white points with a pixel value of 255. The non-corneal region 420 includes missing points indicating non-corneal areas, which are black points with a pixel value of 0. The corneal mask images in the training corneal mask image set are... Figure 4 The first corneal mask image 401 is similar and will not be described again.

[0038] As mentioned above, OCT images contain a significant amount of speckle noise, which can mask smaller features and blur tissue boundaries, thus affecting corneal segmentation results to some extent. To address this issue, median filtering and point-by-point denoising can be applied to the first OCT image to reduce the impact of speckle.

[0039] As mentioned above, additional crosshairs may be generated during the scanning process, which can adversely affect corneal morphology segmentation. To address this issue, in some embodiments, the horizontal and vertical bright lines in the first OCT image can be removed first to obtain the second OCT image.

[0040] For example, the row mean of pixel values ​​for each row in the first OCT image and a first mean of the row mean of pixel values ​​for each row can be determined. For example, the row mean of pixel values ​​for each row can be represented as meanRowN, and the first mean can be represented as rowMeanAll. Subsequently, for rows in the first OCT image where the row mean of pixel values ​​is greater than the first mean, the variance of the pixel values ​​for that row can be determined. Then, the pixels in rows where the variance of pixel values ​​is less than a predetermined reference value can be set to zero. Rows where the variance of pixel values ​​is less than the predetermined reference value are bright rows, that is, horizontal bright lines. This can eliminate horizontal bright lines.

[0041] For example, the column mean of pixel values ​​in each column and the second mean of the column mean of pixel values ​​in each column can be determined in the first OCT image. For example, the row mean of pixel values ​​in each column can be represented as meanColN, and the first mean can be represented as rowColAll. Subsequently, for rows in the first OCT image where the column mean of pixel values ​​is less than a predetermined multiple of the second mean, the pixels in that row are set to zero. The rows where the column mean of pixel values ​​is less than a predetermined multiple of the second mean are bright columns, i.e., vertical bright lines. The predetermined multiple can be, for example, 2 or 3, and can be preset or obtained from statistics of a large number of OCT images. This can eliminate vertical bright lines.

[0042] like Figure 5 As shown, through the above processing, the second OCT image has removed horizontal and vertical bright lines.

[0043] Subsequently, based on the second OCT image, the first corneal mask image can be obtained through an image segmentation neural network model. In this case, the image segmentation neural network model can be trained using an OCT image set with horizontal and vertical bright lines removed as input and the corresponding corneal mask image set as labels.

[0044] This eliminates the influence of extra crosshairs generated during scanning on corneal morphology segmentation, thus improving the accuracy of corneal morphology segmentation.

[0045] Back Figure 2 At box 206, determine the thickness of non-missing points and the mean thickness of non-missing points in each column of the first corneal mask image.

[0046] The non-missing point thickness of each column can be obtained by counting the number of consecutive non-missing points in each column. Then, the non-missing point thicknesses of each column can be added together and divided by the number of columns that include non-missing points to obtain the mean non-missing point thickness.

[0047] like Figure 6 As shown, in addition to the corneal region, the obtained first corneal mask image may also contain some interfering points. In some embodiments, before step 206, multiple standard corneal mask images that meet preset conditions can be acquired, and the average number of points included in the corneal region of the multiple standard corneal mask images can be determined. The standard corneal mask images may, for example, meet the following preset conditions: no missing areas, no corneal deformities or damage, and no additional areas caused by speckles or foreign bodies.

[0048] Next, regions with a number of dots less than a first number can be removed from the first corneal mask image. The first number of dots is the product of the average number of dots and a predetermined first ratio. The predetermined first ratio is, for example, any ratio between 5% and 10%.

[0049] Subsequently, it is determined whether the number of points included in the remaining region in the first corneal mask image is greater than a second number of points, the second number of points being a product of the mean number of points and a predetermined second ratio, the predetermined second ratio being greater than the predetermined first ratio. The predetermined second ratio is, for example, 50%, 45%, 55%, etc.

[0050] If it is determined that the number of points included in the remaining region in the first corneal mask image is greater than the second number of points, the non-missing point thicknesses of the columns in the first corneal mask image and the mean non-missing point thickness are determined. Otherwise, an error can be prompted.

[0051] Thus, the interfering points possibly existing in the first corneal mask image can be removed, facilitating the subsequent fitting of the upper and lower surfaces of the cornea.

[0052] At block 208, based on the non-missing point thicknesses of the columns, the mean non-missing point thickness, and the adjacent relationship between the non-missing points and the missing points, the category of the non-missing points in the columns is determined, the category including normal points, missing points, and abnormal points.

[0053] Reference Figure 6 If the non-missing point thickness of the column in which the non-missing point in the first corneal mask image is located is within a predetermined thickness range centered on the mean non-missing point thickness, the non-missing point is determined to be a normal point. The predetermined thickness range centered on the mean non-missing point thickness includes, for example, but is not limited to, a range in which the mean non-missing point thickness fluctuates by 5%, 6%, 7%, etc.

[0054] If the non-missing point thickness of the column in which the non-missing point in the first corneal mask image is located is outside a predetermined thickness range centered on the mean non-missing point thickness, it is determined whether there is a missing point in the adjacent region of the non-missing point. The adjacent region includes, for example, but is not limited to, a region of adjacent 1-2 pixel points.

[0055] If it is determined that there is a missing point in the adjacent region of the non-missing point, the non-missing point is determined to be a missing point.

[0056] If it is determined that there is no missing point in the adjacent region of the non-missing point, the non-missing point is determined to be an abnormal point.

[0057] At block 210, the normal points of the upper surface of the cornea and the lower surface of the cornea in the first corneal mask image are fitted respectively to obtain a fitting curve of the upper surface of the cornea and a fitting curve of the lower surface of the cornea.

[0058] In some embodiments, the normal points of the upper surface of the cornea in the first corneal mask image can be fitted by cubic to obtain a cubic fitting curve.

[0059] Then, it is determined whether the cubic term coefficient of the cubic fitting curve is less than 0. If the cubic term coefficient of the cubic fitting curve is less than 0, the cubic fitting curve is determined to be the fitting curve of the upper surface of the cornea;

[0060] If the cubic term coefficient of the cubic fitting curve is greater than or equal to 0, a quadratic fitting is performed on the normal points of the upper corneal surface in the first corneal mask image to obtain a quadratic fitting curve.

[0061] Then, it is determined whether the quadratic term coefficient of the quadratic fitting curve is less than 0. If the quadratic term coefficient of the quadratic fitting curve is less than 0, the quadratic fitting curve is determined as the upper corneal surface fitting curve.

[0062] The fitting of the lower corneal surface can also be processed in a similar manner, which will not be described in detail.

[0063] It should be understood that the above fitting method is only a distance description, and the upper and lower corneal surfaces can also be directly fitted with a quadratic curve, or other types of fitting, and the scope of the present disclosure is not limited herein.

[0064] At block 212, based on the categories of the non-missing points of the upper and lower corneal surfaces, respectively, the upper and lower corneal surfaces are reconstructed using the upper and lower corneal surface fitting curves to obtain a corneal segmentation result.

[0065] For the abnormal points of the upper corneal surface, if it is determined that there are normal points in the predetermined adjacent region of the abnormal points, the points corresponding to the column of the abnormal points in the upper corneal surface fitting curve are determined as the upper corneal surface points of the column of the abnormal points with a predetermined first probability.

[0066] The predetermined adjacent region includes, but is not limited to, for example, a region of 8, 9, 10 or the like pixels adjacent to the left and right and above and below. The first predetermined probability includes, but is not limited to, for example, 80%, 75%, 70% or the like. For example, 80% of the points use the fitting curve as the final boundary, and 20% of the points use the original data as the final boundary.

[0067] If it is determined that there are missing points in the predetermined adjacent region of the abnormal points, the points corresponding to the column of the abnormal points in the upper corneal surface fitting curve are determined as the upper corneal surface points of the column of the abnormal points with a predetermined second probability, and the predetermined second probability is greater than the predetermined first probability.

[0068] The second predetermined probability includes, but is not limited to, for example, 90%, 85%, 80% or the like. For example, 90% of the points use the fitting curve as the final boundary, and 10% of the points use the original data as the final boundary.

[0069] For the missing points of the upper corneal surface, the points corresponding to the column of the missing points in the upper corneal surface fitting curve are determined as the upper corneal surface points of the column of the missing points.

[0070] For normal points, the fitting curve result can be directly used, or sampling can be performed, 90% of the points use the fitting curve as the final boundary, and 10% of the points use the original data as the final boundary.

[0071] The sub-corneal surface is also similarly processed and will not be repeated here.

[0072] The obtained corneal segmentation result can be as shown in Figure 7 Compared with the first corneal mask image obtained by inference of the neural network model, the corneal segmentation shape is more complete.

[0073] By setting the markers of different types of points, combining the fitting curves of the upper and lower surfaces of the cornea, and reconstructing the upper and lower surfaces of the cornea, the corneal image can be accurately segmented in the case of incomplete cornea due to foreign matter and congenital damage of the cornea.

[0074] In some embodiments, the corneal thickness can also be determined based on the corneal segmentation result.

[0075] If it is determined that the corneal thickness is greater than the standard machine result, the upper surface or the lower surface of the cornea in the corneal segmentation result is reduced by a predetermined layer of pixels to obtain a thickness-corrected corneal segmentation result.

[0076] If it is determined that the corneal thickness is less than the standard machine result, the upper surface or the lower surface of the cornea in the corneal segmentation result is increased by a predetermined layer of pixels to obtain a thickness-corrected corneal segmentation result.

[0077] Thus, the overall difference of micron level caused by different interpolation methods can be solved by data calibration with the standard machine result.

[0078] In addition, in some embodiments, the thickness-corrected corneal segmentation result can also be subjected to non-uniform difference to obtain a corneal thickness topography.

[0079] As shown in Figure 8 , the right side is the corneal thickness topography reconstructed after correction by the above-mentioned scheme of the present disclosure, and the left side is the corneal thickness topography reconstructed by pure deep learning, for example, based on Figure 4 the first corneal mask image. It can be seen that the corneal thickness topography reconstructed after correction by the above-mentioned scheme of the present disclosure is smoother and more complete.

[0080] Thus, the thickness-corrected corneal thickness topography can be obtained, which is smoother and more complete.

[0081] Figure 9 A block diagram of an electronic device 900 suitable for implementing embodiments of the present disclosure is schematically shown. The device 900 can be used to implement Figure 1The computing device 100 can be implemented with the apparatus 900 as shown in FIG. 9. As shown, the apparatus 900 includes a central processing unit (CPU) 901, which can perform various suitable actions and processes in accordance with computer program instructions stored in a read-only memory (ROM) 902 or computer program instructions loaded into a random access memory (RAM) 903 from a storage unit 908. Various programs and data required by the apparatus 900 for operation are also stored in the RAM 903. The CPU 901, the ROM 902, and the RAM 903 are connected to each other by a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0082] Various components in the apparatus 900 are connected to the I / O interface 905, including an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a magnetic disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the apparatus 900 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0083] The processing unit 901 performs various methods and processes described above, such as performing the method 200. For example, in some embodiments, the method 200 can be implemented as a computer software program stored in a machine-readable medium, such as the storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on the apparatus 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the CPU 901, one or more operations of the method 200 described above can be performed. Alternatively, in other embodiments, the CPU 901 can be configured to perform one or more actions of the method 200 by any other suitable means, such as by means of firmware.

[0084] The present disclosure can be a method, apparatus, system, and / or computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for performing various aspects of the present disclosure.

[0085] Computer readable storage media can be tangible storage media which can retain and store instructions for use by an instruction execution device. Computer readable storage media can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer readable storage media include the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0086] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network can comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0087] Computer readable program instructions for carrying out operations of the present disclosure can be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate array (FPGA), or programmable logic array (PLA) can execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present disclosure.

[0088] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0089] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0090] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0091] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0092] Embodiments of the present disclosure have been described above, and the description is intended to be illustrative, and not restrictive, of the disclosed embodiments. Many modifications and variations of the described embodiments are possible, and all such modifications and variations are intended to be within the scope of the described embodiments. The description used herein is intended to be illustrative, and not restrictive, of the described embodiments. The scope of the described embodiments is not limited to the examples and / or embodiments described herein but only by the claims and their equivalents.

Claims

1. A method for cornea image segmentation, comprising: obtaining a first OCT image comprising a cornea; obtaining a first cornea mask image via an image segmentation neural network model based on the first OCT image, the image segmentation neural network model being trained by an OCT image set comprising a cornea as input and a corresponding cornea mask image set as label, the first cornea mask image comprising missing points indicating non-cornea and non-missing points indicating cornea; determining non-missing point thicknesses of columns in the first cornea mask image and a non-missing point thickness mean; determining a class of non-missing points in the columns based on the non-missing point thicknesses of the columns, the non-missing point thickness mean, and a neighboring relationship between non-missing points and missing points, the class comprising normal points, missing points, and abnormal points; fitting normal points on a cornea upper surface and a cornea lower surface of the first cornea mask image respectively to obtain a cornea upper surface fitting curve and a cornea lower surface fitting curve; and reconstructing the cornea upper surface and the cornea lower surface based on the class of non-missing points on the cornea upper surface and the cornea lower surface respectively, and the cornea upper surface fitting curve and the cornea lower surface fitting curve to obtain a cornea segmentation result; wherein the determining the class of non-missing points in the columns based on the non-missing point thicknesses of the columns, the non-missing point thickness mean, and the neighboring relationship between non-missing points and missing points comprises: if a non-missing point thickness of a column in which a non-missing point in the first cornea mask image is located is within a predetermined thickness range centered on the non-missing point thickness mean, determining the non-missing point as a normal point; if the non-missing point thickness of the column in which the non-missing point in the first cornea mask image is located is outside the predetermined thickness range centered on the non-missing point thickness mean, determining whether a neighboring area of the non-missing point has a missing point; if it is determined that the neighboring area of the non-missing point has a missing point, determining the non-missing point as a missing point; and if it is determined that the neighboring area of the non-missing point has no missing point, determining the non-missing point as an abnormal point. 2.The method of claim 1, wherein the obtaining a first cornea mask image via an image segmentation neural network model based on the first OCT image comprises: removing horizontal bright lines and vertical bright lines in the first OCT image to obtain a second OCT image; and obtaining the first cornea mask image via the image segmentation neural network model based on the second OCT image, the image segmentation neural network model being trained by the OCT image set with the horizontal bright lines and the vertical bright lines removed as input and the cornea mask image set as label. 3.The method of claim 2, wherein the removing horizontal bright lines in the first OCT image comprises: determining a pixel value row mean of each row in the first OCT image and a first mean of the pixel value row means; for a row in the first OCT image in which the pixel value row mean is greater than the first mean, determining a pixel value variance of the row; and ​ ​ ​ Zeroing the pixels in the row whose pixel value variance is less than a predetermined reference value. 4.The method of claim 2, wherein the removing vertical bright lines in the first OCT image comprises: determining a column mean of pixel values of each column in the first OCT image and a second mean of the column means of pixel values of each column; and zeroing the pixels in the row whose column mean of pixel values is less than a predetermined multiple of the second mean. 5.The method of claim 1, wherein the determining the non-missing point thicknesses of each column in the first cornea mask image and a mean of the non-missing point thicknesses comprises: obtaining a plurality of standard cornea mask images satisfying a predetermined condition; determining a mean of the number of points included in the cornea regions in the plurality of standard cornea mask images; deleting the regions in the first cornea mask image whose included number of points is less than a first number of points, the first number of points being a product of the mean of the number of points and a predetermined first ratio; and if it is determined that the number of points included in the remaining regions in the first cornea mask image is greater than a second number of points, determining the non-missing point thicknesses of each column in the first cornea mask image and a mean of the non-missing point thicknesses, the second number of points being a product of the mean of the number of points and a predetermined second ratio, the predetermined second ratio being greater than the predetermined first ratio. 6.The method of claim 1, wherein the fitting the normal points on the upper surface of the cornea in the first cornea mask image to obtain a cornea upper surface fitting curve comprises: performing a cubic fitting on the normal points on the upper surface of the cornea in the first cornea mask image to obtain a cubic fitting curve; if a cubic term coefficient of the cubic fitting curve is less than 0, determining the cubic fitting curve as the cornea upper surface fitting curve; if the cubic term coefficient of the cubic fitting curve is greater than or equal to 0, performing a quadratic fitting on the normal points on the upper surface of the cornea in the first cornea mask image to obtain a quadratic fitting curve; and if a quadratic term coefficient of the quadratic fitting curve is less than 0, determining the quadratic fitting curve as the cornea upper surface fitting curve. 7.The method of claim 1, wherein the reconstructing the upper surface of the cornea based on the category of the non-missing points on the upper surface of the cornea and using the cornea upper surface fitting curve comprises: for the abnormal points on the upper surface of the cornea, performing the following steps: if it is determined that there are normal points in a predetermined adjacent region of the abnormal point, determining, with a predetermined first probability, a point corresponding to the column of the abnormal point in the cornea upper surface fitting curve as a point on the upper surface of the cornea in the column of the abnormal point; and if it is determined that there are missing points in the predetermined adjacent region of the abnormal point, determining, with a predetermined second probability, a point corresponding to the column of the abnormal point in the cornea upper surface fitting curve as a point on the upper surface of the cornea in the column of the abnormal point, the predetermined second probability being greater than the predetermined first probability. 8.The method of claim 1, wherein the reconstructing the upper surface of the cornea based on the category of the non-missing points on the upper surface of the cornea and using the cornea upper surface fitting curve comprises: For the missing point of the corneal upper surface, the point corresponding to the column where the missing point is located in the fitting curve of the corneal upper surface is determined as the corneal upper surface point of the column where the missing point is located.

9. The method of claim 1, further comprising: determining a corneal thickness based on the corneal segmentation result; if it is determined that the corneal thickness is greater than a standard phantom result, reducing a predetermined layer of pixels of the corneal upper surface or the corneal lower surface in the corneal segmentation result to obtain a thickness-corrected corneal segmentation result; and if it is determined that the corneal thickness is less than the standard phantom result, increasing a predetermined layer of pixels of the corneal upper surface or the corneal lower surface in the corneal segmentation result to obtain a thickness-corrected corneal segmentation result.

10. The method of claim 9, further comprising: performing non-uniform difference on the thickness-corrected corneal segmentation result to obtain a corneal thickness topography.

11. An electronic device, comprising: at least one processor; and a memory connected with the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-10. The computer instructions are used to enable the computer to perform the method of any one of claims 1-10. ​ 12. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, ​

Images