Method and device for determining corneal localized cross-linking irradiation position based on deep learning

Abstract

The embodiment of the present application provides a kind of based on deep learning's corneal local crosslinking irradiation position determination method and device, the method comprises: obtaining the corneal topographic image of target object;The corneal topographic image is carried out block segmentation, obtain multiple image block regions;Respectively, image block region is input to the preset image classification model, determines the corneal protruding region and non-protruding region of each image block region;Based on preset constraint condition, the image segmentation of corneal protruding region and non-protruding region is carried out, determines the irradiation position of target object's corneal local crosslinking.The embodiment of the present application carries out the division of corneal protruding region and non-protruding region to the corneal topographic image after block segmentation by image classification model, and by increasing preset constraint condition, the image segmentation of corneal protruding region and non-protruding region is realized, to more accurately determine the irradiation position of target object's corneal local crosslinking, guarantee the realization of corneal local crosslinking.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a method and apparatus for determining the local cross-linking irradiation location of the cornea based on deep learning. Background Technology

[0002] Myopia is a condition caused by an excessively long axial length or relatively strong refractive power of the eye, resulting in blurred vision and decreased visual acuity when parallel light rays enter the eye and focus in front of the retina at rest. Current myopia correction techniques include optical correction, laser corneal refractive surgery, and implantable collamer lenses (ICL). Early-stage myopia (below 300 degrees) is often corrected using collagen cross-linking technology. Ultraviolet riboflavin collagen cross-linking (CXL) is a novel treatment method that uses riboflavin as a photosensitizer and ultraviolet light to mediate cross-linking of collagen fibers, improving mechanical rigidity and biomechanical stability, thereby preventing corneal ectasia and other diseases.

[0003] To achieve ultraviolet light riboflavin-collagen crosslinking, the light source involved in related technical solutions uniformly irradiates the entire cornea to achieve crosslinking. This method makes it difficult to accurately determine the corresponding corneal irradiation location, and thus cannot achieve localized corneal crosslinking, making it impossible to adjust specific surgical parameters for each patient. Summary of the Invention

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the defect in the prior art that it is impossible to accurately determine the local cross-linking irradiation position of the cornea, thereby providing a method and apparatus for determining the local cross-linking irradiation position of the cornea based on deep learning.

[0005] According to a first aspect, embodiments of the present invention provide a method for determining the irradiation location of local corneal crosslinking based on deep learning, comprising the following steps: acquiring a corneal topographic image of a target object; segmenting the corneal topographic image into multiple image block regions; inputting the image block regions into a preset image classification model to determine the corneal convex and non-convex regions of each image block region; the preset image classification model is generated by training based on images labeled with corneal convex and non-convex regions; and performing image segmentation on the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of local corneal crosslinking of the target object.

[0006] Optionally, the image block regions are input into a preset image classification model to determine the corneal convex and non-convex regions of each image block region, including: performing feature extraction and pooling processing on the input image block regions; and inputting the results of feature extraction and pooling processing into an image classifier to determine the corneal convex and non-convex regions.

[0007] Optionally, the results of feature extraction and pooling are input into an image classifier to determine the corneal convex and non-convex regions, including: identifying convex and non-convex regions in an image block region based on preset image markers contained in the results of feature extraction and pooling, and obtaining the coordinate information of the convex and non-convex regions; converting the coordinate information of the convex and non-convex regions into high-dimensional coordinate information respectively; and classifying the corneal convex and non-convex regions of the image block region according to the high-dimensional coordinate information to determine the corneal convex and non-convex regions of the image block region.

[0008] Optionally, image segmentation is performed on the corneal convex and non-convex regions based on preset constraints to determine the irradiation position of the corneal local crosslinking of the target object, including: performing image segmentation based on preset region constraints and the correspondence between the corneal convex and non-convex regions to obtain a first image segmentation result; obtaining a second image segmentation result based on the correspondence between the first image segmentation result and the preset position constraints, and using the second image segmentation result as the irradiation position of the corneal local crosslinking of the target object.

[0009] Optionally, image segmentation is performed based on region constraints and the correspondence between corneal convex and non-convex regions to obtain a first image segmentation result, including: determining the morphological parameters corresponding to corneal convex and non-convex regions according to the region constraints; determining the location information of the corneal convex region according to the determined morphological parameters and the correspondence between corneal convex and non-convex regions, and performing image segmentation based on the location information, and using the image segmentation result as the first image segmentation result.

[0010] Optionally, a second image segmentation result is obtained based on the correspondence between the first image segmentation result and the preset positional constraints, including: obtaining the positional information of the corneal image center point and the positional information of the center points of each corneal convex region; determining the position of the convex region corresponding to the corneal center based on the preset positional constraints, the correspondence between the corneal image center point position information and the positional information of the center points of each corneal convex region, and performing image segmentation based on the position, and using the image segmentation result as the second image segmentation result.

[0011] Optionally, embodiments of the present invention provide a method for determining the irradiation location of local corneal crosslinking based on deep learning, which may include: acquiring information of an adjustable light source; determining adjustable light source parameters based on the correspondence between the adjustable light source information and the irradiation location of local corneal crosslinking of the target object, and using the adjustable light source parameters as corneal local crosslinking parameters.

[0012] According to a second aspect, embodiments of the present invention provide a deep learning-based device for determining the irradiation location of local corneal crosslinking, comprising: a corneal topography image acquisition unit configured to acquire a corneal topography image of a target object; an image segmentation unit configured to segment the corneal topography image into blocks to obtain multiple image block regions; a corneal region determination unit configured to input the image block regions into a preset image classification model to determine the corneal convex and non-convex regions of each image block region; the preset image classification model is generated based on images labeled with corneal convex and non-convex regions; and an irradiation location determination unit configured to perform image segmentation of the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of local corneal crosslinking of the target object.

[0013] According to a third aspect, embodiments of the present invention provide a non-transitory computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the method for determining the corneal local crosslinking irradiation location based on deep learning as described in any embodiment of the first aspect.

[0014] According to a fourth aspect, embodiments of the present invention provide a computer device including at least one processor; and a memory communicatively connected to 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 perform the method for determining the local corneal crosslinking irradiation location based on deep learning as described in any embodiment of the first aspect.

[0015] The technical solution of this invention has the following advantages:

[0016] This invention provides a method and apparatus for determining the irradiation location of local corneal crosslinking based on deep learning. The method includes: acquiring a corneal topographic image of a target object; segmenting the corneal topographic image into multiple image block regions; inputting each image block region into a preset image classification model to determine the corneal convex and non-convex regions of each image block region; the preset image classification model is generated by training based on images labeled with corneal convex and non-convex regions; and performing image segmentation of the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of local corneal crosslinking of the target object. This embodiment of the invention uses an image classification model to divide the segmented corneal topographic image into corneal convex and non-convex regions, and by adding preset constraints, achieves image segmentation of corneal convex and non-convex regions, thereby more accurately determining the irradiation location of local corneal crosslinking of the target object and ensuring the realization of local corneal crosslinking. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a specific example of a method for determining the local cross-linking irradiation location of the cornea based on deep learning, as described in an embodiment of the present invention.

[0019] Figure 2 This is a specific example analysis diagram of the method for determining the local cross-linking irradiation location of the cornea based on deep learning in an embodiment of the present invention;

[0020] Figure 3 This is an analysis diagram of another specific example of the method for determining the local cross-linking irradiation location of the cornea based on deep learning in this embodiment of the invention;

[0021] Figure 4 This is a structural example diagram of the device for determining the local cross-linking irradiation location of the cornea based on deep learning in an embodiment of the present invention;

[0022] Figure 5 This is a structural example diagram of a computer device in an embodiment of the present invention. Detailed Implementation

[0023] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0025] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can also refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0026] In the description of this invention, a hyperplane is a linear subspace in n-dimensional Euclidean space with a co-dimensional dimension of one, that is, a linear subspace of (n-1) dimensions.

[0027] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0028] This embodiment provides a method for determining the local crosslinking irradiation location of the cornea based on deep learning, such as... Figure 1 As shown, Figure 1 This is a flowchart illustrating a specific example of a method for determining the local corneal crosslinking irradiation location based on deep learning, according to an embodiment of the present invention, including:

[0029] S11: Obtain the corneal topography image of the target object.

[0030] Specifically, corneal topography images of the target object can be obtained through an eye-view analyzer or other means. The configuration of the eye-view analyzer and the acquisition of corneal topography images are relatively mature existing technologies, and will not be elaborated further.

[0031] S12: Segment the corneal topography image into multiple image block regions.

[0032] Specifically, the corneal topography image is segmented into multiple image block regions, including: obtaining the center point position of the corneal topography image; and using the center point position of the corneal topography image as the reference position for image block segmentation. Each resulting image block region contains the region corresponding to the center point of the corneal topography image. The center point position of the corneal topography image refers to the position of the corneal pupil in the corneal topography image, and the region corresponding to the center point of the corneal topography image contained in each image block region refers to the corneal pupil position region.

[0033] S13: Input the image patch regions into the preset image classification model to determine the corneal protrusion and non-protrusion regions of each image patch region. The preset image classification model is generated by training based on images labeled with corneal protrusion regions and images with non-protrusion regions.

[0034] Specifically, due to the characteristics of corneal images, the distribution of convex and non-convex regions in corneal topographic images is non-linear. This embodiment of the invention uses an image classification model to classify corneal convex and non-convex regions. The preset image classification model refers to a neural network model, such as the net-16 model, a third-order neural network structure, or other models. Optionally, since the input to the preset image classification model is a segmented image patch region, and each image patch region is smaller than the image input to the preset image classification model, the preset image classification model can use a third-order neural network structure to classify corneal convex and non-convex regions in each image patch region with fewer neural network layers.

[0035] S14: Based on preset constraints, perform image segmentation on the corneal convex and non-convex regions to determine the irradiation location of the corneal local crosslinking of the target object.

[0036] Specifically, image segmentation of the corneal convex and non-convex regions based on preset constraints includes: determining the corneal convex region based on preset region constraints; and determining the irradiation position of the target object's corneal local crosslinking within the determined corneal convex region based on preset position constraints. Here, the preset region constraints refer to the corneal convex region and the corresponding corneal parameters; the preset position constraints refer to the position parameters corresponding to the corneal pupil position region.

[0037] This invention provides a method for determining the irradiation location of local corneal crosslinking based on deep learning, comprising: acquiring a corneal topographic image of a target object; segmenting the corneal topographic image into multiple image block regions; inputting each image block region into a preset image classification model to determine the corneal convex and non-convex regions of each image block region; the preset image classification model is generated by training based on images labeled with corneal convex and non-convex regions; and performing image segmentation of the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of local corneal crosslinking of the target object. This embodiment of the invention uses an image classification model to divide the segmented corneal topographic image into corneal convex and non-convex regions, and by adding preset constraints, achieves image segmentation of corneal convex and non-convex regions, thereby more accurately determining the irradiation location of local corneal crosslinking of the target object and ensuring the realization of local corneal crosslinking.

[0038] In an optional embodiment of the present invention, step S13 above, which involves inputting image block regions into a preset image classification model to determine the corneal convex and non-convex regions of each image block region, includes:

[0039] (1) Perform feature extraction and pooling on the input image block region.

[0040] Specifically, the preset image classification model adopts a three-order neural network structure, with alternating convolutional layers and pooling layers forming a module. The module is cascaded multiple times to achieve feature extraction and pooling processing of the input image block region.

[0041] For example, each image patch region is smaller than the image input to the preset image classification model. The preset image classification model can use a three-order neural network structure to classify the corneal convex and non-convex regions of each image patch region with fewer neural network layers.

[0042] (2) Input the results of feature extraction and pooling into the image classifier to determine the corneal protrusion and non-protrusion areas.

[0043] Specifically, the image classifier is generated through training on images labeled with and without corneal protrusions. Images labeled with corneal protrusions are used as positive sample images, and images without labels (i.e., images without corneal protrusions) are used as negative sample images. These positive and negative sample images are input into a pre-defined image classification model to obtain the model's output. Based on the model's output and the actual output corresponding to the sample images, the parameters of the image classifier are adjusted to obtain the actual image classifier used for image classification.

[0044] In an optional embodiment of the present invention, the results of feature extraction and pooling are input into an image classifier to determine corneal convex and non-convex regions, including:

[0045] (1) Based on the preset image markers contained in the result of feature extraction and pooling, identify the raised and non-raised regions in the image block region and obtain the coordinate information of the raised and non-raised regions.

[0046] Specifically, the pre-defined image labels included in the feature extraction and pooling results refer to the image labels for raised and non-raised regions within the image block region. Each identified image block region has corresponding coordinate information for both raised and non-raised regions.

[0047] For example, such as Figure 2 As shown, Figure 2 This is a specific example analysis diagram of the method for determining the local cross-linking irradiation position of the cornea based on deep learning, according to an embodiment of the present invention. The diagram exemplarily represents the identification results of the raised and non-raised regions within an image patch region based on the coordinate information corresponding to the raised and non-raised regions. Specifically, circular frame 21 represents the non-raised corneal region corresponding to the image patch region, and rectangular frame 22 represents the raised corneal region corresponding to the image patch region. It should be understood that, based on the preset image markers included in the results of feature extraction and pooling processing, identifying the raised and non-raised regions within the image patch region and obtaining the coordinate information of the raised and non-raised regions includes, but is not limited to, […]. Figure 2 The structure shown can be used to identify raised and non-raised regions in an image block region based on preset image markers contained in the result of feature extraction and pooling processing, and to form a two-dimensional image based on the coordinate information of the raised and non-raised regions.

[0048] (2) Convert the coordinate information of the raised area and the non-raised area into high-dimensional coordinate information respectively.

[0049] Specifically, converting coordinate information into high-dimensional coordinate information is a relatively mature existing technology, which will not be elaborated further.

[0050] For example, such as Figure 3 As shown, Figure 3 This is a specific example analysis diagram of the method for determining the corneal local cross-linking irradiation position based on deep learning according to an embodiment of the present invention. It exemplarily shows a schematic diagram formed based on the high-dimensional coordinate information of the convex and non-convex regions after converting the coordinate information of the convex and non-convex regions into high-dimensional coordinate information. In this diagram, the circular frame 31 represents the non-convex corneal region corresponding to the image block region, and the rectangular frame 33 represents the convex corneal region corresponding to the image block region.

[0051] (3) Based on the high-dimensional coordinate information, the corneal protrusion area and non-protrusion area of ​​the image block region are classified to determine the corneal protrusion area and non-protrusion area of ​​the image block region.

[0052] Specifically, based on high-dimensional coordinate information, the corneal convex and non-convex regions of the image patch are classified, including: determining the convex and non-convex regions corresponding to the shortest spatial distance between them, based on the coordinate information of each convex and non-convex region. A hyperplane is inserted based on the determined convex and non-convex regions, and the classification of corneal convex and non-convex regions is achieved based on the inserted hyperplane. The spatial distance from the determined convex region to the hyperplane is equal to the spatial distance from the non-convex region to the hyperplane.

[0053] Specifically, calculating spatial distance based on coordinate latitude information is a relatively mature existing technology, and will not be elaborated further.

[0054] For example, such as Figure 3 As shown, Figure 3 This is a specific example analysis diagram of the method for determining the irradiation position of local corneal crosslinking based on deep learning in an embodiment of the present invention. Circular frames represent non-convex corneal regions corresponding to the image patch region, and rectangular frames represent convex corneal regions corresponding to the image patch region. The spatial distance between circular frame 33 and rectangular frame 31 is the shortest spatial distance between convex and non-convex regions. Circular frame 33 represents the non-convex region corresponding to the shortest spatial distance between convex and non-convex regions, and rectangular frame 31 represents the convex region corresponding to the shortest spatial distance between convex and non-convex regions. Plane 32 represents a hyperplane, and the spatial distance between circular frame 33 and plane 32 is equal to the spatial distance between rectangular frame 31 and the plane. It should be understood that the convex and non-convex corneal regions of the classified image patch region each have coordinate information before conversion to higher dimensions. This coordinate information before conversion to higher dimensions provides a data basis for image segmentation of the convex and non-convex corneal regions and for determining the irradiation position of the local corneal crosslinking of the target object.

[0055] In an optional embodiment of the present invention, based on a preset image classification model, the coordinate information of the raised and non-raised regions in the image block area is converted into high-dimensional coordinate information, and the spatial distance is calculated based on the converted high-dimensional coordinate information. The inserted hyperplane is used as the division between the raised and non-raised regions, thereby realizing the classification of non-linearly distributed raised and non-raised regions in the image block area and improving the classification accuracy of the image classification model for raised and non-raised regions.

[0056] In an optional embodiment of the present invention, step S14 above, which involves image segmentation of the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of the corneal local crosslinking of the target object, includes:

[0057] (1) Based on the preset region constraints and the correspondence between the corneal protrusion region and the non-protrusion region, image segmentation is performed to obtain the first image segmentation result;

[0058] Specifically, the corneal convex and non-convex regions are determined by a pre-defined image classification model, with each region possessing corresponding coordinate information. The first image segmentation result is the segmentation of the corneal convex and non-convex regions based on the coordinate information. The pre-defined region constraints refer to using the morphological parameters corresponding to the convex and non-convex regions as region constraints.

[0059] (2) Based on the correspondence between the first image segmentation result and the preset position constraint, the second image segmentation result is obtained, and the second image segmentation result is used as the irradiation position of the corneal local crosslinking of the target object.

[0060] Specifically, when segmenting the corneal topography image into blocks, each image block region includes the corneal pupil location region. The second image segmentation result is based on the first image segmentation result, separating the image of the convex region at the pupil location from the convex region at the non-pupil location.

[0061] In an optional embodiment of the present invention, the above steps, including image segmentation based on preset region constraints and the correspondence between corneal protrusion and non-protrusion regions to obtain a first image segmentation result, include:

[0062] (1) Determine the morphological parameters corresponding to corneal protrusions and non-protrusions according to the preset area constraints;

[0063] Specifically, the preset region constraint condition refers to using the convex and non-convex regions corresponding to the morphological parameters as region constraint conditions. The morphological parameters include one or more of the following: corneal surface variation index, vertical asymmetry index, keratoconus index, central keratoconus index, minimum radius of curvature, high asymmetry index, and high eccentricity index.

[0064] (2) Based on the determined morphological parameters and the correspondence between the corneal protrusion area and the non-protrusion area, determine the location information of the corneal protrusion area, and perform image segmentation based on the location information, and use the image segmentation result as the first image segmentation result.

[0065] Specifically, the correspondence between the determined morphological parameters and the corneal convex and non-convex regions refers to comparing the morphological parameters that match the corneal convex regions with the morphological parameters of the classified corneal convex and non-convex regions. The region corresponding to the morphological parameters that match the corneal convex regions is taken as the corneal convex region, and the coordinate information corresponding to the corneal convex region is used as the location information of the corneal convex region.

[0066] In an optional embodiment of the present invention, morphological parameters corresponding to corneal protrusions are determined by preset region constraints, and the location of the corneal protrusion region is determined based on the determined morphological parameters. Then, image segmentation is performed based on the determined location information of the corneal protrusion region, with the corneal protrusion region used as the first image segmentation result. In this process, by adding preset region constraints, the accurate location of the corneal protrusion region is determined, effectively separating non-protruding corneal regions, improving the accuracy of the local cross-linked irradiation area of ​​the cornea, and providing a data foundation for further image segmentation based on the corneal protrusion region.

[0067] In an optional embodiment of the present invention, the step of obtaining the second image segmentation result based on the correspondence between the first image segmentation result and the preset positional constraint conditions includes:

[0068] (1) Obtain the location information of the center point of the corneal image and the location information of the center point of each corneal protrusion area.

[0069] Specifically, the center point of the corneal topography image refers to the center point of the corneal pupil location region within the image patch area. The center point location information of each corneal bulge region refers to the location information of the corneal bulge region, using the coordinate information corresponding to the corneal bulge region as the location information of the corneal bulge region.

[0070] (2) Based on the preset position constraints, the correspondence between the corneal image center point position information and the center point position information of each corneal protrusion region, determine the position of the protrusion region corresponding to the corneal center, and perform image segmentation based on the position, and use the image segmentation result as the second image segmentation result.

[0071] Specifically, the preset positional constraint refers to using the distance from the center point of the corneal image to each corneal protrusion region as a constraint. Based on the preset positional constraint and the correspondence between the corneal image center point position information and the center point position information of each corneal protrusion region, the position of the protrusion region corresponding to the corneal center is determined, including: calculating the distance between the corneal image center point and the center point position of each corneal protrusion region; when the distance is less than the preset distance, the coordinate information of the corneal protrusion region corresponding to the distance is used as the position information of the corneal protrusion region.

[0072] Specifically, calculating the distance between the center point of the corneal image and the center points of each corneal convex region refers to calculating the distance based on the coordinate information of the center point of the corneal pupil position region in the image block area and the coordinate information of the corresponding corneal convex region. Calculating distance based on coordinate information is a relatively mature existing technology and will not be elaborated further.

[0073] Specifically, the preset distance can be selected according to the actual working conditions. This application does not impose specific limitations on this, as long as it can be used to determine the protruding area of ​​the pupil position based on the size of the distance.

[0074] In an optional embodiment of the present invention, the distance from the corneal image center point corresponding to the corneal pupil area to each corneal protrusion region is determined by a preset positional constraint condition. If any determined distance is less than a preset distance, the corneal protrusion region corresponding to that distance is designated as a corneal protrusion region within the pupil area. Image segmentation is then performed based on the position of the corneal protrusion regions within the pupil area. In this process, by adding a preset positional constraint condition, protrusion regions located within the pupil area are determined, and these corneal protrusion regions within the pupil area are used as the irradiation positions for local corneal crosslinking of the target object. This effectively eliminates corneal protrusion regions outside the pupil area, thereby accurately locating the corneal local crosslinking irradiation position and ensuring that ultraviolet riboflavin-collagen crosslinking is performed on the corneal protrusion regions located at the lesion location, i.e., the pupil area.

[0075] In an optional embodiment of the present invention, the above-described method for determining the local corneal crosslinking irradiation location based on deep learning may include:

[0076] (1) Obtain information about the adjustable light source.

[0077] Specifically, tunable light source information includes one or more of the following: ultraviolet radiant flux power, ultraviolet illumination shape, and ultraviolet current. Obtaining tunable light source information refers to acquiring historical information about the tunable light source.

[0078] (2) Based on the correspondence between the information of the adjustable light source and the irradiation position of the corneal local crosslinking of the target object, the adjustable light source parameters are determined, and the adjustable light source parameters are used as the corneal local crosslinking parameters.

[0079] Specifically, the parameters of the photocrosslinking surgery include information about the tunable light source, which includes one or more of the following: ultraviolet (UV) radiant flux power, UV irradiation shape, and UV current. The process of determining the tunable light source parameters, based on the correspondence between the tunable light source information and the irradiation location of the corneal local crosslinking in the target subject, involves determining the tunable light source parameters by establishing the historical information of the irradiation location of the corneal local crosslinking in the target subject, and then using these determined tunable light source parameters as the corneal local crosslinking parameters.

[0080] Specifically, based on the correspondence between the information of the adjustable light source and the irradiation position of the local cross-linking of the cornea of ​​the target object, the adjustable light source parameters can be determined as shown in Table 1 below. Table 1 exemplarily shows the information of the adjustable light source, the irradiation position of the local cross-linking of the cornea of ​​the target object, and the correspondence between the adjustable light source parameters.

[0081] Table 1

[0082]

[0083] For example, the correspondence between the irradiation positions of each corneal local cross-link and each adjustable light source can be used to reflect the required treatment intensity for the target object. When the irradiation positions are the same, the stronger the parameters corresponding to the adjustable light source at the irradiation position, the higher the required treatment intensity for the target object.

[0084] In an optional embodiment of the present invention, corneal crosslinking parameters are determined by establishing the correspondence between the irradiation position of the local corneal crosslinking of the target object and the adjustable light source. This allows for adjustments not only to the irradiation position of the local corneal crosslinking during the ultraviolet riboflavin-collagen crosslinking process but also to the adjustable light source, satisfying the specific surgical parameter adjustments for different target objects and further improving the applicability and convenience of local corneal crosslinking.

[0085] This embodiment provides a device for determining the local cross-linking irradiation location of the cornea based on deep learning, such as... Figure 4 As shown, Figure 4 This is a structural example diagram of a device for determining the local cross-linking irradiation location of the cornea based on deep learning, provided in an embodiment of the present invention. It includes: a corneal topography image acquisition unit 41, an image segmentation unit 42, a corneal region determination unit 43, and an irradiation location determination unit 44.

[0086] The corneal topography image acquisition unit 41 is configured to acquire the corneal topography image of the target object; for details, please refer to the relevant description of step S11 in any of the above method embodiments, which will not be repeated here.

[0087] The image segmentation unit 42 is configured to segment the corneal topography image into blocks to obtain multiple image block regions; for details, please refer to the relevant description of step S12 in any of the above method embodiments, which will not be repeated here.

[0088] The corneal region determination unit 43 is configured to input image block regions into a preset image classification model to determine the corneal convex and non-convex regions of each image block region; the preset image classification model is generated by training based on images labeled with corneal convex and non-convex regions; for details, please refer to the relevant description of step S13 in any of the above method embodiments, which will not be repeated here.

[0089] The irradiation position determination unit 44 is configured to perform image segmentation of the corneal convex and non-convex regions based on preset constraints to determine the irradiation position of the corneal local crosslinking of the target object. For details, please refer to the relevant description of step S14 in any of the above method embodiments, which will not be repeated here.

[0090] This invention provides a deep learning-based device for determining the irradiation location of local corneal crosslinking, comprising: a corneal topography image acquisition unit configured to acquire a corneal topography image of a target object; an image segmentation unit configured to segment the corneal topography image into multiple image block regions; a corneal region determination unit configured to input the image block regions into a preset image classification model to determine the corneal convex and non-convex regions of each image block region; the preset image classification model is generated based on images labeled with corneal convex and non-convex regions; and an irradiation location determination unit configured to perform image segmentation of the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of local corneal crosslinking of the target object. This invention uses an image classification model to divide the segmented corneal topography image into corneal convex and non-convex regions. By adding preset constraints, image segmentation of the corneal convex and non-convex regions is achieved, thereby determining the irradiation location of local corneal crosslinking of the target object and ensuring the realization of local corneal crosslinking.

[0091] An embodiment of the present invention also provides a non-transitory computer storage medium storing computer-executable instructions that can execute the methods described in any of the above-described method embodiments. The storage medium may be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0092] One embodiment of the present invention also provides a computer device, such as... Figure 5 As shown, Figure 5This is a schematic diagram of a computer device according to an optional embodiment of the present invention. The computer device may include at least one processor 51, at least one communication interface 52, at least one communication bus 53, and at least one memory 54. The communication interface 52 may include a display screen and a keyboard; optionally, the communication interface 52 may also include a standard wired interface or a wireless interface. The memory 54 may be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. Optionally, the memory 54 may also be at least one storage device located remotely from the aforementioned processor 51. The processor 51 may be combined with... Figure 4 The described apparatus has an application program stored in memory 54, and the processor 51 calls the program code stored in memory 54 to perform the steps of the method described in any of the above method embodiments.

[0093] The communication bus 53 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus 53 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0094] The memory 54 may include volatile memory, such as random-access memory (RAM); the memory may also include non-volatile memory, such as flash memory, hard disk drive (HDD) or solid-state drive (SSD); the memory 54 may also include a combination of the above types of memory.

[0095] The processor 51 can be a central processing unit (CPU), a network processor (NP), or a combination of CPU and NP.

[0096] The processor 51 may further include a hardware chip. This hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0097] Optionally, the memory 54 is also used to store program instructions. The processor 51 can invoke the program instructions to implement the method described in any embodiment of the present invention.

[0098] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for determining the local cross-linking irradiation location of the cornea based on deep learning, characterized in that, Includes the following steps: Obtain corneal topography images of the target object; The corneal topography image is segmented into blocks to obtain multiple image block regions; The image block regions are respectively input into a preset image classification model to determine the corneal convex and non-convex regions of each image block region; The preset image classification model is generated by training on images with and without corneal protrusions. Based on preset constraints, the corneal convex and non-convex regions are segmented to determine the irradiation location of the corneal local cross-linking of the target object. The step of segmenting the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of the local cross-linking of the cornea of ​​the target object includes: Image segmentation is performed based on preset region constraints and the correspondence between corneal convex and non-convex regions to obtain a first image segmentation result. The preset region constraints refer to corneal parameters corresponding to corneal convex regions. Based on the correspondence between the first image segmentation result and the preset position constraint, a second image segmentation result is obtained, and the second image segmentation result is used as the irradiation position for local cross-linking of the cornea of ​​the target object. The preset position constraint refers to the position parameters corresponding to the corneal pupil position region.

2. The method for determining the local corneal crosslinking irradiation location based on deep learning according to claim 1, characterized in that, The step of inputting the image patch regions into a preset image classification model to determine the corneal convex and non-convex regions of each image patch region includes: Feature extraction and pooling are performed on the input image block region; The results of feature extraction and pooling are input into an image classifier to determine the corneal protrusion and non-protrusion regions.

3. The method for determining the corneal local crosslinking irradiation location based on deep learning according to claim 2, characterized in that, The step of inputting the results of feature extraction and pooling into an image classifier to determine the corneal convex and non-convex regions includes: Based on the preset image markers contained in the results of feature extraction and pooling, the raised and non-raised regions in the image block region are identified, and the coordinate information of the raised and non-raised regions is obtained. The coordinate information of the protruding region and the non-protruding region are respectively converted into high-dimensional coordinate information; Based on the high-dimensional coordinate information, the corneal protrusion and non-protrusion regions of the image block region are classified to determine the corneal protrusion and non-protrusion regions of the image block region.

4. The method for determining the local corneal crosslinking irradiation location based on deep learning according to claim 1, characterized in that, The image segmentation based on preset region constraints and the correspondence between corneal convex and non-convex regions to obtain a first image segmentation result includes: Based on the preset region constraints, determine the morphological parameters corresponding to corneal protrusions and non-protrusions respectively; Based on the determined morphological parameters and the correspondence between the corneal convex and non-convex regions, the location information of the corneal convex region is determined, and image segmentation is performed based on the location information. The image segmentation result is used as the first image segmentation result.

5. The method for determining the corneal local crosslinking irradiation location based on deep learning according to claim 4, characterized in that, The step of obtaining the second image segmentation result based on the correspondence between the first image segmentation result and the preset positional constraints includes: Obtain the location information of the corneal image center point and the location information of the center points of each corneal convex region; Based on the preset position constraints and the correspondence between the corneal image center point position information and the center point position information of each corneal protrusion region, the position of the protrusion region corresponding to the corneal center is determined, and image segmentation is performed based on the position. The image segmentation result is used as the second image segmentation result.

6. The method for determining the local corneal crosslinking irradiation location based on deep learning according to claim 1, characterized in that, After performing image segmentation on the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of the corneal local crosslinking of the target object, the method further includes: Obtain information about the tunable light source; Based on the correspondence between the information of the adjustable light source and the irradiation position of the corneal local crosslinking of the target object, the adjustable light source parameters are determined, and the adjustable light source parameters are used as the corneal local crosslinking parameters.

7. A device for determining the local cross-linking irradiation location of the cornea based on deep learning, characterized in that, include: A corneal topography image acquisition unit is configured to acquire a corneal topography image of a target object; The image segmentation unit is configured to segment the corneal topography image into multiple image block regions. The corneal region determination unit is configured to input the image block regions into a preset image classification model to determine the corneal convex and non-convex regions of each image block region; The preset image classification model is generated by training on images with and without corneal protrusions. The irradiation position determination unit is configured to perform image segmentation on the corneal convex region and non-convex region based on preset constraints, and determine the irradiation position of the corneal local crosslinking of the target object; The step of segmenting the corneal convex and non-convex regions based on preset constraints to determine the irradiation location of the local cross-linking of the cornea of ​​the target object includes: Image segmentation is performed based on preset region constraints and the correspondence between corneal convex and non-convex regions to obtain a first image segmentation result. The preset region constraints refer to corneal parameters corresponding to corneal convex regions. Based on the correspondence between the first image segmentation result and the preset position constraint, a second image segmentation result is obtained, and the second image segmentation result is used as the irradiation position for local cross-linking of the cornea of ​​the target object. The preset position constraint refers to the position parameters corresponding to the corneal pupil position region.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the method for determining the corneal local crosslinking irradiation location based on deep learning as described in any one of claims 1-6.

9. A computer device, characterized in that, include: At least one processor; And a memory communicatively connected to 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 perform the method for determining the local corneal crosslinking irradiation location based on deep learning as described in any one of claims 1-6.

Images