Apparatus and method for imaging structures in a transparent medium

Abstract

To image a structure within a transparent medium such as an eye and in particular to quantify and evaluate a structure within gelatinous tissue of an eye for monitoring, evaluating, and / or diagnosing a medical condition.SOLUTION: A medical diagnostic apparatus includes: a receiver circuit that receives three-dimensional data of an eye; and processing circuitry configured to segment the three-dimensional data into regions that include a target structural element and regions that do not include the target structural element to produce a segmented three-dimensional data set. The segmenting is performed using a plurality of segmentation algorithms. Each of the plurality of segmentation algorithms is trained separately on different two-dimensional data extracted from the three-dimensional data. The processing circuitry is further configured to generate at least one metric from the segmented three-dimensional data set, and evaluate a medical condition based on the at least one metric.SELECTED DRAWING: None

Description

Cross - reference to related applications

[0001] This application is a non - provisional application claiming priority to U.S. Provisional Application No. 63 / 168,927, filed on March 31, 2021, the entire contents of which are incorporated herein by reference.

Technical Field

[0002] The present disclosure generally relates to imaging structures within a transparent medium such as an eye, and more particularly to quantifying and evaluating structures within the gelatinous tissue of the eye in order to monitor, evaluate, and / or diagnose medical conditions.

Background Art

[0003] The eye contains various structures. However, with conventional methods, it may be difficult to view those structures in vivo.

[0004]

Summary of the Invention

Means for Solving the Problems

[0005] According to one embodiment of the present invention, a medical diagnostic apparatus includes a receiving circuit that receives three - dimensional data of an eye, and a processing circuit configured to segment the three - dimensional data into a region including target structural elements and a region not including target structural elements using a plurality of segmentation algorithms that are individually trained for different two - dimensional data extracted from the three - dimensional data, thereby generating a segmented three - dimensional data set. The processing circuit is further configured to generate at least one evaluation criterion from the segmented three - dimensional data set, and the processing circuit is further configured to evaluate a medical condition based on the at least one evaluation criterion.

[0006] In the medical diagnostic device, each segmentation algorithm may correspond to a training plane, and each segmentation algorithm may be trained using data corresponding to a plurality of 2D image slices parallel to the corresponding training plane.

[0007] In the medical diagnostic device, the processing circuit includes an axial segmentation algorithm corresponding to the axial plane of the eye, a coronal segmentation algorithm corresponding to the coronal plane of the eye, and a sagittal segmentation algorithm corresponding to the sagittal plane of the eye. Even if the segmentation is performed using the above method good.

[0008] In the medical diagnostic device, the segmentation algorithm may generate weights to be used in a neural network to generate a dataset segmented by plane from all 2D slices parallel to the corresponding training plane in the 3D data, and the datasets segmented by plane obtained from each segmentation algorithm may be combined by averaging or voting the results at each location in the eye to generate the segmented 3D dataset.

[0009] In the medical diagnostic device, the segmentation algorithm may generate procedural parameters used in the processing circuit to generate a dataset segmented for each plane from all 2D slices parallel to the corresponding training plane in the 3D data, and the datasets segmented for each plane obtained from each segmentation algorithm may be combined by averaging or voting the results at each location in the eye to generate the segmented 3D dataset.

[0010] In the medical diagnostic device, the two-dimensional data may be used to train each of the plurality of segmentation algorithms, and the two-dimensional data used to train each of the plurality of segmentation algorithms may further include two-dimensional slices corresponding to positions adjacent to slices in all subsets of two-dimensional slices, which are acquired parallel to the corresponding plane and for which segmentation results are assigned to each coordinate.

[0011] In the medical diagnostic device, the processing circuit may further be configured to perform the segmentation using the plurality of segmentation algorithms, each of which is individually trained on different two-dimensional data extracted from the three-dimensional data and additional three-dimensional data corresponding to one or more eyes other than the eye.

[0012] In the medical diagnostic device, the processing circuit may be configured to generate a skeleton corresponding to the segmented three-dimensional dataset and to generate at least one evaluation criterion based on the characteristics of the skeleton.

[0013] In the aforementioned medical diagnostic device, the medical condition may include at least one of the following: congenital vitreoretinal anomalies, vitreoretinal degenerative diseases, diabetic retinopathy, myopia, pathological myopia, age-related macular degeneration, endophthalmitis and malignant tumors, glaucoma, and myopic neuropathy.

[0014] In the aforementioned medical diagnostic device, the target structural element may include at least one of a vitreous pocket, a cribriform lamina, and an optic nerve head.

[0015] A medical diagnostic method according to one embodiment of the present invention may include receiving three-dimensional data of the eye, segmenting the three-dimensional data into regions containing target structural elements and regions not containing target structural elements using a plurality of segmentation algorithms, each individually trained on different two-dimensional data extracted from the three-dimensional data, generating a segmented three-dimensional dataset, generating at least one evaluation criterion from the segmented three-dimensional dataset, and evaluating the medical condition based on the at least one evaluation criterion.

[0016] In the aforementioned medical diagnostic method, each segmentation algorithm may correspond to a training plane, and each segmentation algorithm is trained using data corresponding to multiple 2D image slices parallel to the corresponding training plane.

[0017] In the medical diagnostic method described above, the segmentation may be performed using an axial segmentation algorithm corresponding to the axial plane of the eye, a coronal segmentation algorithm corresponding to the coronal plane of the eye, and a sagittal segmentation algorithm corresponding to the sagittal plane of the eye.

[0018] In the medical diagnostic method described above, the segmentation algorithm may generate weights used in a neural network to generate a dataset segmented by plane from all 2D slices parallel to the corresponding training plane in the 3D data, and the datasets segmented by plane obtained from each segmentation algorithm are combined by averaging or voting the results at each location in the eye to generate the segmented 3D dataset.

[0019] In the medical diagnostic method, the segmentation algorithm may generate procedural parameters used to generate a plane-segmented dataset from all 2D slices parallel to the corresponding training plane in the 3D data, and the plane-segmented datasets obtained from each segmentation algorithm are combined by averaging or voting the results at each location in the eye to generate the segmented 3D dataset.

[0020] In the medical diagnostic method, the two-dimensional data may be used to train each of the plurality of segmentation algorithms, and includes a subset of two-dimensional slices obtained from the three-dimensional data parallel to the corresponding plane, with segmentation results assigned to each coordinate, and the two-dimensional data used to train each of the plurality of segmentation algorithms further includes two-dimensional slices corresponding to positions adjacent to slices in all subsets of two-dimensional slices obtained parallel to the corresponding plane, with segmentation results assigned to each coordinate.

[0021] In the medical diagnostic method described above, the segmentation may be performed using the plurality of segmentation algorithms, each of which is individually trained on different two-dimensional data extracted from the three-dimensional data and additional three-dimensional data corresponding to one or more eyes other than the eye described above.

[0022] The medical diagnostic method may further include generating a skeleton corresponding to the segmented three-dimensional dataset and generating the at least one evaluation criterion based on the characteristics of the skeleton, wherein the medical condition includes at least one of congenital vitreoretinal anomalies, vitreoretinal degenerative diseases, diabetic retinopathy, myopia, pathological myopia, age-related macular degeneration, endophthalmitis and malignant tumors, glaucoma, and myopic neuropathy.

[0023] In the medical diagnosis method, the target structural element may include at least one of a vitreous pocket, a cribriform plate, and an optic disc.

[0024] A non-temporary computer-readable storage medium storing computer-executable instructions for executing a medical diagnosis method according to a consistent embodiment of the present invention, which includes receiving three-dimensional data of an eye when executed by a computer, and using a plurality of segmentation algorithms individually trained for different two-dimensional data each extracted from the three-dimensional data, segmenting the three-dimensional data into a region including a target structural element and a region not including the target structural element to generate a segmented three-dimensional data set, generating at least one evaluation criterion from the segmented three-dimensional data set, and evaluating a medical condition based on the at least one evaluation criterion.

Brief Description of the Drawings

[0025] The scope of the present disclosure is best understood from the following detailed description of exemplary embodiments when read in conjunction with the accompanying drawings. [Figure 1] FIG. 1 is a diagram of a method and apparatus for imaging the structure of the vitreous according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram of a method and apparatus for obtaining segmented data from 3D OCT data according to an embodiment of the present invention. [Figure 3A] FIG. 3A is a diagram of a method and apparatus for performing segmentation for each plane according to an embodiment of the present invention. [Figure 3B] FIG. 3B is an example of a diagram of a noise-removed 3D OCT volume according to an embodiment of the present invention. [Figure 3C] FIG. 3C is an example showing a frame physically adjacent to a frame in which manual segmentation is performed according to an embodiment of the present invention. [Figure 4]Figure 4 is a diagram of a method and apparatus for performing segmented data steps according to an embodiment of the present invention. [Figure 5A] Figure 5A shows an example of a 3D object that generates a skeletonized display. [Figure 5B] Figure 5B is an example of a skeletonized representation generated from the object in Figure 5A according to an embodiment of the present invention. [Figure 6] Figure 6 shows an example of the configuration of an apparatus according to an embodiment of the present invention. [Modes for carrying out the invention]

[0026] The eye contains various structures such as the vitreous humor, lamina cribriformis, and optic disc. Changes in the optic disc can be a sign of progression of glaucoma or myopic neuropathy.

[0027] For example, the vitreous humor is a transparent, gel-like structure that makes up about 80% of the eyeball's volume. Because the vitreous humor is large, transparent, and a moving mass, it has traditionally been difficult to observe its structure, especially in vivo. Observation of the vitreous humor has mainly been done through laboratory examinations. For example, by injecting ink into the vitreous humor, many ink-filled spaces (called cisterns) can be made visible.

[0028] To date, in vivo imaging using horizontal B-scan images from swept-source optical coherence tomography (SS-OCT) has revealed the posterior vitreous cortical anterior pocket (PPVP), which forms a boat-shaped space in front of the posterior pole of the eyeball, as well as other fluid-filled spaces within the vitreous humor (such as Croquet's tubes and other vessels).

[0029] Embodiments of the present invention can provide improved imaging and analysis of these structures. The following detailed discussion will describe how embodiments can be used to image and evaluate the structure of the vitreous humor, but the present invention can also be applied to any other structure of the eye, including, for example, the structure of the optic nerve head.

[0030] OCT B-scan images record a two-dimensional slice of the vitreous humor at a specific time and location (e.g., the fovea). However, because the vitreous humor is a gelatinous tissue and moves in whole or in part due to eye and body movements, it is difficult to visualize such a dynamic structure in still images. Furthermore, the structure of the vitreous humor may change immediately due to movement, or it may change over time. Similarly, other intraocular structural elements may also be difficult to image in vivo. Embodiments of the present invention provide a real-time 3D representation of the vitreous humor and / or other ocular structural elements, enabling better visualization and evaluation of the dynamic vitreous humor and any changes it may undergo.

[0031] Furthermore, the structure of the vitreous humor, particularly the profile of the vitreous pocket, has been suggested to be an indicator of aging in healthy eyes and may also be an early predictor of eye diseases such as diabetic retinopathy and high myopia.

[0032] Figure 1 shows a flowchart of one embodiment of a method for imaging an intraocular structural element of interest (i.e., a target structural element), such as the vitreous humor, according to the present invention. In step 100, 3D OCT data of the subject's eye is acquired, for example, by receiving data from an OCT scanner or by performing an OCT scan using an OCT scanner. In step 102, the 3D OCT data is segmented. In step 104, the segmented data is quantified. In step 106, the quantified data is displayed, or in step 108, the quantified data is evaluated, for example, to monitor the progression of a medical condition or to compare it with a normative database for anomaly detection. The method in Figure 1 can be executed on one or more general-purpose computers programmed to perform the functions described. Figure 1 and the other figures below are called method flowcharts, but each also represents a corresponding structural diagram of the apparatus for performing the corresponding function. As detailed below, each corresponding function can be implemented using dedicated and / or programmable circuits.

[0033] Figure 2 shows a flowchart of a method for obtaining segmented data from 3D OCT data, similar to step 102 in Figure 1. In step 202, the acquired 3D OCT data (i.e., volume data) 200 is denoised to improve the signal-to-noise ratio. This embodiment may include artificial intelligence (AI) OCT denoising techniques or other known denoising techniques. The denoising in step 202 generates a denoised 3D OCT volume 204.

[0034] The denoised 3D OCT volume 204 is segmented separately in one or more different planes. In the example in Figure 2, the denoised 3D OCT volume 204 is segmented separately in three different planes. For example, segmentation is performed in the axial plane 206, coronal plane 208, and sagittal plane 210 of the eye under examination. However, embodiments of the present invention include, for example, performing segmentation in other possible planes for different reference frames, and segmentation in one or two planes, or any other possible number of planes, such as four or more planes. For each plane, individual OCT images along that plane are selected from the denoised 3D OCT volume 204, and structural elements of interest, such as vitreous pockets, are automatically segmented (i.e., labeled, or otherwise identified) in each OCT image. The segmented structural elements, such as vitreous pockets, from each image are then combined to generate a 3D segmentation result for the corresponding plane 211. The segmentation results for each plane 211 indicate the presence or absence of structural elements of interest (e.g., vitreous pockets) at each location within the imaged volume.

[0035] In step 212, the segmentation results for each plane are averaged, or votes are cast for each location, with the location receiving the most votes being the result for that location. For example, each segmentation result has a binary representation of the presence or absence of a structural element of interest (e.g., a vitreous pocket) at a given location (e.g., a value of 1 indicates the presence of the structural element of interest, and a value of 0 indicates the absence of the structural element of interest). In the averaging step, the average value is calculated for each location from the segmented results for each individual plane (e.g., calculated for the axial, coronal, and sagittal planes). The averaged value is then compared to a threshold, and any comparison result exceeding the threshold is set as indicating the presence of the structural element of interest in the averaged 3D segmentation result 214. In the case of averaging three planes as in this example, the threshold is set to 2 / 3. However, if a different number of planes are segmented separately, other thresholds may be used.

[0036] In step 216, 3D smoothing is performed on the averaged 3D segmentation result 214 to generate the final 3D segmentation result 218. Furthermore, the smoothing process reduces noise and rough edges. Smoothing may be performed using a 3D median filter of size (9,9,9). The filter size may be selected based on the physical properties of the object being segmented. For example, a larger filter size, such as (15,15,15), will produce a smoother contour, but may result in a loss of detail, while a smaller filter size, such as (3,3,3), will largely preserve detail, but may result in a noisy image. For example, (9,9,9) is a suitable filter size for imaging relatively large structures of interest (e.g., large vitreous structures), while (3,3,3) is a suitable filter size for imaging smaller tissue structures, such as pores in cribriform lamina. Furthermore, 3D smoothing can be performed by applying other non-median filters, such as the Laplacian and / or Taubin filter.

[0037] Figure 3A shows a detailed flowchart of plane-by-plane segmentation, as performed in steps 206, 208, or 210 of Figure 2. First, in step 304, data related to the current plane is selected or extracted from the denoised 3D OCT volume 204 to train the AI ​​algorithm. For example, in step 304, data corresponding to multiple 2D images, each representing an image plane parallel to the current plane, is selected or extracted from the denoised 3D OCT volume 204. Next, in step 306, a representative image is selected from the multiple 2D images for manual segmentation along the corresponding plane. For example, the selected image may be one image randomly selected from a predetermined number of images in the multiple 2D images. For example, in step 306, one image may be randomly or sequentially selected for every 10 captured images extracted or selected in step 304. Also in step 306, assignment information for manual segmentation is obtained. For example, in step 306, manual instructions are received indicating whether a certain location contains a structural element of interest (e.g., a vitreous pocket). Manual instructions may be generated by a user interface device (e.g., mouse, keyboard, touchscreen, etc.) operated by a person who manually identifies segmented information within a selected frame. The manual instructions and the corresponding location information are combined to form the manual segmentation result 310. The output of the manual segmentation step 306 includes the selected OCT frame 308 (i.e., data corresponding to the 2D image on which manual segmentation was performed) and the manual segmentation result 310 (i.e., information indicating where the structural element of interest is located).

[0038] Figure 3B shows an example of the denoised 3DOCT volume 204 generated by step 304, the selected OCT frame 308 generated by step 306, and the manual segmentation result 310 also generated by step 306.

[0039] In step 312, the AI ​​algorithm is trained using the selected OCT frames 308 and the manual segmentation results 310. The selected OCT frames 308 are used as input data for training, and the manual segmentation results are used as ground truth data. The AI ​​algorithm can be implemented using a neural network or other machine learning algorithms such as a support vector machine. For example, a deep residual neural network may be used to run the AI ​​algorithm. During training, the weights between nodes of the neural network in the AI ​​algorithm, or the parameters of the procedural algorithm (collectively referred to herein as "weights"), are gradually adjusted by a training process such that the AI ​​algorithm is trained to generate output images based on the input image that best matches the corresponding ground truth image. This training procedure 312 continues until the termination criteria are met. For example, in one embodiment, training is completed when all the training data has run the AI ​​algorithm a predetermined number of times (e.g., 100 times). In another embodiment, training is completed when the output of the test image no longer changes by a predetermined factor. The result of the AI ​​training process 312 is the weights 314 of the trained AI algorithm.

[0040] Once training is complete, the weights 314 of the trained AI algorithm are used by the AI ​​segmentation neural network 316 to automatically perform structural element segmentation of interest (e.g., vitreous pocket segmentation) on 2D image slices other than those used to perform training. For example, all images in the denoised 3D OCT volume 204 parallel to the current plane 304 can be processed by the AI ​​segmentation neural network 316 using the weights 314 of the trained AI algorithm to generate plane-by-plane segmentation results 211. These results include structural element segmentation identification (i.e., identification of whether each coordinate in each slice contains a structural element of interest) performed on each image slide (not just manually segmented image slices) of data from the denoised 3D OCT volume 304 related to the current plane.

[0041] Alternatively, instead of using the denoised 3D OCT volume 204 from which the training data was extracted / selected as selected by step 318, a different denoised 3D OCT volume 320 may be segmented by AI segmentation 316 using the weights 314 of the trained AI algorithm. The other denoised 3D OCT volume 320 may be data corresponding to the same eye from which the denoised OCT volume 204 was obtained, but may be obtained on a different (later or earlier) date or time. Alternatively, the other denoised 3D OCT volume 320 may be data corresponding to a different eye of the same person from whom the denoised 3D OCT volume 204 was obtained. Or, the other denoised 3D OCT volume 320 may be data corresponding to an eye of a different person from the person from whom the denoised 3D OCT volume 204 was obtained.

[0042] If an AI algorithm is trained using data from the eyes of a single subject, the algorithm can perform well on other images of the same subject (those not manually segmented). Including training images from many subjects is advantageous for effectively utilizing the trained AI algorithm on segmented images from other subjects. In other words, if a trained AI algorithm is trained using images from only one subject's eye, it may become specialized for that single eye (or, in some cases, both eyes) but may not be useful for other subjects' eyes. Conversely, an AI algorithm can be trained to achieve high generalizability using a large variation in subject eye data (i.e., data from many subjects' eyes), but it may not achieve the same accuracy for a specific subject compared to a specialized algorithm. Therefore, different training approaches can be applied depending on the ultimate goal (e.g., accuracy vs. automation) and the available training data. For example, when there are limitations on the number of manually annotated images available (i.e., images annotated to identify structural elements of interest, such as vitreous pockets), it is advantageous to train an AI algorithm for each eye and each subject to maximize accuracy.

[0043] By training the AI ​​algorithm with one manually segmented frame for every N frames, the weights 314 of the trained AI algorithm can then be used to automatically perform segmentation on all frames that were not manually segmented (i.e., (N-1) frames). The output of the AI ​​segmentation 316 is a series of images from the corresponding plane containing the segmentation results.

[0044] In the above description, the training input images consist of a single 2D image randomly selected in the corresponding plane. However, according to a further embodiment, in step 308, it is possible to improve the accuracy of the AI ​​algorithm by providing to the AI ​​training 312 not only the selected frame 308 in which the manual segmentation result 310 is set, but also additional adjacent frames, which are frames corresponding to regions physically adjacent to the randomly or sequentially selected frame in which the manual segmentation result is set, within the selected frame 308. The inventors have found that providing additional information about adjacent image slices as part of the AI ​​training 312 improves the training process.

[0045] Figure 3C shows an example of (N-1), N, and (N+1) frames that are physically adjacent within the corresponding axial, sagittal, and coronal planes. In this example, the manual segmentation result is available only for the Nth frame. During training, a 3D image (3 channels) created by combining the (N-1), N, and (N+1) frames is used as input, and manual segmentation for the Nth frame is set as the target. By adding information from adjacent frames to the training data, noise can be reduced and the visibility of the vitreous pocket can be improved, potentially resulting in a higher signal-to-noise ratio for the imaged vitreous pocket.

[0046] Figure 4 shows a detailed flowchart of step 104 of the quantification of segmented data in Figure 1. The final segmentation result 218 is quantified to generate various metrics used for monitoring, evaluating, and diagnosing medical conditions. In step 402, the final segmentation result 218 undergoes 3D skeletonization 402 to generate a 3D skeleton 304 of the structural element of interest (e.g., a vitreous pocket), from which a skeleton quantification process 406 can extract skeleton characteristics including the number of branches 408, the average length of branches 410, the number of branches 412, and the average number of branches per branch 414. Further basic quantification processes 416 extract volume 418 and surface area from the final 3D segmentation result 218. 410、 You may also extract evaluation criteria for structural elements of interest, such as a surface area-to-volume ratio of 412.

[0047] Figures 5A and 5B show an example of skeletonization, a method for quantifying the shape of a 3D object. Figure 5A represents an image of a 3D object 502 (i.e., a horse). Figure 5B shows the skeleton diagram of the horse generated by 3D skeletonization 402. The skeleton diagram includes the arrangement of branches 502 connected by branches 504. For example, 3D skeletonization 402 may include the process of thinning a 3D object into a series of points equidistant from the surface of the 3D object. Together with the distance of each point to the object surface, the result of this 3D skeleton can represent the shape of the 3D object, from which additional measurements and evaluation criteria can be extracted. The complexity of a 3D object can be characterized by the number of branches, with simpler 3D objects tending to have fewer branches. Sphere-likeness is characterized by the number of branches per branch, with a perfect sphere having only one branch and no branches.

[0048] Other evaluation criteria can also be extracted from the final 3D segmentation results 218. These include the height of the structural element of interest (e.g., the vitreous pocket), the width of the structural element of interest along the upward-downward direction, the width of the structural element of interest along the nasal-temporal direction, and the spatial relationships between each space (i.e., the presence or absence of connections between each space).

[0049] These evaluation criteria, obtained individually or in combination or subcombinations (also known as profiles of structural elements of interest), represent indicators of the aging process in healthy eyes. For example, the evaluation criteria for vitreous pockets are also early predictors of ophthalmic diseases such as: congenital vitreous retinal anomalies (e.g., persistent primary vitreous hyperplasia (PHPV), persistent vitreous artery, retinopathy of prematurity (ROP)); vitreoretinal degenerative diseases (e.g., familial exudative vitreoretinopathy (FEVR), blue cone (S cone) enhancement syndrome (ESCS)); diabetic retinopathy (from asymptomatic to progressive stages); myopia and pathological myopia; age-related macular degeneration; endophthalmitis; and malignant tumors (e.g., uveitis, intraocular lymphoma). For instance, the formation of vitreous pockets can be present before pathological myopia and can therefore be used as an early sign for detecting pathological myopia. Optic nerve head structure can be used to monitor and manage patients with glaucoma and myopic neuropathy.

[0050] Without requiring a special protocol, the technology used here does not need to be connected directly to or simultaneously with an OCT scanner and can be applied retrospectively to any existing 3D data.

[0051] In this specification, elements or steps described in the singular and preceded by the word "a" or "an" should be understood not to exclude multiple elements or steps unless such exclusion is expressly stated. Furthermore, references to "one embodiment" of the invention are not intended to be construed as excluding the existence of additional embodiments incorporating the described features.

[0052] The control processing methods and / or systems described herein may be implemented using computer programming or engineering techniques, including computer software, firmware, hardware, or any combination or subset thereof, and the technical effects may include at least the processing of three-dimensional data and diagnostic evaluation criteria as disclosed herein.

[0053] Figure 6 shows a block diagram of a computer capable of carrying out various embodiments described herein. The control processing aspects of this disclosure can be embodied as systems, methods, and / or computer program products. A computer program product may include a computer-readable storage medium on which computer-readable program instructions are recorded causing one or more processors to execute aspects of this embodiment.

[0054] A computer-readable storage medium may be a tangible, non-transient device capable of storing instructions used by an instruction execution unit (processor). A computer-readable storage medium may be, 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 thereof. More specific, non-exclusive examples of computer-readable storage media include flexible disks, hard disks, solid-state drives (SSDs), random-access memory (RAM), read-only memory (ROM), eraseable programmable read-only memory (EPROM or Flash), static random-access memory (SRAM), compact disks (CD or CD-ROM), digital versatile disks (DVDs), magneto-optical disks, memory cards, or sticks, each (or a suitable combination thereof). The computer-readable storage media used in this disclosure are not interpreted as transient signals in themselves, but rather include, for example, radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses passing through optical fiber cables), or electrical signals transmitted through wires.

[0055] Computer-readable program instructions that enable the functions described in this disclosure can be downloaded from a computer-readable storage medium to an appropriate computer device or processing unit, or to an external computer or external storage device, via a global network (i.e., the Internet), a local area network, a wide area network, and / or a wireless network. Networks include copper wires, optical fiber, wireless communications, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface of each computer device or processing unit can receive computer-readable program instructions from the network and transfer the computer-readable program instructions for storage on a computer-readable storage medium within the computer device or processing unit.

[0056] Computer-readable program instructions for performing the operations of the Disclosure may include machine language instructions and / or microcode, which may be compiled or interpreted from source code written in any combination of one or more programming languages, including assembly language, Basic, Fortran, Java, Python, R, C, C++, C#, or similar programming languages. Computer-readable program instructions may be executed entirely on a user's personal computer, notebook computer, tablet, or smartphone, entirely on a remote computer or computer server, or on any combination of these computer devices. The remote computer or computer server may be connected to the user's device or multiple devices via a computer network, including a local area network, a wide area network, or a global network (i.e., the Internet). In some embodiments, for example, an electronic circuit including a programmable logic circuit, a field-programmable gate array (FPGA), or a programmable logic array (PLA) may execute the computer-readable program instructions using information from the computer-readable program instructions to configure or customize the electronic circuit in order to perform an aspect of the Disclosure.

[0057] Aspects of this disclosure are described herein with reference to flowcharts and block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. Those skilled in the art will understand that each block in a flowchart or block diagram, and combinations of blocks in a flowchart or block diagram, can be implemented by computer-readable program instructions.

[0058] Computer-readable program instructions capable of implementing the systems and methods described herein may be provided to one or more processors (and / or one or more cores within a processor) of a general-purpose computer, a dedicated computer, or other programmable device to generate a machine such that instructions executed via the processors of the computer or other programmable device create a system for performing the functions specified in the flowcharts and block diagrams of this disclosure. These computer-readable program instructions may also be stored in a computer-readable storage medium that can instruct a computer, a programmable device, and / or other device to function in a particular manner, and the computer-readable storage medium containing the stored instructions constitutes a product containing instructions that implement the embodiments of the functions specified in the flowcharts and block diagrams of this disclosure.

[0059] Furthermore, computer-readable program instructions can be loaded onto a computer, another programmable device, or other device, and the instructions executed on the computer, another programmable device, or other device can perform a series of operational steps to generate a computer implementation process, such that the instructions implement the functions specified in the flowcharts and block diagrams of this disclosure.

[0060] Figure 6 is a functional block diagram showing a networked system 600 comprising one or more networked computers and servers. In one embodiment, the hardware and software environment illustrated in Figure 6 can provide an exemplary platform for implementing the software and / or methods relating to this disclosure. Referring to Figure 6, the networked system 600 may include, but is not limited to, a computer 605, a network 610, a remote computer 615, a web server 620, a cloud storage server 625, and a computer server 630. In some embodiments, multiple instances may be employed for one or more functional blocks illustrated in Figure 6.

[0061] Further details of computer 605 are also shown in Figure 6. The functional blocks illustrated within computer 605 are provided only to establish exemplary functionality and are not intended to be exhaustive. Details of the remote computer 615, web server 620, cloud storage server 625, and computer server 630 are not shown, but these other computers and devices may include similar functionality to that shown for computer 605. Computer 605 may be a personal computer (PC), desktop computer, laptop computer, tablet computer, netbook computer, personal digital assistant (PDA), smartphone, or other programmable electronic device capable of communicating with other devices on network 610.

[0062] The computer 605 may include a processor 635, a bus 637, memory 640, non-volatile storage 645, a network interface 650, a peripheral interface 655, and a display interface 665. Each of these functions may, in some embodiments, be implemented as an individual electronic subsystem (an integrated circuit chip, or a combination of a chip and associated devices), or in other embodiments, several combinations of functions may be implemented on a single chip (sometimes called a system-on-a-chip or SoC).

[0063] Processor 635 may be one or more single-chip or multi-chip microprocessors designed and / or manufactured by companies such as Intel Corporation, Advanced Micro Devices Corporation (AMD), Arm Holdings (Arm), Apple Computer, etc. Examples of microprocessors include Intel's Celeron, Pentium, Core i3, Core i5, and Core i7; AMD's Opteron, Phenom, Athlon, Turion, and Ryzen; and Arm's Cortex-A, Cortex-R, and Cortex-M. Bus 637 may be a proprietary or industry-standard high-speed parallel or serial peripheral interconnect bus such as ISA, PCI, PCI Express (PCI-e), or AGP.

[0064] Memory 640 and non-volatile storage 645 may be computer-readable storage media. Memory 640 may include any suitable volatile storage device, such as dynamic random-access memory (DRAM) and static random-access memory (SRAM). Non-volatile storage 645 may include one or more of the following: flexible disks, hard disks, solid-state drives (SSDs), read-only memory (ROM), eraseable programmable read-only memory (EPROM or Flash), compact disks (CD or CD-ROM), digital versatile disks (DVDs), and memory cards or sticks.

[0065] The program 648 may be a collection of machine-readable instructions and / or data stored in non-volatile storage 645 and used to create, manage, and control specific software functions described in further detail elsewhere in this disclosure and shown in the drawings. In some embodiments, memory 640 may be considerably faster than non-volatile storage 645. In such embodiments, the program 648 may be transferred from non-volatile storage 645 to memory 640 before execution by processor 635.

[0066] Computer 605 may communicate with and interact with other computers via the network interface 650 and the network 610. The network 610 may be, for example, a local area network (LAN), a wide area network (WAN) such as the Internet, or a combination of both, and may include wired, wireless, or fiber optic connections. Generally, the network 610 can be any combination of connections and protocols that support communication between two or more computers and associated devices.

[0067] The peripheral interface 655 may enable data input and output with other devices that may be locally connected to the computer 605. For example, the peripheral interface 655 may provide a connection to an external device 660. The external device 660 may include devices such as a keyboard, mouse, keypad, touchscreen, and / or other suitable input devices. The external device 660 may also include portable computer-readable storage media such as thumb drives, portable optical or magnetic disks, and memory cards. Software and data used to practice embodiments of this disclosure, for example, program 648, may be stored on such portable computer-readable storage media. In such embodiments, the software may be loaded into non-volatile storage 645, or alternatively, directly into memory 640 via the peripheral interface 655. The peripheral interface 655 may use industry-standard connections such as RS-232 or Universal Serial Bus (USB) to connect to the external device 660.

[0068] The display interface 665 can connect the computer 605 to the display 670. In some embodiments, the display 670 may be used to present a command line or a graphical user interface to the user of the computer 605. The display interface 665 may be connected to the display 670 using one or more proprietary or industry standard connections such as VGA, DVI, DisplayPort, or HDMI®.

[0069] As described above, the network interface 650 provides communication with other computing and storage systems or devices outside of computer 605. The software programs and data described herein may be downloaded to non-volatile storage 645 via the network interface 650 and network 610 from, for example, a remote computer 615, a web server 620, a cloud storage server 625, and a computer server 630. Furthermore, the systems and methods described herein may be executed by one or more computers connected to computer 605 via the network interface 650 and network 610. For example, in some embodiments, the systems and methods described herein may be executed by a combination of a remote computer 615, a computer server 630, or interconnected computers on network 610.

[0070] The data, datasets, and / or databases employed in embodiments of the systems and methods described herein may be stored and / or downloaded from the remote computer 615, web server 620, cloud storage server 625, and computer server 630.

[0071] In light of the above teachings, many modifications and variations of the present invention are possible. Therefore, it should be understood that, within the scope of the appended claims, the present invention can be carried out in ways other than those specifically described herein.

Claims

1. A receiving circuit that receives 3D data from the eye, A processing circuit configured to segment the 3D data into regions containing target structural elements and regions not containing target structural elements, using multiple segmentation algorithms, each individually trained on different 2D data extracted from the 3D data, thereby generating a segmented 3D dataset. Includes, Each segmentation algorithm corresponds to the training plane, Each segmentation algorithm is trained using data corresponding to multiple 2D image slices parallel to their respective training planes. The processing circuit performs the segmentation using an axial segmentation algorithm corresponding to the axial plane of the eye, a coronal segmentation algorithm corresponding to the coronal plane of the eye, and a sagittal segmentation algorithm corresponding to the sagittal plane of the eye. The processing circuit is further configured to generate at least one structural element profile of interest from the segmented three-dimensional dataset, in a medical device.

2. The two-dimensional data used to train each of the aforementioned multiple segmentation algorithms is obtained from the three-dimensional data parallel to the corresponding plane and includes a subset of two-dimensional slices to which the segmentation results are assigned to each coordinate. The two-dimensional data used to train each of the aforementioned multiple segmentation algorithms further includes two-dimensional slices corresponding to positions adjacent to slices in a subset of all two-dimensional slices, acquired parallel to the corresponding plane and to which segmentation results are assigned at each coordinate. The medical device according to claim 1.

3. The processing circuit is further configured to perform the segmentation using the plurality of segmentation algorithms, Each of the aforementioned segmentation algorithms is individually trained on different two-dimensional data extracted from the three-dimensional data and additional three-dimensional data corresponding to one or more eyes other than the aforementioned eye. The medical device according to claim 1.

4. The medical device according to claim 1, wherein the processing circuit is configured to generate a skeleton corresponding to the segmented three-dimensional dataset and to generate the profile of at least one structural element of interest based on the characteristics of the skeleton.

5. The target structural element includes at least one of a vitreous pocket, a lamina cribriformis, and an optic nerve head. The medical device according to claim 1.

6. Receive 3D data from the eye, Using multiple segmentation algorithms, each individually trained on different 2D data extracted from the 3D data, the 3D data is segmented into regions containing the target structural element and regions not containing the target structural element, thereby generating a segmented 3D dataset. Each segmentation algorithm corresponds to the training plane, Each segmentation algorithm is trained using data corresponding to multiple 2D image slices parallel to their respective training planes. The segmentation is performed using an axial segmentation algorithm corresponding to the axial plane of the eye, a coronal segmentation algorithm corresponding to the coronal plane of the eye, and a sagittal segmentation algorithm corresponding to the sagittal plane of the eye. A method for operating a medical device, comprising generating a profile of at least one structural element of interest from the segmented three-dimensional dataset.

7. The two-dimensional data used to train each of the aforementioned multiple segmentation algorithms is obtained from the three-dimensional data parallel to the corresponding plane and includes a subset of two-dimensional slices to which the segmentation results are assigned to each coordinate. The two-dimensional data used to train each of the aforementioned multiple segmentation algorithms further includes two-dimensional slices corresponding to positions adjacent to slices in a subset of all two-dimensional slices, acquired parallel to the corresponding plane and to which segmentation results are assigned at each coordinate. A method for operating the medical device according to claim 6.

8. The segmentation is performed using the plurality of segmentation algorithms, Each of the aforementioned segmentation algorithms is individually trained on different two-dimensional data extracted from the three-dimensional data and additional three-dimensional data corresponding to one or more eyes other than the aforementioned eye. A method for operating the medical device according to claim 6.

9. Furthermore, the method includes generating a skeleton corresponding to the segmented 3D dataset and generating a profile of at least one structural element of interest based on the characteristics of the skeleton. A method for operating the medical device according to claim 6.

10. The target structural element includes at least one of a vitreous pocket, a lamina cribriformis, and an optic nerve head. A method for operating the medical device according to claim 6.

11. When executed by a computer, Receive 3D data from the eye, Using multiple segmentation algorithms, each individually trained on different 2D data extracted from the 3D data, the 3D data is segmented into regions containing the target structural element and regions not containing the target structural element, thereby generating a segmented 3D dataset. Each segmentation algorithm corresponds to the training plane, Each segmentation algorithm is trained using data corresponding to multiple 2D image slices parallel to their respective training planes. The segmentation is performed using an axial segmentation algorithm corresponding to the axial plane of the eye, a coronal segmentation algorithm corresponding to the coronal plane of the eye, and a sagittal segmentation algorithm corresponding to the sagittal plane of the eye. A non-temporary, computer-readable storage medium containing computer-executable instructions for performing a method of operating a medical device that generates at least one structural element profile of interest from the segmented three-dimensional dataset.

Images