Classification of choroidal neovascularization

Abstract

A method for identifying a type of choroidal neovascularization in a retina of a human eye is disclosed, the method comprising: receiving (S1) optical coherence tomography data of the eye in a processor; generating (S2) volume segments of the eye using the optical coherence tomography data and a neural network (3) in the processor; and identifying (S3) a type of choroidal neovascularization in the eye using the volume segments in the processor, the type of choroidal neovascularization comprising one or more of the following: classical choroidal neovascularization (41) and occult choroidal neovascularization (42).

Description

Technical Field

[0001] The present invention relates to a method and apparatus for identifying the type of choroidal neovascularization in a human eye. In particular, the present invention relates to a computer-implemented method, computer, and computer program product for identifying the type of choroidal neovascularization in a human eye. Background Art

[0002] The eye contains many different types of tissues and fluids. The posterior (back) portion of the eye can be broadly divided into the following layers: the neurosensory retina, which contains second- and third-order neurons and photoreceptors; the retinal pigment epithelium, a layer of pigment cells located just outside the neurosensory retina that nourishes these retinal visual cells; and the choroid, the vascular layer of the eye, which contains blood vessels and connective tissue. These layers can, of course, be further subdivided. Choroidal neovascularization (CNV) is the development of new blood vessels in the choroid. These new vessels can lead to lesions. These lesions may contain nonvascular components, such as fibrotic tissue, hemorrhage, pigmentation, or other features that may obscure the boundaries of the CNV.

[0003] Choroidal neovascularization can be divided into two types: classic and occult. Classic CNV lesions typically penetrate the retinal pigment epithelium (RPE) and are therefore located anterior to the RPE, while occult lesions are located inferior to the RPE. Classic CNV is relatively less common than occult or mixed forms (mixed forms are those in which both classic and occult lesions are present). The universal definition of CNV types in use today is based on macular photocoagulation studies (MPS). To determine the type of choroidal neovascularization present, invasive fluorescein angiography (FA) is used, in which a fluorescent dye is added to the patient's circulation, either intravenously in intravenous FA or orally in oral FA. The retina is then illuminated with blue light, and an angiogram is obtained by capturing the fluorescent green light emitted by the dye. Classic CNV is defined as a bright, clearly visible, and well-defined hyperfluorescence in the early stages of angiography. This leakage increases in the later stages of angiography, with the edges of the hyperfluorescence blurred (leakage). Occult CNV is defined by irregular elevation of the RPE with spotty or granular hyperfluorescence after 1–2 minutes. Over time, the borders may or may not reveal leakage. Fluorescein angiography carries risks because the use of fluorescein dye can cause adverse reactions, ranging from nausea to vomiting, urticaria, acute hypotension, and, more seriously, allergic reactions and associated anaphylactoid reactions, leading to cardiac arrest.

[0004] Classic CNVs may occur in exudative age-related macular degeneration (ARMD) but may also arise secondary to other diseases such as ocular histoplasmosis syndrome, pathological myopia, choroidal rupture, angioid streaks, or idiopathic causes.

[0005] Optical coherence tomography (OCT) provides a non-invasive, high-resolution method for creating optical cross-sections of the eye. Currently, OCT is used to evaluate patients with ARMD. This imaging helps physicians and clinicians assess the tomographic sequelae of neovascular membranes and monitor treatment response. The type of light used in OCT allows it to penetrate deep into tissue and examine tissue and fluid layers, even those with high reflectivity. In OCT, many one-dimensional scans (A-scans) are typically performed at several depths within the retina. These A-scans are then combined to form a two-dimensional cross-section of the eye in a so-called B-scan. The B-scans can then be combined to form a three-dimensional image of the eye, also known as a C-scan. Advanced OCT methods such as spectral-domain OCT (SD-OCT) have shortened OCT image acquisition times, and the accuracy of OCT images has also improved with the introduction of technologies that track and compensate for eye movement and blinking.

[0006] WO 2016 / 154485 describes a method for detecting, visualizing, and measuring the extent of retinal neovascularization. The method comprises receiving a set of cross-sectional angiograms, segmenting the set of cross-sectional angiograms into layers, and generating an en face inner retinal angiogram and an en face vitreal angiogram.

[0007] Abhijit Guha Roy et al., “ReLayNet: retinal layer and fluid segmentation of macular optical coherence tomography using fully convolutional networks,” Biomed. Opt. Express 8, 3627-3642 (2017), disclose a method for segmenting 2D retinal layers and fluid in retinal OCT data. The disclosed method uses a neural network to assign each pixel of an OCT B-scan to one of ten classes. Summary of the Invention

[0008] The object of the present invention is to provide a method and apparatus for identifying the type of choroidal neovascularization in a human eye. In particular, the present invention relates to a computer-implemented method, computer and computer program product for identifying the type of choroidal neovascularization in a human eye.

[0009] According to the invention, these objects are achieved by the features of the independent claims.Further advantageous embodiments ensue from the dependent claims and the description.

[0010] According to the present invention, the above-mentioned object is achieved in particular by a computer-implemented method for identifying a type of choroidal neovascularization in a human eye, the method comprising the step of receiving, in a processor of a computer, optical coherence tomography data of an eye, in particular optical coherence tomography data of a retinal region of the eye. In a further step, in the processor, volume segments of the human eye are generated using the optical coherence tomography data and a neural network. In a further step, the processor uses the volume segments to identify a type of choroidal neovascularization in the eye. The type of choroidal neovascularization comprises one or more of the following: classical choroidal neovascularization and occult choroidal neovascularization.

[0011] In an embodiment, generating the volume segments comprises a processor detecting tissue layer segments and fluid segments.

[0012] In an embodiment, generating the volume segments comprises detecting one or more of the following types of tissue layer segments: an internal limiting membrane segment, a retinal pigment epithelium segment, and a Bruch's membrane segment.

[0013] In an embodiment, generating the volume segments comprises detecting one or more of the following types of fluid segments: intraretinal fluid segments, subretinal fluid segments, subretinal highly reflective material segments, central subfield segments, and pigment epithelial detachment segments.

[0014] In an embodiment, identifying the type of choroidal neovascularization includes giving the highest weight to the following types of fluid segments: subretinal highly reflective material segments, subretinal fluid segments, and pigment epithelial detachment segments.

[0015] In an embodiment, identifying the type of choroidal neovascularization includes identifying a lesion in the eye.

[0016] In an embodiment, the method further comprises calculating one or more of the following geometric parameters: height of the volume segments, width of the volume segments, and distance between the volume segments; and identifying the type of choroidal neovascularization further comprises using the geometric parameters.

[0017] In an embodiment, identifying the type of choroidal neovascularization comprises using a decision tree.

[0018] In an embodiment, generating the volume segments comprises training a neural network using machine learning and a neural network training dataset of a large amount of optical coherence tomography data of the eye.

[0019] In an embodiment, identifying the type of choroidal neovascularization includes optimizing a decision tree using a decision tree training dataset of optical coherence tomography data of a large number of eyes and a gradient boosting algorithm.

[0020] In an embodiment, identifying the type of choroidal neovascularization comprises optimizing a decision tree using a catboost algorithm.

[0021] In an embodiment, generating the volume segments using a neural network comprises using a convolutional neural network.

[0022] In an embodiment, generating the volume segments using a neural network includes using a U-net and / or a ResNet.

[0023] In an embodiment, the method further comprises extracting, in the processor, A-scans and / or B-scans from the received optical coherence tomography data, and generating the volume segments further comprises generating the region segments using the A-scans and / or B-scans and a neural network.

[0024] In an embodiment, receiving optical coherence tomography data includes receiving one or more of the following: time-domain optical coherence tomography data, spectral-domain optical coherence tomography data, ultra-high-speed swept-source optical coherence tomography data, ultra-high-resolution optical coherence tomography data, polarization-sensitive optical coherence tomography data, and adaptive optics optical coherence tomography data.

[0025] In an embodiment, identifying the type of choroidal neovascularization further comprises identifying whether the eye has classic choroidal neovascularization, occult choroidal neovascularization, classic and occult choroidal neovascularization, or no choroidal neovascularization.

[0026] In addition to a computer-implemented method for identifying a type of choroidal neovascularization in a human eye, the present invention also relates to a computer for identifying a type of choroidal neovascularization in a human eye, the computer comprising a processor. The processor is configured to receive optical coherence tomography data of the eye, particularly optical coherence tomography data of the retina of the eye. The processor is configured to generate volume segments of the human eye using the optical coherence tomography data and a neural network, and to use the volume segments to identify a type of choroidal neovascularization in the eye. The type of choroidal neovascularization includes one or more of the following: classical choroidal neovascularization and occult choroidal neovascularization.

[0027] In an embodiment, the processor is configured to generate the volume segment by detecting one or more tissue layer segments and / or fluid segments.

[0028] In an embodiment, the processor is configured to generate the volume segments by detecting one or more of the following types of tissue layer segments: an internal limiting membrane segment, a retinal pigment epithelium segment, and a Bruch's membrane segment.

[0029] In an embodiment, the processor is configured to generate the volume segments by detecting one or more of the following types of fluid segments: intraretinal fluid segments, subretinal fluid segments, subretinal highly reflective material segments, central subfield segments, and pigment epithelial detachment segments.

[0030] In an embodiment, the processor is configured to identify the type of choroidal neovascularization by giving the highest weight to the following types of fluid segments: subretinal highly reflective material segments, subretinal fluid segments, and pigment epithelial detachment segments.

[0031] In an embodiment, the processor is configured to identify the type of choroidal neovascularization by identifying a lesion in the eye.

[0032] In an embodiment, the processor is further configured to calculate geometric data comprising one or more of the following geometric parameters: height of the volume segments, width of the volume segments, and distance between the volume segments; and identify the type of choroidal neovascularization by using the geometric parameters.

[0033] In an embodiment, the processor is configured to identify the type of choroidal neovascularization using a decision tree.

[0034] In an embodiment, the processor is configured to generate the volume segments by training a neural network using machine learning and a neural network training dataset of a large amount of optical coherence tomography data of the eye.

[0035] In an embodiment, the processor is configured to identify the type of choroidal neovascularization by optimizing a decision tree using a decision tree training dataset of optical coherence tomography data of a large number of eyes and a gradient boosting algorithm.

[0036] In an embodiment, the processor is configured to identify the type of choroidal neovascularization by optimizing a decision tree using a catboost algorithm.

[0037] In an embodiment, the processor is configured to generate the volume segments using a neural network by using a convolutional neural network.

[0038] In an embodiment, the processor is configured to generate the volume segments using a neural network by using one or more of: U-net and ResNet.

[0039] In an embodiment, the processor is configured to extract A-scans and / or B-scans from the received optical coherence tomography data and generate volume segments by using the A-scans and / or B-scans and a neural network to generate region segments.

[0040] In an embodiment, the processor is configured to receive one or more of the following types of optical coherence tomography data: time-domain optical coherence tomography data, spectral-domain optical coherence tomography data, ultra-high-speed swept-source optical coherence tomography data, ultra-high-resolution optical coherence tomography data, polarization-sensitive optical coherence tomography data, and adaptive optics optical coherence tomography data.

[0041] In an embodiment, the processor is configured to identify the type of choroidal neovascularization by identifying whether the eye has classic choroidal neovascularization, occult choroidal neovascularization, classic and occult choroidal neovascularization, or no choroidal neovascularization.

[0042] In addition to a computer-implemented method for identifying a type of choroidal neovascularization in a human eye and a computer comprising a processor configured to identify a type of choroidal neovascularization in a human eye, the present invention also relates to a computer program product and a computer for identifying a type of choroidal neovascularization in a human eye. The computer program product comprises a non-transitory computer-readable medium having computer program code stored thereon, the computer program code being configured to control a processor of a computer so that the computer performs the following steps: receiving optical coherence tomography data of an eye, in particular optical coherence tomography data of a retina of the eye, in the processor. In another step, generating a volume segment of the human eye using the optical coherence tomography data and a neural network in the processor. In another step, the processor identifies a type of choroidal neovascularization in the eye using the volume segment. The type of choroidal neovascularization comprises one or more of the following: classical choroidal neovascularization and occult choroidal neovascularization. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention will be explained in more detail by way of example with reference to the accompanying drawings, in which:

[0044] Figure 1 : Shows a block diagram schematically illustrating a computer including a processor and a memory.

[0045] Figure 2 : A flow chart illustrating a series of steps for identifying the type of choroidal neovascularization in the eye is shown.

[0046] Figure 3 : shows a block diagram illustrating a hierarchy of types of volume segments.

[0047] Figure 4 : A flow chart illustrating a series of steps for identifying the type of choroidal neovascularization in the eye is shown.

[0048] Figure 5 : A flowchart illustrating a series of steps for generating a volume segment of an eye is shown.

[0049] Figure 6 : A block diagram illustrating an alternative approach for identifying the type of choroidal neovascularization in the eye is shown.

[0050] Figure 7 : Diagram showing a B-scan before and after segmentation.

[0051] Figure 8 : Shown are experimental optical coherence tomography data of two eyes characterized by classical neovascularization, specifically showing the central B-scan with pixel mask for volume measurements, frontal projection, and thickness map.

[0052] Figure 9 : Shown are experimental optical coherence tomography data of two eyes characterized by occult neovascularization, specifically showing the central B-scan, en face projection, and thickness map with pixel mask for volume measurements. DETAILED DESCRIPTION

[0053] exist Figure 1 , reference numeral 1 denotes a computer comprising one or more processors 11. The computer 1 may further comprise various components, such as a memory 12, a communication interface and / or a user interface. The components of the computer 1 may be connected to each other via a data connection mechanism so that they can transmit and / or receive data.

[0054] The term "data connection mechanism" refers to a mechanism that facilitates data communication between two components, devices, systems, or other entities. The data connection mechanism can be wired, such as a cable or a system bus. The data connection mechanism can also include wireless communication. The data connection mechanism can also include communication via a network such as a local area network, a mobile radio network, and / or the Internet. Depending on the implementation, the Internet may include intermediate networks.

[0055] The processor 11 may include a system on a chip (SoC), a central processing unit (CPU), and / or other more specific processing units, such as a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), a reprogrammable processing unit such as a field programmable gate array (FPGA), and a processing unit specifically configured to accelerate certain applications, such as an AI (artificial intelligence) accelerator for accelerating neural network and / or machine learning processes.

[0056] Memory 12 includes one or more volatile and / or non-volatile storage components. Storage components may be removable and / or non-removable, and may also be integrated in whole or in part with processor 11. Examples of storage components include RAM (random access memory), flash memory, a hard drive, a data storage device, and / or other data storage devices. Memory 12 includes a non-transitory computer-readable medium having stored thereon computer program code configured to control processor 11 so that computer 1 performs one or more steps and / or functions as described herein. Depending on the embodiment, the computer program code is compiled or non-compiled program logic and / or machine code. Thus, computer 1 is configured to perform one or more steps and / or functions. The computer program code defines and / or is part of a discrete software application. Those skilled in the art will appreciate that the computer program code may also be distributed across multiple software applications. In embodiments, the computer program code further provides an interface, such as an API, that enables remote access to the functions and / or data of computer 1, for example, via a client application or a web browser.

[0057] In an embodiment, computer 1 is implemented as a server computer accessible by one or more client computing devices via a network.

[0058] Figure 2 A flow chart illustrating a series of steps for carrying out the present invention is shown.

[0059] In step S1, computer 1 receives optical coherence tomography (OCT) data of an eye. Specifically, processor 11 receives OCT data of a human eye using a data connection mechanism. In the previous step, the OCT data was generated by the OCT system using optical interferometry as known to those skilled in the art.

[0060] In an embodiment, the OCT data includes one or more of the following: time-domain optical coherence tomography (TD-OCT) data, spectral-domain optical coherence tomography (SD-OCT) data, ultra-high-speed swept-source optical coherence tomography data, ultra-high-resolution optical coherence tomography data, polarization-sensitive optical coherence tomography data, and adaptive optics optical coherence tomography data. In a preferred embodiment, SD-OCT data is used.

[0061] The OCT data is stored in a separate server's memory or directly in memory 12 on computer 1. In the case where the OCT data is stored in memory 12, processor 11 can retrieve the OCT data directly from memory 12. In the case where the OCT data is stored in a separate server's memory, processor 11 must first retrieve the OCT data. Processor 11 uses the OCT data to generate a three-dimensional image or model of the eye or a portion of the eye. In particular, processor 11 generates a three-dimensional image or model of a portion of the retina of the eye.

[0062] In step S2, a volume segment 2 is generated by the processor 11 using a neural network 3, which takes the OCT data as input. The neural network 3 is stored in the memory 12. The processor 11 uses the neural network 3 to output a volume segment 2, which is a region of space in the eye, particularly the retinal region. The volume segment 2 is a region of space within a three-dimensional image or model of a portion of the retina of the eye generated by the processor 11 using the OCT data. In particular, the volume segment 2 is a region of space that has significant extension in all three dimensions, so that the volume segment 2 does not correspond to a two-dimensional slice of only nominal thickness. In an embodiment, a given tissue layer or fluid layer of the retinal region of the eye is represented in a single volume segment 2, such that a surjective function exists between the tissue layer and the fluid layer and the volume segment 2. Preferably, consecutive volume segments 2, i.e., volume segments 2 that are adjacent to each other, do not have adjacent tissue layers or fluid layers of the same type.

[0063] In an embodiment, each volume segment 2 corresponds to a specific type of tissue or fluid of the eye, in particular a tissue layer or fluid layer of the retinal region, and each tissue layer or fluid layer corresponds to a specific volume segment 2, such that a bijective function exists between the volume segment 2 and the layers of the eye (i.e., the tissue layers and the fluid layers).

[0064] In an embodiment, the volume segments 2 are marked with features. The features include one or more of a plurality of types of volume segments 2, as described below. Figure 3 As explained in the description.

[0065] In an embodiment, the neural network 3 utilizes convolutional layers. Preferably, the neural network 3 is adapted to efficiently process three-dimensional input data. Preferably, the neural network 3 comprises a U-Net and / or ResNet architecture, as both types of neural networks 3 can be used and / or adapted to achieve good results when generating volume segments 2 in three-dimensional data. The neural network 3 is trained to generate volume segments 2 using machine learning and a large neural network training dataset of OCT data of eyes. This neural network training dataset includes defined volume segments 2 of OCT data, allowing the neural network 3 to be trained using known methods to generate volume segments 2 of OCT data of previously unseen eyes. The neural network training dataset includes eyes with classic choroidal neovascularization 41 and occult choroidal neovascularization 42 (defined by the corresponding eye data). In an embodiment, the neural network training dataset further includes eyes without choroidal neovascularization (healthy eyes) (defined by the corresponding eye data).

[0066] In an embodiment, the volume segments 2 of the neural network training dataset are further annotated with features including types and descriptors, and training the neural network 3 includes training the neural network 3 to label the generated volume segments 2 with one or more features. Thus, the trained neural network 2 assigns features to each generated volume segment 2. The descriptors may include geometric parameters 6 associated with the volume segment 2, such as Figure 4 As explained in more detail in the description.

[0067] Each volume segment 2 is a three-dimensional image, model, or representation defined by a set of voxels of the three-dimensional OCT data belonging to that particular volume segment 2. Alternatively, the volume segment 2 can be defined by a mesh such as a polygonal mesh. The volume segments 2 can overlap each other, or they can be non-overlapping. As described above, the volume segments 2 can be labeled with features.

[0068] In step S3, the processor 11 uses the generated volume segment 2 to identify the type of choroidal neovascularization 40 of the eye. The identified neovascularization types include classic choroidal neovascularization 41 and occult choroidal neovascularization 42. The processor 11 uses one or more models, algorithms, and / or functions to identify the type of choroidal neovascularization 40 present in the retina of the eye.

[0069] For example, using volume segments 2 rather than multiple 2D slices to identify the type of neovascularization allows processor 11 to improve its performance in identifying the type of choroidal neovascularization. More specifically, using volume segments 2 enables processor 11 to identify the type of choroidal neovascularization with greater specificity and sensitivity. This is because the retina of the eye is a three-dimensional physical structure, and therefore a three-dimensional representation is more naturally suitable. In particular, the three-dimensional properties of the retinal region of the eye and the various tissue and fluid layers, such as the geometric parameters of a given tissue or fluid layer, or the geometric relationships between a given tissue or fluid layer and other tissue or fluid layers, are more directly represented in a set of volume segments 2 and their associated features. This more direct and natural representation of a given fluid or tissue layer as a single volume segment 2, rather than a set of regions in 2D space from a set of 2D scans, is exemplified because it allows for simpler and less complex functions (particularly surjective or bijective functions) to exist between the volume segment 2 (representation) and the fluid and tissue layers (reality). In particular, fewer volume segments 2 are required to represent the retinal region of the eye than would be required for a set of 2D scans.

[0070] In an embodiment, the processor 11 is further configured to use one or more models, algorithms, and / or functions to identify the absence of choroidal neovascularization 40 and, therefore, determine whether the eye is healthy. In particular, the processor 11 is configured to identify a healthy eye, an eye with classic choroidal neovascularization 41, and an eye with occult choroidal neovascularization 42.

[0071] In an embodiment, the processor 11 uses a nearest neighbor algorithm to identify the type of choroidal neovascularization 40 of the eye in question. The nearest neighbor algorithm is a supervised classification algorithm that uses a labeled classification training dataset. The classification training dataset includes volume segments 2 of a large number of eyes with known types of choroidal neovascularization 40. The nearest neighbor algorithm checks which eye in the classification training dataset has a volume segment 2 that most closely corresponds to the volume segment 2 of the eye in question, and assigns the type of choroidal neovascularization 40 of the eye in question to the eye in question.

[0072] In an embodiment, the processor 11 uses logistic regression to identify the type of choroidal neovascularization 40 in the eye in question. Logistic regression is a statistical model used to analyze a dataset comprising one or more independent variables, where the output is a dichotomous variable. The processor 11 uses logistic regression and a labeled classification training dataset (comprising a large number of eye volume segments 2 with known types of choroidal neovascularization 40) to generate a model that can identify the type of choroidal neovascularization 40 in the eye in question.

[0073] In an embodiment, the processor 11 uses a support vector machine to identify the type of choroidal neovascularization 40 present in the eye in question. The support vector machine uses a classification training dataset (comprising a large number of volume segments 2 of eyes with known types of choroidal neovascularization 40) to generate an optimal hyperplane, which is used by the processor 11 to identify which type of choroidal neovascularization 40 is present in the eye in question.

[0074] In an embodiment, the processor 11 uses a separate neural network to identify the type of choroidal neovascularization 40 in the eye. The neural network uses a classification training dataset (comprising a large number of volume segments 2 of the eye with known types of choroidal neovascularization 40) to optimize a set of internal weights and biases, which are then used by the processor 11 to identify which type of choroidal neovascularization 40 is present in the eye in question.

[0075] In an embodiment, the processor 11 uses the decision tree 4 to identify the type of choroidal neovascularization 40 in the eye, as in Figure 6 As explained in more detail in the description.

[0076] In an embodiment, the processor 11 also identifies whether there are any choroidal neovascularization 40. The processor 11 may also identify whether both classic choroidal neovascularization 41 and occult choroidal neovascularization 42 are present.

[0077] In an embodiment, the processor 11 identifies a lesion in the eye using the volume segment 2. The identified lesion is used by the processor 11 to identify the type of choroidal neovascularization 40.

[0078] Figure 3 A block diagram illustrating a hierarchy of types of volume segments 2 is shown. The volume segments 2 are further classified into tissue layer segments 21 and fluid segments 22. The tissue layer segments 21 are further classified into at least the following categories: inner limiting membrane segments 211, retinal pigment epithelium segments 212, and Bruch's membrane segments 232. The fluid segments 22 are further classified into at least the following categories: intraretinal fluid segments 221, subretinal fluid segments 222, subretinal highly reflective material segments 223, central subfield segments 224, and pigment epithelium detachment segments 225. These listed types of volume segments 2 are not limiting, and other types of volume segments 2 and hierarchical structures are possible.

[0079] As mentioned above, in Figure 2In step S2, the processor 11 uses the neural network 3 to generate volume segments 2 of the eye using the OCT data. Furthermore, the processor 11 uses the neural network to assign one or more of the aforementioned types of volume segments 2 to each generated volume segment 2. Overall, when combining the different types of volume segments 2 and the possible features that each volume segment 2 may have associated therewith, more than 100 features were found to be clinically relevant. These features are either qualitative, such as corresponding to the type of volume segment 2, or quantitative, such as relating to the location of the volume segment 2 in the eye, or geometric parameters 6 of the eye (not shown), as explained in more detail below.

[0080] Figure 4 A flow chart illustrating a series of steps for performing the present invention is shown. An additional step S21 of generating features of the volume segment 2 is shown between step S2 and step S3. In step S21, geometrical parameters 6 of the volume segment 2 of the eye are calculated. As described above in Figure 1 As described in the description of FIG, in step S2, the processor 11 uses the neural network 3 to generate volume segments 2 of the eye. The processor 11 then calculates geometric parameters 6, as shown in step S21. These geometric parameters 6 relate to the volume segments 2 and their size and shape, their position relative to the geometric center of the retina of the eye, and also to the spatial arrangement of the volume segments 2 in the retina of the eye. In particular, the geometric parameters 6 include one or more from the following list: the height 61 of a given volume segment 2, the width 62 of a given volume segment 2, the distance 62 between a given volume segment 2 and other volume segments 2, the volume of a given volume segment 2, the thickness of a given volume segment 2, a measure of the sphericity of a given volume segment 2, and a measure of the surface smoothness of a given volume segment 2. These geometric parameters 6 are features assigned to each of the volume segments 2 and are used in subsequent steps, in particular in step S3, in which the processor 11 identifies the type of choroidal neovascularization 40 in the eye.

[0081] Figure 5 A flow chart illustrating a series of steps for performing the present invention is shown. In step S1 ', the step is as described above. Figure 1Alternatively to step S1 described above, the OCT data received by processor 11 may include A-scans 71 and / or B-scans 72. If the OCT data received by processor 11 includes A-scans 71, in optional step S11, processor 11 combines a subset of A-scans 71 to generate a sequence of B-scans 72. An A-scan 71 is a one-dimensional scan of the eye's retina at a given depth. A-scans are sufficiently dense in both the lateral and depth directions to allow for subsequent high-resolution reconstruction of both a two-dimensional cross-section (B-scan 72) and a three-dimensional image (C-scan 73) of the eye. If the OCT data includes B-scans 72, processor 11 proceeds directly to step S12.

[0082] In step S12, processor 11 generates regional segments of a two-dimensional B-scan 72 (received in the OCT data or generated by processor 11) using neural network 3. Because B-scans 72 are two-dimensional images, they are well-suited for segmentation using convolutional neural networks, particularly neural networks 3 including U-Net and / or ResNet. U-Net is a convolutional neural network developed for biomedical image segmentation, where the architecture is designed to produce precise and accurate results using fewer training images than classic fully connected convolutional neural networks. It achieves this by having, in addition to a contraction component that detects small features, an upsampling component that increases the resolution of the output image and receives contextual information from many contraction layers, thereby generating a high-resolution output. ResNet is a neural network architecture that uses shortcut or skip connections to enable efficient training of deep convolutional neural networks.

[0083] In an embodiment, the tissue layers present in the B-scan 72 are segmented using an algorithm that is different from the fluid layer. Using the algorithm, the following seven tissue layer types of the retina of the eye are segmented into two-dimensional tissue layer segments: inner limiting membrane, outer plexiform layer-Hanle fiber layer, boundary between the myoid and inner ellipsoid segments, IS / OS junction, inner boundary retinal pigment epithelium, outer boundary retinal pigment epithelium, and Bruch's membrane. Each B-scan 72 and two-dimensional tissue layer segments in the B-scan 72 generated by the above algorithm are labeled with the appropriate tissue layer type.

[0084] In an embodiment, the processor 11 uses the neural network 3 as described above to segment the fluid layers present in the B-scan 72 into two-dimensional fluid segments. In particular, the intraretinal fluid, subretinal fluid, subretinal highly reflective material segments, and pigment epithelial detachment layers are detected and marked in the B-scan 72, and their corresponding regions are defined as two-dimensional fluid segments.

[0085] The generated two-dimensional area segments of the consecutive B-scans 72 are used to generate volume segments 2 of the retina of the eye in step S2. Specifically, the processor 11 uses the B-scans 72, which include two-dimensional fluid segments and / or two-dimensional tissue layer segments and their respective assigned characteristics, to generate a three-dimensional image (C-scan 73) of the retina of the eye, including the volume segments 2. Each segment has an assigned characteristic based on its type of tissue layer segment or fluid layer segment. Each pixel of each two-dimensional B-scan 72 either belongs to a two-dimensional area segment or does not belong to any two-dimensional area segment. If a pixel of a two-dimensional B-scan 72 belongs to a two-dimensional area segment, it is assigned a label based on its type, as listed above. Each pixel of a given two-dimensional B-scan is then mapped to a voxel of a C-scan 73 with its assigned characteristic. In an embodiment, the volume segments 2 are generated such that adjacent volume segments 2 do not contain adjacent fluid layer regions or tissue layer regions of the same type. Specifically, the volume segments 2 are generated such that each volume segment 2 corresponds to a single tissue layer segment 2 or a single fluid segment 22.

[0086] Figure 6 A flowchart illustrating a series of steps for performing the present invention is shown. In step S3', the processor 11 identifies the type of choroidal neovascularization 40 in an eye using the volume segments 2, their features, and the decision tree 4. The decision tree 4 is trained or optimized using a decision tree training dataset of a large number of eye OCT data. The decision tree training dataset includes a large number of volume segments 2 and the type of choroidal neovascularization 40 present in each eye. Preferably, in addition to the large number of volume segments 2 of eyes with choroidal neovascularization, the decision tree training dataset further includes a large number of volume segments 2 of healthy eyes, i.e., eyes without any type of choroidal neovascularization, as control or reference data. Preferably, the decision tree training dataset includes data for at least one thousand eyes. Preferably, each volume segment 2 of an eye in the decision tree training dataset further includes associated assigned features that represent the volume segment type and additional descriptors. These features are either categorical, in that they place their associated volume segment 2 into one or more of a given set of categories, or numerical, in that they describe some geometric property of the volume segment 2, such as its volume or extent in a particular direction, as explained in more detail below.

[0087] Processor 11 optimizes decision tree 4 using a decision tree training dataset so that decision tree 4 accurately identifies the type of choroidal neovascularization 40 as defined in the decision tree training dataset. During a specific training iteration, processor 11 predicts the type of choroidal neovascularization 40 present in an eye from the decision tree training dataset and then checks the accuracy of the predicted type against the actual type defined in the decision tree training dataset. A logarithmic loss function is used to determine the accuracy of the prediction, and the parameters of decision tree 4 are iteratively optimized to minimize the logarithmic loss function. Preferably, a gradient boosting algorithm is used to optimize decision tree 4. In an embodiment, a collection of decision trees 4 is used in conjunction with the gradient boosting algorithm. Using decision tree 4 to identify the type of choroidal neovascularization 40 is advantageous because decision tree 4 can be easily visualized and understood as a flowchart-like structure with transparent and easily understood decision rules. To validate the performance of decision tree 4 and ensure it has not overfitted, a separate decision tree validation dataset of at least one hundred eyes, previously unseen during training, is used. Furthermore, cross-validation was used to evaluate performance by combining the training and validation datasets and then randomly partitioning them into new training and validation datasets. After processor 11 had trained decision tree 4, it was further evaluated and manually adjusted or simplified to become more understandable, allowing it to handle edge cases and outliers. The trained decision tree 4 accurately identified whether an eye had choroidal neovascularization. The trained decision tree 4 enabled excellent discrimination between classic choroidal neovascularization 41 and occult choroidal neovascularization 42, with an area under the receiver operating characteristic curve (AUROC) of 0.91 and a 95% confidence interval of 0.89-0.94.

[0088] In an embodiment, the processor 11 uses a catboost algorithm to optimize the decision tree 4. The catboost algorithm is advantageous because it is robust to overfitting, and because many of the above features are categorical, for which the catboost algorithm is particularly well suited.

[0089] In an embodiment, the decision tree 4 weights the following types of volume segments 2 higher than other types of volume segments 2: subretinal high-reflective material segments, subretinal fluid segments, and pigment epithelial detachment segments. In particular, subretinal high-reflective material segments and pigment epithelial detachment segments are weighted higher than other types of volume segments 2 because they have been shown to be the most discriminative characteristics. This means that the presence of a volume segment 2 having characteristics that are one of the types of volume segments 2 mentioned above has a greater impact on determining whether an eye is classified as having classic choroidal neovascularization 41 or occult choroidal neovascularization 42 than other types of volume segments 2. The decision tree 4 also takes into account additional features, such as the size or volume of a given volume segment 2.

[0090] In an embodiment, multiple decision trees 4 are used in a random forest model to identify the type of choroidal neovascularization 40. A random forest model is a collection of decision trees whose results are aggregated. A random forest model has the advantage of reducing errors due to bias and errors due to variance.

[0091] Figure 7 An illustration of a B-scan 72 (left) and a segmented B-scan 73 (right) is shown, wherein two-dimensional fluid segments are represented in grayscale. Overlaid on both images is an illustration of a U-Net used to perform image segmentation.

[0092] Figure 8 Two samples are shown, one of the retina 51 (top) and one of the retina 52 (bottom), with the retina 51 showing a typical case of classic choroidal neovascularization 41 and the other of the retina 52 also showing a typical case of classic choroidal neovascularization 41. B-scans 72 of the respective eyes 51, 52, acquired by SD-OCT, are on the left, indicating subretinal highly reflective material 511, intraretinal fluid 512, and fluid segments of subretinal fluid 513. The middle panel 514 is a frontal projection of the B-scan, and the rightmost panel 515 is a thickness map of the respective eyes 51, 52.

[0093] Figure 9 Two samples are shown of the retina 53 (top) of the eye showing a typical case of occult choroidal neovascularization 42 and the retina 54 (bottom) of the eye, which also shows a typical case of occult choroidal neovascularization 42. A B-scan 72 of the respective eyes 53, 54 acquired by SD-OCT is on the left, indicating a segment of pigment epithelial detachment 531. The middle panel 532 is a frontal projection of the B-scan, and the rightmost panel 532 is a thickness map of the respective eyes 53, 54.

[0094] It should be noted that in the description, the order of the steps has been presented in a specific order, however, those skilled in the art will understand that the order of at least some of the steps may be changed without departing from the scope of the present invention.

Claims

1. A computer-implemented method for identifying a type of choroidal neovascularization (40) in a human eye, the method comprising: Receiving (S1) optical coherence tomography data of an eye in a processor (11); generating (S2) a volume segment (2) of a human eye in a processor (11) using optical coherence tomography data and a neural network (3), wherein generating (S2) the volume segment (2) comprises detecting one or more of the following types of fluid segments (22): intraretinal fluid segments (221), subretinal fluid segments (222), subretinal highly reflective material segments (223), central subfield segments (224), and pigment epithelial detachment segments (225); and The type of choroidal neovascularization (40) in the eye is identified (S3) in a processor (11) using volume segments (2), the type of choroidal neovascularization (40) comprising one or more of the following: classical choroidal neovascularization (41) and occult choroidal neovascularization (42), wherein identifying the type of choroidal neovascularization (40) comprises giving the highest weighting to the following types of fluid segments (22): subretinal high reflective material segments (223), subretinal fluid segments (222), and pigment epithelial detachment segments (225).

2. The method according to claim 1, wherein generating (S2) the volume segment (2) comprises detecting a tissue layer segment (21) and a fluid segment (22).

3. The method according to claim 1 or 2, wherein generating (S2) the volume segment (2) comprises detecting one or more of the following types of tissue layer segments (21): an internal limiting membrane segment (211), a retinal pigment epithelium segment (212) and a Bruch's membrane segment (213).

4. The method according to claim 1 or 2, wherein: Identifying the type of choroidal neovascularization (40) includes identifying a lesion in the eye.

5. The method according to claim 1 or 2 further includes calculating (S21) one or more of the following geometric parameters (6): the height (61) of the volume segment, the width (62) of the volume segment and the distance (63) between the volume segments; and identifying (S3) the type of choroidal neovascularization (40) further includes using the geometric parameters (6).

6. The method according to claim 1 or 2, wherein identifying (S3) the type of choroidal neovascularization (40) comprises using a decision tree (4).

7. The method according to claim 1 or 2, wherein generating the volume segments (S2) comprises training the neural network (3) using machine learning and a neural network training dataset of a large amount of optical coherence tomography data of the eye.

8. The method according to claim 1 or 2, wherein identifying (S3) the type of choroidal neovascularization (40) comprises optimizing the decision tree (4) using a decision tree training dataset of optical coherence tomography data of a large number of eyes and a gradient boosting algorithm.

9. The method according to claim 1 or 2, wherein generating (S2) the volume segments using a neural network (3) comprises using a convolutional neural network.

10. The method according to claim 1 or 2, wherein the method further comprises extracting (S11) A-scans (71) and / or B-scans (72) from the received optical coherence tomography data in the processor (11) and generating (S2) volume segments further comprises generating (S12) regional segments using the A-scans (71) and / or B-scans (72) and the neural network (3).

11. The method according to claim 1 or 2, wherein: Identifying (S3) the type of choroidal neovascularization (40) further includes identifying whether the eye has classic choroidal neovascularization (41), occult choroidal neovascularization (42), classic choroidal neovascularization (41) and occult choroidal neovascularization (42), or no choroidal neovascularization.

12. A computer (1) for identifying (S3) a type of choroidal neovascularization (40) in a human eye, the computer (1) comprising a processor (11) configured to perform the method according to any one of claims 1 to 11.

13. A computer program product for identifying (S3) a type of choroidal neovascularization (40) in a human eye, comprising a non-transitory computer-readable medium (12) having computer program code stored thereon, the computer program code being configured to control a processor (11) of a computer (1) such that the computer (1) performs the method according to any one of claims 1 to 11.

Images