Method for determining autofluorescence of the fundus using optical coherence tomography

Abstract

Method for predicting autofluorescence of the fundus of an eye, comprising the following procedural steps: a) Acquiring a fundus using optical coherence tomography and providing a captured three-dimensional image (1) of the fundus showing multiple retinal layers to an evaluation unit, b) Dividing the three-dimensional image (1) into a multitude of two-dimensional cross-sections (2) per retinal layer, c) Combining several cross-sections (2) into at least one reflectivity map (3) and at least one layer thickness map per retinal layer, wherein the reflectivity map includes the reflectivity of the retinal layer and the layer thickness map includes the layer thickness of the retinal layer, d) Dividing the reflectivity maps (3) and the layer thickness map into several pixels (4), e) Determine at least one reflectivity value and layer thickness per pixel (4), f) Prediction of an autofluorescence value based on the reflectivity value and the layer thickness per pixel (4).

Description

[0001] The invention relates to a method for predicting autofluorescence of the fundus of an eye using optical coherence tomography.

[0002] Fundus autofluorescence (FAF) is a natural phenomenon in which light emitted from the fundus after excitation with light of different wavelengths can be captured and further analyzed. FAF plays a crucial role in ophthalmology for the diagnosis and monitoring of retinal diseases.

[0003] In clinical examinations, FAF is primarily performed using short-wavelength light (blue) or long-wavelength light (near-infrared). Fluorescein angiography, in which a suitable dye is injected into the bloodstream, is also a known technique. Blue FAF images are time-consuming, use very bright, dazzling blue light, and require a high number of averaging steps due to the weakness of the autofluorescence signal. Current methods for measuring FAF are complex and involve long measurement times, making these procedures not only time-consuming but also uncomfortable for the patient.

[0004] The article “Structural changes in optical coherence tomography underlying spots of increased autofluorescence in the perilesional zone of geographic atrophy” by Oishi, Maho et al. in Investigative Ophthalmology & Visual Science describes the combination of fundus autofluorescence imaging (FAF) and optical coherence tomography (OCT) to analyze structural changes in the retina in geographic atrophy (GA) as part of age-related macular degeneration.

[0005] US Patent 2004 / 0036838A1 describes a multi-channel optical mapping device that can deliver one or at least two images with different depth resolutions simultaneously, or images with different depth resolutions sequentially, or a combination of these images, or a single image with adjustable depth resolution.

[0006] Based on this, the object of the invention is to provide a method for predicting the autofluorescence of the fundus of the eye that is particularly quick and easy to perform and is ubiquitous.

[0007] This problem is solved by the subject matter of claim 1. Preferred embodiments are found in the dependent claims.

[0008] According to the invention, a method for predicting autofluorescence of the fundus of an eye is provided. The method comprises the following steps: a) Capturing the fundus using optical coherence tomography and providing a captured three-dimensional image of the fundus showing multiple retinal layers to an evaluation unit, b) Dividing the three-dimensional image into a multitude of two-dimensional cross-sections per retinal layer, c) Combining several cross-sections into at least one reflectivity map and at least one thickness map per retinal layer, wherein the reflectivity map includes the reflectivity of the retinal layer and the thickness map includes the thickness of the retinal layer, d) Dividing the reflectivity maps and the layer thickness map into several pixels each, e) Determine at least one reflectivity value and the layer thickness per pixel, f) Prediction of an autofluorescence value based on the reflectivity value and the layer thickness per pixel.

[0009] The term "autofluorescence" specifically refers to pseudo-autofluorescence. Pseudo-autofluorescence is predicted indirectly via reflectivity, meaning that autofluorescence is not directly detected but rather derived as pseudo-autofluorescence.

[0010] When the term "fundus" is used here, it refers to the area behind the transparent vitreous humor on the inner wall of the eyeball. The fundus includes, in particular, the retina, the retinal pigment epithelium (RPE), and Bruch's membrane.

[0011] The AF signal is primarily caused by bisretinoids in the granules of the retinal pigment epithelium (RPE). In addition to their autofluorescent properties, these granules also exhibit light-reflecting characteristics. This plays a crucial role in structural imaging of the retina using spectral domain optical coherence tomography (SD-OCT).

[0012] SD-OCT is a fast, safe, and easy-to-perform examination technique that visualizes the individual retinal layers at near-histological resolution. The RPE, in particular, exhibits highly reflective properties due to its more than 1500 granules within the cell body. Different layers reflect the incoming light differently, allowing a cross-section of the examined tissue to be calculated and displayed using the information obtained. This type of reconstruction is called tomography.

[0013] Procedures such as retinal endoscopy and fluorescein angiography provide a "top-view" view of the retina and its structures. However, they do not show the three-dimensional structure of the tissue under examination.

[0014] The coherence tomography measurement method does not strain the eye, nor is the measurement affected by the vitreous humor of the eye.

[0015] Autofluorescence is determined based on this reflectance measurement from an SD-OCT scan. It has been shown that a high reflectance corresponds to a high number of intracellular granules, and changes in reflectance in the RPE lead to changes in autofluorescence.

[0016] A key aspect of the invention is therefore that the reflectivities in the retinal layers can be determined using coherence tomography, which is quick and easy to perform, and autofluorescence can be predicted based on this.

[0017] From a three-dimensional volume macular scan, several two-dimensional cross-sections per retinal layer are generated, each with a color-coded layer thickness. The retinal layers preferably have a predetermined thickness. Alternatively, the retinal layer thickness is preferably determined based on the individual characteristics of a patient. The retinal layer thickness is then determined using white and black lines in the histological scan that have a histological correlate. Two-dimensional reflectivity maps and layer thickness maps are created from the cross-sections. In particular, all cross-sections are combined into several reflectivity maps. The layer thickness of the retinal layer is graphically represented in a two-dimensional layer thickness map using a grayscale scale. The reflectivity map is generated by averaging the layer thicknesses.The reflectivity values, which indicate the reflectivity of the retinal layer, are also graphically represented using a grayscale. In particular, the thicker the layer for grayscale layer thickness maps and the higher the reflectivity for grayscale reflectivity maps, the whiter the representation. This process generates so-called en-face maps, preferably four en-face maps per retinal layer. The reflectivity maps are divided into a multitude of pixels. At least one reflectivity value and one layer thickness are determined for each pixel.

[0018] Autofluorescence is predicted for each pixel using the reflectivity value and the layer thickness.

[0019] In this context, an "en-face map" refers to a frontal view of an originally three-dimensional image, whereby one dimension, namely depth, is eliminated by the frontal view through the aggregation or averaging of depth information. In other words, the three-dimensional image is reduced by one dimension by averaging the reflectivity in height or depth and graphically representing the length.

[0020] When we speak of "reflectivity" here, we are referring to the property of the back of the eye or its tissues to reflect incoming light. The higher the reflectivity value, the greater the proportion of light that is reflected.

[0021] According to a preferred embodiment of the invention, the method comprises the following further process steps: g) Providing the predicted autofluorescence values ​​to a visualization unit, h) Visualizing the predicted autofluorescence values ​​in at least one autofluorescence map using a grayscale.

[0022] Using the reflectivity value and the layer thickness, autofluorescence is predicted for the majority of reflectivity maps per pixel. The resulting autofluorescence values ​​are initially provided numerically. A visualization unit then graphically represents these numerical autofluorescence values ​​on a grayscale, allowing this graphical representation of the autofluorescence to be evaluated by a medical professional.

[0023] According to a preferred embodiment of the invention, the reflectivity value comprises a reflectivity maximum per pixel and / or a reflectivity minimum per pixel and / or a reflectivity value averaged over the layer thickness per pixel. Thus, for each retinal layer, up to four parameters are available per pixel for predicting autofluorescence: the layer thickness, the average reflectivity, the reflectivity maximum, and the reflectivity minimum.

[0024] According to a preferred embodiment of the invention, the prediction of autofluorescence values ​​is carried out using weighted decision trees based on previously recorded target values. "Target values" refers in particular to conventional FAF values. These indicate the autofluorescence in a healthy fundus and were statistically recorded in preparation for the method according to the invention.

[0025] Decision trees are specifically ordered, directed trees used to represent decision rules. Their graphical representation as tree diagrams illustrates hierarchically sequential decisions. They are important in numerous fields where automatic classification is performed or where formal rules are derived or represented from experiential knowledge. Decision trees are a method for the automatic classification of data objects and thus for solving decision problems. A decision tree always consists of a root node and any number of internal nodes, as well as at least two leaves. Each node represents a logical rule, and each leaf represents an answer to the decision problem.

[0026] The decision problem here is the presence and / or magnitude of autofluorescence. Based on different weightings of the nodes, the autofluorescence values ​​can be predicted using the reflectivity values ​​and the layer thickness per pixel.

[0027] According to an alternative embodiment of the invention, the autofluorescence values ​​are predicted using an artificial neural network. The artificial neural network is trained to generate the required parameters from the three-dimensional scan using optical coherence tomography and, based on these parameters, to estimate the autofluorescence.

[0028] The number of cross-sections per reflectivity map can preferably be freely selected. According to a preferred embodiment of the invention, each reflectivity map comprises 40 aggregated cross-sections of a retinal layer averaged over 10 layers. The number of cross-sections generated from the volume macula scan is preferably more than 100. For each reflectivity map, an autofluorescence value is predicted pixel by pixel, so that the autofluorescence values ​​are aggregated and displayed via at least one autofluorescence map or predicted pseudo-FAF map (inferred BAF).

[0029] Furthermore, the use of reflectivity values ​​obtained from optical coherence tomography of the fundus is planned for predicting fundus autofluorescence. Surprisingly, it has been shown that fundus reflectivity can be used to make predictions about autofluorescence. A change in reflectivity leads to changes in autofluorescence.

[0030] The invention will now be explained in more detail with reference to the drawings and a preferred embodiment.

[0031] The drawings show Fig. 1 schematically a method for predicting autofluorescence according to a preferred embodiment of the invention, Fig. 2 schematically the different stages of the process according to a preferred embodiment of the invention.

[0032] Out of Fig. Figure 1 schematically shows a method for predicting autofluorescence according to a preferred embodiment of the invention. First, the fundus of an eye is acquired using optical coherence tomography, so that the acquired three-dimensional image 1 of the volume macule scan can be provided to an evaluation unit a).

[0033] The three-dimensional image 1 is then divided into a multitude of two-dimensional cross-sections 2 showing the different retinal layers (B-scan) b). From these cross-sections, 10 maps are created for each retinal layer, representing the retinal layer thickness and reflectivity in grayscale c). This results in several reflectivity maps 3 and one layer thickness map per retinal layer.

[0034] These reflectivity maps 3 are each divided into several pixels 4 d). Subsequently, a reflectivity value is determined for each reflectivity map 3 and each pixel 4 e). This can be the reflectivity averaged over the layer thicknesses and / or a reflectivity maximum or minimum. Based on the predicted reflectivity values ​​and the average layer thickness, an autofluorescence value is determined for each pixel f).

[0035] The numerical autofluorescence values ​​are then provided to a visualization unit g) so that they can be visually represented via a grayscale h) and evaluated by a medical professional.

[0036] Fig. Figure 2 shows the individual stages of the procedure described above. First, a three-dimensional image 1 of the fundus, the so-called volume macule scan, is available.

[0037] The three-dimensional image 1 is then divided into a multitude of two-dimensional cross-sections 2. These cross-sections, or so-called OCT B-scans, are converted from the preset logarithmic representation into the reflectivity patterns in raw format 2'. The reflectivity maps 3 can be created from this raw data.

[0038] The reflectivity maps 3 each comprise several cross-sections 2'. The raw data are averaged along the layer thickness. This means that a two-dimensional reflectivity map 3 is generated from the several cross-sections with a specific image depth. The reflectivity maps 3 are each divided into several pixels 4. Fig. 2. The pixel sizes are not drawn to scale. The pixels are significantly smaller than in Fig. 2 shown.

[0039] For each pixel, an autofluorescence value is predicted based on the reflectivity values ​​and the average layer thickness. The numerical autofluorescence values ​​are summarized in an autofluorescence map 5, which graphically represents the predicted autofluorescence values. This image can be used to determine pathological changes in the fundus. Reference symbol list 1 three-dimensional image of the fundus 2 cross-sections in logarithmic representation 2' cross-sections in raw format 3 Reflectivity map 4 pixels 5 Autofluorescence map

Claims

[1] Method for predicting autofluorescence of the fundus of an eye, comprising the following method steps: a) Acquiring a fundus using optical coherence tomography and providing a captured three-dimensional image (1) of the fundus showing multiple retinal layers to an evaluation unit, b) Dividing the three-dimensional image (1) into a multitude of two-dimensional cross-sections (2) per retinal layer, c) Combining several cross-sections (2) into at least one reflectivity map (3) and at least one layer thickness map per retinal layer, wherein the reflectivity map includes the reflectivity of the retinal layer and the layer thickness map includes the layer thickness of the retinal layer, d) Dividing the reflectivity maps (3) and the layer thickness map into several pixels (4), e) Determine at least one reflectivity value and layer thickness per pixel (4), f) Prediction of an autofluorescence value based on the reflectivity value and the layer thickness per pixel (4). [2] The method of claim 1 comprising the following further process steps: g) Providing the predicted autofluorescence values ​​to a visualization unit, h) Visualizing the predicted autofluorescence values ​​in at least one autofluorescence map (5) using a grayscale. [3] Method according to claim 1 or 2, wherein the reflectivity value comprises a reflectivity maximum per pixel (4) and / or a reflectivity minimum per pixel (4) and / or a reflectivity value averaged over the layer thickness per pixel (4). [4] Method according to any one of claims 1 to 3, wherein the prediction of the autofluorescence values ​​is carried out using weighted decision trees based on previously recorded target values. [5] Method according to any one of claims 1 to 3, wherein the prediction of the autofluorescence values ​​is carried out using an artificial neural network. [6] Method according to one of the preceding claims, wherein 40 cross-sections (2) are combined to form a reflectivity map (3).

Images