Diagnostic method

Abstract

To provide diagnostic methods using images of cell death and / or activation states in the eye.SOLUTION: The present invention relates to a method for determining the stage of a disease, in particular an ocular neurodegenerative disease such as Alzheimer's disease, Parkinson's disease, Huntington's disease and glaucoma, comprising the step of identifying the status of microglial cells in the retina and relating said status to the stage of the disease. Also provided are methods of identifying cells in the eye, labeled markers and uses thereof.SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] The present invention relates to diagnostic methods that use images of cell death and / or activation states, particularly in the eye. Regarding the law. [Background technology]

[0002] Cell death and neuronal loss are common in Alzheimer's disease (AD), Parkinson's disease, Huntington's disease, and AD is a major pathological driver of neurodegeneration in conditions such as Alzheimer's disease, Alzheimer's disease, and glaucoma. , the number of affected Americans will increase from 4 million to 12 million over the next 20 years. Glaucoma is the most common single form of dementia, with a predicted irreversible mortality rate worldwide. It is the leading cause of global blindness, affecting 2% of people over the age of 40. Its silent and progressive nature causes significant morbidity and often leads to delays in diagnosis and treatment.

[0003] Live-cell imaging can be used to investigate neuronal dysfunction in cultured cells in vitro. It is widely used to detect various cellular activities and specific molecular sites, along with fluorescent multiplexing. The inventors used labeled apoptosis markers to visualize the localization pattern. It is possible to observe retinal ganglion cell death using this method (Patent Document 1), and it is also possible to diagnose certain pathological conditions using this method. The present inventors have previously reported the usefulness of monitoring cell death using this method (Patent Document 2). The activation states of other cell types, especially microglial cells, can also be observed in the Furthermore, the inventors have discovered that the state of cells can be accurately monitored over a given period of time. We found that it is possible to monitor this. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2009 / 077790 [Patent Document 2] International Publication No. 2011 / 055121 Summary of the Invention

[0005] A first aspect of the present invention provides a method for determining the stage of a disease, particularly a neurodegenerative disease, comprising: The method identifies the activation state of microglial cells in a subject's eye and correlates the state of the cells with disease. The step of identifying the activation state includes a step of associating the activation state with the microglial cells. The method may include generating an image of

[0006] Microglial cells are found throughout the brain and spinal cord. These cells can be reactive or resting. Reactive microglia can be in a (ramified) state. Activated microglia include amoeboids, which are microglia that can become inflammatory cells. , antigen-presenting, cytotoxic, and inflammatory-mediating signaling capabilities, and phagocytosis of foreign substances. In addition, amoeboid microglia can also phagocytose foreign substances, but they lack antigen-presenting activity. Ramified microglia are not capable of phagocytosis.

[0007] The inventors found that reactive microglia could be distinguished from ramified microglia. Furthermore, the present inventors have demonstrated that amoeboid microglia, ramified microglia, or activated microglia The number and / or location of activated microglia can be used to provide an indication of the stage of the disease. The presence of activated microglia is generally associated with young and / or healthy individuals. On the other hand, the presence of amoeba is associated with disease, which may be predicted based on the subject's age or health status. A lower or lower percentage of activated microglia and / or a higher or higher number of activated microglia than If a percentage of amoeboid microglia is found, the subject has a neurodegenerative disease. These findings are indicative of a person's susceptibility to developing a neurodegenerative disease.

[0008] The above-described method detects activated microglia, ramified microglia, and and / or counting the number of amoeboid microglia. In addition, the above-described method can detect activated microglial cells, branched microglial cells, and The number or percentage of glial cells or amoeboid microglial cells was determined from the data obtained so far. or activated microglia, ramified microglia, or amoeboid microglia. This may include comparing with the expected number or percentage of glia. Expected numbers of differentiated, ramified, or amoeboid microglia or Percentages are predicted microglia numbers based on previous images of the same subject. or percentage, or detection in similar subjects or a large number of subjects of similar age The average number or average percentage of these microglia can also be taken.

[0009] The present inventors have also demonstrated that activated microglia, ramified microglia, and / or amoebae Microglial patterns can be linked to disease states and specific disorders. For example, the inventors have determined that healthy subjects have activated microglia throughout the retina. While subjects with neurodegenerative diseases have a regular and normal distribution of retinal activity, subjects with neurodegenerative diseases have a regular and normal distribution of retinal activity. Microglia with a diffuse, irregular pattern or numerous amoeboid microglia The present inventors have found that phagocytic microglia are likely to be involved in glaucoma. In AMD, these are located around the perimacula. Therefore, the pattern or patterns detected in the image The change indicates that the subject has a neurodegenerative disease or is indicative of a worsening or amelioration of the disease. The above-described method can identify patterns of cellular status in the eye and correlate the patterns with disease. The method may include associating the state of the

[0010] The status of microglia in the eye can be assessed by administering markers, particularly labeled markers, to the subject. Therefore, the subject can be identified by administering a labeled marker. Alternatively, the above method can also be used to target a labeled marker. The markers can be administered by any suitable method, particularly intravenously. It can be administered by injection, topically, or nasally.

[0011] The present inventors have confirmed that the above-mentioned labeled markers may be apoptosis markers. The term "apoptotic marker" refers to cells undergoing apoptosis. It refers to a marker that can distinguish cells from live cells and preferably necrotic cells. Apoptosis markers include, for example, the annexin family of proteins. is a protein that reversibly binds to cell membranes in the presence of cations. The annexins used can be natural or recombinant. The proteins mentioned above are whole proteins. Proteins or functional fragments, i.e., antibodies that specifically bind to the same molecule as the whole protein. It can be a fragment or part of a nexin. Functional derivatives may also be used. A variety of annexins are available, such as those described in U.S. Pat. No. 6,239,693. Preferred annexins is annexin 5, which is well known in the art. It can be used as an apoptosis marker. Other annexins that may be involved include annexins 1, 2, and 6. Other apoptosis markers include The C2A domain of synaptotagmin-I, the C2A domain of duramycin, isatin, non-peptide isatin sulfonamide analogs such as WC-Il-89, and Posens, e.g., NST-732, DDC, and ML-10 (Saint-Hubert el ai,2009).

[0012] The apoptotic marker is preferably labeled with a visible label. In particular, the label is preferably The term "wavelength-optimized label" refers to a label that emits light in response to excitation. It is a substance that has high sensitivity while complying with light exposure safety standards to avoid phototoxic effects. It is used for its improved signal-to-noise ratio and therefore improved image resolution and sensitivity. The optimized wavelengths include infrared and near-infrared wavelengths. Such labels are well known in the art and include IRDye700, IRDye800, D-776 and These dyes also include dyes such as D-781 and D-782. The optimized wavelength also includes fluorescent substances formed by conjugating the fluorescent substance to a molecule. Preferably, the wavelength-optimized label produces little or no irritation when applied. The pigment is D-776, which is safe for the eyes, whereas other pigments can cause inflammation. This is because it has been shown to cause little or no inflammation in The optimized dye preferably has a level of fluorescence that is detectable histologically and in vivo. This study reveals a close correlation between the level of histological and in vivo fluorescence. It is particularly preferred that there is a substantial correlation, especially a 1:1 correlation, between

[0013] In a particular embodiment, the marker is Annexin 5 labeled with D-776. The annexin 5 can be wild-type annexin 5 or modified annexin 5. In morphology, annexin 5 is a single molecule of annexin that allows accurate cell counting. The antibody has been modified to ensure conjugation with one molecule of a label.

[0014] Labeled apoptosis markers are conjugated with wavelength-optimized labels and marker compounds. Such labels can be prepared using standard techniques. The label can be obtained from well-known sources such as: Suitable techniques are known in the art and can be obtained from the label manufacturer.

[0015] The advantage of using apoptosis markers is that the above method is useful for detecting apoptosis and microglia. The inventors have also linked markers to To further distinguish between apoptotic cells and microglial cells that have phagocytosed the marker, To their surprise, they found that apoptotic cells that had bound markers could Generally, they appear as a circular ring with a hole in the center. They appear in the form of amoeboid microglia and can be recognized by their multiple processes. It is larger than activated microglia.

[0016] The step of generating an image of the state of the cells generates an image of apoptotic cells. The above-mentioned method may further include counting the number of apoptotic cells. and / or observing the pattern of apoptotic cells. The above method can be used to determine the number or pattern of apoptotic cells in a manner that is predictive of the number or pattern, or the number or pattern of apoptotic cells in images previously generated from the subject. The apoptotic cells may be, in particular, retinal ganglion cells. Retinal neurons, including retinal glial cells (RGCs), bipolar cells, amacrine cells, horizontal cells, and photoreceptor cells In one embodiment, the cells are retinal ganglion cells. The combined use of both apoptotic retinal neurons and microglial activation status allows for diagnostic Improve your performance.

[0017] By comparing specific cells over time, disease progression or the effectiveness of treatment can be monitored. It is particularly preferred that the method described above can be used to convert an image obtained at an earlier time point into a The method may further comprise comparing the image or more than one image of the subject's eye with the image or more than one image of the subject's eye. The above method can detect activated microglia and / or amyloid in a single image. This involves comparing the number or pattern of splenic microglia with previous images. and / or identify specific cells in one image relative to previous images. This can include comparing the same cells in the previous and subsequent images. Changes in the activation state of microglial cells during the course of the disease may indicate disease progression. The above method can be used to detect the number or pattern of apoptotic cells in an image, or The method compares specific cells to the same cells in previous images and also tracks disease progression or treatment. The method may include monitoring the effect of the placement of activated microglia or amoeboid microglia. Changes in the number or pattern of chromosomes and / or apoptotic cells may provide clinicians with a diagnosis of the disease. Amoeboid microglia and / or apoptosis, which can provide information about disease progression An increase in the number of cytosed cells may indicate progression of the disease. A decrease in the number of amoeboid or apoptotic cells may be seen. The number of cells seen in one image or one or more further images may be used. Comparison with the image can be used to identify the stages.

[0018] When comparing specific cells, one image can be precisely overlaid with the other. Advantageously, the above method can be used to generate one, two, three or more additional images. This step can be included with the following page.

[0019] The above-mentioned disease is preferably an ocular neurodegenerative disease. The term "ocular neurodegenerative disease" refers to It is well known to those skilled in the art and refers to a disease caused by the gradual and progressive loss of the ophthalmic nerve. is a common cause of glaucoma, diabetic retinopathy, AMD, Alzheimer's disease, Parkinson's disease, and multiple sclerosis. Including but not limited to sclerosis.

[0020] To generate an image of a cell, the labeled marker can be administered, for example, by intravenous injection, local administration, or The area of ​​interest to be imaged, the eye, is then imaged using an ophthalmoscope, particularly Located within the detection field of a medical imaging device such as a confocal scanning laser ophthalmoscope The emission wavelength of the labeled marker is then imaged to obtain a map of the area of ​​cell death. The generation of the image can be repeated over a period of time. This can be monitored in real time.

[0021] Specific treatment courses can be selected and optionally monitored, allowing for staging or diagnosing the disease. Therefore, the above method can be used to treat glaucoma or other neurological disorders. Treatment of degenerative diseases includes the treatment of glaucoma, which is well known in the art. Examples of treatments for the disorder are described in the detailed description of the invention. Other treatments may also be appropriate and should be considered by the skilled clinician. can be selected without difficulty.

[0022] The present invention also provides a method for identifying a microglial activation state, comprising administering to a subject a subject, comprising administering to a subject a method for identifying a microglial activation state, comprising administering to a subject ... We provide labeled apoptotic markers as described above.

[0023] The inventors have made further improvements to the method for identifying cells in retinal images. For example, the inventors have used ophthalmoscopes to visualize the state of cells in images generated using ophthalmoscopy. Improvements in the methods for monitoring have been made. In particular, cells are labeled with wavelength-optimized labels as described herein. Cell types of interest include, for example, microglia and retinal ganglion cells. The above method preferably comprises: (a) providing an image of a subject's retina; (b) Identify one or more spots in each image as potential labeled cells. Steps and (c) filtering the selection; and, optionally, (d) normalizing the results for intensity variations.

[0024] The spots can be identified in any suitable manner. Known methods such as template matching by convolution, thresholding (static or Dynamic (post-processing) connected component analysis, watershed detection, Laplacian-Gaussian, generalized Hough transform, and In one embodiment, the spots include a template mask and a spoke filter. It is identified by the timing.

[0025] The step of filtering the selection may, for example, optionally be performed using an autoencoder. static fixed threshold filters, decision trees, which can be calculated automatically , support vector machines, and random forests. Select based on the image or use the whole image and automatically calculated features to Using deep learning methods such as Lenet, Vgg16, ResNet, and Inception This can be done by any suitable method, including screening using

[0026] The above-described method may provide more than one image of the subject's retina, e.g., at different time intervals. The method may include providing an image captured within a few milliseconds. They can be taken seconds, minutes, hours, days, or even weeks apart. When using two image sequences, the method described above also allows cells seen in one image to be seen in another image. Steps to align the image to ensure alignment with the cells seen in the image. The inventors have used imaging to monitor the state of individual cells over time. We found that it is important to align the images of the retina. Therefore, it is extremely difficult to keep the retina in exactly the same orientation for each image. It is also necessary to accommodate physical differences in patient and eye position and orientation. Images taken at different times can be aligned to reveal changes in individual cells. To my surprise, I discovered that the step of aligning the images A stacking step may be included.

[0027] The above method may further include steps to accommodate known variations that may cause erroneous candidates to be identified. This reduces the possibility of misidentifying labeled cells, making it possible to identify specific retinal This means that the eigenvalues ​​or other variations can be taken into account. Power change, optical blur, alignment blur, and low-light noise, as well as choroidal vasculature patterns Including biological complexities such as turns, blood vessels, and blurring due to cataracts.

[0028] The steps of the above-described methods may be performed by any suitable mechanism or means. For example, this can be done manually or using automated methods. , in particular by automated means, for example using artificial neural networks. This can be done.

[0029] When the steps of the above-described method are performed by automated means, the automated means The method described above can be automated and trained to improve subsequent results. Spots identified or classified by such means may be reassessed by human observation or other automated means. and compare the identified or classified spots to better identify the candidate labeled cells. and further comprising using the results to train the first automated mechanism. can be done.

[0030] Step (a) may include imaging the subject's retina. can be imaged, for example, once, twice, three times, four times, five times or more.

[0031] The labeled cells may be microglial cells, retinal neurons, especially retinal ganglion cells, or both. This can be done in a more flexible manner.

[0032] In another aspect, the present invention also provides a method for detecting retinal markers, for example, to determine the stage of a disease. A computer-implemented method for identifying a state of a cell is provided, the method comprising: Like, (a) providing an image of a subject's retina; (b) Identify one or more spots in each image as potential labeled cells. Steps and (c) filtering the selection; and, optionally, (d) normalizing the results for intensity variations.

[0033] In another aspect, the present invention provides a method for detecting cells in the retina, e.g., to determine the stage of disease. A computer program for identifying the state is further provided, When executed by a processing system, causes the processing system to: (a) providing an image of the subject's retina; (b) Using template matching, candidate labeled cells are identified in each image. identifying one or more spots in the (c) Use object classification filters to identify objects without template matching. Screening the selections made, and optionally, (d) Normalize the results for intensity variations.

[0034] The methods for determining the stage of a disease described herein may include, in a processing system or processor, These methods can be implemented using computer processes operating in accordance with the above. A computer program, particularly on a carrier or It can be expanded to a computer program in a carrier. Non-transitory source code, object code, code intermediate source, and partially compiled In the form of object code, such as in the form of a file or in the implementation of the processes described herein. The carrier may be in any other non-transitory form suitable for use in the The program may be any entity or device that can include a program. For example, The above-mentioned carrier may be a solid-state drive (SSD) or other semiconductor-based RAID. M, ROM, such as CD ROM or semiconductor ROM, magnetic recording medium, such as floppy disk This may include storage media such as disks or hard disks, optical memory devices in general, etc. .

[0035] In embodiments, when executed by a processing system, the processing system is configured to stage the disease. A non-transitory computer-readable storage device that stores a set of computer-readable instructions that cause the device to perform a method for determining a computer-readable storage medium, the method comprising: Identifying one or more spots in one or more images of the subject's retina as candidate cells. Step 1: Using object classification filters, template matching A step of filtering the selections made and a step of normalizing the results for intensity variations. The examples given include computer software stored in (non-transitory) memory. may be implemented at least in part by a processor, or hardware, or Physically stored software and hardware (and physically stored firmware) The above-described method may be carried out by a combination of the above-described methods. The method may include providing an image of the

[0036] The invention will now be described in detail, by way of example only, with reference to the accompanying drawings, in which: [Brief explanation of the drawings]

[0037] [Figure 1] FIG. 1 shows microglia in both eyes (OHT and IVT) of naive rats and glaucoma model rats. [Figure 2] Figure 2 shows microglia in the aged and IVT Alzheimer's 3xTG mouse model. [Figure 3] FIG. 3 shows DARC and Alzheimer 3×TG mouse models of middle age and IVT. [Figure 4] FIG. 4 shows microglial staining with Annexin V. [Figure 5] FIG. 5 shows the results of intranasal DARC in the DARC and Alzheimer 3×TG mouse model. [Figure 6] FIG. 6 is a Consort schematic showing glaucoma and control cohort subjects and DARC image analysis. [Figure 7]Figure 7 is a CNN-assisted algorithm flowchart showing the analysis stages of the DARC image. [Figure 8] Figure 8 shows a representative retinal image of possible spot candidates. Candidate spots were detected using template matching and correlation maps. Local maxima were selected and filtered by thresholding the correlation coefficient and intensity standard deviation (corresponding to the spot brightness). This threshold was set very low, generating many more spot candidates (approximately 50-1) than the number of spots observed by humans. [Figure 9] Figure 9 shows the CNN training and validation stages. The CNN training (A) and validation (B) curves are shown. While good accuracy was obtained at 200 epochs (training cycles), training was extended to 300 epochs to verify stability. The matching validation accuracy also shows similar accuracy without any signs of overtraining. The accuracy was found to be 97%, with a sensitivity of 91.1% and a specificity of 97.1%. [Figure 10] Figure 10 shows a representative comparison of DARC spots between human observation and the CNN algorithm. Spots found by the CNN and by at least two human observers are shown in the original retinal image. (A) Patient 6, left eye. Progressive glaucoma (measured by OCT total RFNL 3.5 rings). (B) Patient 31, left eye. Stable glaucoma. Green circles represent human observation only (false negatives), blue circles represent CNN-assisted algorithm only (false positives), and cyan circles represent agreement between the algorithm and human observation (true positives). [Figure 11]Figure 11 shows receiver operating characteristic (ROC) curves for glaucoma progression in analyses of human observation and CNN algorithm. To test the predictive value of glaucoma progression at 18 months, receiver operating characteristic (ROC) curves were constructed for both the CNN-assisted algorithm (A) and human observation with at least two agreements (B). The rate of progression (RoP) was calculated from Spectralis OCT whole retinal nerve fiber layer (RNFL) measurements at 3.5 mm of the optic disc in 18-month follow-up of glaucoma patients after DARC. Patients with a significant (p<0.05) negative slope were defined as progressing compared to patients without a significant negative slope, who were defined as stable. The highest sensitivity (0.85%) and specificity (71.43%) were obtained with an AUC of 0.79 at a DARC count of 12. The highest sensitivity (90.0%) and specificity (85.71%) were obtained with the CNN algorithm with an AUC of 0.89 at a DARC count of 23, relative to the human observation counts, demonstrating that the CNN-assisted algorithm performed well. [Figure 12] Figure 12 shows that CNN DARC counts were significantly increased in patients with progressing glaucoma compared with patients with stable glaucoma. (A) Using a CNN-assisted algorithm, CNN DARC counts were significantly higher in patients with progressing glaucoma at 18 months (mean 26.13) compared with patients with stable glaucoma (mean 9.71) (p=0.02). DARC counts were determined as the number of ANX776-positive spots in retinal images at 120 minutes after subtraction of baseline spots. (B) The trend was similar to that of manual observation (≥2 agreement), but there was no significant difference between patients with progressing glaucoma at 18 months (mean 12.25) compared with patients with stable glaucoma (mean 4.38) (p=0.0692). Boxplots showing individual data points in patients with and without significant RoP as measured by OCT total RFNL3.5 rings are shown. Asterisks indicate the level of significance according to the Mann-Whitney test. Horizontal lines indicate the median and minimum and maximum ranges and represent all individual data points. DETAILED DESCRIPTION OF THE INVENTION

[0038] Example 1 Labeled annexin V was prepared as described in WO 2009 / 077750 The labeled annexin was analyzed as described in Cordeiro MF, Guo L, Luong V, et al.,Real-time imaging of single nerve cell apoptosis in retinal neurodegenera tion.Proc Natl Acad Sci USA 2004; 101: 1 The test was administered as described in J. Med. Chem. Soc. 1999, 3352-13356.

[0039] Iba-1 (ionized calcium-binding adaptor molecule 1 (Iba1)) has been reported in the art. It was used as a marker for microglia using known techniques.

[0040] Brn3a was used as a marker for retinal ganglion cells using techniques known in the art. did.

[0041] The animals used were naive rats, glaucoma model rats (OHT), and Alzheimer's disease model rats. These included mouse models of glaucoma, and mouse models of glaucoma. Such models are well known in the art. An example is given in WO 2011 / 055121.

[0042] Figure 1 shows (a) naive control and (b) one eye with surgically elevated IOP. (c) from the opposite eye of a rat (ocular hypertension OHT model), (d) from the OHT eye of the same animal. The results of immunostaining (Iba1) in whole mount rat retina are shown. and amoeboid microglia can be identified.

[0043] Figure 2 shows the network of a 16-month-old Alzheimer's triple transgenic mouse. Iba1 was used to identify microglia in membrane whole mounts. After intravitreal (IVT) injection, the morphology of microglia was characterized as amoeboid microglia. In contrast, uninjected eyes (no IVT) at the same age show activated morphology.

[0044] As can be seen in Figure 3, we used the same stain, labeled annexin. We found that both retinal ganglion cells and microglia could be identified using this method. As shown, both RGC and microglial staining appeared to colocalize with Annexin 5. In Figure 4, annexin 5 is a 488 fluorescent full-cell marker detectable by histological microscopy. The RGCs are fluorescently labeled with fluorophores. Annexin staining occurs around the cells, while intracellular staining occurs within the cells. do not have.

[0045] As can be seen in Figures 4 and 5, the staining of microglia with Annexin V was The annexins are either intracellular or, as seen in RGCs, external to the cell membrane. be.

[0046] Example 2 Artificial intelligence is increasingly being used in medicine, especially in ophthalmology (Popl (Ting et al., 2018) (Ting et al., 2019). Algorithms have become important analytical aids in retinal imaging, and their use has led to diagnostic Diabetic retinopathy, advanced retinopathy, and other conditions that are believed to optimize both sensitivity and specificity for diagnosis and surveillance. It is often recommended in the management of age-related macular degeneration and glaucoma (Seba stian A Banegas et al.,2015)(Quellec et al. al.,2017;Schmidt-Erfurth,Bogunovic,et al. ., 2018;Schmidt-Erfurth,Waldstein,et al. ,2018; Orlando et al.,2019). Deep learning in blindness pathology Its use has been heralded as an advancement to reduce its health and socio-economic impact. However, its accuracy is hampered by the size of the dataset and inadequate reference standards. (Orlando et al., 2019).

[0047] Glaucoma is a progressive, slowly progressing ocular neurodegenerative disease that is a worldwide irreversible It is the leading cause of blindness, affecting 60.5 million people and a growing elderly population is expected to double by 2040 as n, 2006; Tham et al., 2014). The main objective is to identify people at risk of rapid progression and blindness. This includes: Structural (optical coherence tomography (OCT), papillary imaging) and functional (visual field or standard automated visual field) The study included methods involving multiple levels of data, including assessment of the effectiveness of the test (SAP). However, in some studies, clinicians have used SAP, OCT, and optic nerve mammograms. There is wide variability in opinion regarding progress using standardized assessments, including stereo cephalography. (AC Viswanathan et al., 2003) (More no-Montanes et al.,2017)(Sebastian A Ban egas et al., 2015). However, clinical grading is Considered the gold standard in clinical practice and in deep learning datasets (Jiang et al., 2018; Kucur, Hollo and Sznitman,2018;Asaoka et al.,2019a;Ian J C MacCormick et al.,2019;Medeiros, Jammal and Thompson,2019a;Thompson, Jammal and Medeiros, 2019; Wang et al., 2019). Furthermore, OCT Both the retinal ganglion cells (RGCs) and the SAP are altered only after the significant death of a large number of retinal ganglion cells (RGCs). It has been recognized that the There is an unmet need for earlier markers.

[0048] Recently, a method called DARC (detection of apoptotic retinal cells) has been developed in the human retina. reported a new method for visualizing apoptotic retinal cells in vitro (Cordeiro et al., 2011). The molecular marker used in the above technique is fluorescently labeled annexin. A5, which appears on the surface of stressed cells undergoing the early stages of apoptosis. It has a high affinity for phosphatidylserine, which is a phospholipid. The number of DARC-positive staining cells in the membrane fluorescence image was used to assess glaucoma disease activity. It may also be possible to correlate future glaucoma disease progression, even in a small number of patients. DARC has recently shown further promise in Phase 2 clinical trials. It has been studied in subjects with

[0049] Herein, a CNN was used to develop and train a control cohort of subjects. After developing the drug, DARC was tested in glaucoma patients in a Phase 2 clinical trial. We describe an automatic method for spot detection. CNNs are widely used in medicine, including medical image classification. It has demonstrated high performance in computer vision tasks.

[0050] Materials and Methods [Contributor] The Phase 2 clinical trial of DARC is being conducted at Western Eye Hospital, Imperial College Healthcare At Scare NHS Trust, between February 15, 2017 and June 30, 2017, Each elephant received a single intravenous injection of fluorescent annexin 5 (ANX776, 0.4 mg). This was a single-center, open-label study. Both glaucoma and glaucoma patients were recruited for the study, and the study was approved by the Brent Research Ethics Committee. After the study, informed consent was obtained in accordance with the Declaration of Helsinki (ISRCTN107518 59).

[0051] All glaucoma subjects were already receiving treatment at the Glaucoma Department at Western Eye Hospital. The subjects were free of any ocular or systemic diseases other than glaucoma and had retinal optical coherence tomography (Spectr alis SD OCT, software version 6.0.0.2; H eidelberg Engineering, Inc., Heidelberg,G ermany) and Swedish interactive threshold a Standard automated perimetry (SAP) using the Igorithm Standard 24-2 HFA 640i, Humphrey Field Analyzer; Carl Ze A minimum of three consecutive evaluations by the NIH Meditec, Dublin, CA Patients were considered for inclusion in the study if they had previously undergone the procedure. Patient eligibility is summarized in Tables 1 and 2. Evidence of progressive disease in at least one eye for any parameter progression is considered present if found to be present, and progression is considered to be a significant ( *p<0.05; **p<0.01) determined by negative slope. SAP parameter OCT parameters included visual field index (VFI) and mean deviation (MD). The results were compared with the retinal images at three different optic disc diameters (3.5, 4.1, and 4.7 mm). This includes the RNFL measurements and the minimum rim width (MRW) at the edge of the Bruch membrane. The pre-intervention period of evaluation allowed for machine-in-built software to determine the progression rate. If software is not available, the linear rate of change for each parameter over time can be determined as best as possible. Calculated using the least squares method (Wang et al., no date; Pathak , Demirel and Gardiner, 2013).

[0052] Healthy volunteers were initially assigned to accompany patients to the clinic and to the PIC. Healthy volunteers were recruited from local optometry services that function as a networking From the database of healthy volunteers at Royal College Healthcare NHS Trust Potential collaborators were contacted and invited to participate. The research team contacted PIC participants who agreed to participate in the study and told them about the study. A series of tests were conducted according to the inclusion and exclusion criteria selected by the inventors. Once participants were deemed eligible, they were enrolled. Briefly, healthy subjects: If there is no ocular or systemic disease as recognized by their GP, the optic disc, RNFL (retinal Any glaucoma with either a glaucoma-like defect (nerve fiber layer) or visual field abnormality and normal IOP (intraocular pressure) In cases where there is no evidence of a sexual process and where repeatable and reliable imaging and visualization are required If yes, accepted.

[0053] [DARC Image] All participants were given anesthetic medication after pupil dilation (1% tropicamide and 2.5% phenylephrine). Patients received a single dose of 0.4 mg of ANX776 via intravenous injection, with the same treatment regimen as in Phase 1. The protocol was evaluated using the protocol (Cordeiro et al., 2017). In other words, the retinal image is captured by cSLO (HRA + OCT Spectralis, H eidelberg Engineering GmbH, Heidelberg, Ge rmany) in high resolution mode, ICGA infrared fluorescence setting (diode The images were acquired using a laser with 786 nm excitation and a photodetector with an 800 nm barrier filter. Baseline infrared autofluorescence images were taken before ANX776 administration, and then after ANX776 administration. Images were acquired at the time of injection and at 15, 120, and 240 minutes after injection. An average image of the sequence was recorded at each time point. All images were analyzed before any analysis was performed. For the development of the CNN algorithm, baseline data from control and glaucoma subjects were collected. Only images from the 120 minute video were used.

[0054] The classification of the analyzed images is shown in the "Consort" diagram in Figure 6. Therefore, 73 control eyes at 120 minutes were available for analysis. Among 20 glaucoma patients who received ANX776, baseline and 120-minute time points Images were available in 27 eyes.

[0055] [Analysis of human observations] The anonymized images were randomly displayed on the same computer under the same lighting conditions. ImageJ(National Institutes of Mental Health Image review was performed by five blinded personnel using a 3D scanner (Illegible, USA). (ImageJ, undated). ImageJ's "multipoint" tool allows the observer to Use this to identify each structure in the image that you want to label as an ANX776-positive spot. Each positive spot was identified by vector coordinates. The spots from different observers were compared if they were within 30 pixels of each other. If there was agreement between two or more observers, this was considered to be the same spot. Used within automated applications as a spot reference for training and comparing programs did.

[0056] Automated Image Analysis Overview (Figure 7) To detect DARC-labeled cells, candidate spots were identified in the retinal image and analyzed. Using an algorithm trained using candidates and spots identified by observation The samples were then classified as either "DARC" or "non-DARC." Figure 7 shows an overview of the process.

[0057] (A) Image optimization For each eye, 120 min of images were transformed using an affine transformation followed by a non-rigid transformation. The image was aligned with the baseline image. The image was then aligned with the alignment arch. The image was then cropped to remove the artifacts. To accommodate this, the intensity was normalized by z-scoring each image. The noise was removed from the image by Gaussian blurring with a sigma of 5 pixels.

[0058] (B) Spot candidate detection Template matching, especially zero normalized cross correlation (ZNCC), identifies candidate spots. A simple method for finding the 30x30 pixel area of ​​the spot identified by human observation. The images are combined using the average image function to create a spot template. This template was applied to the retinal image to create a correlation map. The maximum value is selected, and the correlation coefficient and the intensity standard deviation (corresponding to the brightness of the spot) are thresholded. These thresholds were set sufficiently high to include all spots seen by human observation. Some of the human observations were extremely faint (probably spot on). In the perfectly differentiated spots, the correlation was low due to their proximity to blood vessels. This is because the threshold is set very low and there are many more spots than observed by humans. This means that we had to generate candidates (approximately 50-1).

[0059] As can be seen from Fig. 8, the spot candidates cover most of the retinal image, which is Reduces the number of points to classify by a factor of 1500 (compared to looking at every pixel). Using the maximum value, each candidate detection typically appears as a spot with the brightest spot in the center. It is object-centered, meaning that classification does not need to tolerate off-center spots. This also means that the accuracy of the classification measurement depends on the randomness of the DARC spots in the image. Not only can it distinguish DARC spots from other spot-like objects, but it can also distinguish DARC spots from other spot-like objects. This means that the quality of the work is more meaningful because it reflects the ability to

[0060] (C) Spot classification To determine which spot candidates are DARC cells, the spots were analyzed using the Mobile It uses a well-known convolutional neural network (CNN) called Net v2. (Sandler et al., 2018; Chen et al., 20 19;Pan, Agarwal and Merck,2019;Pang et al. .,2019). This CNN processes over 400 spot images in a single batch. Each batch is expected to have approximately four DARC spots. This allows us to deal with imbalanced data of 50-1.

[0061] We used the MobileNet v2 architecture, but modified the first and last layers. The first layer takes a 64x64 pixel spot candidate image, so the input is 64x64x1. layer (this size is used to give some context to the network) (We chose to include additional regions around the image.) The final layer is a binary classification layer rather than a multi-classification layer. Replaced with stratum densa containing sigmoid activation to allow (DARC spot or not) We found that the α value of 0.85 for MobileNet was optimal, and the filter The number of arrows has been adjusted appropriately.

[0062] (D) Training Only the control eye was trained. Briefly, the retinal image was recorded at 50% Randomly selected from 120 minutes of images from control patients. The DARC was trained using candidate spots that were identified as spots by human observation. 58,730 spot candidates were extracted from these images (1022 (including DARC spots observed by two people who agreed). 70% of these spots 30% were used for training and 30% for validation. The remaining 50% were retinal images from control patients. The data were used to test the classification accuracy (48610 candidate spots, of which 8 98 were observations by two concordant people).

[0063] Increase network tolerance with rotation, reflection, and intensity changes of spot images To correct for the 50-1 imbalanced data, the DARC The spot class weights were set to 50 for spots and 1 for other objects.

[0064] Additionally, the training validation accuracy increased, and the matching validation accuracy remained similarly accurate without any signs of overtraining. As the training curve shows (see Figure 9), good accuracy was obtained in 200 epochs. While training was performed for 300 epochs to verify stability.

[0065] Three training steps were performed to create three CNN models. For inference, we used the three models. Combined, each spot is classified based on the average probability obtained by each of the three models. did.

[0066] (E) Testing in Glaucoma DARC Images Developing a CNN-assisted algorithm has demonstrated that baseline The test was conducted on images acquired at 120 minutes and 120 minutes. DARC counts were determined by the 12-well platelet count after subtracting the baseline spot. It was defined as the number of ANX776-positive spots seen in the retinal image at 0 min.

[0067] [Glaucoma progression evaluation] Progression rates were calculated from serial OCT scans of glaucoma patients 18 months after DARC. (p<0.05) Those patients with a negative slope were defined as stable. Patients were defined as having progressed compared to those without the condition. and optic disc measurements were performed by five blinded clinicians.

[0068] [result] [Patient demographics] Sixty glaucoma patients were screened according to established inclusion / exclusion criteria and progressed. Active glaucoma (significant (p<0.05) of any parameter in at least one eye) Twenty patients with 05) (defined by a negative slope) received intravenous DARC. The baseline characteristics of these glaucoma patients are shown in Table 2. Thirty-eight eyes were eligible for inclusion. Three of them had no images available for human observation counting, and two were low resolution. One had images acquired in the Fluorescence Imaging mode, and the other two had significant intrinsic autofluorescence. All patients except for 1 were followed up at the ophthalmology clinic and data were collected for post-hoc assessment of progression. This has been made available for use.

[0069] [Spot classification test] The results in Figure 9 show the 50% of controls that were reserved for testing (and not used for training). When testing the CNN-assisted algorithm on the eyes of a roll, an accuracy of 97% was obtained. The sensitivity was 91.1% and the specificity was 97.1%.

[0070] Sensitivity and specificity are important, especially when data from observations by trained and tested personnel demonstrate a high level of observer accuracy. This was encouragingly high, as it showed variability between A typical example of an image and human observation / algorithm spot is shown in FIG.

[0071] [Classification study in a glaucoma cohort] OCT total RNFL progression rate (RoP 3.5 rings) was performed at 18 months to define progression. ) was used to divide the glaucoma cohort into progressive and stable groups. Inter-observer agreement is not good, therefore, a single OCT scan is not suitable for objective and simple The parameters were used. Those patients with a significant (p<0.05) negative slope were included. , defined as progressing compared to patients without it who were defined as stable. 3a. Of the 29 glaucoma eyes analyzed, 8 progressed by this definition. It was found that 21 was stable.

[0072] Using this definition of glaucoma progression, DARC counts were used to determine whether glaucoma progression occurred at 18 months. To examine whether the CNN-assisted algorithm predicts the Both rhythm and two-person agreement of human observations were constructed and shown in Figure 6. 71.4%) and specificity (87.5%) for DARC counts greater than 12. The highest sensitivity (85.7%) and specificity for human observation was obtained with an AUC of 0.79. The CNN algorithm achieved a high degree of accuracy (91.7%) for DARC counts above 24. The CNN-assisted algorithm performed well, with an AUC of 0.88. showed.

[0073] [DARC count as a predictor of glaucoma progression] DARC in both stable and progressive glaucoma groups with a CNN-assisted algorithm Counts are shown in Figure 7a, and human DARC counts (consensus between two observers) are shown in Figure 7b. DARC counts were calculated using a CNN-assisted algorithm to identify patients with stable disease (average 9 0.71) compared with patients who were later found to be progressing at 18 months (mean 26.1 3) was found to be significantly more prevalent (p=0.02, Mann Whitney). Then, the DARC counts of human observations (two or more people agreeing) stabilized (average 4.3 8) Compared with glaucoma patients, patients with progression at 18 months (mean 12.25) Although the results were higher than those of the control group, this did not reach statistical significance (p=0.0692, Man n Whitney).

[0074] [Consider] The primary goal in managing glaucoma is to prevent vision loss. Progression is gradual, and the current gold standard for assessing change takes time to develop. Not only that, but also after significant structural and functional impairment has already occurred (Cordeiro e (t al., 2017). Evaluate the risk of future progression and the effectiveness of treatment in glaucoma. There is an unmet need for reliable measures of fman, 2009, 2011). Herein, markers of retinal cell apoptosis When combined with DARC, which is a marker of RNFL thinning, OCT showed significant improvement after 18 months. A new CNN-assisted algorithm is described that can predict the progression of defined glaucoma. This method, when used in conjunction with DARC, provides an automated, objective biomarker. It is possible.

[0075] The development of surrogate markers is primarily focused on cancer, where they are used as predictors of clinical outcome. In glaucoma, the most common clinical outcome measure is visual acuity loss, which evaluates treatment efficacy. and subsequent decline in quality of life. This would allow for longer-term treatment and also shorter, and therefore more cost-effective, clinical trials. However, to be a valid surrogate marker, the measurement must be shown to be accurate. For example, OCT has shown good repeatability in widespread use (DeLeon Ortega et al.,2007)(Tan et al.,2012), Author It has a sensitivity and specificity of 83% and 88%, respectively, for detecting small RNFL abnormalities ( Chang et al., 2009). In comparison, the CNN of this application The algorithm had a sensitivity of 85.7% and a specificity of 91.7% for glaucoma progression. was.

[0076] Phase 1 results suggested some expected levels of DARC This was done at various doses of Anx776, 0.1, 0.2, 0.4, and 0.5 mg. There were a maximum of four glaucoma cases per eye, with only three in the 0.4 mg group. This was done on a very small data set (Cordeiro et al., In this study, all subjects received 0.4 mg of Anx776. , 27 eyes were analyzed.

[0077] In clinical practice, glaucoma patients are at risk for progression due to factors such as age, elevated intraocular pressure (IOP), , too high in such individuals), ethnicity, positive family history of glaucoma, stage of disease, and severity The assessment is based on identifying the presence of risk factors, including myopia (Jonas et al., 2011). (t al., 2017). The risk of more severe disease is increased by a cup:nipple ratio of >0.7, 0. The standard deviation of the visual field pattern per 2 dB increase, binocular involvement and disc asymmetry, and disc hemorrhage and These included the presence of eczema and pseudoexfoliation (Gordon et al., 2002, 2003 ;Budenz et al.,2006;Levine et al.,2006;M However, none of these measures are related to the individual's progress. Cannot be used to reliably predict rows.

[0078] Objective assessment is difficult due to the variability of opinion among clinicians, even with technical support. is increasingly recognized as important in glaucoma. has been shown to determine patient progress using visual fields, OCT, and stereo imaging. (AC Viswanathan et al., 2003) (Sebasti an A. Banegas et al.,2015)(Blumberg et al. .,2016)(Moreno-Montanes et al.,2017). In this study, five blinded senior glaucoma specialists (co-authors) were asked to perform optic disc evaluation, O Using CT and clinical judgment based on the field of view, the patient's progress is graded. Unfortunately, there was a wide range of opinions among them (see unpublished data). data). Therefore, a single objective measure of progression rate (Tatham and Medei ros, 2017) to define the groups used to test the CNN-assisted algorithms. I agreed.

[0079] The progression analysis was post hoc and included 18 months of follow-up followed by guiding the treating clinician. As with the oral memantine trial (Weinreb et al. al., 2018), leaving patient management, especially with regard to IOP reduction, to the discretion of the glaucoma specialist. However, despite this, the OCT showed a complete R Using FNL3.5 ring RoP, 8 of 29 eyes had progressed at 18 months .

[0080] Due to variations in opinion among clinicians in identifying progression, optic disc photography has become increasingly common in recent years. (Jiang et al.,2018)(Ian JCMacCormick e t al., 2019)(Thompson, Jammal and Medeiros , 2019), visual field (Pang et al., 2019) (Kucur, Hollo and Sznitman, 2018), and OCT (Asaoka et al., 2 019b)(Medeiros, Jammal and Thompson, 2019b) There is a great interest in the use of artificial intelligence to assist in glaucoma diagnosis and prognosis using AI, along with A recent study by Medeiros et al. The estimated RNFL thickness was predicted by training a CNN using the values. An algorithm for evaluating bottom photographs was described (Medeiros, Jammal and d Thompson, 2019b). At 95% specificity, the predicted measurement was 76 % sensitivity, while actual SD OCT measurements had a sensitivity of 73%. At a specificity of 90%, the predicted measurements are 90% more accurate than OCT measurements, which have a sensitivity of 90%. The authors concluded that their method was effective in extracting progression information from photographs of the optic disc. However, as in the present study, long-term data They state that further validation with datasets is needed.

[0081] Template matching is commonly used for tracking cells in microscopy. Similar evaluations are required for long-term in vivo single-cell analysis in this study. For template matching of fine writing, a 30x30 pixel template is used, and the CNN The image used was 64x64 pixels. The reason for this difference in size is that the template match Since the chip is highly sensitive to blood vessels, a small template is used to reduce the possibility of including blood vessels. In CNN, the number of pixels around the spot that can be useful for classification is Larger images are useful because they give the CNN more context about the region.

[0082] The algorithm performed well, detecting progressive glaucoma 18 months earlier than other methods. While this provides a viable method for achieving this, we believe there are areas where it can be optimized. The parts are listed below.

[0083] MobileNet such as Support Vector Machine (SVM) or Random Forest Alternative classification algorithms to V2 include nonlinear intensity changes, optical blur, and alignment blur. Complexities due to image acquisition, such as low-light noise, and the pattern of choroidal vasculature The need to address biological complexities such as blurring caused by glaucoma, blood vessels, and cataracts It requires "handcrafted" features that are difficult to create. The network There is some bias in the intensities of the samples. Attention to intensity normalization and a larger data set may be necessary. By extending the data with more realistic intensity variations using a set of We believe that this can improve the performance of other networks such as VGG16. When I wrote that I found that BileNetV2 performed best, I evaluated this. We are continuing to evaluate whether our network is best suited to this requirement. The CNN VGG16 can be limited to 64 spots in a batch, which This may mean that the stencil does not have DARC spots, preventing training. Detect and classify spots in a single step using the YOLO3 algorithm. There are other similar methods that are more effective and useful with more data. Although we believe this could be a viable method, at this stage, the best accuracy obtained with YOLO is based on this paper. It is not as good as the method described in

[0084] [Conclusion] In this study, retinal images of patients with glaucoma showed that markers of retinal cell apoptosis and We describe a CNN-assisted algorithm to analyze DARC using the It allows the calculation of DARC counts, which when tested in patients, range from 18 to OCT RNFL was found to successfully predict glaucoma progression after 1 month. We support the use of this method to provide an automated, objective biomarker with potential clinical applications. It is something that is attached.

[0085] [Table 1] [Table 1b] [Table 2] [Table 3a] [Table 3b]

[0086] References TIFF2026021537000007.tif217152TIFF2026021537000008.tif224159TIFF2026021537000009.tif20715 0TIFF2026021537000010.tif215155TIFF2026021537000011.tif211150TIFF2026021537000012.tif59159

Claims

1. 1. A labeled apoptosis marker for use in a method for determining the stage of a disease, particularly a neurodegenerative disease, comprising: The method comprises: (a) generating an image of the activation state of microglial cells in the eye of a subject; (b) correlating said state of said cell with a stage of disease; A labeled apoptotic marker, wherein the subject is a subject to which the labeled marker has been administered.

2. 1. Use of a labeled apoptosis marker in the preparation of a diagnostic agent for use in a method for determining the stage of a disease, in particular a neurodegenerative disease, comprising: The method comprises: (a) generating an image of the activation state of microglial cells in the eye of a subject; (b) correlating said state of said cell with a stage of disease; The subject is a subject to which a labeled marker has been administered.

3. The method comprises: (c) counting the number of activated microglia, ramified microglia, and / or amoeboid microglia in the generated image; (d) comparing the number or percentage of activated microglial cells, ramified microglial cells, or amoeboid microglial cells detected in the image with a previously obtained image or with an expected number or percentage of activated microglia, ramified microglia, or amoeboid microglia.

4. 4. The labeled apoptosis marker of claim 1 or 3 or the use of claim 2 or 3, wherein the method further comprises identifying a pattern of cellular status in the eye and correlating the pattern with a disease state.

5. The labeled apoptosis marker according to any one of claims 1, 3 and 4 or the use according to any one of claims 2 to 4, wherein the labeled marker is an apoptosis marker, in particular a labeled annexin, more particularly annexin 5.

6. A labeled apoptosis marker according to any one of claims 1 and 3 to 5 or a use according to any one of claims 2 to 5, wherein the label is a visible label, in particular a wavelength-optimized label, more particularly D-776.

7. The labeled apoptotic marker of any of claims 1 and 3 to 6 or the use of any of claims 2 to 6, wherein the method further comprises the steps of generating an image of apoptotic cells, and optionally counting the number of apoptotic cells and / or observing the pattern of apoptotic cells, and optionally comparing the number of apoptotic cells or the pattern of apoptotic cells with an expected number or pattern, or with a number or pattern of apoptotic cells in images previously generated from the subject.

8. The method comprises: comparing the image with one or more images of the subject's eye obtained at a previous time point; comparing the number or pattern of activated microglia and / or amoeboid microglia in one image with previous images; comparing a particular cell in one image with the same cell in a previous image; and The labeled apoptosis marker of any of claims 1 and 3 to 7 or the use of any of claims 2 to 7, further comprising one or more steps of comparing the number or pattern of apoptotic cells, or specific cells, in one image with the same cells in a previous image.

9. 9. A labeled apoptotic marker or use according to claim 8, wherein the method comprises the step of overlaying one image with one, two, three or more further images.

10. The labeled apoptosis marker according to any one of claims 1 and 3 to 9 or the use according to any one of claims 2 to 9, wherein the disease is an ocular neurodegenerative disease.

11. 11. A labeled apoptosis marker according to any of claims 1 and 3 to 10 or a use according to any of claims 2 to 10, further comprising the step of determining an appropriate treatment for said subject and / or administering to said subject a treatment, in particular for glaucoma or other neurodegenerative diseases.

12. Labeled apoptotic markers for use in identifying microglial activation states.

13. 1. A method for determining the stage of a disease, particularly a neurodegenerative disease, comprising: (a) generating an image of the activation state of microglial cells in the subject's eye; (b) correlating said state of said cell with a stage of disease.

14. (c) counting the number of activated microglia, ramified microglia, and / or amoeboid microglia in the generated image; 14. The method of claim 13, further comprising one or both of the steps of: (d) comparing the number or percentage of activated, ramified, or amoeboid microglial cells detected in the image with a previously obtained image or an expected number or percentage of activated, ramified, or amoeboid microglial cells.

15. 15. The method of claim 13 or 14, further comprising identifying a pattern of cellular status in the eye and relating the pattern to a disease state.

16. The method according to any one of claims 13 to 15, wherein the subject is a subject to which a labeled marker has been administered.

17. The method of any of claims 13 to 15, further comprising administering to the subject a labeled marker.

18. 18. The method according to claim 16 or 17, wherein the labeled marker is an apoptosis marker, in particular a labeled annexin, more particularly annexin 5.

19. 19. The method of claim 16, 17 or 18, wherein the label is a visible label, in particular a wavelength-optimized label, more particularly D-776.

20. 20. The method of any of claims 13 to 19, further comprising the steps of generating an image of apoptotic cells, and optionally counting the number of apoptotic cells and / or observing the pattern of apoptotic cells, and optionally comparing the number or pattern of apoptotic cells to an expected number or pattern, or to a number or pattern of apoptotic cells in images previously generated from the subject.

21. comparing the image with one or more images of the subject's eye obtained at a previous time point; comparing the number or pattern of activated microglia and / or amoeboid microglia in one image with previous images; comparing a particular cell in one image with the same cell in a previous image; and 21. The method of any of claims 13 to 20, further comprising one or more of the steps of comparing the number or pattern of apoptotic cells, or specific cells, in one image with the same cells in a previous image.

22. 22. The method of claim 21, comprising superimposing one image with one, two, three or more additional images.

23. The method according to any one of claims 13 to 22, wherein the disease is an ocular neurodegenerative disease.

24. 24. The method of any of claims 13 to 23, further comprising the step of determining an appropriate treatment for the subject and / or administering to the subject a treatment, particularly for glaucoma or other neurodegenerative diseases.

Images