Hyperspectral Imaging Guided Raman Ocular Imaging for Alzheimer's Disease Pathology

A non-invasive ocular light-based detection system using hyperspectral imaging and Raman spectroscopy addresses the limitations of current AD diagnosis methods by providing early and specific detection of AD-related pathologies in the eye, facilitating rapid screening and treatment monitoring.

JP7595132B2Active Publication Date: 2024-12-05RETISPEC INC
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
JP2023151505
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2017-11-27
Filing Date
2023-09-19
Publication Date
2024-12-05
Estimated Expiration
2038-11-27

AI Technical Summary

Technical Problem

Current methods for diagnosing Alzheimer's disease (AD) are invasive, costly, complex, and difficult to access, making early detection and screening of asymptomatic stages challenging.

Method used

A non-invasive ocular light-based detection system using hyperspectral imaging and Raman spectroscopy to identify AD-related pathologies in the eye, without the need for exogenous fluorescent agents or tracers, by analyzing retinal reflections and Raman scattered light to detect specific chemical compositions indicative of AD.

Benefits of technology

Provides a highly sensitive and specific method for early detection and tracking of AD-related pathologies, enabling rapid pre-screening and monitoring of treatment effectiveness with minimal invasiveness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007595132000001
    Figure 0007595132000001
  • Figure 0007595132000002
    Figure 0007595132000002
  • Figure 0007595132000003
    Figure 0007595132000003
Patent Text Reader

Abstract

SOLUTION: A non-invasive ocular light-based detection device for detecting Alzheimer's Disease (AD)-associated pathologies in the retina of the eye. The device uses two imaging modalities where the first imaging modality guides the particular regions to be interrogated by the second imaging modality. A hyperspectral reflectance imaging unit detects light reflected and / or backscattered off of the retina from a broadband light source, which is used to determine one or more regions of interest. A Raman spectroscopy unit detects light that is re-emitted by the retina by Raman scattering phenomenon from a laser that is targeted onto a region of interest that is determined from hyperspectral reflectance information. The detected information from the hyperspectral reflectance imaging unit and the Raman spectroscopy unit are used to determine the presence of the one or more AD-associated pathologies.EFFECT: The detection device allows for the identification of at-risk populations, diagnosis, and tracking of patient response to treatments.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] Exemplary embodiments generally relate to detecting, locating, and detecting Alzheimer's disease-related pathology in the eye. This invention relates to an ocular light-based diagnostic detector for identification and quantification.

[0002] (CROSS REFERENCE TO RELATED APPLICATIONS) This application was filed on November 27, 2017. R SCANNER FOR DETECTION OF ALZHEIMER'S D U.S. Provisional Patent Application No. 62 / 590,8 entitled ISEASE PATHOLOGIES No. 36, the contents of which are incorporated herein by reference in their entirety. It is incorporated into the detailed description of the embodiment. [Background technology]

[0003] Alzheimer's disease (AD) is a fatal neurodegenerative disease. Identification of the body is generally performed postmortem. There are several existing conventional systems for diagnosis. These procedures involve highly invasive procedures or are inaccessible to imaging equipment due to cost or complexity. or use harmful radioactive tracers.

[0004] Several conventional biomarker methods are used to identify AD-related pathologies and are clinically This could help doctors detect AD early and distinguish its symptoms from other forms of dementia. These techniques are considered adjunctive tools that can be used to detect amyloid brain deposits or downstream neuronal damage. For example, amyloid beta (Aβ) and phosphorylated tau (Tau-1) are often evaluated. Neurofibrillary tangles (NFT) in cerebrospinal fluid (CSF) Spinal fluid measurement, amyloid beta or fluorodeoxyglucose (FDG) Positron emission tomography study of rodeoxyglucose uptake (metabolism in parietal and temporal lobes) PET (positron emission tomography), and magnetic resonance imaging (MRI) of brain atrophy. This includes GNATIC resonance imaging. Summary of the Invention [Problem to be solved by the invention]

[0005] However, many of these techniques are highly invasive and time-consuming (e.g., in external laboratories). They require validation, are expensive, complex, difficult to access, and are difficult for many clinicians to training is required and is insufficient to identify early or asymptomatic stages of AD. It is.

[0006] Therefore, patient responses to early detection, diagnosis, and prevention or treatment of AD-related pathologies have been investigated. Easy to operate and access for clinicians to screen patient populations for response tracking The objective of the present invention is to provide a non-invasive light-based detection system that can detect exogenous The goal is to perform detection without the use of fluorescers, dyes, or tracers.

[0007] To more specifically characterize AD-related pathology, specific characteristics of the chemical composition of parts of the eye are Detection is the purpose of the system. [Means for solving the problem]

[0008] Exemplary embodiments include a non-invasive ocular photodetector for detecting AD-related pathology in the eye. The present invention relates to a detection device based on the present invention, for optically examining a part of the fundus, such as the retina. This detection device can be used to identify populations at risk for AD, diagnose, and treat and intervene. Light-based procedures provide an accessible, non-invasive procedure for tracking the effectiveness of It is a step. The detector uses two imaging modalities, the first imaging modality being the second The detector uses the first imager to detect retinal reflection and / or detects light from a scattered broadband light source and further detects one or more regions of interest (R Determine the location and size of OIs (regions of interest). The detection system is initiated by the incidence of laser light on each ROI using a second imaging device. The detector detects the light re-emitted by the Raman scattering process that occurs when the light is scattered by the detector. The Raman spectroscopic information is detected and characterized to be specific to one or more chemical components of AD-related pathologies. The counts of specific wave number shifts can be detected.

[0009] The detection device is a non-invasive device with the sensitivity and specificity for the detection of one or more AD-related pathologies. It is a valuable tool that can be used for pre-screening, diagnosis, and tracking the effectiveness of treatments and interventions. Conventional optical methods for non-invasive detection can suffer from a lack of specificity and sensitivity. or may rely on exogenous fluorescers, dyes, or tracers.

[0010] Two imaging modalities are used in sequence to determine the presence of AD-related pathology indicative of AD. In an exemplary embodiment, the first imaging device includes a light source (e.g., a broadband lamp or Using a monochromatic patterned light (PLL), the researchers used hyperspectral imaging to measure the subject's A wide field-of-view reflected light-based image of the retina is acquired. The first imaging device may be a protein oligomer or aggregate based on physical properties. The method enables detection of potentially abnormal regions and identifies the location and size of one or more ROIs. These can be further imaged by a second imaging device using a second light source, such as a monochromatic laser. A monochromatic laser probes each ROI, and in the exemplary embodiment, Using NMR spectroscopy, we have identified specific wavenumber shifts characteristic of the chemical components of these AD-related pathologies. Check whether the test results are correct. Raman spectroscopy has revealed protein aggregates and other features characteristic of AD or its precursors. Raman spectroscopy is a highly specific method of detecting the target of interest (e.g. , protein aggregates and other features) re-emit light characteristic of the chemical moiety (Raman scattering) The Raman scattered light is collected by a detector and reacts to a monochromatic laser. The results are then spectroscopically analyzed to detect chemical signatures of AD-related pathologies.

[0011] The detection system does not rely on exogenous fluorescent agents, dyes, or radioactive tracers. It is completely non-invasive and acts synergistically to treat tauopathy, soluble and / or insoluble amylolytic enzymes. Not only AD-related pathologies such as id beta species and amyloid precursor protein (APP), Not only is the detection highly sensitive, but so are the surrounding neuritis and glial pathology and vascular features. ,Two different imaging modalities are utilized, yielding high ,specificity.

[0012] In some examples, the detection device may be configured to operate the detection device and retrieve a retina from the fundus of the subject. Machine learning algorithms are used to classify the acquired optical information. , regarding AD disease, diagnosis, and tracking the effectiveness of treatments and interventions (positive or negative response), This allows for rapid and non-invasive pre-screening of at-risk populations. Many current noninvasive optical methods of detection rely on the use of exogenous fluorescent agents. The detection device allows for highly specific detection of AD-related pathologies without the use of exogenous fluorescent agents. To do this, the intrinsic optical contrast and Raman resonance in the eye are used.

[0013] In some examples, the machine learning algorithm is performed by the detection device in two steps: First, a hyperspectral microscope is used to guide the Raman spectrometer laser to those ROIs. Based on the spectral reflectance information, regions of interest are identified. Then, the detection of these specific ROIs is performed. The Raman spectra and hyperspectral reflectance information returned from the scans can be used to identify AD-related pathology. These two spectroscopy modalities and machine learning algorithms are When combined with circadian rhythm, it can be used to prescreen, diagnose, and treat AD in at-risk populations. Providing a highly sensitive, specific, non-invasive detection tool to track the effectiveness of treatments and interventions .

[0014] In some cases, machine learning algorithms use validated training data. The validated training data are from subjects known to have AD. Adjacent slices of ex vivo tissue samples from subjects were One slice of the subject's tissue is then imaged using hyperspectral imaging. The slices are then analyzed using CT and Raman spectroscopy, and adjacent slices are stained and imaged using a microscope or other imaging techniques. The results are verified by histological diagnosis using a single slice imaging device. To verify the pathology of D. cerevisiae, hyperspectral imaging and Raman spectroscopy were used to Slices can be analyzed in their corresponding locations, allowing for validated training of machine learning algorithms. This can be used as training data.

[0015] The detection device comprises a non-invasive device for detecting one or more AD-related pathologies from an eye of a subject. An in vivo eye light-based detection device, comprising a broadband light source and a hyperspectral camera. a hyperspectral reflectance imaging unit including a laser and a spectrometer; a Raman spectroscopy unit including a memory; and one or more processors configured to execute the following instructions stored in a memory: and a processor controls the hyperspectral reflectance imager to capture a wideband light source. to illuminate a wide field of the fundus of the eye and determine hyperspectral reflectance information. A hyperspectral camera is used to detect reflected and / or backscattered light from the eye. From the hyperspectral reflectance information, one or more ROIs can be identified as potential AD-related pathologies. and configuring a Raman spectroscopy unit to illuminate each of the one or more ROIs using a laser. Control and use a spectrometer to determine Raman spectroscopic information obtained from the laser The system detects Raman scattered light from the target and uses hyperspectral reflectance information and Raman spectroscopy information. In order to classify subjects as having one or more AD-related conditions, The pathology involves protein aggregates, which are tau neurofibrillary tangles, amitriptyline, and tau protein aggregates. amyloid beta deposits, soluble amyloid beta aggregates, or amyloid precursor protein Includes at least one of the following:

[0016] Another exemplary embodiment is directed to non-invasive biopsy of one or more AD-related pathologies from a subject's eye. A method for in-vivo detection, comprising controlling a hyperspectral reflectance imager to use a broadband light source. To illuminate a wide field of the fundus of the eye using a hologram and determine hyperspectral reflectance information. To achieve this, a hyperspectral camera is used to detect light from the eye that originates from a broadband light source. and extracting from the hyperspectral reflectance information using one or more processors: Identifying the location of one or more ROIs as potential AD-related pathologies and controlling a Raman spectroscopy unit to illuminate each of the one or more ROIs using the Raman To determine spectroscopic information, a spectrometer is used to measure the Raman scattering from the eye obtained from the laser. Detecting turbulent light and using one or more processors to extract hyperspectral reflectance information Using the genomic and Raman spectroscopic information, subjects were identified as having one or more AD-related conditions. and classifying the one or more AD-related pathologies as comprising protein aggregates and protein Protein aggregates are expressed as tau neurofibrillary tangles, amyloid beta deposits, and soluble amyloid beta. The protein comprises at least one of the following: amyloid aggregates, amyloid precursor protein, or amyloid precursor protein.

[0017] Another exemplary embodiment is a computer program using a machine learning training process. a computer program product stored on a non-volatile computer readable medium; When the image is stored in the memory and executed by the computer, the image is transmitted to the computer from the subject's eyeball. Instructions for non-invasive in vivo detection of one or more Alzheimer's disease (AD)-related pathologies The machine learning training process uses one or more processors to training a computer program using the training data; Validated training data was obtained by slicing ex vivo tissue samples from subjects. 2. Prepare tissue slices and place the tissue slices on slides. Place the first tissue slice on the first slide. Stained slices were placed adjacent to the first tissue slice in the ex vivo tissue sample, and unstained slices were placed adjacent to the first tissue slice in the ex vivo tissue sample. Provide a second slide having a second tissue slice of color and stained using a histology diagnostic. Verify that the first tissue slice has one or more AD-related pathologies and verify that the second slice Activating at least one imaging device in the image forming apparatus to obtain imaging information, and processing the imaging information by Obtained by classifying as one or more AD-related pathologies, one or more AD-related pathology involves protein aggregates, which are then transformed into tau neurofibrillary tangles. amyloid beta deposits, soluble amyloid beta aggregates, or amyloid precursor proteins. It contains at least one of the following proteins:

[0018] Another exemplary embodiment, when executed by a computer, causes the computer to non-invasive in vivo detection of one or more Alzheimer's disease (AD)-related pathologies from the subject's eye. Machine learning training of computer programs stored in memory that execute the output The method includes: training a computer program using the training data, Training data was generated by slicing ex vivo tissue samples from subjects. 1. Place the tissue slices on a slide and stain the first tissue slice on the first slide. The stained and ex vivo tissue samples are adjacent to the first tissue slice and the second unstained. and providing a second slide having a tissue slice of the stained first slide using a histopathologist. Verify that the tissue slice has one or more AD-related pathologies and activating at least one imaging device to obtain detection information; and converting the detection information into one or more The AD-related pathology is obtained by classifying one or more of the AD-related pathologies as Protein aggregates include tau neurofibrillary tangles, amyloid beta deposits, soluble amyloid beta aggregates, or amyloid precursor protein The method further includes storing the trained computer program in a memory. It is provided to store.

[0019] Reference will now be made, by way of example, to the accompanying drawings which illustrate exemplary embodiments. [Brief description of the drawings]

[0020] [Figure 1] FIG. 1 illustrates in schematic form a non-invasive ocular light-based ocular detection device for detecting AD pathology in the eye, according to an exemplary embodiment. [Diagram 2] 2 is a schematic side view of an interface unit of the detection device of FIG. 1. [Diagram 3] FIG. 2 is a top-down schematic diagram of the Raman spectroscopy section of the detection apparatus of FIG. 1. [Figure 4] FIG. 1 shows Raman maps of unstained formalin-fixed paraffin-embedded (FFPE) brain tissue from a postmortem AD patient. [Diagram 5] FIG. 5 shows broadband Raman spectra of AD plaques corresponding to pixels from the Raman map of FIG. 4. [Figure 6] FIG. 5 shows broadband Raman spectra of a pixel from the Raman map of FIG. 4 including background tissue. [Figure 7] FIG. 13 shows another Raman map of unstained FFPE brain tissue from a post-mortem AD patient. [Figure 8] FIG. 8 shows broadband Raman spectra corresponding to pixels from the Raman map of FIG. [Figure 9] FIG. 8 shows broadband Raman spectra of a pixel from the Raman map of FIG. 7 including background tissue. [Figure 10] FIG. 1 illustrates a hyperspectral imaging map of patient tissue for identifying regions of interest for subsequent Raman spectroscopy, according to an exemplary embodiment. [Figure 11] FIG. 1 is a flow diagram of a method for detecting AD pathology in an eye, according to an exemplary embodiment. [Figure 12] 2 is a flow diagram of a method for determining training data for a machine learning algorithm of the detection device of FIG. 1 in accordance with an exemplary embodiment. [Figure 13] FIG. 1 illustrates a system for detecting AD pathology in the eye, according to an exemplary embodiment. [Figure 14] FIG. 5 shows a polarized microscope image containing plaques (e.g., stained red dots) corresponding to the Raman map in FIG. 4. [Figure 15] FIG. 8 shows two different magnifications (10x, left, and 40x, right) of polarized light microscope images of a stained slide showing plaque next to a blood vessel, corresponding to the Raman map (mirror image) in FIG. [Figure 16] FIG. 16 shows a hyperspectral image (left) of an unstained slide containing the same vessels seen in FIG. 15 in adjacent slices, and its polarized microscope image (right, mirror image). [Figure 17A] FIG. 17 shows a white light image of the same unstained slide as in FIG. 16 taken with a Raman spectroscopy unit, illustrating that the same blood vessels in FIG. 16 can be located using a Raman spectroscopy unit. [Figure 17B] FIG. 17B illustrates the white light image of FIG. 17A and shows the Raman mapped area. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] Similar reference numbers may be used in different figures to indicate similar components.

[0022] FIG. 1 illustrates a non-invasive method for detecting AD-related pathology in the eye, according to an exemplary embodiment. The detection device 100 is based on the active intraocular biophoton. The detection device 100 can be used to perform some optical inspections. Point-of-care procedures that provide accessible, non-invasive procedures to identify at-risk populations The detection device 100 is a point-of-care (POC) tool. The detection device 100 also detects one or more AD-related conditions. In order to detect the presence of vitreous matter with high specificity, the vitreous matter emitted from the fundus in response to irradiation by a monochromatic laser is The Raman scattered light can then be detected, which can indicate the presence of one or more AD-related pathologies. Based on this, at-risk populations can be identified.

[0023] The detection device 100 is interposed between the Raman base station 1 and the subject under study. and an interface unit 2. The subject may be, for example, a human or an animal. The device 100 includes a hyperspectral reflectance imaging portion and a Raman spectroscopy portion. The spectral reflectance imaging unit is located inside the interface unit 2. The Raman spectroscopy unit is The system is defined by the components of a base station 1 and an interface unit 2. .

[0024] The subject placed the chin on the chin rest 5 and placed the interface against the rubber eye cup 4. The Raman base station 1 and the interface unit 2 are placed in front of the The Raman base station 2 is connected via an optical fiber 3. It serves to provide monochromatic laser light in a narrow beam configuration from unit 1 to interface unit 2. Fulfill your quota. The laser light is coherent light at 532 nm in one example, or 785 nm in another example. In other examples, other laser wavelengths can be used. Through the fiber optic 3, the Raman process also detects specific and collecting the light re-emitted by the area or part of the subject's eye. , for detection by a suitable photodetector (e.g., a spectrometer), at the Raman base station 1 Supply to. In another exemplary embodiment, the Raman base station 1 and the interface unit The two may be combined into a single device, as would be apparent to one of ordinary skill in the art in light of the teachings herein. or may be further separated. The computer 6 controls the Raman base station 1 and the interface unit 2. The computer 6 is used for interfacing (control and communication) with the memory. 30 and a processor 32. The electrical cable 7 connects the Raman base station 1 to the A coaxial cable 8 connects the interface unit 2 to the computer 6. It transmits information between the computer 6. The computer 6 uses a machine learning algorithm, as described in more detail herein, to The computer 6 processes the received information and outputs the output of the machine learning algorithm or other control The information is transmitted to the interface unit 2 via the electrical cable 9, The unit 2 uses the received information to detect a specific area or region of the subject's eye from the optical fiber 3. For example, the computer 6 analyzes the received imaging information and One or more dedicated image analysis tools capable of processing information in real time, either partially or completely Chips (e.g., graphics processing units or GPUs) )).

[0025] FIG. 2 shows the interface unit 2 in more detail. The interface unit 2 comprises: It controls the operation of the interface unit 2, the computer 6 and the Raman base station The system may include one or more controllers or processors (not shown) for communicating with the system 1. The target under investigation can be placed in front of the rubber eyecup 4. When taking an image, the subject faces the eyeballs toward the eyecup 4 and places his / her chin on the chin rest 5. This allows the subject to be aligned with the optical path of the interface unit 2. The interface unit 2 is used for in vivo imaging of animals and ex vivo imaging of tissues. g), or any other suitable target, and connecting the stage to the interface unit The target can be placed in a suitable position (e.g., in the focal plane of an optical system) by mounting it on a Other components for supporting and positioning the target may be used in other examples. stomach.

[0026] The interface unit 2 includes a fundus camera capable of taking a wide-field image of the subject's fundus. A light sensor 10 capable of detecting and identifying different wavelengths of light can capture an image. The optical sensor is used to capture the image of the surroundings. This may take the form of a vector camera, a red, green, blue camera, or a monochrome camera. A broadband light source 11 covering the visible and near infrared spectrum (400 nm to 1100 nm) is used to illuminate the subject's retina. A broadband light source 11 passes through two beam splitters 12, 13 and enters a lens assembly 1. In another exemplary embodiment, the retina is focused on the subject's eye via a focusing element such as 4. It will be appreciated that other focusing and beam shaping elements may be present to adjust the light distribution. Once directed at the subject's eye, at least a portion of the broadband light strikes the retina or eye. Some of this light is reflected and / or backscattered from other areas of the interface unit. 2 where it is collected by lens assembly 14 and beam splitter 13 The position and orientation of the hyperspectral camera 10 are Other suitable configurations of reflected and / or backscattered light collection geometries will be apparent to those skilled in the art. Perhaps.

[0027] As directed by the lens assembly 14, the entire field of view is captured in one shot. The sensor 10 detects the entire image. For example, in the case of the hyperspectral camera 10, Wavelength information is detected simultaneously across the entire field of view. The image section is contrasted with a raster, which scans rows or columns across the field of view, i.e., one at a time. detecting wavelength bands (e.g., multispectral imaging) or line hyperspectral imaging. requires a vector camera, or coherence (e.g., optical coherence tomography) Contrast with illumination.

[0028] In this example, the central region of the subject's retina is the imaging target that occupies the entire visual field. In this example, other regions of the fundus serve as imaging targets.

[0029] The coaxial cable 8 transmits the hyperspectral signal detected by the hyperspectral camera 10. The information is then transmitted to computer 6 for processing in real time. As we will explain in detail, the machine learning algorithms in Computer 6 Using the training information, we estimate the size of one or more ROIs based on previously acquired training data. Check the location and size. In some instances, the size of each ROI is determined by measuring a circular area centered at this location (e.g., radius or diameter) or a rectangular area (e.g., M x N pixels) Once one or more ROIs have been identified, the Raman spectrometer can be used to measure the Other imaging techniques, such as NMR spectroscopy, can be implemented by the detector 100. A second light source such as the light source 18 (FIG. 3) is housed within the Raman spectroscopic section. Light from a monochromatic laser 18 is directed to the appropriate ROI by mirrors 15, 16, and 17. The mirrors 15, 16, and 17 are driven by electromechanical motors via the interface unit 2. The hyperspectral reflectance imager was controlled by the The focused laser beam was aimed at the appropriate ROI on the subject's retina, as previously identified from the retinal information. do. The electrical cable 9 transmits signals to a computer to control the angles of the mirrors 15, 16, and 17. The laser light is transmitted from the imager 6 to the interface unit 2. The laser light interacts with the retina in the ROI. Through the Raman phenomenon, the light is then converted into specific wavenumber shifts characteristic of the chemical composition of the examined tissue. This re-emitted light is collected via lens assembly 14 and The light passes through the beam splitter 13, and then passes through the beam splitter 12 and the mirrors 15, 16, 17 and then coupled into the optical fiber 3. The optical fiber then The emitted Raman light is transmitted back to the Raman base station 1 for detection.

[0030] Raman spectroscopy is used to characterize one or more specific chemical components for each identified ROI. This is performed to determine whether there is a significant wavenumber shift. The spectrum of the ROI is obtained by Raman spectroscopy using a lens assembly 14 and / or an aperture. The pixel information may comprise one or more specific pixels in the organizational environment. In some instances, counts at a particular wavelength can be detected. , and one or more wavenumber sequences can be obtained therefrom by calculating the difference with the known wavelength of the monochromatic laser 18. The weight is calculated.

[0031] For example, the Raman spectroscopic information of each identified ROI having a location and a size may be The ROI location and size are detected by one of the monochromatic lasers. For example, the lens assembly 14 may be stimulated by an instance of The Raman spectrometer uses a monochromatic laser 18 can be used to control the size of the ROI stimulated by the laser light from In some examples, a diaphragm, iris, or collimation device (not shown) is used. It is also possible to control the size of the ROI stimulated by the monochromatic laser 18 .

[0032] In another example, each pixel of the ROI is stimulated by a monochromatic laser 18. The Raman spectroscopic information is obtained by the Raman spectroscopic unit over each pixel of the ROI. A Raman map of the ROI is generated or an integrated spectroscopic result over the ROI is acquired. It will be appreciated that it is not necessary to perform a Raman scan over a large field of the retina.

[0033] In various examples described in more detail herein, the Raman spectroscopic section includes an optical filter 2 0 (Figure 3) to detect specific wavelengths or bands of interest prior to detection by the Raman spectrometer. Similarly, digital filtering can be used to filter out specific wavelengths or wavelengths of interest. This can be performed by computer 6 for the bands.

[0034] Next, the operation of the hyperspectral camera 10 to perform hyperspectral imaging The hyperspectral camera 10 is sensitive to light in the visible and near infrared ranges. The photo-sensor includes a two-dimensional array of photo-sensors that are identified by pixels. A two-dimensional filter array is disposed over the array of photosensors. Each individual filter selectively transmits light of a given wavelength, which is then transmitted to the sensor array. The pattern of the filter array is detected by dedicated pixels across the photosensor. This is repeated over and over so that light from every point in the field of view is filtered and detected by the sensor. It is served. In this way, all wavelength / frequency information from all regions of the field of view is captured in one go. are acquired simultaneously. This detects only different wavelengths of light across a one-dimensional line in the field of view. This is different from a line hyperspectral camera, which can distinguish between two points by using multiple filters. It also differs from the common multispectral approach, which uses sequential In other words, first capture the "red" information, then insert another filter to capture the "green" information. This is different from an approach that captures information such as "

[0035] FIG. 3 shows an exemplary embodiment of the Raman base station 1 in more detail. A monochromatic laser 18 in the near infrared wavelength range is housed inside the Raman base station 1. The laser 18 emits 532 nm coherent light in this example, or 785 nm in another example. m. Laser 18 may emit other specific wavelengths in other examples. The laser output from the laser 18 is directed through a beam splitter 19 and then fed to a fiber adapter 2. The laser beam is coupled to an optical fiber 3 via an interface unit 2. As described above, the interface unit 2 transmits this laser light (high the subject's retina (as identified by computer 6 based on perspectral imaging) Orient each ROI. In this example, the size (radius) of the laser light for each ROI is determined by the lens assembly 14 and / or In another example, the laser light can be controlled using a 3D filter or an aperture. The laser light interacts with the tissue in these regions and, via the Raman phenomenon, Light is scattered from the tissue with a change in wavelength that is characteristic of the tissue being examined. This Raman scattered light is shaped and directed by collection optics, such as one or more additional lenses (not shown). Therefore, it can be efficiently coupled into an optical fiber 3 and returned to the Raman base station 1 . Beam splitter 19 serves to redirect this returning light to spectrometer 21 for detection. The optical filter 20 filters at 534 nm (greater than the laser wavelength from the laser 18). The optical path may include a long pass filter having a small offset to receive back reflections along the optical path. It is used to filter out any direct laser light returning to the Raman base station 1. It is used. In another example, the optical filter 20 may filter out a particular wavelength of the laser (e.g., 532 nm in one example). Notch filter with narrow filter for coherent light (785 nm in another example) The optical filter 20 is configured to filter only the light from the Raman phenomenon into the spectrometer 21. and the light from the original laser 18 is filtered out by an optical filter 20. The spectrometer 21 separates the individual wavelength components and transmits them to the optical sensor 22. The optical system includes a refractive element that projects onto separate pixels. The spectral information measured by the spectrometer 21 is then coupled to an electrical cable for further processing. The data is sent to computer 6 via cable 7. Computer 6 may alternatively or physically In conjunction with the optical filter 20, further filtering algorithmically (digital The Raman base station 1 can perform Raman base 1 and communicates with the computer 6 and the interface unit 2. The system may include one or more controllers or processors (not shown) for performing the operations.

[0036] In some examples, the illumination and light collection system may include an adaptive optics (AO) system and method. This can be done by using

[0037] Second, a machine learning algorithm is trained on the Raman spectra of one or more substances. is executed by the computer 6 and is a characteristic of the chemical component of the source of the Raman signal. Identifying a specific wavenumber shift, and using this to identify AD-related proteins in the eye Identifying the presence of lipid aggregates or other pathologies. Identification is achieved by measuring the intensity of one or more wavenumber shifts. This may include counting stances and / or other formulas. Exemplary protein aggregates in the fundus that may be detected by the detection device 100 include tau neurogenic aggregates. Fibril changes (e.g., soluble or insoluble tau oligomers or tau fibrils), amyloid beta Deposits (e.g., soluble amyloid beta aggregates or insoluble amyloid beta plaques, Amyloid beta oligomers, or amyloid beta precursors, and amyloid precursor proteins Contains protein (APP). Detection of this Raman signal allows for a more accurate understanding of AD-related pathology compared to hyperspectral imaging alone. The specificity of detecting AD-related pathology can be significantly improved. To assess the classification of AD or other AD outcomes, radiological data obtained at different times from the same patient were used. The Raman spectroscopic information is compared. For example, the Raman spectra of potential plaques in a particular ROI are compared. The values ​​(or ratios or other characteristics) of the serotonin receptor agonists may increase over time in AD subjects. In some instances, machine learning algorithms combine Raman spectroscopy information with hyperspectral Using both the reflectance and reflectance information, we can more appropriately conclude that AD-related pathology or other AD-related conditions Classify.

[0038] Computer 6 interprets the Raman spectroscopy information and uses machine learning algorithms to generate the RO Classifying I as including or not including one or more AD-related pathologies, such as protein aggregates In an exemplary embodiment, the subject is classified as having AD or a precursor to AD. have been prescreened for potential AD and are not eligible for further investigation. It may also be the conclusion of the AD as to whether an examination is necessary. The data is output to a screen, stored in local memory, or transmitted to the server 204, the client 206, or an EMR (electronic medical record) server 214 (FIG. 13), etc. The device can be programmed to transmit to another device.

[0039] Figures 4 to 10 show the images used as training data for the machine learning algorithm. Similarly, in FIG. 4 to FIG. 10, the imaging information indicates how the detection device 100 can be used in vivo to classify AD-related pathology in a given subject’s eye. Both scenarios will be described with reference to Figs.

[0040] Figure 4. Unstained formalin-fixed paraffin-embedded (FFPE) brain from a postmortem AD patient. A Raman map 400 of tissue is shown. Bright spots correspond to the locations of amyloid beta plaques. This was independently verified through histological examination of adjacent ex vivo tissue slices from the same subjects. This map 400 shows that every pixel (beta sheet protein structure) (corresponding to the Raman vibrational resonance of -1 Plot the signal strength of 20 00cm -1 and 2500cm -1 Subtract the linear background of the Raman signal between The axis is generated by the computer 6 by the physical distance of the tissue slice. The bright spot 402 corresponds to the pixel (55 in FIG. 4) μm, 46 μm). In this example, the specific Raman capture is done by the Raman unit spectroscopy unit. For one capture, two or more pixels of the ROI are illuminated by the laser 18. In another example, each pixel of the ROI can be illuminated by a monochromatic laser 18. The Raman spectrometer scans over each pixel of the ROI, with each pixel stimulated by the Raman spectroscopy information is obtained to generate a Raman map of the ROI or Calculate the integrated spectroscopic result over

[0041] Figure 5 shows the AD plaques corresponding to the bright spot 402 pixels (55 μm, 46 μm) in Figure 4. 5 shows a broadband Raman spectrum graph 500 of the pixel. , which indicates the wavenumber shift corresponding to the count of the received Raman scattered light. -1 and 1663cm -1 Peaks 502 and 504 in the figure are alpha helix and vein helix, respectively. Raman spectra of the double-pleated sheet protein conformation, the so-called amide I band This corresponds to dynamic resonance. Most of the remaining peaks correspond to the presence of paraffins. -1 and 1663cm -1 A peak at indicates the presence of protein at this location, These peaks are due to the low background signal present at these wavelengths in the neighboring regions of the field of view (Figure 6 , 602 and 604). This confirmed the localized presence of proteins characteristic of amyloid beta plaques. It is recognized. 1800cm -1 There is no Raman signal from the beta sheet at 1800c m -11663cm -1 The map showing the ratio of the Raman signals of amyloid beta proteins Note that the hot spots are shown in the corresponding locations of the larks. For background signal ("noise"), 1800 cm -1 1663cm -1 of Criteria can be set according to the signal ratio. For example, a signal-to-noise ratio of 3:1 can be used to measure amyloid. The presence of idobeta plaque can be confirmed.

[0042] Figure 6 shows the broadband Raman spectrum of the background tissue pixels (10 μm, 15 μm) in Figure 4. Shows rough 600. 1600cm -1 and 1663 cm -1 Note that there is no peak in This is the first study of the conformational properties of alpha-helical and beta-pleated sheet proteins. This is consistent with histological findings from adjacent ex vivo tissue slices from the same subject. In some instances, the Raman spectrograph 600 or Other Raman spectroscopic information on background tissues was used as control information for training machine learning algorithms. can be used as information (as a negative classification or as a value to be subtracted / divided).

[0043] In some examples, a Raman spectrograph 600 obtained from a ROI of the present subject matter may This is then used by machine learning algorithms to classify plaques or AD-related pathologies. For example, the computer 6 may obtain a Raman spectrograph 600 of the subject's background tissue. , and a comparison is made with a Raman spectrograph 500 (FIG. 5) of potential plaque in the subject. The Raman spectrograph 600 was taken from the same subject as the Raman spectrograph 500. The comparison provides useful results because the results are based on plaque or other AD. To classify as relevant medical conditions, machine learning algorithms, comparisons, formulas, calculations, tables, subtractions, This may include a ratio, or other comparison performed by computer 6.

[0044] In some instances, the Raman maps of the validated training data are Integrate over the vector region and plot this integrated quantity for every pixel. It is generated by: Figure 7 shows such an example, where each pixel represents 1663 cm of the area. -1 and 1 698cm -1 Encode the cumulative count between 2000cm and -1 and 2500c m -1 The linear background signal based on the Raman spectrum between The bright spots 702 in FIG. 7 are easily identifiable and may be useful for tissue diagnosis of adjacent tissue slices. corresponds to amyloid beta plaques, as independently confirmed through

[0045] In other examples, chemometrics are used to identify the spectrum that best corresponds to AD-related pathology. The Raman spectrum can be analyzed by algorithmic analysis. Perform statistical analyses to identify features specific to AD-related pathology that are not readily apparent can be done.

[0046] In some instances, rather than a single broad spectral range, the acquired Raman signal is One or more narrow spectral regions centered on the spectral region of interest, such as those identified in 5. It can consist of rule regions or bands.

[0047] For example, Raman spectroscopy does not perform a raster scan over an extended area. , is performed on one or more specified ROIs. The result in these cases is Instead of a complete image consisting of Raman spectra, a single Raman spectrum, as shown in Figure 5, is In some cases, the incident laser beam is used to cover a larger area. The beam can be expanded to a larger diameter and the Raman scattered light is diffused by the expanded laser spot size. The data is collected from the ROIs covered by the

[0048] FIG. 10 illustrates an exemplary hyperspectral imaging map 1000 of patient tissue, which , a hyperspectral image that may be used to identify one or more ROIs for subsequent Raman spectroscopy. 10 illustrates an example of spectral imaging information. The hyperspectral imaging maps 1000 each include 10. A plurality of images representing a map of counts of particular detected wavelengths from the hyperspectral camera 10. The individual hyperspectral image maps 1000a, 1000b, . . . , 1000e of It can include. If a pixel in a particular hyperspectral image map has a high (or low) count, This means that the pixel should be investigated further using Raman spectroscopy. From the hyperspectral imaging map 1000, the computer 6 executes a machine learning algorithm to Use one or more pixels that require further investigation by Raman spectroscopy. An upper ROI 1002 (one shown) can be determined. In other examples, each hyperspectral map 1000 may be a single map of a particular wavelength, rather than a single wavelength. A range of wavelengths can be represented and the counts are for that particular range of wavelengths. In yet another example, the hyperspectral map 1000 uses a linear combination of specific wavelengths. The method may be used to generate a genomic analysis of AD-related pathology that best encapsulates the characteristics of the pathology. It will be made into a system.

[0049] Another exemplary representation of hyperspectral reflectance imaging information is for each pixel or region of a subject. The spectral graph for the pixel is shown in FIG. 1 illustrates the counts of received light versus wavelength. It can also be used to train learning algorithms and for classification using machine learning algorithms.

[0050] For example, the ROI may be a hyperspectral imaging map 1000 or a hyperspectral imaging As shown in the image spectrogram, it can be determined from the hyperspectral imaging information. can.

[0051] With further reference to FIG. 10, several wavelengths of a hyperspectral imaging map 1000 Computer 6 uses machine learning algorithms to provide better results than others. , which hyperspectral reflectance maps 1000 and their corresponding wavelengths are processed. For example, a high-power laser that covers the entire visible and near-infrared spectrum is It is not necessary to analyze the perspectral imaging map 1000, but rather the hyperspectral imaging. One or more particular wavelengths of the image map 1000 are then sent to the computer 6 for further processing. is selected. In some other examples, the computer 6 may be configured to determine the ROI for Raman spectroscopy. 1000 hyperspectral imaging maps with low correlation to AD pathology of interest Less weight is given to the more relevant hyperspectral imaging maps 1000 is given greater weight.

[0052] In one example, the hyperspectral imaging of interest used by computer 6 Map 1000 or Hyperspectral Imaging Spectrograph is a system that measures the visible-near infrared (VNI) R) wavelength range (400-1400 nanometers), specifically, 460 nm to 6 00nm optical wavelength range or 650nm to 950nm optical wavelength range, It is more suitable for detecting protein aggregates such as idobeta deposits. In another exemplary embodiment, based on the disease state of the group and the machine learning algorithm, different or A more specific wavelength range is used.

[0053] Referring again to FIG. 4, the hyperspectral reflectance map 1 is generated by the computer 6. 000 (or a hyperspectral reflectance spectrograph) is generated and used to The location and size of one or more specific ROIs of the subject that are to be further analyzed using NMR spectroscopy. The size of the ROI is determined. The ROI can contain one or more pixels. In this example, I is the bright spot at pixel (55 μm, 46 μm) in FIG. Therefore, it is not necessary to perform Raman scanning of the entire field of view for a hyperspectral reflectance map. However, in localized regions such as the pixels (55 μm, 46 μm) in Figure 4, the Raman spectroscopic information was detected by the Raman spectroscopy unit in one capture. This is the ROI in Figure 10. In some cases, some of the areas around the bright pixel are pixels, or a defined pixel radius around a bright pixel, in one capture. It can be analyzed using Raman spectroscopy.

[0054] FIG. 7 shows another Raman map 700 of unstained FFPE brain tissue from a postmortem AD patient. Again, through histological examination in adjacent tissue slices from ex vivo subjects. As independently verified, the bright spots correspond to the locations of amyloid beta plaques. 702 is located at pixel (17 μm, 31 μm) in FIG. 0cm -1 and 2500cm -1 Linear background subtraction of the Raman signal between After that, 1663 cm -1 and 1698cm -1 Integrated Raman signal between all pixels The axis is in units of the physical distance of the tissue slice (millimeters). It corresponds to the unit of measurement (micrometer).

[0055] Figure 8 shows the broadband Raman scattering corresponding to the 702 pixels (17 μm, 31 μm) of the bright spot in Figure 7. Spectral graph showing 800. 1600cm -1 and 1663 cm -1 Peak 8 in 02,804 are Raman vibrational resonances of the amide I band, i.e., alpha helical and The remaining peaks correspond to the para- and beta-pleated sheet conformations. This corresponds to the presence of

[0056] Figure 9 shows the broadband Raman spectrum of the background tissue pixels (45 μm, 10 μm) in Figure 7. Roughly 900. 1600cm -1 and 1663 cm -1 It is noted that there is no peak in This is due to the alpha-helical and beta-pleated sheet protein conformations. This is due to the lack of a clear histological profile in adjacent tissue slices. The Raman spectrograph 900 of the background tissue pixel or Other Raman spectroscopic information serves as a control (negative classification) for training the machine learning algorithm. The Raman spectrograph 900 of the background tissue pixels or other Raman The spectral information is compared to a spectrograph 800 (FIG. 8) to classify the plaque. The data can be used for comparison or other calculations.

[0057] Referring again to FIG. 7, one or more of the hyperspectral reflectance maps 1000 (FIG. 10) Above, the subject's eyeball that is further investigated by computer 6 using Raman spectroscopy. In this example, the ROI is the location and size of the Bright spots at pixels (17 μm, 21 μm). Therefore, the entire field of view of the Raman map 700 in FIG. 7 is of interest when assessing AD pathology. It is not necessary to perform a man-scan. Rather, localization such as pixels (17 μm, 21 μm) in FIG. The region measured by Raman spectroscopy was the same location on the subject as ROI 1002 in Figure 10. In one example, Raman spectroscopy of the ROI is performed in a single capture by the Raman spectrometer. In another example, the ROI is scanned pixel by pixel and a Raman map of the ROI is generated. or calculate the integrated counts for a particular wavelength of interest.

[0058] The results in Figures 4 to 10 show that the subjects were trained in classification and detection of amyloid beta plaques. In another exemplary embodiment, the training data is: Instead of or in addition to amyloid beta plaques, other AD-related pathologies are classified. For example, if the AD-related pathology is tau neurofibrillary tangles, the Raman resonance of interest The wavelength remains the same (1600 cm for phosphorylated tau). -1 ~1700cm - 1 ), which is now found intracellularly. Further Raman studies are being conducted on other AD-related pathologies. The resonant wavelengths can be used to classify and detect AD-related pathologies.

[0059] FIG. 13 illustrates a method for detecting AD pathology in a subject's eye, according to an exemplary embodiment. 2 shows a system 200 for performing a test on a subject. In some examples, the system 200 The system 200 implements a machine learning algorithm to operate the detection device 100. The output device 100, the server 204, the client station 206, and the electronic medical record (EMR) The system 200 includes a MR server 214. There may be multiple devices of each type in the system 200. The devices of the system 200 can communicate over a network 202. The client stations 206 may be computers, laptops, mobile phones, The network 202 may be a local area network, LAN, Wireless Wide Area Network (WWAN), Private Network, and the Internet. The computer 6 of the detection device 100 (FIG. 1) A communications subsystem for communicating over a network 202 .

[0060] In FIG. 13, the server 204 is typically separate from the detection device 100 and runs the machine learning algorithms. The server 204 is configured to train rhythm. The system may include one or more cloud servers, or one or more server instances. In some examples, the server 204 may be an Amazon® AWS, Microsoft oft(registered trademark) Azure, Google(registered trademark) Cloud, and IBM(registered trademark) (Trademark) Watson or other third-party machine learning platforms. can be accessed. The server 204 includes a machine learning module 218 and a database of verified training data. Memory 2 for storing the base and the trained neural network 16. The server 204 executes instructions stored in the memory 216. The system may include one or more controllers or processors (not shown) configured to do.

[0061] The EMR server 214 is used to store, archive, and retrieve patient electronic medical records. The EMR server 214 is a data repository for patient data. The EMR server 214 may include, in one example, a third-party server. The EMR server 214 may contain medical, demographic, and physical information of the patient. The EMR server 214 may, in some examples, include verified training data. can be done.

[0062] The memory 216 or EMR server 214 may store previous hyperspectral imaging data for a particular patient. information or Raman spectroscopy information, thereby enabling previous hyperspectral imaging The information or Raman spectroscopic information is compared with other Raman spectroscopic information of the patient obtained at other times. The computer 6 or server 204 can then perform the AD conclusion for a particular patient. For example, time-separated hyperspectral imaging information or the same RO Compare the Raman spectroscopic information of the same patient with the individual history of the same patient to confirm progression (regression) The patient's progression (regression) can be compared to other population cohorts and their historical progression (regression). It is also possible.

[0063] The server 204 executes machine learning algorithms using one or more neural networks. Machine learning algorithms include logistic regression, variational autoevolution, and Encoding, Convolutional Neural Networks, or Identifying and Differentiating AD-Related Pathologies Machine learning algorithms can include other statistical techniques used to Use pre-validated Raman scattering models, other scattering models, or optical physics models A neural network can contain multiple layers, some of which are defined and some are undefined (or hidden). Neural Networks is a supervised learning neural network.

[0064] In some examples, the neural network consists of a neural network input layer, The neural network may include intermediate hidden layers, and an output layer of the neural network. Each layer of a neural network contains multiple nodes (or neurons). The nodes of a neural network layer are usually connected in series. The output of each node in the neural network layer is the input to one or more nodes in the subsequent neural network layer. Connected to the input. Each node is assigned a weight based on the input, its weights (if any), and its bias coefficients (if any). Activation functions (also called transfer functions) are used to transform or manipulate data and generate output. The activation function of each node is a logic programming unit that executes the This produces a particular output depending on the particular inputs, weights, and bias coefficients. The inputs of each node can be scalars, vectors, matrices, objects, data structures, and / or Each node can be, or may be a reference to, other items. Each node can be, independently of other nodes, The respective activation functions, weights (if any), and bias coefficients (if any) In some exemplary embodiments, one of the neural network output layers may be stored. The determination of one or more output nodes may be based on previously determined weights, as is understood in the art. using weighting and bias factors, and using scoring functions and / or decision tree functions It can be calculated or determined.

[0065] The server 204 receives input from the practitioner to the client station 206, and validates the Training the neural network using the training data 208 Additional training data sets can be obtained from the EMR server 214 or from the detection device. This can be obtained from the operation of the device 100 itself. For example, hyperspectral reflectance information and Raman spectroscopic information may be obtained from the operation of the detection device 100. The information is stored in the server 204 or the EMR server 214. Additional Raman capture can be performed later to obtain more Raman spectroscopic information. - Historical trends in spectral reflectance information and Raman spectroscopy information will indicate AD at a later date. These may be examined as precursors of AD or as a precursor to AD. Subjects were diagnosed with AD, and this diagnosis was based on previous hyperspectral reflectance information and Raman spectroscopic information can be classified as AD or pre-AD. Similarly, for some subjects, EMR information may be updated in later years and there may be no AD. In some cases, postmortem tissue diagnosis may be used to identify AD information for patients. Tissue diagnosis can be performed using a microscope or other imaging device.

[0066] In some examples, the server 204 may implement two neural networks. As is understood in the art, each neural network can itself It is possible to have one or more neural networks in parallel, serial, or other configurations. The first neural network takes the hardware as input to the first neural network. Based on the hyperspectral reflectance information, the output of the first neural network is: Used to identify one or more ROIs. The second neural network computes the metric returned from examining these particular ROIs. The spectra were then classified together with the Raman spectroscopic information as input to a second neural network. The output of the second neural network is used to calculate the tangents for each of the ROIs. The classification is whether or not the patient has one or more AD-related pathologies of interest, such as protein aggregates. do.

[0067] In some instances, the classification (the output of the second neural network) determines whether the subject has AD. have AD, have a precursor to AD, or have been pre-screened for potential AD The results may be reviewed and one or more ADs may conclude whether further investigation is required. be. Such AD conclusions are then classified by a second neural network, e.g. Determined or calculated using combined weighted scores, scorecards, or probabilistic decisions The present invention can be based on one or more of the AD pathologies that are associated with the pathology of AD. For example, The presence or probabilistic classification of neurofibrillary tangles may lead to a higher probability conclusion of AD. There is a gender. In some instances, a conclusion about AD can be made by comparing, for example, the patient's previous Raman spectroscopy information. In some instances, the method may be based on changes in the patient's physiology over time. Hyperspectral reflectance information is a second neural network that further aids in the classification of AD pathology. It is also used as input information for the network.

[0068] Next, more details on training the neural network using the server 204 will be described. The verified training data 208 is stored in the client station 20. 6 and then transmitted by the client station 206 to the server 204. can be. In an exemplary embodiment, the validated training data 208 is a set of adjacent exercise data for subjects. By comparing so-called in vivo tissue slices, one slice was analyzed to determine the Hyperspectral reflectance and Raman spectroscopic information is acquired and adjacent slices are analyzed for histology. Hyperspectral reflectance information verified by cross-section and verified Raman scattering Optical information is obtained. For training the first neural network, we used a validated hyperspectral model. The reflectance information 210 is input to the client station 206. The collected hyperspectral reflectance information 210 may include a particular The wavelength counts are correlated with one or more AD-related pathologies. Verified Raman spectroscopy information 212 is available to client status for training purposes. In one example, the verified Raman spectroscopic information 212 is input to the ROI or correlates counts of specific wavelengths in the Raman maps to one or more AD-related pathologies.

[0069] In some instances, hyperspectral reflectance information may be used simply to determine the ROI. ,First, it can be used beyond training neural networks. For example, hyperspectral reflectance information could be used to aid in the classification of certain AD-related pathologies. It can also be used to train a second neural network to The reflectance information can be used together with Raman spectroscopy information to classify specific AD-related pathologies. Sometimes weighting or additional assurance is given. Similarly, machine learning algorithms could be used to classify specific AD-related pathologies. The correlations and relationships between the spectral information and the Raman spectroscopic information can be determined. Run the trained neural network on the hyperspectral images for classification. When using both reflectance and Raman spectroscopy information, to align the same ROI, The hyperspectral reflectance information and the Raman spectroscopic information are processed by a computer 6. Co-registration can be performed digitally.

[0070] In some instances, the ROI may include a group of pixels covering the plaque. In one example, the size of the plaque (e.g., the circular area indicated by the radius, or M The rectangular area represented by N × N pixels is used to classify AD-related pathologies. In some examples, the Raman spectroscopic information 212 may include a specific wavelength at the center of the ROI. It is possible to have higher counts in the vicinity of the ROI and lower counts at the periphery (but depending (still higher than background tissue). In some instances, individual counts in different pixels within the ROI indicate AD-related It can be used to classify disease states. In another example, the cohesive (integration) characteristics of a group of pixels within an ROI The properties can be used to classify plaques, for example, in a single Raman capture. Therefore, the size of the plaque ROI can be used as additional information for classifying plaques. It may also be part of the training of a second neural network that is used.

[0071] In some instances, the subject's background tissue Raman spectroscopic information 212 is also collected using validated training data. The Raman spectroscopic information of the background tissue of a given patient is used to identify the The comparison of the background tissue with the ROI can be used to identify AD-related pathologies. This may be part of training a second neural network to classify the shape of the Logistic regression, variational autoencoding, convolutional neural networks, and Other algorithms or calculations, including other statistical approaches, may be used to train the second neural network. It can be used for supervised training of

[0072] After the server 204 trains the neural network, the server 204 To run the neural network trained by the The trained neural network can be transmitted to the detection device 100. Computer 6 is then informed of the criteria that should be used to assess the AD-related condition of interest. Be known. Training updates to the neural network are periodically provided by the server 204. Run in real-time or whenever there is more training data available The updated neural networks can then be transmitted to the detection device 100.

[0073] In another example, at least some or all of the neural network may be stored on the server 204. The hyperspectral reflectance information and the Raman spectroscopic information detected by the , and control information is communicated between the server 204 and the computer 6. The server 204 receives the hyperspectral reflectance information from the computer 6 and By instructing the computer 6 as to what the ROI of the neural network spectroscopic section is, The server 204 receives Raman spectroscopic information from the computer 6. and classify AD-related symptoms or AD conclusions.

[0074] FIG. 11 is a block diagram of a method for detecting AD-related pathology in a subject's eye, according to an exemplary embodiment. 11 shows a flow diagram of a method 1100 implemented by the detection device 100 for detecting a smeared object. In an embodiment, the computer 6 of the detection device 100 performs at least some of the method 1100. Use neural networks for this. In step 1102, the detection device 100 includes a hyperspectral reflectance imager ( 1) and receives hyperspectral reflectance information from the hyperspectral reflectance imaging unit. By receiving the signal, wide-field imaging of the subject's fundus is performed. In step 1104, the computer 6 executes a hyperspectral reflectance imager and The hyperspectral reflectance information from the first neural network is used to further 2. Determine the location and size of one or more ROIs for subjects requiring testing. In step 1106, the detection device 100 controls the Raman spectroscopic section (FIG. 1) to determine The ROI is stimulated at a specified location and size in one capture, and Raman images are captured from the Raman spectrometer. By receiving the spectroscopic information, Raman spectroscopy is performed on the ROI. In step 1108, the second neural network calculates the Raman spectroscopic portion of the ROI. The Raman spectroscopic information obtained from the hyperspectral reflectance imaging unit and the hyperspectral The torr reflectance information is used to classify one or more AD-related conditions or a conclusion of AD. In step 1110, the detection device 100 outputs the classification to an output device (e.g., a display screen). , memory, or another computer. , 1110 are executed by the server 204. In an exemplary embodiment, where multiple AD-related pathologies are of interest to be detected In the method 1100, all AD-related pathologies are performed on one or more ROIs (parallel The same sample was used to determine whether all AD-related pathologies of interest were detected. If there is a positive diagnosis by the surgeon, histologist, pathologist, etc., the client station The method includes, but is not limited to, determining whether a subject has one or more AD-related symptoms or AD by such a practitioner. can be used to affirmatively (and independently) verify that the Validation serves as further training data to improve machine learning algorithms. It can be used in machine learning algorithms (primary and secondary neural networks).

[0075] In some examples, after step 1106, the detection system 100 further comprises a Raman spectrometer. Additional ROs identified by the hyperspectral reflectance imager that may require further investigation. To determine the Raman spectroscopy of I, a loop 1112 is included which returns to step 1102. It can be configured as follows. Loop 1112 is a loop for the same session, for example when the user is still holding the chin rest 5. For example, in step 1104, Data 6 may determine more than one ROI for a subject that may require further testing. So a loop 1112 is performed to search those other ROIs. The classification in step 1108 is performed using a hyperspectral reflectance imaging unit and a Raman spectroscopy unit. Provide a conclusion based on multiple different individual captures of the same subject taken by It is possible. In other examples, loop 1112 is not performed and the determination is made from hyperspectral reflectance imaging information. One Raman capture is performed for one ROI with a specific position and size. Only the command is executed.

[0076] In some examples, the computer 6 may include a processor 61 that receives the hyperspectral reflectance from the hyperspectral reflectance imager. - Uses spectral reflectance information to determine the ) to select a baseline ROI that is relevant to parts of the eye that are not subject to potential AD-related pathology. The baseline ROI can be analyzed using Raman spectroscopy. In step 1108, the computer 6 determines whether or not the patient is at risk for one or more AD-related conditions or AD. To classify conclusions, a baseline ROI is analyzed using Raman spectroscopy. It can be compared with one or more of I.

[0077] In some examples, in step 1104, computer 6 may One or more potential AD-related events of interest for the patient (validated from a known patient population) The series of lesions (or specific ROIs) were prestored. For example, using the detection device 100, a previous session may detect one or more potential AD-related The pathology of the joints was stored in advance. Specific landmarks were used to identify arteries, blood vessels, optic nerves, etc. One or more underlying AD-related pathologies in a particular patient may be identified. The hyperspectral reflectance imager and the hyperspectral image from the first neural network Using the spectral reflectance information, computer 6 calculates the patient's pre-stored potential A D Identify the relevant pathology (or specific ROI), determine the appropriate ROI, and then use the computer The data 6 is further adapted for a user to record Raman images while holding the chin rest 5, all during the same session. Survey the appropriate ROI using the spectroscopic section.

[0078] FIG. 12 illustrates a validated training of a neural network in accordance with an exemplary embodiment. 12 shows a flow diagram of a method 1200 for determining the data 208. The scanning data 208 is obtained by subjecting adjacent ex vivo tissue slices to a hyperspectral One slice was analyzed by reflectance imaging and Raman spectroscopy, and the other by tissue diagnosis. The slices are then compared to adjacent slices that have been verified using the same method. Exemplary results of the method 1200 include Raman spectroscopic information 212 shown in FIGS. Hyperspectral reflectance shown in the hyperspectral map 1000 shown in FIG. The rate information is 210. Additionally, in some instances, in vivo imaging from operation of the detection device 100 may be used to further It is also possible to obtain training data that is different from the training data.

[0079] In method 1200, ex vivo human brain tissue (cortex) from a deceased, confirmed AD patient is obtained. The samples were fresh frozen and formalin-fixed paraffin-embedded (FFPE). Both were used. In step 1202, the sample is sliced ​​and placed on a slide. Using a rotomy or cryostat, cut 12 μm thick slices of the sample. A series of adjacent slices were cut and placed on a microscope slide. . In step 1204, every third slice in the series is determined to contain amyloid beta stained with Congo red, which binds to erythrocytes, or by similar staining procedures, e.g., immunostaining. The remaining intermediate slides are left unstained. In step 1206, standard polarized light microscopy or other methods of tissue diagnosis are used. In this study, amyloid beta plaques were identified on stained slides. Histological diagnosis was based on clinical The method may be performed manually by a physician, automatically by a computer, or both. The typical size of brain plaques is greater than 20 μm in diameter. A line will likely span multiple 12 μm slices.

[0080] At step 1208, each stained slide having one or more plaques is The slides are then co-registered with adjacent unstained slides. This can be done automatically using the data, manually, or both. Co-registration of adjacent slides allowed identification of plaque locations on unstained slides. The location can be identified. Coregistration is achieved by identifying multiple features at various size scales. Cortical folds are large-scale structures that are used for the general orientation of two adjacent slices. Vessels co-register with neighboring slices at finer size scales. It uses multiple blood vessels in an image and groups them together to create small features that are used to When co-located in adjacent slices, a given plaque can be located within a few micrometers. By superimposing images of two adjacent slices, The alignment of the vessels allows for coregistration of the images in one example.

[0081] In step 1210, adjacent unstained slides are stained at corresponding locations on the stained slide. An imaging device is activated to the co-registered locations of the slide to image characteristics of AD-related pathology. In other instances, the entire adjacent unstained slide is stained to determine the identity of the stained tissue, e.g., background tissue, other stains, etc. Use imaging devices to obtain additional information about AD-related pathology, macro-structure, etc. In an exemplary embodiment, the imaging device includes: It can be hyperspectral reflectance imaging or Raman spectroscopy. In step 1212, after the imaging information of the plaque is acquired, the imaging information (or processed The plaque is classified as a plaque. The information 210, in this example, correlates with amyloid beta plaques. An example of the hyperspectral reflectance information 210 is the hyperspectral image shown in FIG. The Raman spectroscopic information 212 verified also correlates with amyloid beta plaques in this example. See FIG. 7. Examples of Raman spectroscopic information 212 are shown in FIG. 4 and FIG. 8, respectively. The Raman spectroscopic information shown is 500, 800.

[0082] The verified training data 208 is input to the client station 206; The data can be sent to the server 204 for training of the neural network.

[0083] The method 1200 further includes, to obtain the verified training data 208, This could be repeated for other AD-related pathologies, such as other protein aggregates and vascular characteristics. The non-plaque areas of background tissue detected using the imaging device are considered to be neural networks. The background tissue can be used as control information or negative classification information for training the network. It is also used to subtract or subtract from the count of an ROI to classify the ROI as being plaque. Other non-plaque regions can be used, for example, to determine the relative location of the area of ​​interest within the subject. It is possible to have macro structures for coregistration, such as for information. Texture detection can also be used to train neural networks.

[0084] Method 1200 may involve the collection of tissue samples from a single subject, such as both brain tissue and eye tissue. The method 1200 may be repeated for multiple tissue samples from a human subject. The assay can be repeated with tissue samples from different subjects. By using it, we determine the validated training data208 for machine learning algorithms. A sufficient sample set is used to measure the baseline or control training data. The data is obtained by acquiring hyperspectral reflectance information 210 and Raman spectroscopy information 212. Therefore, it can also be obtained from healthy (non-AD) subjects.

[0085] FIG. 14 illustrates a plaque 1402 (e.g., stained 14 shows a polarized light microscope image 1400 including a red dot (a red dot in the center of the image). The polarized light microscope image 1400 can be used to The Raman map 700 includes Raman spectroscopic information including the plaque 1402 in the same ROI. can be confirmed.

[0086] Figure 15 shows polarized light microscope images of stained slides at two different magnifications (10x, left; 7 (mirror image), corresponding to Raman map 700 in FIG. 7 (mirror image). The images show the spots 1502, 1504 (e.g., stained red dots). , the Raman map 700 is a Raman spectroscopic image including plaques 1502 and 1504 in the same ROI. It can be confirmed that the information is included.

[0087] FIG. 16 shows an unstained slice containing the same blood vessel 1602 seen in FIG. 15 in an adjacent slice. Hyperspectral image 1600 of Ride (left) and the same vessel in the same adjacent slice. 1604 and a polarizing microscope image of the same (right).

[0088] The Raman spectroscopic unit 1 can also be configured to take white light images. The same unstained slide as shown in FIG. 16 (same adjacent slide) was photographed by the spectroscopy section. FIG. 17A shows a white light image 1700 of the same blood vessel 17 shown in FIG. 02 can be located and captured by the detection device 100 using the Raman spectroscopy section. This indicates that.

[0089] FIG. 17B shows a white light image 1700 of FIG. 17A, captured using Raman spectroscopy. The region 1704 of the blood vessel 1702 is shown. 1600, and may have a particular location and size as determined by the label image 1600. In another example, the region 1704 can be captured in one capture by the Mammography section. Raman analysis to generate maps or calculate integrated counts for specific wavelengths The optical section can be used to scan pixel by pixel.

[0090] Referring again to FIG. 1, in some examples, the detection apparatus 100 detects potential Other types of imaging devices are being used to determine the specific wavelengths that are characteristic of the chemical components of AD pathology. In one embodiment, the imaging device may include a broadband light source and a second The light source is used as a coherent single-wavelength light source of a specific wavelength, using the inverse Raman effect. will be done.

[0091] In another example, the imaging device uses stimulated Raman scattering (SR) with two coherent lasers at specific wavelengths. In another example, a pulsed light coherent illumination source is used to The imaging device is implemented using automated fluorescence measurements at several different wavelengths.

[0092] In some instances, alternative imaging devices such as white light non-hyperspectral fundus cameras may be used. This additional imaging device can be used to position the wide field of view of the hyperspectral reflectance imager. Such positioning can be used to guide the positioning of the image from this additional imaging device. The information can be used to automatically perform and / or operate the computer 6. The computer 6 may be implemented using machine learning and one or more Neural networks can be used to perform the positioning automatically.

[0093] In an exemplary embodiment, several components of the system 200 are subject to relative motion, vibration, and and to reduce the amount of extraneous electromagnetic radiation entering the system 200, Confined and surrounded.

[0094] In an exemplary embodiment, computer 6, server 204, and any of system 200 The device includes one or more communication subsystems (wired or wireless) and one or more control subsystems. A controller may include a controller that is responsible for a particular application, component, or function. Depending on the function, hardware, software, or a combination of hardware and software It can be provided with: In some exemplary embodiments, the one or more controllers may be analog or digital. The present invention may include components such as one or more processors, one or more non-volatile storage media, such as memory, that store executable instructions; and / or one or more The above analog circuit components may be included.

[0095] Exemplary embodiments include a method for detecting one or more AD-related pathologies from an eye of a subject. A non-invasive in vivo ocular light-based detection device comprising a broadband light source and a hyperspectral camera. a hyperspectral reflectance imaging section including a laser; and a Raman spectroscopy section including a laser and a spectrometer. a memory and one or more processors configured to execute the following instructions stored in the memory: and a processor for controlling the hyperspectral reflectance imager to obtain a broadband light The source is used to illuminate a wide field of the fundus of the eye and determine hyperspectral reflectance information. To this end, a hyperspectral camera is used to measure reflected and / or backscattered light from the eye. Detect and identify potential AD-related disorders from hyperspectral reflectance information in one or more ROIs. The pathology of the lesion is determined by illuminating each of one or more ROIs using a laser. A spectrometer is used to control the spectroscopic section and determine Raman spectroscopic information obtained from the laser. The Raman scattered light from the eyeball is detected, and the hyperspectral reflectance information and Raman spectroscopic information are obtained. The information was used to classify subjects as having one or more AD-related conditions and to determine whether one or more AD-related pathology involves protein aggregates, which are then transformed into tau neurofibrillary tangles. amyloid beta deposits, soluble amyloid beta aggregates, or amyloid precursor proteins. It contains at least one of the following proteins:

[0096] In any of the above exemplary embodiments, the subject is a person having one or more AD-related symptoms. The classification as being based on hyperspectral reflectance information is further based on hyperspectral reflectance information.

[0097] In any of the above exemplary embodiments, the subject is a person having one or more AD-related symptoms. Classifying a subject as having a high level of cognitive impairment is done by examining the subject's previous high level of cognitive impairment stored in memory or another device. It is further based on perspectral reflectance information and / or Raman spectroscopy information.

[0098] In any of the above exemplary embodiments, the subject is a person having one or more AD-related symptoms. Classification as a hazard is based on hyperspectral reflectance information of subjects over time and / or is further based on the change in Raman spectroscopic information.

[0099] In any of the above exemplary embodiments, the subject is a person having one or more AD-related symptoms. Classifying a subject as having more than one AD-related condition does not classify a subject as having more than one AD-related condition. This includes:

[0100] In any of the above exemplary embodiments, the one or more processors may further include a hypervisor. Baseline ROIs from spectral imaging information were placed on dorsal surfaces that did not contain potential AD-related pathology. The Raman spectrometer is controlled to detect the base line of the eyeball using a laser. Use a spectrometer to illuminate the ROI and determine the Raman spectroscopic information of the background tissue. The method further includes detecting and classifying light from the eye generated by the laser, the light being detected by the eye being detected by the laser. Based on the Raman spectroscopy of typical AD-related pathologies compared with that of background tissue, Made.

[0101] In any of the above exemplary embodiments, the one or more AD-related pathologies are tau neurogenic Includes two or more AD-related pathologies, including fibrotic changes.

[0102] In any of the above exemplary embodiments, the one or more AD-related symptoms are including bulbar neuritis or glial cell pathology, or vascular characteristics of the subject's ocular bulb or choroid.

[0103] In any of the above exemplary embodiments, the amyloid-beta vector is administered to one or more AD-related pathologies. When data deposits are included, classification is made between alpha helical and beta pleated. The Raman vibrational resonance at 1600 cm -1 From 1700cm -1 Wave number shift in the range The technique is based on analyzing Raman spectroscopic information at 100 nm.

[0104] In any of the above exemplary embodiments, the one or more AD-related pathologies are characterized by tau neuronal origin. If the protein contains a fiber change, the classification is based on the Raman vibrational resonance of phosphorylated tau. 600cm -1 From 1700cm -1 By analyzing Raman spectroscopic information at wavenumber shifts in the range Based on.

[0105] In any of the above exemplary embodiments, the one or more processors may include one or more ROs. To determine I or to classify a subject as having one or more AD-related conditions The method uses machine learning algorithms for one or both of the following:

[0106] In any of the above exemplary embodiments, the machine learning algorithm may be based on a verified traceability system. Use learning data.

[0107] In any of the above exemplary embodiments, the verified training data is collected from the subject. The ex vivo tissue samples were sliced ​​into tissue slices, and the tissue slices were then sliced. Place the first slide of one of the tissue slices in the slide holder and stain and extract the tissue slices. In vivo tissue samples, a second unstained tissue slice was placed adjacent to the first tissue slice. providing a second slide having a tissue profile and using tissue diagnostics to determine whether the first slide has one or more Verify that the subject has AD-related pathology and detect at least one imaging device in the second slide. Activating the device to obtain imaging information and classifying the imaging information as one or more AD-related pathologies. It is obtained by:

[0108] In any of the above exemplary embodiments, at least one imaging device is configured for Raman spectroscopy. a hyperspectral reflectance imaging unit, or both.

[0109] In any of the above exemplary embodiments, the machine learning algorithm may include one or more Use the global network.

[0110] In any of the above example embodiments, the one or more processors further comprise: (ii) have one or more AD-related medical conditions; and Further train the machine learning algorithm using independent validation of the It is configured as follows.

[0111] In any of the above exemplary embodiments, the one or more processors may include a processor for processing the Raman spectroscopy information. The information provided by this study indicates that AD, or a precursor to AD, or a potential pre-existing condition that warrants further investigation. Classification of subjects as screened in or responsive to a treatment or intervention The present invention is further configured to

[0112] In any of the above exemplary embodiments, for classification of one or more AD-related pathologies, No exogenous fluorescers, dyes, or tracers are required.

[0113] In any of the above exemplary embodiments, the detection device further comprises: one or more optical filters for filtering out the wavelengths of the laser before .

[0114] In any of the above exemplary embodiments, the one or more processors may Determine the respective size of each of the one or more ROIs from the total reflectance information and perform Raman spectroscopy. The eyeball ROIs are each configured to emit a laser beam onto the ROIs of the eyeball having a respective size by controlling the It is done.

[0115] In any of the above exemplary embodiments, the Raman spectroscopic unit is In order to measure the intensity of each of the one or more ROIs, the laser of the Raman spectroscopy section is used to measure the intensity of each of the ROIs. It is controlled by one or more processors to perform scanning of the ROI.

[0116] In any of the above exemplary embodiments, the hyperspectral camera includes: It consists of a two-dimensional array of photosensors that are sensitive to a range of wavelengths of light, and each individual filter detects a specific wavelength. and a two-dimensional filter array overlaying the array of photosensors that selectively transmit light. .

[0117] Another exemplary embodiment is directed to non-invasive biopsy of one or more AD-related pathologies from a subject's eye. A method for in-vivo detection, comprising controlling a hyperspectral reflectance imager to use a broadband light source. To illuminate a wide field of the fundus of the eye using a hologram and determine hyperspectral reflectance information. To achieve this, a hyperspectral camera is used to detect light from the eye that originates from a broadband light source. and extracting from the hyperspectral reflectance information using one or more processors: Identifying the location of one or more ROIs as potential AD-related pathologies and controlling a Raman spectroscopy unit to illuminate each of the one or more ROIs using the Raman To determine spectroscopic information, a spectrometer is used to measure the Raman scattering from the eye obtained from the laser. Detecting turbulent light and using one or more processors to extract hyperspectral reflectance information Using the genomic and Raman spectroscopic information, subjects were identified as having one or more AD-related conditions. and classifying the one or more AD-related pathologies as comprising protein aggregates and protein Protein aggregates are expressed as tau neurofibrillary tangles, amyloid beta deposits, and soluble amyloid beta. The protein comprises at least one of the following: amyloid aggregates, amyloid precursor protein, or amyloid precursor protein.

[0118] Another exemplary embodiment is a computer program using a machine learning training process. a computer program product stored on a non-volatile computer readable medium; When the image is stored in the memory and executed by the computer, the image is transmitted to the computer from the subject's eyeball. Instructions for non-invasive in vivo detection of one or more Alzheimer's disease (AD)-related pathologies The machine learning training process uses one or more processors to training a computer program using the training data; Validated training data was obtained by slicing ex vivo tissue samples from subjects. 2. Prepare tissue slices and place the tissue slices on slides. Place the first tissue slice on the first slide. Stained slices were placed adjacent to the first tissue slice in the ex vivo tissue sample, and unstained slices were placed adjacent to the first tissue slice in the ex vivo tissue sample. Provide a second slide having a second tissue slice of color and stained using a histology diagnostic. Verify that the first tissue slice has one or more AD-related pathologies and verify that the second slice Activating at least one imaging device in the image forming apparatus to obtain imaging information, and processing the imaging information by Obtained by classifying as one or more AD-related pathologies and one or more AD-related The pathology includes protein aggregates, which include tau neurofibrillary tangles, amyloid Amyloid beta deposits, soluble amyloid beta aggregates, or amyloid precursor protein sacs Includes at least one of the following:

[0119] In any of the above exemplary embodiments, the verified training data is a first set A further result is obtained by co-registering the tissue slice with a second tissue slice.

[0120] In any of the above exemplary embodiments, at least one imaging device is configured for Raman spectroscopy. A Raman spectrometer and a hyperspectral reflectance imaging unit, or a Raman spectrometer and a hyperspectral reflectance imaging unit Both the image part and the

[0121] In any of the above exemplary embodiments, activating at least one imaging device are a set of small sets of images that are collectively used to classify imaging information as one or more AD-related pathologies. This involves operating at least two imaging devices.

[0122] Another exemplary embodiment, when executed by a computer, causes the computer to ,Perform non-invasive in vivo detection of one or more AD-related pathologies from a subject's eyeball.,Note 1. A method for machine learning training of a computer program stored in a memory, comprising: The method uses one or more processors to perform a test using the validated training data. training a computer program, the verified training data comprising: Slicing an ex vivo tissue sample from a subject into tissue slices; The first tissue slice on the first slide was stained and the ex vivo tissue was stained. A second unstained tissue slice is adjacent to the first tissue slice in the tissue sample. Provide a second slide with the stained first tissue slice and use a histology diagnostic to 1. Verify that the patient has at least one AD-related condition on the second slide. Activating the imaging device to obtain detection information, and identifying the detection information as one or more AD-related pathologies. The classification is based on the presence of one or more AD-related pathologies, including protein aggregates. Protein aggregates include tau neurofibrillary tangles, amyloid beta deposits, and soluble amyloid. and at least one of amyloid beta aggregates, amyloid precursor protein, and further comprising storing the trained computer program in a memory. .

[0123] Although some of the present embodiments are described in terms of methods, those skilled in the art will appreciate that the present embodiments are in the form of hardware components, software, or any combination of the two, or in any other manner, Components for carrying out at least some of the aspects and features of the methods described herein may be provided. The present invention is also directed to a variety of devices, such as processors, circuitry, and controllers, including You will understand.

[0124] In the figures, where appropriate, at least some or all of the illustrated subsystems or blocks may a processor that executes instructions stored in a memory or a non-volatile computer readable medium; Combinations and subcombinations of any of the above. Modifications may be made to some of the exemplary embodiments, which may include: are merely examples and are not meant to limit the scope of the present disclosure in any way. Variations of the innovations described herein will be apparent to those skilled in the art having the benefit of the illustrative embodiments. Obviously, such variations are within the intended scope of the present disclosure. Features from one or more of the embodiments described herein may be combined in alternative embodiments comprising subcombinations of features. may be selected to create a form, which may not be explicitly described above. Furthermore, features from one or more of the above embodiments may be selected and combined to achieve the above-described To create alternative embodiments consisting of combinations of features that may not be expressly described. Features appropriate for such combinations and subcombinations may be included in any number of combinations and subcombinations, as considered in their entirety. The subject matter described herein is intended to be within the skill of the art and should be readily apparent to those of ordinary skill in the art. It is intended to encompass and include all changes that may occur.

[0125] Certain adaptations and modifications of the described embodiments may be made. The embodiments are considered to be illustrative and not restrictive.

Claims

1. A computer program product from a machine learning training process, comprising: instructions stored on a non-volatile computer readable medium and that, when executed by a computer, cause the computer to classify the subject as having one or more Alzheimer's Disease (AD) related pathologies using hyperspectral reflectance information or Raman spectroscopy information obtained from the subject's eye; The machine learning training process includes: training a computer program using the validated training data, using one or more processors; The verified training data is Slicing an ex vivo tissue sample from the subject into tissue slices; placing the tissue slice on a slide; staining a first tissue slice of a first slide; providing a second slide having a second unstained tissue slice adjacent to the first tissue slice within the tissue sample; verifying that the stained first tissue slice has the one or more AD-related pathologies using a tissue diagnosis; activating at least one imaging device at the second slide to obtain imaging information; the imaging information includes the hyperspectral reflectance information or the Raman spectroscopy information, and the imaging information is obtained by classifying the one or more AD-related pathologies; The one or more AD-related pathologies include protein aggregates, the protein aggregates comprising at least one of tau neurofibrillary tangles, amyloid beta deposits, soluble amyloid beta aggregates, or amyloid precursor protein.

2. 2. The computer program product of claim 1, wherein the validated training data is further obtained by co-registering the first tissue slice and the second tissue slice.

3. 10. The computer program product of claim 1, wherein the at least one imaging device is a Raman spectroscopy unit, a hyperspectral reflectance imaging unit, or both the Raman spectroscopy unit and the hyperspectral reflectance imaging unit.

4. 2. The computer program product of claim 1, wherein operating the at least one imaging device comprises operating at least two imaging devices that are collectively used to classify the imaging information as the one or more AD-related pathologies.

5. 1. A method for machine learning training of a computer program stored in a memory, which, when executed by a computer, causes the computer to classify a subject as having one or more Alzheimer's Disease (AD) related pathologies using hyperspectral reflectance or Raman spectroscopy information obtained from an eye of the subject, comprising: training said computer program using validated training data, using one or more processors; The verified training data is Slicing an ex vivo tissue sample from the subject into tissue slices; placing the tissue slice on a slide; staining a first tissue slice of a first slide; providing a second slide having a second unstained tissue slice adjacent to the first tissue slice within the tissue sample; verifying that the stained first tissue slice has the one or more AD-related pathologies using a tissue diagnosis; activating at least one imaging device at the second slide to obtain detection information; the detection information includes the hyperspectral reflectance information or the Raman spectroscopy information, and is obtained by classifying the detection information as the one or more AD-related pathologies; The one or more AD-related pathologies include protein aggregates, the protein aggregates comprising at least one of tau neurofibrillary tangles, amyloid beta deposits, soluble amyloid beta aggregates, or amyloid precursor protein, and further comprising storing a trained computer program in the memory.

Citation Information

Patent Citations

  • Spectral data analysis device, biological substance detection system and biological substance detection method

    JP2012032368A

  • Optical methods for detecting Alzheimer's disease

    JP2012503012A

  • Systems and methods using composite optical probes

    JP2017534321A

  • Methods of Identifying Disease Biomarkers in the Lense of the Eye

    US20100105098A1

  • Methods and magnetic imaging devices to inventory human brain cortical function

    US20170281071A1