WO2010129711A1 - Dispositifs, systèmes et procédés d'évaluation de la vision, et diagnostic et compensation de l'altération de la vision - Google Patents

Dispositifs, systèmes et procédés d'évaluation de la vision, et diagnostic et compensation de l'altération de la vision Download PDF

Info

Publication number
WO2010129711A1
WO2010129711A1 PCT/US2010/033780 US2010033780W WO2010129711A1 WO 2010129711 A1 WO2010129711 A1 WO 2010129711A1 US 2010033780 W US2010033780 W US 2010033780W WO 2010129711 A1 WO2010129711 A1 WO 2010129711A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
data
image
cells
lgn
Prior art date
Application number
PCT/US2010/033780
Other languages
English (en)
Inventor
Paul Sajda
Jianing Shi
R. Theodore Smith
James Wielaard
Original Assignee
The Trustees Of Columbia University In The City Of New York
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by The Trustees Of Columbia University In The City Of New York filed Critical The Trustees Of Columbia University In The City Of New York
Publication of WO2010129711A1 publication Critical patent/WO2010129711A1/fr

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • Macular disease such as age-related macular degeneration (AMD), diabetic retinopathy (DR), and macular dystrophy (MD) account for the overwhelming majority of blindness in the United States Approximately 13 million people in the U S ha ⁇ e some signs of macular degeneration, with the rate of new cases dramatically increasing
  • the disclosed subject matter provides mechanisms for quantifying impairment of vision and for using quantitative information to provide various benefits which are disclosed Embodiments were conceived in light of the need identified above, among other things
  • the embodiments can provide, for example, a psychophysics paradigm for quantitat ⁇ e assessment of visual perception, a framework to predict functional perception based on retinal images, and a computational system that can be integrated into existing ophthalmology devices
  • Embodiments of the disclosed subject matter employ a model of the vision system to predict the impact of drusen or other kinds of impairments of the retina on the ability to recognize objects
  • the disclosed subject matter provides a means for quantifying the effects on vision
  • the ability of different parts of the retina to be used for visual discrimination, such as face recognition is determined using the visual system model
  • Various other applications and techniques are disclosed for determining how the visual cortex responds to the phenotype of retinal impairment, for instance scotoma, and distortion of retinal input and how this changes the ability to recognize objects
  • the embodiments of the disclosed subject matter may be applied to evaluate and predict vision loss for patients with macular disease, and can also be applied to patients with other types of vision impairment, such as glaucoma Glaucoma is another big market for low- vision treatments, apart from age-related macular degeneration (AMD)
  • AMD age-related macular degeneration
  • An example of a vision system model may accept an impaired retinal image, for example, a 256 by 256 pixel image with 8 bit per pixel luminance
  • the impaired retinal image may be obtained by taking an image of a target and modifying by masking with an image, derived from a patient, of a drusen distribution, for example
  • the impaired retinal image consequently serves as input to an anatomically and physiologically realistic model of Vl
  • the neuronal activity of the Vl model may be reduced to a recogition result, such as discriminating a car from a face, using, for example, a linear decoder that exploits the spatial- temporal dynamics generated by the Vl model neurons For experiments, this was demonstrated using a 11 -regularized logistic regression model In the experiments, the decoding accuracy of the system against psychophysics experimental results was demonstrated
  • Embodiments may include approaches for e ⁇ aluating patients using a psychophysics testing system and predicting vision loss and/or treatment outcome using a computational model that transforms data representing physical characteristics of patients into a quantitative output value predictive of vision loss and/or treatment outcome
  • the approaches may be used independently or in parallel
  • the psychophysical approach includes carrying out a battery of psychophysics experiments on patients to determine a patient's visual impariment or competence
  • the computational approach models the impact of patient's retinal image distortion on visual perception using a model of the visual system that includes the primary visual cortex For example, impairment may be derived from retinal fundus images obtained by a retinal imaging device
  • the psychophysics approach and the computational approach can be used to validate and fine tune the system reciprocally, referred to as "tuning mode" above
  • the psychophysics testing can provide ground truth to validate the output from the s> stem computational model described above
  • the system computational model can provide output data that can be used to tune the psychophysics paradigm and identify a battery of psychophysics experiments that elicit the perceptual responses that are senstitive, specific and robust to the disease process in study
  • Embodiments may include generating output data used to identify a relationship between population neuronal actmty, ⁇ isual input and visual reception Also, embodiments can generate output results for linking retinal pathology, resulting from macular disease, with visual function assessed via psychophysical test result data obtained from patients
  • the psychophysics tests can be carried out on a system having a computer (or processor), a patient response input de ⁇ ice, and an experimental paradigm adapted to test patients and to probe retinal disease and obtain results that can be used to vet output from a system computational model
  • the system may present a series of images to a patient
  • the patient can be asked to identify the object in the image, for example is the image of a face or a car?
  • the patient presses a button or other user input device on the response device and the computer records the patient answers
  • the computer can also record the reaction time (or length of time it took the patient to answer), which can be used in evaluating the vision of the patient
  • An exemplary input to the model is retinal fundus images of patients
  • the computational model or system can generate a prediction for vision perception as output
  • the output can be used in a number of ways
  • An exemplary system or method can include two primary stages or modes.
  • the system can operate in a tuning mode in which it can gather data and refine or "tune" its responses through analysis, transformation, and evaluation of input data — e g , retinal imagery — and output data
  • the system can operate in an evaluation mode in which it is processes input data to provide a prediction of vision loss Even while operating in the evaluation mode, the system can continue to collect data and tune or learn from the collected data sets Weights in the regression can be adjusted to provide the best prediction results
  • the system tuning affords greater accuracy with respect to the predictive results generated as part of the evaluation mode This dual-mode aspect can be included with any and all of the embodiments described herein
  • the system computational model may link the specific pathology characteristics in the patient's retina into vision loss, and can rule out irrevelant characteristics in the retina that are not part of the disease in study Therefore, the computational approach can validate the specificity of the psycho physics testing with respect to the patient ' s disease, and consequently be used as feedback to optimize the psychophysics testing regarding its disease specificity
  • Such a feedback mechansim can be used to design a batter)' of ps> chophysics experiments for evaluating patients' ⁇ ision loss
  • system embodiments may provide a quantitative test for the effectiveness of a clinical treatment for macular disease
  • a longitudinal study can be conducted on patients having macular disease in which the system can be used to track the vision changes (improvement or decline) with clinical treatment, especially drug treatment
  • Clinical usage of the system may accelerate and improve the quality of drug development during the clinical trial stage
  • embodiments may be used as a stand-alone system or integrated into a traditional or functional ophthalmology imaging device to add a functional component
  • the psychophysics aspect of the embodiments may be used as a clinical testing procedure that detects early signs of macular disease, which can provide clinicians impro ⁇ ed opportunity to control the disease at its early stage rather than late stage
  • early symptoms of AMD can be probed using certain psychophysics testing
  • the psychophysics test may also provide new information to the disease assesssment at late stage, thereby adding a dimension to the clinical treatment decision
  • the computational aspect of the embodiments provides predictions a quantitative prediction of patients' visual performance This prediction can provide valuable information to the physician and add a dimension to the clinical database for improved clinical treatment decisions
  • the prediction data can be used to suggest at what stage a patient's vision loss is currently.
  • the embodiments can help identify endpoints for treatment outcome in clinical trials, with respect to specificity and robustness. Such a decision may requireidentification of clinical tests that are specific to the disease process in study and quantitative metrics that are robust. The ability of using our system in both tuning mode and evaluation mode serves this purpose.
  • any of the embodiments described herein can include a psychophysics response curve or psychometric curve derived from psychophysics experiment responses, and a simulated neurometric curve derived from retina image data, to perform comparisons.
  • a goal may be to adjust the system to match the neurometric curves with the psychometric curves.
  • retinal image data for patients with macular disease can be introduced in the evaluation mode to provide vision loss predictions Changes in a patient's vision over time can be represented by shifts in the neurometric curve.
  • the embodiment regarding the batter ⁇ ' of psychophysics experiments may be used as a diagnostic tool to probe early symptoms of AMD, especially face recognition.
  • the test for face recognition proceeds the detection of AMD using traditional clinical metrics such as visual acuity test.
  • Such information may be used to define endpoints for clinical trials for gene therapies, especially when such therapies aim to slow down and/or stop AMD progress from early stage to late stage.
  • the model of Vl is anatomically and physiologically realistic, and incorporates information both spatially in the visual stimuli and temporally during the patient's viewing.
  • Conductance based description of the neuron makes it feasible to implement such a model not only on computers but also on hardwares such as neuromorphic chips.
  • the computational system of Vl model may be used as a tool to optimize retinal prosthesis device.
  • Retinal prosthesis surgery is at the frontier of treating retinitis pigmentosa (RP) and blindness, by implanting an artificial retina to restore vision and to eventually allow patients to read and recognize faces
  • RP retinitis pigmentosa
  • the retinal prosthesis involves expensive and high-risk surgery. It remains unclear what is the optimal way to stimulate the retina.
  • Our technology of mapping retinal implant stimulation into perception can be used to optimize the device design of the artificial retina and thus to improve the surgical outcome by predicting visual outcome a prion. This breakthrough will become a milestone for the next generation of retinal implant.
  • Another key component in the computational system is a linear classifier that has the ability to automatically determine informative dimensions, in other words, to carry out feature selection within high dimensional data. This is achieved by mathematically using Ll-regularization, a sparsifying operator, combined with novel numerical algorithms to efficiently and accurately carry out optimization.
  • the sparse classifier provides attractive benefits, including coping with high dimensionality in the data providing robustness for classification in face of noise, identifying important features.
  • Such a sparse classifier may be used in other applications related to feature selection, such as gene classification, image classification and neural imaging data analysis.
  • a sparse classifier may be used to identify important gene factors regulating disease progress from early AMD to late AMD, for example, and hence enable treatment to slow down and/or stop the disease by targeting identified gene factors through gene therapy
  • a sparse classifier may be used to analyze neuro imaging data recorded via electroencephalography (EEG), while a subject is performing various cognitive tasks, such as discrimination, recognition, triage, or attention.
  • EEG electroencephalography
  • the sparse classifier can identify important neural signatures during the subject's decision making process that can be used to optimize visual perception.
  • Another application of the sparse classifier is that it may accelerate the identification of important images.
  • Fig. 1 shows systems for quantifying vision impairment including a first system for quantitatively predicting recognition competence based on pathology such as a retinal scan and a second system for performing psychophysical tests to make the same predictions about a particular patient; both systems being usable together to assess one system against the other, to provide alternative or combinable predictions, or being usable as separate systems; the predictions being usable for treating a patient, for evaluating compensation strategies, for screening or comparing treatment efficacies for drugs, or for any other purpose.
  • pathology such as a retinal scan
  • a second system for performing psychophysical tests to make the same predictions about a particular patient
  • both systems being usable together to assess one system against the other, to provide alternative or combinable predictions, or being usable as separate systems
  • the predictions being usable for treating a patient, for evaluating compensation strategies, for screening or comparing treatment efficacies for drugs, or for any other purpose.
  • Fig. 2 shows a method for determining a compensation strategy for a patient with impaired vision.
  • Fig 3 shows simulating cortical activity in the Magno cortex (MO) under drifting grating stimuli
  • MO Magno cortex
  • a The average firing rates over the drifting grating c ⁇ cle for the Magno cortex, simulated for a normal subject
  • b The average firing rates over the drifting grating cycle for the Magno cortex under retinal impairment, simulated with one example drusen mask A decrease in the overall firing rates is observed
  • c The distribution across all neurons The lowest-peaking curve denotes the CV distribution for the normal subject, while other curves with other colors indicate the CV distribution for 5 AMD patients The distribution of CV is consistently shifted to the right for the AMD patients, compared to the normal subject, indicating a reduction in the orientation selectivity of the neurons
  • Fig 4 shows simulated cortical activity in the Parvo cortex (PO) under drifting grating stimuli
  • a The average firing rates over the drifting grating cycle for the Parvo cortex, simulated for a normal subject
  • b The average firing rates over the drifting grating cycle for the Parvo cortex under retinal impairment, simulated with one example drusen mask
  • c The distribution across all neurons The first curve denotes the distribution for the normal subject, while the other curves indicate the distribution for 3 AMD patients The distribution remains unshifted for the AMD patients, compared to the normal subject, indicating a retention in the orientation selectivity of the neurons
  • Fig 5 shows psychometric curves for both control subjects (upper curve) and AMD patients (lower curve), constructed from beha ⁇ ioral data Both curves represent averaged psychometric performance across subjects The errorbar indicates standard error Clearly the AMD patients suffer from lower discrimination accuracy compared to control subjects The degradation of patient performance is more pronounced at higher phase coherences
  • Fig 6 compares predicted neurometric curves with psychometric curves (a) Average psychometric curve for the control subjects (lower curve) is plotted, together with the average neurometric curve (upper curve) based on the simulated model with normal vision (b) Average psychometric curve for the AMD patients (lower curve) is plotted, together with the average neurometric curve (upper curve) based on the model with retinal impairment In all curves, the errorbar indicates standard error
  • Fig. 7 illustrates three patient cases, where the neurometric curves are good predictor for the psychometric curves (a) RF fundus image of the patient (b) Binary retinal mask used as input to the Vl model (c) Comparison of the neurometric curve (upper curve) and the corresponding psychometric curve (lower curve) of one patient, plotted against the individual psychometric curves (dashed) for all the patients
  • Fig 8 illustrates three patient cases, where the neurometric curves are different from the psychometric curves (a) RF fundus image of the patient (b) Binar> retinal mask used as input to the Vl model (c) Comparison of the neurometric curve (upper curve) and the corresponding psychometric curve (lower curve) of one patient, plotted against the individual psychometric curves (dashed) for all the patients
  • Fig. 9 shows statistical analysis for establishing the correlation between the fundus image, model prediction and behavior data
  • Fig 10 illustrates a perceptual decision making experimental design, a two- alternative forced choice paradigm for face versus car discrimination Images are flashed for 50 ms, followed by an interval of 200 ms with the same mean luminance as the stimulus By varying the phase coherence one can manipulate the evidence in the stimuli for face or car The same set of stimuli is used for the human psychophysics experiment and Vl model simulation
  • Fig 11 illustrates image processing and analysis for experiments
  • Fig 12 illustrates an exemplary model architecture
  • A A select cluster of ON (circles) and OFF (enlarged dots) M-LGN cells that feed into one cortical cell Receptive field centers of LGN cells are organized on a square lattice (array of smaller dots)
  • B Select M-LGN axons in our model-Vl Points in the circular clusters are cortical cells that connect to the same LGN cell
  • C Pmwheel structure and ocular dominance columns for MlO model, constructed from averaged responses in the spirit of optical imaging experiments Samples used to analyze the two extraclassical phenomena are made up of cells from within the white dashed rectangle
  • Fig 13 shows select classical response properties of the MO version of the model, other configurations yield similar results
  • A Spatial distribution of orientation selectivity, expressed in the circular variance (CV). for cells within the white rectangle in Fig. 12(c) Black pixels are cells that do not show sufficient response for this monocular stimulation
  • B Spatial frequency tuning curves for select MO cells Thick black curve refers to the LGN cells, which are all identical in this respect
  • C Distribution of preferred spatial frequencies for the MO cells
  • D Temporal frequency tuning curves of some MO cells
  • Thick black curves refer to the LGN cells, which are all identical in this respect
  • E Distribution of preferred temporal frequencies for the MO cells Histograms show fraction of cells, arrows indicate mean ⁇ alue
  • Fig 14 shows response modulations in the the model for (large) drifting grating stimuli with average optimal spatial and temporal frequency of the cortical cells
  • A Response waveforms for a simple and a complex cell, with optimal spatial and temporal frequencies close to the grating values, for a number of different grating orientations (angles).
  • B Distributions (normalized to peak value one) of Fls/FOs for spike responses and Flv/FOv for membrane potential responses
  • C Spatial distribution of Fls/FOs (spike train) for cells within the white rectangle in Fig 12(c) Black pixels are cells that do not show sufficient response for this monocular stimulation
  • Fig 15 shows response modulations in the model for (large) standing grating (contrast reversal) stimuli with the average optimal spatial and temporal frequency of the cortical cells
  • A Response waveforms for a simple and a complex cell with preferred orientation, optimal spatial and temporal frequencies, close to the grating values, for a number of different spatial phases
  • B Distributions (normalized to unity) of the phase averaged F2s/Fls ratio for spike tram (top) and F2v/Flv for membrane potential (bottom) responses to a contrast reversal grating at the preferred orientation
  • Sample contained 1200 cells
  • Fig 16 shows a summary of surround suppression and receptive field growth at low contrast in the model
  • A Responses (FIs, FIv) as function of aperture size r A for a cell from the MO model Shown are firing rate (first and third curves from (deg) axis) and membrane potential (second and fourth curves from (deg) axis) for high (squares) and low (circles) contrast Standard errors are negligibly small
  • B Responses (FIs) for another cell from the MO model Shown are firing rate as function of the aperture size r A (top two curves) and the response to the inverse stimulus (bottom two curves), i e a stimulus where the aperture is "blank" and is surrounded by drifting grating Responses are again shown at high (solid) and low contrast (dashed) Stimulation solely from outside the cell's receptive field r + does not e ⁇ oke any response, believed to be a signature of extraclassical responses (C &
  • D Receptive field and surround sizes for the PlO model at high (unfilled) and low (shaded) contrast.
  • the diversity of responses produced by the model is similar to what is seen in experimental data (E) Distribution for the MO model of the suppression index 57, at high (unfilled) and low (shaded) contrast All suppression is exclusively cortical in origin and due solely to short-range connectivity (F) Distributions for the MO model of the ratios of the receptive field and surround sizes at low and high contrast, r Ir + (shaded) and R IR + (unfilled) (Wilcoxon test on ratio larger than unity p ⁇ O 001 for both receptive field and surround growth) In panels C-F, histograms give fraction of cells, arrow indicate means, solid arrow correspond to shaded histograms, dashed arrow correspond to unfilled histograms
  • FIG 17 shows relations between some key response measures for the MO configuration of the model, other cases yield qualitatively similar results
  • B Scatter plot of surround suppression at low and high contrast expressed in the two different suppression indices SI 1 (black) and SI 2 (grey)
  • C Scatter plot of surround suppression in spike tram and membrane potential at high (black) and low (gray) contrast
  • D First harmonic (Fl) of spike responses, membrane potential responses, and cortical conductances as function of aperture size for a model simple cell which shows about 50% surround suppression (in spike train) Notice the surround suppression of the conductances
  • Fig 18 shows (A) Joint distribution of high and low contrast receptive field sizes. r + and r , based on spike responses All scales are logarithmic (base 10) All distributions are normalized to peak value one Receptive field growth at low contrast is clear Average growth ratio is 1 9 and is significantly greater than unity (Wilcoxon test, p ⁇ 0 001) (B & C) Joint distributions of receptive field growth and growth of spatial summation extent of excitation (B) and inhibition (C) (computed as ratios) There is practically no correlation between receptive field growth and the growth of the spatial summation extent of excitatory or inhibitory inputs (see text) For cells in the sample with larger receptive field growth (factor of : 2 and greater) this growth is always considerably larger than the growth of their excitatory and inhibitory inputs
  • Fig 19 shows two example cells, an MO simple cell which receives LGN input (left) and an MlO complex cell which does not receive LGN input (right)
  • C & D Conductances for high contrast at consecutive apertures near the maximum responses.
  • E & F Conductances for low contrast at consecutive apertures near the maximum responses.
  • Panels C and E each consist of nine sub-panels
  • panels D and F each consist of eleven sub-panels, giving the cycle-trial averaged conductances as function of time (relative to cycle) and aperture size.
  • Asterisks indicate corresponding apertures of maximum response in A-B
  • Fig. 20 illustrates transfer of LGN surround suppression to cortical cells.
  • B-E
  • F Distributions of the suppression index of LGN inputs g LGN
  • G Distributions of the suppression index for spike responses of cortical cells in the model.
  • H Prevalence of the suppression mechanisms in the model.
  • I Distributions of the ratio of spatial summation extent at high and low contrast of LGN input g LGN .
  • J Distributions of receptive field growth of cortical cells in the model based on spike responses. Histograms give fractions of cells, arrows indicate means, solid arrow correspond to shaded histogram, dashed arrow correspond to unfilled histogram.
  • MO configuration yields similar results
  • A Simulated optical image of ocular dominance and orientation preference, in the spirit of optical imaging experiments (compare Fig 12)
  • B Distribution of the suppression index SI 1 at high (unfilled) and low (shaded) contrast
  • C Distributions of the receptive field and surround growth ratios, r Ir + (shaded) and R IR + (unfilled) Histograms give fractions of cells, arrows indicate means, solid arrow correspond to shaded histogram, dashed arrow correspond to unfilled histogram
  • Fig 22 illustrates some examples of the approximations given by Eq (30) and (31) for a simple cell (left) and a complex cell (right) for apertures around the aperture of the maximum response Fig layout is as Fig 7 of the paper
  • C & D Comparison of model responses ⁇ v(t,r A ) > with the Ohmic (slaving) approximation V(t,r A ) (Eq 30), indicating the model's operation at high conductance states
  • Fig 23 shows a summary of extraclassical spatial summation for the four model configuration
  • unshaded histograms are for gratings at high contrast ( + ), shaded at low contrast(-)
  • solid arrows refer to shaded histograms
  • dashed arrows to unfilled histograms
  • ROW 1 Distributions of the suppression index SI 1 All suppression is exclusively due to short-range cortical connectivity
  • ROW 2 Receptive field size distributions (r + ).
  • Fig 25 shows an llustration of the Armijo-hke line search, comparing the iterative shrinkage algorithm with (right column) and without (left column) line search
  • the gray bars under the iteration" axes highlight the difference between the number of iterations— the gray bar in (a) represents the same number of iterations as the gray bar in (b)
  • the step length without line search is bounded by 2 to ensure convergence
  • a varying utol especially tightening utol as we mo ⁇ e along the path
  • a continuation path of length 8 starting from 0 128 and ending at 0.001.
  • Fig. 27 Comparison of our hybrid iterative shrinkage (HIS) method with several other existing methods in literature. Benchmark data were taken from the UCI machine learning repository, including 10 publicly available data sets, (a) Distribution of computation time across 10 data sets, (b) Distribution of cardinality for the solution across 10 data sets, averaged over a regularization path.
  • HIS hybrid iterative shrinkage
  • Fig. 28 Comparison for the random benchmark data, between the HIS algorithm and the IP algorithm, (a) Speed profile for these two approaches: (upper curve) shows the speed profile for the IP algorithm, and (lower curve) shows the speed profile for the HIS algorithm as a function of the data dimension, (b) An example of the solutions using the IP algorithm (upper graph) and the HIS algorithm (lower graph).
  • Fig. 29 Solution w evolves along a regularization path, following a geometric progression from 10 -1 to 10 -4 Data is ionosphere from UCI machine learning repository. As the ⁇ becomes smaller, the cardinality of the solution goes up.
  • Fig. 31 A diagram of a proposed hybrid iterative shrinkage (HIS) algorithm
  • the HIS algorithm is comprised of two phases: the iterative shrinkage phase and the interior point phase.
  • the iterative shrinkage is inspired by a fixed point continuation method, which is computationally fast and memory friendly.
  • the interior point method is based on a second- order truncated Newton method.
  • Our hybrid approach takes advantage of different computational strengths of the two methods and uses them for optimal algorithm acceleration while attaining high accuracy. Black dots indicate the nonzero dimensions, gray dots indicate dimensions that are eliminated, and the size of the dots show the error that each dimension contributes to the final solution.
  • An exemplary embodiment can include a system and method for predicting and evaluating patient visual perception Retinal imaging data for a patient — e g , a retinal fundus photograph — is collected using an imaging device connected to the system Alternatively, an externally obtained retinal image can be provided as input to the system Automatic segmentation of the retinal image is performed in order to prepare a retinal mask The retinal mask is then fed into a processing engine that includes a computational model The retinal mask acts as a filter, obscuring portions of a target image (e.g , a picture of a face) in a binary manner i e , each image dot or pixel has an on/off or visible/obscured status Using the computational model, the processing engine simulates and predicts how the patient would visually perceive the target image during an actual patient test The simulated patient visual perception can be used to generate
  • An exemplary embodiment can include a system and method for predicting and evaluating relationships between population activity, visual input and visual perception.
  • Population groups can be identified based on selected vision-intensive activities or occupations — e g , driving, computer-based processing, manufacturing — and studied for long-term effects on visual perception.
  • the system and method described above can be used to generate predictive results for use with various population group studies These results can be used to correlate macular disease related retinal pathology with certain occupations and/or activities among targeted population groups. Further, the macular disease related retinal pathology can be correlated with visual function assessed via psychophysics testing of the population groups. Such correlation can be performed with any and all of the embodiments described herein.
  • An exemplary embodiment can include a system and method for predicting and evaluating the effectiveness of clinical treatment. Using the system and method, one or more modes of treatment can be tested on different population groups. In a clinical trial, for example, an embodiment can provide a quantitative test for the effectiveness of a clinical treatment. As an example, a longitudinal study can be conducted on patients having macular disease in which the system can be used to track the vision changes — improvement or decline — with clinical treatment, especially one involving pharmaceuticals. The computational model can generate output for predicting the results each of the pharmaceuticals will have, respectively, on each of the test population groups. Psychophysics testing can be performed in order to generate population data for each of the population groups.
  • the predictive results generated by the computational model can be vetted against the population data from each of the population groups.
  • the computational model can be optionally calibrated for greater accuracy, using the population data resulting from the psychophysics testing.
  • the predictive results generated by the computational model can be used to evaluate the efficacy of the various pharmaceuticals being screened. Based on these evaluations, the composition of the pharmaceuticals can be adjusted and the computational modeling can be iterated to evaluate the efficacy of the modified pharmaceuticals.
  • Parallel psychophysics testing may be included with the modeling iterations for additional model calibration. Correlations between the computational model and population data are much greater than correlations between retinal imaging data and population data.
  • the system and method described herein allows for a spiral-type approach to drug screening, where modifications to the pharmaceutical composition and the subsequent efficacy trials can be iterated using a relatively short cycle time. This may ultimately lead to increased effectiveness and an overall reduction in the time required to develop and obtain federal (e.g., FDA) approval for pharmaceuticals.
  • An exemplary embodiment can include a system and method for prescribing, evaluating, and optimizing patient rehabilitation programs.
  • patients coping with macular disease for example, can learn to fixate on a location within their visual field of view that exhibits relathely less visual impairment, thus enabling the patient to enjoy a more satisfying visual experience.
  • retinal imaging data for the patient e.g., a retinal fundus photograph
  • an image that is obtained externally i.e., outside the system of the exemplary embodiment — can be provided as input to the system. Automatic segmentation of the retinal image is performed in order to prepare a retinal mask.
  • the retinal mask is then fed as input to the system's processing engine which includes a computational model
  • the system ' s processing engine is used to move a target image off-fovea — i.e., away from the area responsible for central vision — to various selected locations and to generate a map representing the patient's relative visual acuity at each of the selected locations.
  • the system's psychophysics testing component can be performed in parallel to also move a target image off-fo ⁇ ea to varying locations and to solicit patient responses for each of the varying locations, in order to map the patient's relative visual acuity at each of the varying locations.
  • the results of the parallel mapping can be correlated, and the mapped location corresponding to the patient's relatively best vision is selected as the preferred retinal location.
  • the patient is then placed in a training program that includes a series of exercises to help train the patient's viewing habits such that their visual focus is shifted off-fovea (or off-center) and on the newly selected preferred retinal location
  • a training program that includes a series of exercises to help train the patient's viewing habits such that their visual focus is shifted off-fovea (or off-center) and on the newly selected preferred retinal location
  • Periodic modeling and testing can be performed to evaluate the effectiveness of the training program.
  • the training program can be modified based on the modeling and testing results, in order to yield optimal rehabilitation results in the patient
  • the system and method described herein can be adapted for use in a clinic, home, or other remote setting.
  • a home training program can be designed where the processing engine and computational model are adapted for home computer or iPhone use.
  • the target image moving and visual acuity mapping functions of the processing engine can be embodied in a computer readable medium, which can be read by a home computer that can carry out these functions in accordance with a set of instructions on the computer readable medium
  • the processing engine and computational model can be embodied on a chip, integrated circuit, or special purpose processor, in order to minimize the processing demand otherwise placed on a home computer or other remote computing device
  • Psychophysics testing can also be administered via a home computing device that performs a series of test instructions embodied, for example, in a computer readable medium
  • images can be displayed to the patient via display devices such as an iPhone, LCD monitor, and home projection system
  • An embodiment can include a system and method for incorporating high-level images into retinal mapping to study, for example, the effects of macular disease
  • a high- level image such as a face or car, for example, used as stimuli
  • retinal images, structural and/or functional, selected for automatic segmentation in order to prepare a retinal mask, which is provided as input to the system computational model
  • the model can map the disease on the retina to a representation which may provide a substantially better prediction of high-level perceptual performance than that provided by more traditional clinical metrics such as visual sensitivity test measured by micrope ⁇ metry, which measure a patient's ability to detect dots of varying brightness against a dark background
  • Fig 1 shows an exemplary embodiment of a system as described herein
  • An embodiment includes a processor, display device, patient response input device and a database
  • the database can contain software instructions, which can include instructions and data for operating the computational model and for carrying out specially designed psychophysics experiments as described herein
  • the database can also contain result data for the psychophysics experiments, retinal imaging data, and vision loss prediction output data e g , neuromet ⁇ c curve data Any of the data in the database can be provided as output in any suitable form, such as.
  • connection can be wired or wireless and can be direct or network connections as appropriate to meet the requirements for a particular embodiment
  • system can be contained within a single processing system or may reside on a distributed processing system
  • exemplary system shown in Fig 1 can be used with any and all of the embodiments described herein
  • the psychophysics tests can be carried out on a system such as that shown in Fig 1, having a computer (or other processing device), a display device, a patient response input device, and an experimental paradigm stored in a database and adapted to test patients and obtain results that can be used to vet output from a system computational model
  • the system may present a series of images to a patient via the display device The patient can be asked to identify the object in the image, e g , is the image that of a face or a car?
  • the patient Upon reaching a decision, the patient can press a button or activate some other feature on the patient response device to proude their response, which is received and recorded by the computer
  • the computer can also record the reaction time (or length of time it took the patient to answer) and accuracy of the patient response, which can be used in generating psychometric data and curves that characterize the patient's visual perception.
  • an embodiment such as that shown in Fig 1 can be used as a stand-alone system or integrated into a traditional or functional ophthalmology imaging device to add a functional component, based on a software that generates visual stimuli, and a computational model stored in and operated via the database and processor
  • the processor can process imaging data from the existing imaging device and provide a quantitative prediction of patient visual performance This prediction can provide valuable information to the physician and add a dimension to the clinical database for improved clinical treatment decisions For example, the prediction data can be used to suggest at what stage a patient's vision loss is currently
  • One embodiment is a method for quantitative assessment of macular disease
  • the method includes obtaining a retinal image of a patient
  • the method also includes performing a psychophysics experiment battery on the patient and collecting response data representing visual perception of the patient
  • the method includes providing the retinal image as input data to a processor executing a computational model
  • the processor through the computational model, transforms the input data into output data representing a vision loss prediction for the patient
  • the operations further include generating psychometric and neuromet ⁇ c curves representing psychophysics response data and computational model output data
  • Another embodiment is a s ⁇ stem for quantitatively assessing perceptual consequences of macular disease
  • the system includes a processor, display device, patient response input device and a database
  • the processor is operatively coupled to the display device, the patient response input device, and the database
  • the database contains computer software instructions to cause the processor to perform a predetermined psychophysics experiment on the patient and to record patient responses
  • the display device provides images for viewing by the patient
  • the patient response input device receives physical input from the patient and generates data representing that physical input for processing by the processor.
  • the database base also contains retinal images of the retinal fundus of the patient.
  • the processor executes a computational model to process input data including the retinal images and generate an output representing visual perception of the patient. Patient progress during a course of treatment can be quantitatively assessed by comparing output from the computational model for the patient over time
  • the retinal images can be obtained by a directly connected retinal imaging device or received from another system.
  • the retinal images can be generated by various retinal imaging devices, including structural imaging data and functional imaging data.
  • Another embodiment includes a computer program product or computer readable medium that has stored thereon software instructions for causing a processor to perform operations including obtaining retinal images of a patient's retinal fundus, and performing a psychophysics experiment battery on the patient and collecting response data representing visual perception of the patient.
  • the operations also include providing the retinal image as input data to a computational model being executed by the processor
  • the operations also include transforming the input data into output data representing a vision loss prediction for the patient.
  • the operations further include generating psychometric and neurometric curves representing psychophysics response data and computational model output data.
  • Another embodiment includes an upgrade (or retrofit) system to add quantitative macular disease assessment functions to an existing retinal imaging device.
  • the upgrade system includes a processor, display device, patient response input device and a database.
  • the processor is operatively coupled to the display device, the patient response input device, and the database.
  • the database contains computer software instructions to cause the processor to perform a predetermined psychophysics experiment on the patient and to record patient responses.
  • the display device provides images for viewing by the patient.
  • the patient response input device receives physical input from the patient and generates data representing that physical input for processing by the processor.
  • the database base also contains retinal images of the retinal fundus of the patient obtained by the existing retinal imaging device.
  • the processor executes a computational model to process input data including the retinal images and generate an output representing visual perception of the patient. Patient progress during a course of treatment can be quantitatively assessed by comparing output from the computational model and/or output from the psychophysics test for the patient over time
  • the upgrade ystem may use suitable existing components of the functional retinal imaging device, such as the processor, database, display device, and patient response input device
  • a diagnostic tool may be integrated into ophthalmic imaging devices for eye examination
  • the combination tool may be achieved, for example, by retrofitting components to an existing imaging device or as a unitary machine
  • the systems and devices of the disclosed subject matter may be employed for quantitative assessment of the efficacy of alternative drugs to accelerate ophthalmological drug screening or patient progress monitoring and evaluation.
  • the disclosed subject matter can also be used in low vision assessment and rehabilitation, especially in optimizing targeted retinal locus, which is a component in ophthalmic rehabilitation devices for low vision patients
  • a predictive tool may be employed using the visual system model to determine the potential effectiveness of different subregions of the retina for visual discrimination tasks
  • a patient's retina may be scanned or some other disto ⁇ on may be imposed on candidate images for evaluation purposes
  • the retma representation is divided into candidate regions to which a patient's attention has been found to be possible to train for recognition purposes
  • the typical region on which an image is projected when a person looks at an object may be the foveal region
  • other regions of the retina may be usable if a patient is trained to direct his line of sight accordingly
  • Candidate regions corresponding to the possible directions of sight (and possibly sizes thereof determined by distance from the object) may be identified
  • the computation methods described herein may be used to quantify the efficacy of the candidate regions for object recognition as indicated at S3
  • Other factors may be considered, as indicated at S4, for example, how difficult it is for a patient to be trained to use the alternative region for recognition tasks
  • the result of S4 may be a score based on this consideration as well as other possible considerations For
  • Age-related macular degeneration is the major cause of blindness in the developed world
  • AMD age-related macular degeneration
  • Macular diseases such as age-related macular degeneration (AMD), diabetic retinopathy (DR), and macular dystrophy (MD) account for the overwhelming majority of blindness in the United States Approximately 13 million people in the United States have some signs of macular degeneration, with the rate of new cases dramatically increasing due to longer life expectancies and the aging baby-boomer population Macular disease is not limited to older generations, also being a common problem for diabetics in the United States, 7 million of which suffer from diabetic retinopathy
  • Retinal imaging does not necessarily provide a complete picture for the nature of the expected vision loss It is important to consider how the visual cortex responds to the resulting scotoma and distortion of the retinal input and how this relates to perception.
  • Psychophysics has played an important role in characterizing the effects of retinal scotomata on visual perception.
  • Early efforts focused on perceptual processes for filling in a scene across the retinal blind spot. For instance, one study showed perceptual filling-in of background color, bars, and geometric patterns at the blind spot. Other studies have shown that the filling-in of the blind spot in one eye can influence perception arising from the other eye.
  • the computational model may provide a basis for linking retinal pathology, resulting from macular disease, with visual function assessed via psychophysical tests for a population of low-vision patients
  • r( ⁇ ) is the mean firing rate at orientation ⁇ .
  • CV is a measurement of orientation selectivity. A smaller value for CV indicates a higher orientation selectivity.
  • Fig. 3 illustrates the cortical responses in the Magno cortex.
  • the firing rates for the neuron population in the Magno system have seen an o ⁇ erall reduction. Very little distortion of the spatial distribution of the firing rates is observed. The distribution is consistently shifted to the right, indicating that the neurons become less orientation selective.
  • Fig. 4 shows the cortical responses of the Parvo cortex. Unlike the Magno cortex, there are obviously " holes" of activity in the Parvo cortex, indicating inactivation of neural activity. Noticeably the spatial distribution of such inactivation is correlated with the location of drusen in the visual field. The level of the overall firing rates is maintained at the cost of neuronal inactivation. Orientation selectivity of the neurons, on the other hand, is not affected.
  • Behavioral performance was evaluated by constructing psychometric curves for the control subjects and AMD patients.
  • Fig. 5 shows the psychometric curves for control (upper curve) and AMD (lower curve) patients.
  • Clearly AMD patients suffer from lower discrimination accuracy compared to the control population.
  • AMD patients clearly diverge from the control population in the psychometric curves at higher phase coherences.
  • the intersubject variability is small for the control subjects at higher phase coherences, while intersubject variability is much bigger for AMD patients, a natural consequence of the diversities of retinal pathology among patients.
  • Fig. 12 shows the predicted neurometric curves along with the psychometric curves, for both the control subjects and AMD patients. It can be noted that the neurometric curves are a good predictor of the psychometric curves for the patient cases.
  • Fig. 6. compares predicted neurometric curves with psychometric curves, (a) Average psychometric curve for the control subjects (lower curve) is plotted, together with the average neurometric curve (upper curve) based on the simulated model with normal vision, (b) Average psychometric curve for the AMD patients (lower curve) is plotted, together with the average neurometric curve (upper curve) based on the model with retinal impairment. In all curves, the errorbar indicates standard error.
  • Fig. 8 illustrated are three patient cases, where the neurometric curves are different from the psychometric curves, (a) RF fundus image of the patient, (b) Binary retinal mask used as input to the Vl model, (c) Comparison of the neurometric curve (upper curve) and the corresponding psychometric curve (lower curve) of one patient, plotted against the individual psychometric curves (dashed) for all the patients.
  • FIG. 9 shown is statistical analysis for establishing the correlation between the fundus image, model prediction and behavior data
  • Statistical analysis was carried out to investigate the relationship between the fundus image, model prediction and behavioral data.
  • Fig. 9(a) shows the psychometric Az values versus neurometric Az values, across all the phase coherences.
  • Vl retinal impairment
  • the sparse decoder subsequently maps the cortical activity into behavior, under a face versus car discrimination task.
  • the combination of the Vl model and the decoder provides a computational framework to examine the cortical and perceptual consequences of vision loss caused by AMD.
  • Color fundus photography is regarded as the traditional method documenting structural changes in retinal disorders over time and can be used in patients regardless of fixation.
  • color fundus and red-free (RF) fundus photographs were either taken using a film-based photographs acquired on with a Topcon TRC-50EX (Topcon Medical Systems Inc, Paramus, New Jersey, USA) and digitized on the Nikon CoolScan V (Nikon Corp, Tokyo, Japan) or were acquired digitally on the Zeiss FF 450 Plus Fundus Camera (Carl Zeiss Meditec Inc. Jena, Germany)
  • FIG 11 (a) Standardized image, gray channel, gray scale, slightly contrast-enhanced for visualization (b) The Otsu double thresholds in each region have provided estimates for vessels (pixel values below the lower threshold), background (pixel values between the two thresholds), and drusen (pixel values above the higher threshold) The mathematical model fit to the estimated background in A, displayed as a contour graph with gray scale levels in side bar (c) The image in A has been leveled by subtracting the background variability of the math model in C (result slightly contrast-enhanced). Note that the background is much more uniform, (d) Final drusen segmentation by uniform threshold.
  • the wet macular degeneration is a more severe form of AMD than the dry form, and accounts for approximately 10 % of all AMD but 90 % of all blindness from the disease.
  • the wet form is characterized by choroidal neovascularization (CNV), the development of abnormal blood vessels beneath the retinal pigment epithelium (RPE) layer of the retina. These vessels can bleed and cause macular scarring that further result in profound loss of central vision, which will be beyond the scope of our retinal model.
  • CNV choroidal neovascularization
  • RPE retinal pigment epithelium
  • Our model consists of 8 ocular dominance columns and 64 orientation hypercolumns (i.e. pinwheels), representing a 16 mm 2 area of a macaque Vl input layer ACa or AC ⁇ .
  • the model contains approximately 65,000 cortical cells and the corresponding appropriate number of LGN cells.
  • Our cortical cells are modeled as conductance based integrate-and-fire point neurons, 75% are excitatory cells and 25% are inhibitory cells.
  • Our LGN cells are half-wave rectified spatio-temporal linear filters.
  • the model is constructed with isotropic short-range cortical connections ( ⁇ 500/m ), realistic
  • Dynamic variables of a cortical model-cell z are its membrane potential v x (t) and its spike train where t is time and t t k is its k th spike time.
  • Membrane potential and spike train of each cell obey a set of N equations of the form
  • geometric parameters which define and relate the model's geometry in visual space and cortical space. Geometric properties are different for the two input layers 4Ca, ⁇ and for the two eccentricities. As said, the two extraclassical phenomena we seek to explain are observed to be largely insensitive to those differences. In order to verify that our explanations are consistent with this observation, we have performed numerical simulations for four sets of parameters, corresponding to the ACa, ⁇ layers at para-foveal eccentricities ⁇ 5° and at eccentricities around 10° . These different model configurations are referred to as MO, MlO and PO, PlO in the text. Reported results are qualitatively similar for all four configurations unless otherwise noted.
  • LGN neurons are modeled as half- wave rectified center-surround linear spatio-temporal filters.
  • a cortical cell, j t P x ( ⁇ ) is connected to a set Nff of left eye LG ⁇ cells, or to a set of right eye LG ⁇ cells, ⁇ ) I( )] (9)
  • LG ⁇ kernels are the spatial and temporal LG ⁇ kernels respectively
  • y t is the receptive field center of the £ th left or right eye LG ⁇ cell, which is connected to the j th cortical cell
  • I(y, s) is the visual stimulus.
  • the parameters g° represent the maintained activity of LG ⁇ cells and the parameters measure their responsiveness to visual stimuli Their numerical values are taken to be identical for all LG ⁇ cells in the model, and
  • the LG ⁇ kernels are of the form
  • k is a normalization constant
  • ⁇ c t and ⁇ s t are the center and surround sizes respectively
  • K 1 is the integrated surround-center sensitivity.
  • v l is the cortical magnification factor
  • ⁇ c is the LGN receptive field size (center size)
  • i c is a characteristic length scale for the excitatory cortical connectivity
  • the stimulus used to analyze surround suppression and contrast dependent receptive field size is a drifting grating confined to a circular aperture, surrounded by a blank (mean luminance) background
  • the luminance of the stimulus is given by and for with average luminance I 0 , contrast ⁇ , temporal frequency ⁇ , spatial wave vector k , and aperture radius r A .
  • the aperture is centered on the receptive field of the cell and varied in size, while the other parameters are kept fixed at close to preferred values for the cell.
  • Surround suppression is typically characterized by comparing the neuron's maximum firing rate to its firing rate at large aperture sizes.
  • the aperture size for which the response reaches its maximum (f max ) is sometimes referred to as "the classical receptive field" size
  • We define the asymptotic response f ⁇ as the average response beyond R .
  • We define the suppression index SI 1 as the relative surround suppression, where / 0 is the response to a blank stimulus.
  • the suppression index SI 1 is different from the integrated suppression index (see below).
  • SI 2 can be larger than one, indicating surround suppression beyond the background response.
  • the limitations of the DOG model can be made more apparent by noting that, given the validity of the half- wave rectification model Eq. 13, one
  • Equation 15 can be rewritten such that the numerator (N) and denominator (D) represent a half- wave rectified weighted difference of the excitatory and inhibitory conductances, and the total conductance g ⁇ , respectively.
  • the ROG model is then obtained by the substitutions
  • FIG. 14 Shown in Fig 14A are averaged response waveforms of spike train and membrane potential in response to a drifting grating These are responses of a simple and a complex cell, for several grating orientations, at the cells' preferred spatial and temporal frequencies
  • the modulation in the spike train at the preferred orientation is frequently used to classify simple and complex cells in Vl A cell is complex" whenever Fl s/FOs ⁇ 1 , and simple" otherwise, where FIs is the first harmonic of the spike response and FOs the mean
  • FIs is the first harmonic of the spike response and FOs the mean
  • FIG. 14B top
  • Our model contains about an equal number of simple and complex cells
  • the bimodal shape of the distribution of Fls/FOs agrees with experimental data
  • the a ⁇ ailabihty of this distribution provided us with a useful criterion for setting the cortical interaction strength parameters in the model
  • FIG 15 Averaged response wa ⁇ eforms of spike train and membrane potential in response to a standing (contrast reversal) grating at the preferred orientation are shown in Fig 15 A
  • Fig 15 A Shown are the responses of a simple and a complex cell in the model for se ⁇ eral spatial phases ⁇ of the grating Simple cells perform an approximately linear spatial summation, that is, their responses contain a dominant first harmonic (FIs, FIv) and the spatial phase dependence of their response waveform is similar to the spatial phase dependence of the magnitude of the intensity modulations of the stimulus at any g ⁇ en fixed position
  • Complex cells respond nonhnearly, their response waveform is relatively insensitive to spatial phase and contain a dominant second harmonic (F2s, F2v)
  • the distribution of the ratio of second to first harmonic of the response, a ⁇ eraged over the spatial phase ⁇ is shown in Fig 15B
  • the distribution of F2s/Fls displays
  • the average suppression index (over all eccentricities) is SI 1 : 0 2 and this is about half of what is observed experimentally
  • the receptive field and surround growths (Fig 16F) are expressed as ratios, r Ir + and R IR + respectnely
  • the indices + and - refer to high and low contrast respectively
  • Average growth ratio for the excitatory space constant is ⁇ E l ⁇ E + : 1 5 (both DOG and ROG, averaged over all eccentricities)
  • Equations (17) and (18) allow us to base our analysis directly on the (cycle-trial averaged) conductances as a function of the aperture radius r 4 and time In what follows we drop the averaging notation ⁇ • >, assuming it unless stated otherwise Given Eq.
  • surround suppression is caused by (A) an increase in the inhibitory conductance, or (B) a decrease in the excitatory conductance, or (C) both (A) and (B) simultaneously
  • a change in the spatial summation extent of g E and/or g I is just one of the many ways to change the behavior of G and consequently the receptive field size.
  • some other possibilities are illustrated by the two cells in Fig. 19. These cells show, both in spike and membrane potential responses, a receptive field growth of a factor of 2 (left) and 3 (right) at low contrast.
  • the spatial summation extent of excitation at low contrast is one aperture less than at high contrast.
  • the optimal spatial frequency of the LGN cells in our model is substantially smaller than the average optimal spatial frequency k c of the cortical cells (see Fig. 13B). This is why, although our model LGN cells of course do show surround suppression at their optimal spatial frequency, they do not show surround suppression at the higher cortically optimal spatial frequency k c . This is illustrated in Fig.
  • the parameters are interaction strengths that define the density and length scale invariant contribution of population to the conductance of a cell in population
  • the change in membrane potential of cell due to a single spike of cell is proportional to where is the cell density of population P k ( ⁇ ) and
  • the time constants are based on experimental observations [?].
  • the inhibitory kernels G 7 1 have a fast (GABA) component set by a I t , chosen from a uniform distribution between 3 ms and 6 ms, and a slow component [?] defined by , while The constants are normalization constants.
  • axon and dendrite parameters are (axons of excitatory neurons), (axons of inhibitory neurons), and (dendrites of all neurons) respectively.
  • excitatory connections both on excitatory and inhibitory cells
  • characteristic length scale 200 ⁇ m
  • inhibitory connections both on excitatory and inhibitory cells
  • the set of strength parameters we provide is obtained by adhering to a few general principles regarding the LGN input and cortico-cortical interactions. These general principles are: (a) No distinction between LGN input in the excitatory and inhibitory cell populations, (b) Cells with LGN input, both excitatory and inhibitory cells, receive their excitation in about equal amounts from LGN input, cells with LGN input, and cells without LGN input, (c) Cells without LGN input, both excitatory and inhibitory cells, receive most of their excitation from cells with LGN input (d) Cells with (without) LGN input, both excitatory and inhibitory cells, receive most of their inhibition from cells with (without) LGN input.
  • these classical response properties are (i) Absence of any global phase-locked oscillations and synchrony, both under visual stimulation and without visual stimulation. This condition limits the overall maximum size of the strength parameters (ii) Distribution of activity (firing rates) over the cell population, with and without visual stimulation.
  • the noise strengths ⁇ u ° are drawn from a uniform distribution between 1 and 5 if / e P 0 (E) , are equal to 2 if i G P 1 (E) , are drawn from a uniform distribution between 0 and 30 if i e P 0 (I) , and are drawn from a uniform distribution between 16 and 46 if i e P 1 (I) .
  • the sets are constructed as follows.
  • Our 4x4 mm model cortex is partitioned into 8 parallel bands (0.5x4 mm 2 ), which alternate representation between the two eyes.
  • initial retinotopic maps for each eye were defined as the identity map plus scatter as follows.
  • the parameter v is 0.2° /mm for 0° eccentricity (M0,P0) and
  • the scatter parameter ⁇ is 0.3 ° , 0.1 ° , 0 35 ° and 0.35 ° for MO, PO, MlO and PlO respectively. These scatter values are in the experimentally observed range and moreover assure a more or less uniform initial distribution of cortical recepti ⁇ e fields.
  • each cell in P 1 ( ⁇ ) (for each eye) is assigned a template for the organization of the ON and OFF subfields of its initial RF, which is randomly chosen from the 4 basic symmetry configurations seen experimentally.
  • LGN-cortical connections are initially made so as to best approximate the assigned initial receptive field center and template for each cell. Then, the LGN-cortical connections are rearranged by disconnecting and reconnecting cells, so as to achieve LGN axon sizes that agree with the anatomical findings for macaque.
  • Responses to contrast-reversal stimuli are obtained in a similar way
  • the apertures are centered on a rectangular grid (5x5 or 6x6, grid spacing about half of the a ⁇ erage RF size) which itself is centered on the visual field covered by our model cortex
  • the angle, spatial and temporal frequencies of the grating are kept fixed during this experiment
  • the temporal frequencies are set to the averaged preferred values for the case under consideration
  • the preferred parameters and the receptive field centers are from averaged spike
  • receptive field growth can occur without any change in the spatial summation extent (r y ) of the conductances.
  • receptive field growth can be induced, for instance, by an overall increase (X) or an overall decrease (Z) in relative gain G(r A ) as shown in Fig. 24 A (dashed line).
  • Receptive field growth also can be caused by more drastic changes in G so that the transitions X -» Y.
  • X ⁇ Z orY ⁇ Z transition occurs in about 70% of the cells with significant receptive field growth, while about 20% of the cells with significant receptive field growth (Fig. 24B, bottom) have r E ⁇ r s ⁇ /7 at high contrast and /7 ⁇ r s ⁇ r E at low contrast
  • Fig. 24C also demonstrates the presence of a rich diversity in relative gain changes in our model, since all transitions (for all cells, unfilled histograms) occur with some reasonable probability. Finally, Fig. 24C establishes that there is a relationship between the difference in the prevalence of the surround suppression mechanisms at high and low contrast and receptive field growth. To see this, first note that for the redefined Y and Z classes with respect to r s and /7 , the surround suppression, if any, may be caused by mechanisms B or C.
  • I x -regularized logistic regression also know ⁇ as sparse logistic regression, is widely used in machine learning, computer vision, data mining, bioinformatics and neural signal processing.
  • regularization attributes attractive properties to the classifier. such as feature selection, robustness to noise, and as a result, classifier generality in the context of supervised learning.
  • a sparse logistic regression problem has large-scale data in high dimensions, it is computationally expensive to minimize the non-differentiable -norm in the objective function.
  • We propose a no ⁇ el hybrid algorithm based on combining two types of optimization iterations one being very fast and memory friendly while the other being slower but more accurate.
  • the resulting algorithm is comprised of a fixed point continuation phase and an interior point phase.
  • the first phase is based completely on memory efficient operations such as matrix- vector multiplications, while the second phase is based on a truncated Newton's method.
  • various optimization techniques including line search and continuation, can significantly accelerate convergence.
  • the algorithm has global convergence at a geometric rate (a Q-linear rate in optimization terminology).
  • Logistic regression is an important linear classifier in machine learning and has been widely used in computer vision, bioinformatics, gene classification, and neural signal processing.
  • I 1 -regularized logistic regression or so-called sparse logistic regression where the weight ⁇ ector of the classifier has a small number of nonzero values, has been shown to have attractive properties such as feature selection and robustness to noise.
  • sparse logistic regression For supervised learning with many features but limited training samples, overfitting to the training data can be a problem in the absence of proper regularization. To reduce overfitting and obtain a robust classifier, one may find a sparse solution
  • the basic form of logistic regression seeks a hyperplane that separates data belonging to two classes.
  • the conditional probability for the classifier label b based on the data, according to the logistic model takes the following form,
  • the classifier parameters w and v can be determined by minimizing the average logistic loss function,
  • Such an optimization can also be interpreted as a MAP estimate for classifier weights M ⁇ and intercept v
  • problem (2) is in general NP -hard. Due to this computational complexity, I x regularization has become a popular alternative, and is subtly different than I 0 regularization, in that the -norm penalizes large coefficients/parameters more than small ones.
  • the I x -regularized logistic regression problem (38) is a convex and non- differentiable problem A solution always exists but can be non-unique These characteristics postulate some difficulties in solving the problem Generic methods for nondifferentiable convex optimization, such as the ellipsoid method and various sub-gradient methods, are not designed to handle instances of (38) with data of very large scale There has been very active development on numerical algorithms for solving the I x -regularized logistic regression, including LASSO, GIl ce.
  • the IRLS-LARS (iteratively reweighted least squares least angle regression) algorithm uses a quadratic approximation for the average logistic loss function, which is consequently solved by the LARS (least angle regression) method
  • the BBR (Bayesian logistic regression) algorithm uses a cyclic coordinate descent method for the Bayesian logistic regression Glmpath, a solver for I x - regula ⁇ zed generalized linear models using path following methods, can also handle the logistic regression problem
  • MOSEK is a general purpose primal-dual interior point solver, which can solve the I x -regularized logistic regression by formulating the dual problem, or treating it as a geometric program SMLR, algorithms for various sparse linear classifiers, can also solve sparse logistic regression
  • the algorithm takes truncated Newton steps and uses preconditioned conjugated gradient iterations. This interior-point solver is efficient and provides a highly accurate solution.
  • the truncated Newton method has fast convergence, but
  • the first phase is based on a new algorithm called iterative shrinkage, inspired by a fixed point continuation (FPC), which is computationally fast and memory friendly; the second phase is a customized interior point method.
  • FPC fixed point continuation
  • Fig. 31 shows a diagram of our hybrid algorithm, termed Hybrid Iterative Shrinkage (HIS) algorithm.
  • HIS Hybrid Iterative Shrinkage
  • Our algorithm uses less memory and, on mid/large-scale problems, runs faster than the interior point method.
  • the iterative shrinkage phase only performs matrix-vector multiplications in size of X , as well as a ⁇ ery simple shrinkage operation, and therefore uses minimal memory consumption.
  • Q-linear convergence we prove Q-linear convergence and show that the signs of w opt (hence, the indices of nonzero elements) are obtained in a finite number of steps, typically much earlier than convergence. Based on the latter result, we propose a hybrid algorithm that is even faster and results in highly accurate solutions.
  • our algorithm predicts the sign changes in future shrinkage iterations, and when the signs of w k are likely to be stable, switches to the interior point method and operates on a reduced problem that is much smaller than the original
  • the interior point method achieves high accuracy in the solution, making our hybrid algorithm equally accurate.
  • the iterative shrinkage algorithm used in the first phase is inspired by a fixed point continuation algorithm.
  • the gradient descent step h reduces f(x) by moving along the negative gradient direction of f and the shrinkage step s reduces the I x -norm by shrinking" the magnitude of each nonzero component in the input vector.
  • step length be bounded by [0232]
  • the iterative shrinkage algorithm for sparse logistic regression is ), forwcomponent, forvcomponent, (48) which is a composition of two mappings h and s from P " to P " , where the gradient operator is
  • ⁇ u k ⁇ is shown to have a limit point x , that is, a subsequence converging to u , due to the compactness of ⁇ and (10). can be proven to converge globally to u .
  • u is a limit point
  • ?? fixed point with respect to (??)
  • a linear convergence result with a certain convergence rate can also be obtained.
  • the sequence converges to x * R -linearly, and converges to g -ii near iy
  • R -linear convergence can be strengthened to Q -linear convergence by using the fact that the minimal eigenvalue of at x * is strictly greater than O.
  • step length ⁇ at each iteration is the step length ⁇ at each iteration.
  • Shrinkage step Obtain search direction: max line search attempts" Armijo-like condition is met Accept line search step, update Keep backtracking ⁇
  • Fig. 25 illustrates the computational speedup using the line search.
  • the top panel shows the evolution of the objective function as a function of iterations. Tested on the benchmark data from the UCI repository, we see that our algorithm results in a speedup of 40 (6000 iterations without line search vs. 150 iterations with line search).
  • the bottom panel shows the step length used in the algorithm. In the absence of the line search, we specify that the step length satisfy .
  • the Armijo-like line search we illustrate both the
  • step lengths can be on the order of 100 times larger for line search vs. no line search.
  • the goal of a continuation strategy is to construct a path with different rate of convergence, with which we can speed up the whole algorithm.
  • the solution obtained from a previous subpath associated with ⁇ l ⁇ is used as the initial condition for the next subpath for A 1 .
  • Fig. 26 shows the continuation path using fixed utol and a varying utol following geometric progression.
  • the solver spends a lot of time evolving slowly.
  • the objective function shows a fairly flat reduction at earlier stages of the path.
  • we can accelerate the computation shown in (b).
  • the choice of utol and gtol seems to be data dependent in our experience, and the result we show in (b) might be suboptimal. Further optimization of the continuation path can potentially accelerate the computation even more.
  • HIS HIS
  • iterative shrinkage algorithm described previously to enforce sparsity and identify the support in the data, followed by a subspace optimization via an interior point method.
  • Corollary 2 implies an important fact' there are two phases in the fixed point continuation algorithm
  • the first phase the number of nonzero elements in the x evolve rapidly, until after a finite number of iterations, when the support (non-zero elements in a vector) is found. Precisely, it means that for all k > K .
  • the nonzero entries in u k include all true nonzero entries in u with the matched signs However, unless k is large, u k typically also has extra nonzeros
  • the fixed point continuation reduces to the gradient projection, starting the second phase of the algorithm
  • the zero elements in the vector stay unaltered, while the magnitude of the nonzero elements (support) keeps evohing.
  • the abo ⁇ e observation is a general statement for any / that is convex Recall
  • a logarithmic barrier function, smooth and convex, is further constructed for the bound constraints
  • the hybrid algorithm leverages the computational strengths of both the iterative shrinkage solver and the interior point solver
  • Fig 28 shows the benchmark result using data from the UCI machine learning repository All numerical results shown are averaged over a regularization path The parameters for the regularization path are calculated according to each data set, where the maximal regularization parameter is calculated as follows
  • Table 3 Speed comparison of the HIS algorithm with the IP algorithm, based on simulated random benchmark data. Shown here is the computation speed as a function of dimension. Data used here are generated by sampling from two Gaussian distributions. Note that in the simulation, the continuation path used in the iterative shrinkage may or may not be optimal, which means that the speed profile for the HIS algorithm can be accelerated even more.
  • Table 3 summarizes the computational speed for the HIS algorithm and the IP algorithm. It is noteworthy that the HIS algorithm improves the efficiency of computation, while maintaining comparable accuracy to the IP algorithm.
  • Fig. 28 plots the computation result as a function of dimension for better illustration In (a) one can clearly see the speedup we gam from the HIS algorithm (lower curve), compared to the IP algorithm (upper curve) We also show the solution quality in (b), where the weights we get from both solvers, is comparable
  • the regularization parameter ⁇ affects the number of iterations to converge for any solver As ⁇ becomes smaller, the cardinality of the solution increases, and the computation time needed for com ergence also increases Therefore when one seeks a solution with less sparsity (small ⁇ ), it is more computationally expensi ⁇ e
  • Fig 29 shows the evolution of solution along the regularization path, using a small data set (ionosphere) from the UCI machine learning repository This explores sparsity of different degrees in the solution, and one can determine the optimal sparsity for the data This is an attractive property of this model, where one can search in the feature space the most informative features about discrimination
  • Fig 30(a) The speedup of the HIS algorithm compared to the IP algorithm is shown in Fig 30(a), where dots indicate the computation time of the IP algorithm, and asterisks show the HIS algorithm
  • the HIS algorithm results in a significant speedup over the IP algorithm, without loss of accuracy
  • Fig 30(b) illustrates the classification result as a function of the cardinality of the solution
  • the HIS algorithm then transits into the second phase, using a more accurate interior point solver.
  • the HIS algorithm also scales very well with dimension of the problem, making it attractive for solving large-scale problems.
  • HIS algorithm There are several ways to extend the HIS algorithm. One is to extend it beyond binary classification, allowing for multiple classes. The other is to further improve the regularization path.
  • the disclosed subject matter includes a method for evaluating accuracy of visual perception given a pathology.
  • the method includes applying a plurality of images of a predetermined set of recognizable objects to an input of a nonlinear computational model of a vision system of an animal, the model running on a processor.
  • the images are modified responsively to pathology of an individual patient, or a derivative of a population of patients.
  • the applying may include receiving data representing the pathology and combining the pathology data with a selected one of the plurality of images to form electromagnetic signals representing a feature ⁇ ector and storing the signals as data to make the data accessible to the processor for generation of the model
  • the method may further include collecting data representing a first output vector from activity of the model and then using the model to generate a confidence estimate of a recognition by the model of the object represented by the selected image.
  • the using operation may include using a software classifier running on one or more processors to generate from activity of a model of a primary cortex of an animal to generate the confidence estimate
  • the method may further include modifying the image in a graded fashion
  • the activity may include temporal activity
  • the generating may include performing a sparse logistic regression on the first output vector and generating a second vector output indicating an estimate of a recognized one of the predetermined set of recognizable objects
  • the second vector may include at least one confidence estimate
  • the pathology may be, or include, a pathology of a retina
  • the pathology may include material that blocks light falling on a retina
  • the model may include an input representing a matrix of pixels representing a projection of the selected one of the plurality of images on a retina This matrix may incorporate a model of the retina to account for the varying distribution of receptor cells in a target subject
  • the projection may include a masking of the image by a drusen distribution in a retina
  • the operations of the foregoing methods may be repeated iteratively to allow tuning of a set of weights of a logistic regression to modify the prediction
  • the drusen distribution may be obtained from microperimetry or a retinal image of a single patient or one obtained from a population of patients, where the operations are iterated for each member of the population
  • the method may include identifying an optimum treatment by measuring the pathology after a treatment and determining a change in a confidence estimate resulting therefrom, and doing this iteratively to evaluate multiple treatments
  • the treatment may include a surgical treatment, gene therapy, prosthesis or administration of a drug
  • the drug may provide a treatment for retinal disease
  • the model may include actual or simulated neurons that are substantially more numerous than a size of any of the plurality of images
  • the model may include features of animal anatomy and is implemented on both computer processors and hardware such as neuromorphic chips
  • the embodiments include devices or systems for implementing any of the foregoing methods [0295]
  • the disclosed subject matter includes a method of determining a strategy for ameliorating the effect of a visual impairment.
  • the method includes projecting an image on a plurality of alternative regions of a retina.
  • the method further includes estimating the ability of a visual system to identify objects represented by the images for each alternative region.
  • the method further includes developing, responsively to the estimating, a training regimen to teach a person to use a region of the retina for image recognition responsively to the estimating.
  • the projecting and estimating may be done by one or more processors and include reading a test from a data store, displaying instructions on a display, reading an image from an image database, automatically projecting an image at candidate angles relative to a predetermined axis of view, receiving data indicating an object recognition confidence estimate, storing the estimate, and generating output comparing the different candidate angles.
  • the projecting and estimating may be done by one or more processors and include generating an estimate of a retinal image that combines a particular pathology with the image for candidate angles; applying the resulting retinal image to a computer generated model of a primary visual processing system of an animal, and generating confidence estimates from the response of the primary visual processing system
  • the projecting and estimating may be done by one or more processors and include reading a test from a data store, displaying instructions on a display, reading an image from an image database, automatically projecting an image at candidate angles relative to a predetermined axis of view, receiving data indicating an object recognition confidence estimate, storing the estimate, and generating output comparing the different candidate angles.
  • the one or more processors may run a program generating a model of the primary visual cortex of a human, the model including a number of interconnected neurons which are substantially greater in number than a size of the image.
  • the estimating may further include reducing a dimensional size of the activity of the model by means of a regression to generate a sparse output vector whose dimensional size is approximately equal to a number of the objects to be recognized.
  • the embodiments may include a device or system configured for implementing any of the above methods.
  • Embodiments of the disclosed subject matter including a method for quantifying the impact on visual perception of a predetermined condition.
  • the method includes modifying a visual stimulus responsively to measurement data representing a measured condition to create input data indicating a physical impact of the condition on the visual stimulus as received by sensory neurons of an animal, the input data forming a feature vector to a machine recognition process.
  • the method includes storing the input data in a machine readable data store
  • the method includes accessing the input data with a processor
  • the method includes running a simulation of a visual perceptual process on the processor and responsive to the input data to generate activity whose dimensional size is substantially greater than a dimensional size of the input data
  • the method includes deriving from the activity a sparse output indicative of a perceptual result response to the input data, the output indicating a perceptual impact of the measured condition
  • the condition may alter an image on the retina of an animal
  • the condition may include a drusen distribution.
  • the output may represent one of a set of possible objects represented by the stimulus, the stimulus including an image projected on a retina
  • the simulation may be a non-linear encoding of the input data and the deriving may include a linear decoding of the encoded input data
  • the embodiments include a method as in any of the foregoing discussed ones where a predicted effect of a pathology on visual perception is calculated as quantitative data and output on a display or stored in a database sto ⁇ ng like data
  • the embodiments include a device for implementing any of the methods to form a diagnostic tool that is combined with a device for detecting the pathology and configured to represent the pathology as pathology data
  • the pathology data may be applied as digital signals to a machine recognition process for accomplishing the recited method
  • the model may include a model of a retina, the model of a retina representing an effect of sensitivity of receptors to light and a density and type of receptor cells of a retina, such that the physical effect of these factors of a retina are combined with the image data for application to a model
  • the disclosed subject matter includes a method for optimizing a retinal locus for a person with impaired vision, the method detecting pathological features of a retma of a patient
  • the method includes combining the detected features with image data representing an object
  • the method further includes applying combined retina pathology and image data to a machine image classifier to compute confidence estimates of predicted recognition competence of the patient for multiple retinal loci
  • the method may further include indicating an optimum one of the retinal loci corresponding to a highest confidence estimate of the machine image classifier
  • the machine image classifier may include a computational neural modeling of early visual processing
  • the pathological features may include a retinal image and the detected features include data representing an attenuation of light falling on a retina
  • the method may include optimizing the retinal locus responsively to computed confidence estimates
  • the method may further include displaying a confidence estimate
  • the embodiments include a device including a processor, memory, and a data store for implementing any of these methods
  • the embodiments include a method of quantifying a vision impairment due to a pathology
  • the method includes measuring a pathology and converting it to data and using a processor implementing a computational neural model of early visual processing a model to recognize images responsively to the data.
  • the method includes generating at least a confidence estimate from the activity of the model and providing the estimate or a derivative thereof on an output such as a display

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Eye Examination Apparatus (AREA)

Abstract

Cette invention concerne un système et un procédé d'évaluation et de prévision de la perception visuelle chez un sujet animal. Des données d'image rétinienne ou autres données représentant la pathologie d'un sujet animal sont saisies en tant que données d'un modèle informatique non linéaire du cortex visuel primaire du sujet. Plusieurs images d'un groupe prédéfini d'objets reconnaissables sont appliquées aux données saisies pour le modèle informatique, puis mises à la disposition d'un processeur pour générer le modèle. Une estimation de confiance d'une reconnaissance par le modèle de l'objet représenté par l'image sélectionnée est générée. Le système et le procédé permettent de quantifier l'influence sur la perception visuelle d'une pathologie prédéterminée, ainsi que la détermination d'une stratégie pour améliorer l'effet d'une altération visuelle. Les facteurs de pondération de régression du modèle informatique peuvent être ajustés pour obtenir de meilleures prévisions, notamment par optimisation itérative, par exemple à l'aide de données psychométriques.
PCT/US2010/033780 2009-05-05 2010-05-05 Dispositifs, systèmes et procédés d'évaluation de la vision, et diagnostic et compensation de l'altération de la vision WO2010129711A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US17568109P 2009-05-05 2009-05-05
US61/175,681 2009-05-05

Publications (1)

Publication Number Publication Date
WO2010129711A1 true WO2010129711A1 (fr) 2010-11-11

Family

ID=43050447

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2010/033780 WO2010129711A1 (fr) 2009-05-05 2010-05-05 Dispositifs, systèmes et procédés d'évaluation de la vision, et diagnostic et compensation de l'altération de la vision

Country Status (1)

Country Link
WO (1) WO2010129711A1 (fr)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104331634A (zh) * 2014-11-22 2015-02-04 张姬娟 检测数据的分析处理系统和方法
CN104361245A (zh) * 2014-11-22 2015-02-18 张姬娟 检测数据处理系统和方法
CN104392129A (zh) * 2014-11-22 2015-03-04 张姬娟 数据处理系统和方法
EP3111828A1 (fr) * 2015-06-29 2017-01-04 Carl Zeiss Vision International GmbH Dispositif et procédé pour l'apprentissage d'un locus rétinien préféré de fixation
US9575930B2 (en) 2014-02-20 2017-02-21 International Business Machines Corporation Conjugate gradient solvers for linear systems
CN109471053A (zh) * 2018-10-18 2019-03-15 电子科技大学 一种基于双约束的介电特性迭代成像方法
CN110957042A (zh) * 2020-01-17 2020-04-03 广州慧视医疗科技有限公司 一种基于领域知识迁移的不同条件下眼部疾病的预测和模拟方法
CN111652317A (zh) * 2020-06-04 2020-09-11 郑州科技学院 基于贝叶斯深度学习的超参数图像分割方法
CN112022642A (zh) * 2020-09-16 2020-12-04 杭州集视智能科技有限公司 一种基于视野中心损伤的边缘视野训练设备及训练方法
CN115841644A (zh) * 2022-12-29 2023-03-24 杭州毓贞智能科技有限公司 基于物联网的城市基础建设工程设备的控制系统及其方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4927260A (en) * 1988-10-11 1990-05-22 Orville Gordon Apparatus and method for the precision evaluation of visual function in the fovea centralis (macula) area of the retina
US20030027745A1 (en) * 2001-06-13 2003-02-06 Repine John E. Diagnostic and prognostic method for evaluating ocular inflammation and oxidative stress and the treatment of the same
WO2007037432A1 (fr) * 2005-09-30 2007-04-05 Matsushita Electric Industrial Co., Ltd. Dispositif d’aide au recouvrement de la vue
US20070197932A1 (en) * 2006-02-17 2007-08-23 Feke Gilbert T Non-invasive methods for evaluating retinal affecting neurodegenerative diseases
US20070269804A1 (en) * 2004-06-19 2007-11-22 Chondrogene, Inc. Computer system and methods for constructing biological classifiers and uses thereof
WO2009055818A1 (fr) * 2007-10-25 2009-04-30 Research Foundation Of State University Of New York Bio-marqueur spectral et algorithme pour l'identification et la détection de cellules souches et progénitrices neurales et leur utilisation dans l'étude des cerveaux mammaliens

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4927260A (en) * 1988-10-11 1990-05-22 Orville Gordon Apparatus and method for the precision evaluation of visual function in the fovea centralis (macula) area of the retina
US20030027745A1 (en) * 2001-06-13 2003-02-06 Repine John E. Diagnostic and prognostic method for evaluating ocular inflammation and oxidative stress and the treatment of the same
US20070269804A1 (en) * 2004-06-19 2007-11-22 Chondrogene, Inc. Computer system and methods for constructing biological classifiers and uses thereof
WO2007037432A1 (fr) * 2005-09-30 2007-04-05 Matsushita Electric Industrial Co., Ltd. Dispositif d’aide au recouvrement de la vue
US20090244481A1 (en) * 2005-09-30 2009-10-01 Matsushita Electric Industrial Co., Ltd. Vision recovery training device
US20070197932A1 (en) * 2006-02-17 2007-08-23 Feke Gilbert T Non-invasive methods for evaluating retinal affecting neurodegenerative diseases
WO2009055818A1 (fr) * 2007-10-25 2009-04-30 Research Foundation Of State University Of New York Bio-marqueur spectral et algorithme pour l'identification et la détection de cellules souches et progénitrices neurales et leur utilisation dans l'étude des cerveaux mammaliens

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9575930B2 (en) 2014-02-20 2017-02-21 International Business Machines Corporation Conjugate gradient solvers for linear systems
US9582472B2 (en) 2014-02-20 2017-02-28 International Business Machines Corporation Conjugate gradient solvers for linear systems
CN104361245A (zh) * 2014-11-22 2015-02-18 张姬娟 检测数据处理系统和方法
CN104392129A (zh) * 2014-11-22 2015-03-04 张姬娟 数据处理系统和方法
CN104331634A (zh) * 2014-11-22 2015-02-04 张姬娟 检测数据的分析处理系统和方法
CN107787213B (zh) * 2015-06-29 2020-10-09 卡尔蔡司光学国际有限公司 用于训练优选视网膜注视点的装置及计算机程序
EP3111828A1 (fr) * 2015-06-29 2017-01-04 Carl Zeiss Vision International GmbH Dispositif et procédé pour l'apprentissage d'un locus rétinien préféré de fixation
WO2017001335A1 (fr) 2015-06-29 2017-01-05 Carl Zeiss Vision International Gmbh Dispositif et programme informatique pour l'apprentissage d'un locus rétinien préféré de fixation
CN107787213A (zh) * 2015-06-29 2018-03-09 卡尔蔡司光学国际有限公司 用于训练优选视网膜注视点的装置及计算机程序
US11826306B2 (en) 2015-06-29 2023-11-28 Carl Zeiss Vision International Gmbh Device and computer program for training a preferred retinal locus of fixation
CN109471053A (zh) * 2018-10-18 2019-03-15 电子科技大学 一种基于双约束的介电特性迭代成像方法
CN110957042A (zh) * 2020-01-17 2020-04-03 广州慧视医疗科技有限公司 一种基于领域知识迁移的不同条件下眼部疾病的预测和模拟方法
CN111652317A (zh) * 2020-06-04 2020-09-11 郑州科技学院 基于贝叶斯深度学习的超参数图像分割方法
CN111652317B (zh) * 2020-06-04 2023-08-25 郑州科技学院 基于贝叶斯深度学习的超参数图像分割方法
CN112022642A (zh) * 2020-09-16 2020-12-04 杭州集视智能科技有限公司 一种基于视野中心损伤的边缘视野训练设备及训练方法
CN115841644A (zh) * 2022-12-29 2023-03-24 杭州毓贞智能科技有限公司 基于物联网的城市基础建设工程设备的控制系统及其方法
CN115841644B (zh) * 2022-12-29 2023-12-22 吕梁市经开区信息化投资建设有限公司 基于物联网的城市基础建设工程设备的控制系统及其方法

Similar Documents

Publication Publication Date Title
WO2010129711A1 (fr) Dispositifs, systèmes et procédés d'évaluation de la vision, et diagnostic et compensation de l'altération de la vision
Lesmes et al. Bayesian adaptive estimation of the contrast sensitivity function: the quick CSF method
Hope et al. Predicting outcome and recovery after stroke with lesions extracted from MRI images
Geisler et al. Separation of low-level and high-level factors in complex tasks: visual search.
Kauppi Eye fundus image analysis for automatic detection of diabetic retinopathy
US10925481B2 (en) Systems and methods for measuring visual function maps
US20190313895A1 (en) System and method for automatic assessment of disease condition using oct scan data
Patri et al. Transient disruption of the inferior parietal lobule impairs the ability to attribute intention to action
WO2018094381A1 (fr) Système et procédé pour l'évaluation automatique d'affection à l'aide de données de balayage oct
US20190096277A1 (en) Systems and methods for measuring reading performance
Ghaderi-Kangavari et al. A general integrative neurocognitive modeling framework to jointly describe EEG and decision-making on single trials
CN117133404B (zh) 一种针对胸廓出口综合征的智能化康复护理装置
Leopold et al. Deep learning for retinal analysis
KR20200114837A (ko) 녹내장 진단용 피쳐 생성 방법 및 장치, 그것을 이용한 녹내장 진단 방법 및 장치
Bar-David et al. Elastic deformation of optical coherence tomography images of diabetic macular edema for deep-learning models training: how far to go?
Shyamalee et al. Automated Tool Support for Glaucoma Identification with Explainability Using Fundus Images
Pin et al. Retinal diseases classification based on hybrid ensemble deep learning and optical coherence tomography images
Naramala et al. Enhancing Diabetic Retinopathy Detection Through Machine Learning with Restricted Boltzmann Machines
Yadav et al. Unraveling human perception of facial aging using eye gaze
Shirokanev et al. Intelligent system for coagulates preliminary plan analysis for assessment of the diabetic retinopathy treatment effectiveness
Sutradhar et al. Blind-spot network for image anomaly detection: A new approach to diabetic retinopathy screening.
US11813406B2 (en) Software and methods for controlling neural responses in deep brain regions
Koenig Remembering What We've Seen: The Hippocampus and Relational Memory
Mazar Pasha et al. Diabetic Retinopathy Severity Categorization in Retinal Images Using Convolution Neural Network.
Vance et al. Investigation into sub-receptive fields of retinal ganglion cells with natural images

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 10772792

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 10772792

Country of ref document: EP

Kind code of ref document: A1