SYSTEM AND METHOD FOR AUTOMATED DETECTION OF AGE RELATED MACULAR DEGENERATION AND OTHER RETINAL ABNORMALITIES

• US 20110242306A1
• Filed: 12/22/2009
• Published: 10/06/2011
• Est. Priority Date: 12/19/2008
• Status: Active Grant

• Current Assignee: Johns Hopkins University
• Original Assignee: Johns Hopkins University
• Inventors: Burlina, Philippe Martin, Freund, David Eric, Bressler, Neil
• Granted Patent: US 8,896,682 B2
• Time in Patent Office: Days
• US Class Current: 348/78
• CPC Class Codes: A61B 3/0025 characterised by electronic signal processing, e.g. eye models
• CPC Class Codes: A61B 3/12 for looking at the eye fundus, e.g. ophthalmoscopes A61B3/13 takes precedence

Abstract

A system and method for automated detection of age related macular degeneration and other retinal abnormalities which may have a retinal scanner capable of obtaining retinal data from a subject. The retinal scanner may be coupled to a central processing unit (CPU) which may have memory storing CPU-executable instructions which may detect retinal abnormalities. When the CPU receives retinal data from the retinal scanner, the system may perform CPU-executable instructions for detecting retinal abnormalities. The system may analyze the retinal data to determine one or more healthy areas and, based on the analyzed healthy area, the system may detect abnormalities in the retinal data.

Claims

1. A retinal abnormalities detection system, comprising:

a retinal scanner constructed to obtain retinal data;

a central processing unit (CPU) in communication with the retinal scanner, the CPU comprising memory-storable CPU-executable instructions for detecting retinal abnormalities;

wherein the CPU performs the following in response to receiving retinal data based on the memory-storable CPU-executable instructions;

an analysis of the retinal data to determine a healthy area; and

a detection of an abnormality based on the determined healthy area.

2. The system according to claim 1, wherein the detection of an abnormality comprises an algorithm based on:

a machine learning single class classifier;

or

an anomaly detector algorithm.

3. The system according to claim 1, wherein the detection of an abnormality comprises an algorithm based on:

a support vector data description (SVDD);

a support vector machine (SVM);

a relevance vector machine (RVM);

a neural network;

neural analysis;

a large margin classifier;

a kernel based classifier;

a probability density function (PDF) estimator;

a Parzen PDF estimator;

a Bayesian classifier;

a Constant False Alarm Rate (CFAR) detector;

or a fuzzy logic based classifier.

4. The system according to claim 1, wherein the central processing unit is remotely located from the retinal scanner.

5. The system according to claim 1, wherein the central processing unit and the retinal scanner are integrated together in a kiosk.

6. The system according to claim 1, wherein the detection of an abnormality comprises detection of age related macular degeneration, macular and retinal degenerations, a diabetic retinopathy, a diabetic macular edema, a retinal vein occlusion, a vitreomacular interface abnormality, a macular hole, an epiretinal membrane, retinal vasculopathies, maculopathies, an optic nerve pathology, or glaucomatous optic nerve damage.

7. The system according to claim 1, wherein said system forms a retinal image based on the retinal data.

8. The system according to claim 7, wherein the analysis of the retinal data comprises detecting and excluding a non-pathological anatomical feature from the retinal image.

9. The system according to claim 7, wherein the analysis of the retinal data comprises:

detecting and excluding from the retinal image an optic disk region near an optic nerve, a non-pathological pigmentation, a lesion, or a blood vessel;

selecting from a parsed local region in order to identify anomalies;

or

selecting from the parsed local region in order to exclude anomalies.

10. The system according to claim 7, wherein the analysis of the retinal data comprises:

selecting a subset of normal pixels for training from;

a prescribed region of the retinal image;

a random selection of pixels from the retinal image;

or

a set of preexisting normal training images.

11. The system according to claim 7, wherein the analysis of the retinal data comprises:

using full red, green, and blue (RGB) color components of the retinal image, extracting a green channel from the retinal image, extracting a red channel from the retinal image, extracting a blue channel from the retinal image, performing intensity remapping of the retinal image, performing contrast stretching of the retinal image, performing image enhancement, performing thresholding of the retinal image, performing image registration and matching, performing remapping of an original (RGB) color space into another color space, performing stochastic classification of pixels of the retinal image into different classes, performing best feature selection using principal component analysis (PCA) on the retinal image, performing edge detection on the retinal image, segmenting retinal anatomical features in the retinal image, segmenting retinal vasculature in the retinal image, segmenting fovea in the retinal image, creating a feature vector, performing connected components on anomalous pixels of the retinal image to obtain anomalous blobs, performing intensity equalization on the retinal image, performing texture analysis of the retinal image, obtaining a wavelet signature of pixels of the retinal image, or performing multi-resolution processing of the retinal image.

12. The system according to claim 1, wherein the analysis of the retinal data comprises:

linear filtering, spectral filtering, nonlinear filtering, low pass filtering, median filtering, performing histogram equalization, performing adaptive histogram equalization, normalizing the retinal data, whitening the retinal data, performing band selection on the retinal data, performing dimensionality reduction on the retinal data, performing best feature and classifier selection using Boosting, computing a gradient, computing a Hessian, performing global static window analysis, performing local static window analysis, performing local sliding window analysis, comparing probability distribution functions (PDFs) of retinal data features using histograms, comparing PDFs of retinal data features using mutual information, comparing PDFs of retinal data features using correlation, comparing PDFs of retinal data features using Kullback-Leiber (KL) distance, using segmentation of the retinal data, creating a feature vector composed of original single band data, creating a feature vector composed of original multi band data, creating a feature vector composed by downselecting original multi band data, creating a feature vector composed of output from filtering original data, creating a feature vector composed of a wavelet analysis of original data, creating a feature vector composed of a statistical measure derived from a single or multiple bands of the retinal data, creating a feature vector composed of morphological features of the retinal data, creating a feature vector composed of shape features of anomalous blobs, creating a feature vector composed of an area of detected objects, using binary level mathematical morphological operations, using grey level mathematical morphological operations, using binary level mathematical morphological filters, using grey level mathematical morphological filters, or creating a feature vector.

13. The system according to claim 1, wherein the retinal data is acquired from a fundus imager, a non-mydriatic fundus imager, a mydriatic fundus imager, an optical coherent tomography system, a monochromatic imager, a portable optical coherence tomography machine, a multispectral camera, a hyperspectral sensor, a scanning laser ophthalmoscopy system, an eye fundus angiography system, a stereoscopic imaging system, or a multi-modality system.

14. The system according to claim 1, wherein the detection of an abnormality comprises detecting an abnormality based on:

a global anomaly detection analysis;

or

a local anomaly detection analysis.

15. The system according to claim 7, wherein the detection of an abnormality comprises detecting an abnormality based on:

color, gradient, texture, intensity derived features, depth, area, or shape derived features.

16. The system according to claim 7, wherein the detection of an abnormality comprises detecting an abnormality based on a longitudinal study comprising:

comparing the retinal image with a second retinal image and detecting an abnormality in the retinal image based on;

counting pixels;

or

change detection.

17. The system according to claim 1, wherein the detection of an abnormality comprises incorporation of demographic information, age data, race data, ocular information, ph level, ocular pressure, arterovenous ratio, systemic factors, body mass index data, blood pressure data, genetic information, family history data, or a patient severity scale.

18. The system according to claim 7, wherein the detection of an abnormality further comprises:

analyzing the retinal image for a quality criterion; and

requesting new retinal data when the retinal image does not meet the quality criterion.

19. The system according to claim 18, wherein the quality criterion comprises:

sharpness, focus, contrast, or presence of artifacts due to illumination.

20. A method for automated detection of retinal abnormalities, comprising:

receiving retinal data of the retina of a subject from a retinal scanner, wherein the retinal scanner is in communication with a computer;

characterizing a healthy area of the retina from the retinal data using the computer; and

determining an area of abnormality in the retinal data that deviates from the characterized healthy area using the computer.

21. The method according to claim 20, wherein the determining an area of abnormality comprises using an algorithm based on:

a machine learning single class classifier;

or

an anomaly detector algorithm.

22. The method according to claim 20, wherein the determining an area of abnormality comprises using an algorithm based on:

a support vector data description (SVDD), a support vector machine (SVM), a relevance vector machine (RVM), a neural network, neural analysis, a large margin classifier, a kernel based classifier, a probability density function (PDF) estimator, a Parzen PDF estimator, a Bayesian classifier, a Constant False Alarm Rate (CFAR) detector, or a fuzzy logic based classifier.

23. The method according to claim 20, wherein the determining an area of abnormality comprises detection of age related macular degeneration, macular and retinal degenerations, a diabetic retinopathy, a diabetic macular edema, a retinal vein occlusion, a vitreomacular interface abnormality, a macular hole, an epiretinal membrane, retinal vasculopathies, maculopathies, an optic nerve pathology, or glaucomatous optic nerve damage.

24. The method according to claim 20, further comprising forming a retinal image based on the retinal data.

25. The method according to claim 24, further comprising detecting and excluding a non-pathological anatomical feature of the retinal image using the computer.

26. The method according to claim 24, further comprising:

detecting and excluding from the retinal image an optic disk region near an optic nerve, a non-pathological pigmentation, a lesion, or a blood vessel using the computer;

selecting from a parsed local region in order to identify anomalies using the computer;

or

selecting from the parsed local region in order to exclude anomalies using the computer.

27. The method according to claim 24, further comprising:

selecting a subset of normal pixels for training from;

a prescribed region of the retinal image;

a random selection of pixels from the retinal image;

or

a set of preexisting normal training images.

28. The method according to claim 24, further comprising:

using full red, green, and blue (RGB) color components of the retinal image, extracting a green channel from the retinal image, extracting a red channel from the retinal image, extracting a blue channel from the retinal image, performing intensity remapping of the retinal image, performing contrast stretching of the retinal image, performing image enhancement, performing thresholding of the retinal image, performing image registration and matching, performing remapping of an original (RGB) color space into another color space, performing stochastic classification of pixels of the retinal image into different classes, performing best feature selection using principal component analysis (PCA) on the retinal image, performing edge detection on the retinal image, segmenting retinal anatomical features in the retinal image, segmenting retinal vasculature in the retinal image, segmenting fovea in the retinal image, creating a feature vector, performing connected components on anomalous pixels of the retinal image to obtain anomalous blobs, performing intensity equalization on the retinal image, performing texture analysis of the retinal image, obtaining a wavelet signature of pixels of the retinal image, or performing multi-resolution processing of the retinal image.

29. The method according to claim 20, further comprising:

linear filtering, spectral filtering, nonlinear filtering, low pass filtering, median filtering, performing histogram equalization, performing adaptive histogram equalization, normalizing the retinal data, whitening the retinal data, performing band selection on the retinal data, performing dimensionality reduction on the retinal data, performing best feature and classifier selection using Boosting, computing a gradient, computing a Hessian, performing global static window analysis, performing local static window analysis, performing local sliding window analysis, comparing probability distribution functions (PDFs) of retinal data features using histograms, comparing PDFs of retinal data features using mutual information, comparing PDFs of retinal data features using correlation, comparing PDFs of retinal data features using Kullback-Leiber (KL) distance, using segmentation of the retinal data, creating a feature vector composed of original single band data, creating a feature vector composed of original multi band data, creating a feature vector composed by downselecting original multi band data, creating a feature vector composed of output from filtering original data, creating a feature vector composed of a wavelet analysis of original data, creating a feature vector composed of a statistical measure derived from a single or multiple bands of the retinal data, creating a feature vector composed of morphological features of the retinal data, creating a feature vector composed of shape features of anomalous blobs, creating a feature vector composed of an area of detected objects, using binary level mathematical morphological operations, using grey level mathematical morphological operations, using binary level mathematical morphological filters, using grey level mathematical morphological filters, or creating a feature vector.

30. The method according to claim 20, wherein the retinal data is acquired from a standard fundus imager, a non-mydriatic imager, a mydriatic fundus imager, an optical coherent tomography system, a monochromatic imager, a portable optical coherence tomography machine, a multispectral camera, a hyperspectral sensor, a scanning laser ophthalmoscopy system, an eye fundus angiography system, a stereoscopic imaging system, or a multi-modality system.

31. The method according to claim 20, wherein the determining an area of abnormality comprises detecting an abnormality based on:

a global anomaly detection analysis;

or

a local anomaly detection analysis.

32. The method according to claim 24, wherein the determining an area of abnormality comprises detecting an abnormality based on:

color, gradient, texture, intensity derived features, depth, area, or shape derived features.

33. The method according to claim 24, wherein the determining an area of abnormality comprises detecting an abnormality based on a longitudinal study comprising:

comparing the retinal image with a second retinal image and detecting an abnormality in the retinal image based on;

counting pixels;

or

change detection.

34. The method according to claim 20, further comprising:

incorporating demographic information, age data, race data, ocular information, ph level, ocular pressure, arterovenous ratio, systemic factors, body mass index data, blood pressure data, genetic information, family history data, or a patient severity scale using the computer.

35. A computer readable medium storing executable instructions for execution by a computer having memory, the medium storing instructions for:

receiving retinal data from a retinal scanner coupled to the computer;

characterizing a healthy area of the retina from the retinal data; and

determining an area of abnormality in the retinal data that deviates from the characterized healthy area.

36. The computer readable medium according to claim 35, wherein the determining an area of abnormality comprises an algorithm based on:

a machine learning single class classifier;

or

an anomaly detector algorithm;

37. The computer readable medium according to claim 35, wherein the determining an area of abnormality comprises an algorithm based on:

a support vector data description (SVDD);

a support vector machine (SVM);

a relevance vector machine (RVM);

a neural network;

neural analysis;

a large margin classifier;

a kernel based classifier;

a probability density function (PDF) estimator;

a Parzen PDF estimator;

a Bayesian classifier;

a Constant False Alarm Rate (CFAR) detector;

or a fuzzy logic based classifier.

38. The computer readable medium according to claim 35, wherein the determining an area of abnormality comprises detection of age related macular degeneration, macular and retinal degenerations, a diabetic retinopathy, a diabetic macular edema, a retinal vein occlusion, a vitreomacular interface abnormality, a macular hole, an epiretinal membrane, retinal vasculopathies, maculopathies, an optic nerve pathology, or glaucomatous optic nerve damage.

39. The computer readable medium according to claim 35, wherein the retinal data comprises a retinal image.

40. The computer readable medium according to claim 39, further comprising instructions for:

detecting and excluding a non-pathological anatomical feature from the retinal image.

41. The computer readable medium according to claim 39, further comprising instructions for:

detecting and excluding from the retinal image an optic disk region near an optic nerve, a non-pathological pigmentation, a lesion, or a blood vessel;

selecting from a parsed local region in order to identify anomalies;

or

selecting from the parsed local region in order to exclude anomalies.

42. The computer readable medium according to claim 39, further comprising instructions for:

selecting a subset of normal pixels for training from;

a prescribed region of the retinal image;

a random selection of pixels from the retinal image;

or

a set of preexisting normal training images.

43. The computer readable medium according to claim 39, further comprising instructions for:

using full red, green, and blue (RGB) color components of the retinal image, extracting a green channel from the retinal image, extracting a red channel from the retinal image, extracting a blue channel from the retinal image, performing intensity remapping of the retinal image, performing contrast stretching of the retinal image, performing image enhancement, performing thresholding of the retinal image, performing image registration and matching, performing remapping of an original (RGB) color space into another color space, performing stochastic classification of pixels of the retinal image into different classes, performing best feature selection using principal component analysis (PCA) on the retinal image, performing edge detection on the retinal image, segmenting retinal anatomical features in the retinal image, segmenting retinal vasculature in the retinal image, segmenting fovea in the retinal image, creating a feature vector, performing connected components on anomalous pixels of the retinal image to obtain anomalous blobs, performing intensity equalization on the retinal image, performing texture analysis of the retinal image, obtaining a wavelet signature of pixels of the retinal image, or performing multi-resolution processing of the retinal image.

44. The computer readable medium according to claim 35, further comprising instructions for:

linear filtering, spectral filtering, nonlinear filtering, low pass filtering, median filtering, performing histogram equalization, performing adaptive histogram equalization, normalizing the retinal data, whitening the retinal data, performing band selection on the retinal data, performing dimensionality reduction on the retinal data, performing best feature and classifier selection using Boosting, computing a gradient, computing a Hessian, performing global static window analysis, performing local static window analysis, performing local sliding window analysis, comparing probability distribution functions (PDFs) of retinal data features using histograms, comparing PDFs of retinal data features using mutual information, comparing PDFs of retinal data features using correlation, comparing PDFs of retinal data features using Kullback-Leiber (KL) distance, using segmentation of the retinal data, creating a feature vector composed of original single band data, creating a feature vector composed of original multi band data, creating a feature vector composed by downselecting original multi band data, creating a feature vector composed of output from filtering original data, creating a feature vector composed of a wavelet analysis of original data, creating a feature vector composed of a statistical measure derived from a single or multiple bands of the retinal data, creating a feature vector composed of morphological features of the retinal data, creating a feature vector composed of shape features of anomalous blobs, creating a feature vector composed of an area of detected objects, using binary level mathematical morphological operations, using grey level mathematical morphological operations, using binary level mathematical morphological filters, using grey level mathematical morphological filters, or creating a feature vector.

45. The computer readable medium according to claim 35, wherein the retinal data is acquired from a fundus imager;

a non-mydriatic imager;

a mydriatic fundus imager;

an optical coherent tomography system, a monochromatic imager, a portable optical coherence tomography machine, a multispectral camera, a hyperspectral sensor, a scanning laser ophthalmoscopy system, an eye fundus angiography system, a stereoscopic imaging system, or a multi-modality system.

46. The computer readable medium according to claim 35, wherein the determining an area of abnormality comprises detecting an abnormality based on:

a global anomaly detection analysis;

or

a local anomaly detection analysis.

47. The computer readable medium according to claim 39, wherein the determining an area of abnormality comprises detecting an abnormality based on:

color, gradient, texture, intensity derived features, depth, area, or shape derived features.

48. The computer readable medium according to claim 39, wherein the determining an area of abnormality comprises detecting an abnormality based on a longitudinal study comprising:

comparing the retinal image with a second retinal image and detecting an abnormality in the retinal image based on;

counting pixels;

or

change detection.

49. The computer readable medium according to claim 35, further comprising instructions for incorporating demographic information, age data, race data, ocular information, ph level, ocular pressure, arterovenous ratio, systemic factors, body mass index data, blood pressure data, genetic information, family history data, or a patient severity scale.

50. The computer readable medium according to claim 39, wherein the determining an area of abnormality further comprises:

analyzing the retinal image for a quality criterion; and

requesting new retinal data when the retinal image does not meet the quality criterion.

51. The computer readable medium according to claim 50, wherein the quality criterion comprises:

sharpness, focus, contrast, or presence of artifacts due to illumination.