System and method for detecting retinopathy

1. A system for classifying fundus image features for detecting retinopathy, the system comprising:

a memory storing program instructions;

a processor configured to execute program instructions stored in the memory and configured to;

classify one or more fundus images based on a predetermined stage of retinopathy;

apply one or more transformations to the one or more classified fundus images;

extract a set of features from each of the transformed fundus images using multiple Convolutional Neural Networks (CNNs),wherein each of the multiple CNNs extracts the set of features corresponding to a specific class respectively from the classified transformed fundus images, and wherein the multiple CNNs includes:

a first CNN including a multiclass classification model extracting a first set of features corresponding to a first set of class probabilities from each of the images;

a second CNN including a binary classification model extracting a second set of features corresponding to a second set of class probabilities from each of the images;

a third CNN including a binary classification model extracting a third set of features corresponding to a third set of class probabilities from each of the images;

a fourth CNN including a binary classification model extracting a fourth set of features corresponding to a fourth set of class probabilities from each of the images; and

a fifth CNN including a binary classification model extracting a fifth set of features corresponding to a fifth set of class probabilities from each of the images;

and

determine a stage of retinopathy by classifying the extracted first, second, third, fourth and fifth set of features based on pre-stored features, wherein the pre-stored features are extracted from one or more training fundus images by the multiple CNNs, and further wherein each pre-stored feature corresponds to a class which is associated with a predetermined stage of retinopathy.

2. The system of claim 1 , wherein the one or more transformations comprise at least one of: contrast stretching and hue transformation.

3. The system of claim 1 , wherein the one or more features are extracted from the one or more transformed fundus images by passing the one or more transformed fundus images through one or more layers of the multiple CNNs thereby extracting details corresponding to one or more points of interest within the one or more transformed fundus images.

4. The system of claim 1 , wherein the one or more extracted features are classified by matching each of the one or more extracted features with the pre-stored features and further wherein the stage of retinopathy is determined based on the predetermined stage of retinopathy associated with the class corresponding to a maximum matched pre-stored features.

5. The system of claim 1 wherein the one or more fundus images are of one or more patients.

6. The system of claim 1 , wherein the predetermined stage of retinopathy is one of: absence of retinopathy, mild retinopathy, moderate retinopathy, severe retinopathy and proliferative retinopathy.

7. A computer-implemented method for classifying fundus image features for detecting retinopathy, via program instructions stored in a memory and executed by a processor, the computer-implemented method comprising:

classifying one or more fundus images based on a predetermined stage of retinopathy;

applying one or more transformations to the one or more classified fundus images;

extracting a set of features from each of the transformed fundus images using multiple Convolutional Neural Networks (CNNs), wherein each of the multiple CNNs extracts the features corresponding to a specific class respectively from the classified transformed fundus images, and wherein the multiple CNNs includes;

a first CNN including a multiclass classification model extracting a first set of features corresponding to a first set of class probabilities from each of the images;

a second CNN including a binary classification model extracting a second set of features corresponding to a second set of class probabilities from each of the images;

a third CNN including a binary classification model extracting a third set of features corresponding to a third set of class probabilities from each of the images;

a fourth CNN including a binary classification model extracting a fourth set of features corresponding to a fourth set of class probabilities from each of the images; and

a fifth CNN including a binary classification model extracting a fifth set of features corresponding to a fifth set of class probabilities from each of the images; and

determining a stage of retinopathy by classifying the extracted first, second, third, fourth and fifth set of features based on pre-stored features, wherein the pre-stored features are extracted from one or more training fundus images by the multiple CNNs, and further wherein each pre-stored feature corresponds to a class which is associated with a predetermined stage of retinopathy.

8. The computer-implemented method of claim 7 , wherein the one or more transformations comprise at least one of: contrast stretching and hue transformation.

9. The computer-implemented method of claim 7 , wherein the one or more features are extracted from the one or more transformed images by passing the one or more transformed images through one or more layers of the multiple CNNs thereby extracting details corresponding to one or more points of interest within the one or more transformed images.

10. The computer-implemented method of claim 7 , wherein the one or more extracted features are classified by matching each of the one or more extracted features with the pre-stored features and further wherein the stage of retinopathy is determined based on the predetermined stage of retinopathy associated with the class corresponding to a maximum matched pre-stored features.

11. The computer-implemented method of claim 7 , wherein the predetermined stage of retinopathy is one of: absence of retinopathy, mild retinopathy, moderate retinopathy, severe retinopathy and proliferative retinopathy.

12. A computer program product for classifying fundus image features for detecting retinopathy, the computer program product comprising:

a non-transitory computer-readable medium having computer-readable program code stored thereon, the computer-readable program code comprising instructions that when executed by a processor, cause the processor to:

classify one or more fundus images based on a predetermined stage of retinopathy;

apply one or more transformations to the one or more classified fundus images;

extract a set of features from each of the transformed fundus images using multiple Convolutional Neural Networks (CNNs), wherein each of the multiple CNNs extracts the features corresponding to a specific class respectively from the classified transformed fundus images, and wherein the multiple CNNs includes;

a first CNN including a multiclass classification model extracting a first set of features corresponding to a first set of class probabilities from each of the images;

a second CNN including a binary classification model extracting a second set of features corresponding to a second set of class probabilities from each of the images;

a third CNN including a binary classification model extracting a third set of features corresponding to a third set of class probabilities from each of the images;

a fourth CNN including a binary classification model extracting a fourth set of features corresponding to a fourth set of class probabilities from each of the images; and

a fifth CNN including a binary classification model extracting a fifth set of features corresponding to a fifth set of class probabilities from each of the images; and

determine a stage of retinopathy by classifying the extracted first, second, third, fourth and fifth set of features based on pre-stored features, wherein the pre-stored features are extracted from one or more training fundus images by the multiple CNNs and further wherein each pre-stored feature corresponds to a class which is associated with a predetermined stage of retinopathy.

13. The system of claim 1 , wherein the processor is configured to apply the one or more transformations to at least two versions of the one or more classified fundus images; and

extract the one or more features from the at least two versions of each of the transformed fundus images using the multiple Convolutional Neural Networks (CNNs).

14. The system of claim 1 , wherein a predetermined number of class probabilities are extracted corresponding to each CNN of the multiple CNNs.

15. The computed-implemented method of claim 7 , wherein the step of applying comprises applying the one or more transformations to at least two versions of the one or more classified fundus images; and

extracting the one or more features from the at least two versions of each of the transformed fundus images using the multiple Convolutional Neural Networks (CNNs).

16. The computed-implemented method of claim 7 , wherein a predetermined number of class probabilities are extracted corresponding to each CNN of the multiple CNNs.

17. The system of claim 1 , wherein the second set of features is same in number as the first set of features and the second set of class probabilities is less in number than the first set of class probabilities, the third set of features is same in number as the first set of features and the third set of class probabilities is less in number than the second set of class probabilities, the fourth set of features is same in number as the first set of features and the fourth set of class probabilities is same in number as the third set of class probabilities, and the fifth set of features is same in number as the first set of features and wherein the fifth set of class probabilities is same in number as the second set of class probabilities.

18. The computer-implemented method of claim 7 , wherein the second set of features is same in number as the first set of features and the second set of class probabilities is less in number than the first set of class probabilities, the third set of features is same in number as the first set of features and the third set of class probabilities is less in number than the second set of class probabilities, the fourth set of features is same in number as the first set of features and the fourth set of class probabilities is same in number as the third set of class probabilities, and the fifth set of features is same in number as the first set of features and wherein the fifth set of class probabilities is same in number as the second set of class probabilities.