Artificial intelligence-based system and methods for corneal diagnosis

Patent metadata

Field	Value
Publication number	US-10468142-B1
Application number	US-201816047944-A
Country	US
Kind code	B1
Filing date	Jul 27, 2018
Priority date	Jul 27, 2018
Publication date	Nov 5, 2019
Grant date	Nov 5, 2019

Abstract

A method of predicting a disease or condition of a cornea or an anterior segment of an eye includes inputting input data into an AI model, processing the input data, and generating a set of scores and outputting a prediction. The input data may be representative of a cornea or anterior segment of an eye. Processing the input data may include processing the data through the plurality of convolutional layers, the fully connected layer, and the output layer. Each score of the set of scores may be generated by a corresponding node in the output layer. The output prediction may be related to the cornea or anterior segment of the eye represented by the input data processed through the AI model. The prediction may be determined by at least one score of the set of scores.

First claim

What is claimed is:

1. A system for facilitating neural-network-based determinations of corneal conditions, the system comprising:
a neural network comprising a plurality of layers, the plurality of layers comprising:
at least one input layer configured to take, as associated inputs, both an image of one eye and an image of another eye; and
at least one hidden layer configured to (i) obtain a first prediction related to presence of a corneal condition in the one eye, (ii) obtain a second prediction related to presence of the corneal condition in the other eye, and (iii) adjust at least one of the first or second predictions based on the other one of the first or second predictions; and

one or more processors executing computer program instructions that, when executed, cause the one or more processors to:
for each eye of at least one subject, provide multiple frames of a same image cut from a corneal scan of the eye to the neural network to train the neural network, the neural network generating at least one prediction related to presence of at least one corneal condition for the eyes of the at least one subject based on the multiple frames of the same image cut;
subsequent to the training of the neural network, provide, as associated inputs, an image of a first eye of a subject and an image of a second eye of the subject to the neural network; and
obtain, via the neural network, one or more predictions related to presence of at least one corneal condition for the first and second eyes of the subject, the neural network generating the one or more predictions based on the images of the first and second eyes.

2. The system of claim 1 , wherein the at least one hidden layer is configured to perform the adjustment by applying a weight associated with the first prediction to the second prediction to adjust the second prediction.

3. The system of claim 1 , wherein the at least one hidden layer is configured to perform the adjustment by adjusting the second prediction based on the first prediction indicating that the corneal condition is present in the one eye.

4. The system of claim 1 , wherein the at least one input layer is configured to take, as associated inputs, the images of the one eye and the other eye and one or more thickness maps, and wherein the neural network is configured to generate the first and second predictions based on the images of the one eye and the other eye and the one or more thickness maps.

5. The system of claim 1 , wherein providing the multiple frames comprises providing multiple raw frames of the same image cut from the corneal scan to the neural network such that each frame of the multiple raw frames of the same image cut comprises different noise patterns from other frames of the multiple raw frames.

6. The system of claim 1 , wherein the one or more processors are caused to:
provide multiple raw frames of a first image cut of the first eye and multiple raw frames of a second image cut of the second eye to the neural network to obtain the one or more predictions,
wherein the neural network generates the one or more predictions based on the multiple raw frames of the first and second image cuts.

7. The system of claim 1 , wherein the one or more predictions relate to presence of at least one of corneal ectasia, Keratoconus, corneal graft rejection episode and failure, dry eye syndrome (DES), Fuchs' dystrophy, corneal limbal stem cell deficiency, cataract, or glaucoma.

8. The system of claim 1 , wherein the one or more predictions relate to presence of two or more of corneal ectasia, Keratoconus, corneal graft rejection episode and failure, dry eye syndrome (DES), Fuchs' dystrophy, corneal limbal stem cell deficiency, cataract, or glaucoma.

9. A method implemented by one or more processors executing computer program instructions that, when executed, perform the method, the method comprising:
providing a prediction model comprising a plurality of layers, the plurality of layers comprising:
at least one input layer configured to take, as associated inputs, corneal images associated with one another;
at least one output layer configured to output one or more cornea-related predictions; and
at least one hidden layer between the at least one input layer and the at least one output layer, the at least one hidden layer being configured to (i) obtain a first prediction related to presence of a corneal condition, (ii) obtain a second prediction related to presence of the corneal condition, and (iii) adjust at least one of the first or second predictions based on the other one of the first or second predictions, wherein the first prediction is derived from one of the associated corneal images, and the second prediction is derived from another one of the associated corneal images;

providing, as associated inputs, a first corneal image associated with a subject and a second corneal image associated with the subject to the prediction model; and
obtaining, via the prediction model, one or more predictions related to presence of at least one corneal condition for the subject, the prediction model generating the one or more predictions based on the first and second corneal images.

10. The method of claim 9 , wherein the at least one hidden layer is configured to perform the adjustment by applying a weight associated with the first prediction to the second prediction to adjust the second prediction.

11. The method of claim 9 , wherein the at least one hidden layer is configured to perform the adjustment by adjusting the second prediction based on the first prediction indicating that the corneal condition is present in an eye of the subject.

12. The method of claim 9 , wherein the at least one input layer is configured to take, as associated inputs, the corneal images and one or more thickness maps, and wherein the prediction model is configured to generate the first and second predictions based on the corneal images and the one or more thickness maps.

13. The method of claim 9 , further comprising:
providing multiple frames of a same image cut from a corneal scan to the prediction model to train the prediction model, the prediction model generating at least one prediction related to presence of at least one corneal condition based on the multiple frames of the same image cut.

14. The method of claim 13 , wherein providing the multiple frames comprises providing multiple raw frames of the same image cut from the corneal scan to the prediction model such that each frame of the multiple raw frames of the same image cut comprises different noise patterns from other frames of the multiple raw frames.

15. The method of claim 9 , further comprising:
providing multiple raw frames of a first image cut from a first corneal scan and multiple raw frames of a second image cut from a second corneal scan to the prediction model to obtain the one or more predictions,
wherein the prediction model generates the one or more predictions based on the multiple raw frames of the first and second image cuts.

16. The method of claim 9 , wherein the one or more predictions relate to presence of at least one of corneal ectasia, Keratoconus, corneal graft rejection episode and failure, dry eye syndrome (DES), Fuchs' dystrophy, corneal limbal stem cell deficiency, cataract, or glaucoma.

17. The method of claim 9 , wherein the one or more predictions relate to presence of two or more of corneal ectasia, Keratoconus, corneal graft rejection episode and failure, dry eye syndrome (DES), Fuchs' dystrophy, corneal limbal stem cell deficiency, cataract, or glaucoma.

18. T

Assignees

• Univ Miami

Inventors

• Abou Shousha Mohamed
• Elsawy Amr Saad Mohamed

Classifications

• G06T2207/30041

  Eye; Retina; Ophthalmic · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/10101

  Optical tomography; Optical coherence tomography [OCT] · CPC title

• G06T2207/10024

  Color image · CPC title

• G06T7/0012

  Biomedical image inspection · CPC title