Assessment of fundus images
An example method for automating a quality assessment of digital fundus image can include: obtaining a digital fundus image file; analyzing a first quality of the digital fundus image file using a model to estimate an optimal time to capture a fundus image; and analyzing a second quality of the digital fundus image file using the model to estimate a disease state.
1. A method of assessing a fundus image, the method comprising:
obtaining the fundus image;
analyzing a first quality of the fundus image using a first trained model;
analyzing a second quality of the fundus image using a second trained model; and
using the first and second qualities to generate an output, wherein the output includes an automated image capture for a subsequent fundus image.
2. The method of claim 1 , wherein the output further includes an automated disease diagnosis.
3. The method of claim 1 , wherein the first and second qualities are selected from the group consisting of overall optical quality of the funds image, optical quality in an optic disc area, optical quality in a macula area, vessel count in a predetermined region, and field position.
4. The method of claim 3 , wherein the field position is macular centered or optic disc centered.
5. The method of claim 1 , wherein the first and second trained models are convolutional neural networks.
6. The method of claim 1 , further comprising:
classifying an overall optical quality of the image;
when the overall optical quality is classified as readable, determining whether the image depicts a retina;
when the image is determined as depicting the retina, detecting fovea and optic disc regions in the image; and
determining an optical quality in the optic disc region.
7. The method of claim 6 , further comprising:
determining a blood vessel count for a predetermined region of the retina;
determining a field position of the image; and
providing an assessment of a disease state.
8. A device for assessing a fundus image, comprising:
at least one processing unit; and
a memory storing instructions which, when executed by the at least one processing unit, cause the device to:
obtain the fundus image;
analyze a first quality of the fundus image using a first trained model;
analyze a second quality of the fundus image using a second trained model; and
use the first and second qualities to generate an output, wherein the output includes an automated image capture for a subsequent fundus image.
9. The device of claim 8 , wherein the device is embedded in a fundus imaging system that captures the fundus image.
10. The device of claim 8 , wherein the device is located outside of a fundus imaging system that captures the fundus image.
11. The device of claim 8 , wherein the output further includes an automated disease diagnosis.
12. The device of claim 8 , wherein the first and second qualities are selected from the group consisting of overall optical quality of the funds image, optical quality in an optic disc area, optical quality in a macula area, vessel count in a predetermined region, and field position.
13. The device of claim 12 , wherein the field position is macular centered or optic disc centered.
14. The device of claim 8 , wherein the first and second trained models are convolutional neural networks.
15. The device of claim 8 , wherein the instructions further cause the device to:
classify an overall optical quality of the image;
when the overall optical quality is classified as readable, determine whether the image depicts a retina;
when the image is determined as depicting the retina, detect fovea and optic disc regions in the image; and
determine an optical quality in the optic disc region.
16. The device of claim 15 , wherein the instructions further cause the device to:
determine a blood vessel count for a predetermined region of the retina;
determine a field position of the image; and
provide an assessment of a disease state.
17. A non-transitory computer storage medium storing computer readable instructions configured for execution by at least one processing unit, the computer readable instructions causing the at least one processing unit to:
obtain a fundus image;
analyze a first quality of the fundus image using a first trained model;
analyze a second quality of the fundus image using a second trained model; and
use the first and second qualities to generate an output, wherein the output includes an automated image capture for a subsequent fundus image.
18. The non-transitory computer storage medium of claim 17 , wherein the first and second qualities are selected from the group consisting of overall optical quality of the funds image, optical quality in an optic disc area, optical quality in a macula area, vessel count in a predetermined region, and field position.