Method and system for local adaptive detection of microaneurysms in digital fundus images
View Patent ↗A local adaptive method is proposed for automatic detection of microaneurysms in a digital ocular fundus image. Multiple subregions of the image are automatically analyzed and adapted to local intensity variation and properties. A priori region and location information about structural features such as vessels, optic disk and hard exudates are incorporated to further improve the detection accuracy. The method effectively improves the specificity of microaneurysms detection, without sacrificing sensitivity. The method may be used in automatic level-one grading of diabetic retinopathy screening.
1. A method for detection of candidate microaneurysms in a digital ocular fundus image, the method comprising:
subdividing the image into a plurality of subregions;
image enhancing each subregion to correct for background variations within the subregion, wherein each subregion is assigned a threshold value equal to a lesser of a locally adapted threshold value based on properties of the subregion and a pre-defined maximum threshold value, and the candidate microaneurysms are identified in the subregions; and
segmenting the candidate microaneurysms from a background image and from other structures in each subregion using shape features of microaneurysms and directional morphological operations.
2. The method of claim 1 , further comprising:
separating the subregions into overlapping regions; and
discounting a false microaneurysm by determining that a structure detected in a first overlapping region is a part of a structure other than the candidate microaneurysms detected in a second overlapping region.
3. The method of claim 2 , wherein the digital ocular fundus image is 1024×1280 pixels, the subregions are 120 pixels per side and the regions overlap by 10 pixels per side.
4. The method of claim 1 , wherein image enhancing each subregion to correct for background variations within the subregion, further comprises:
correcting for shading effect in each subregion.
5. The method of claim 4 , wherein correcting for shading effect in each subregion, further comprises:
estimating a background image in the subregion; and
subtracting the estimated background image from an original image.
6. The method of claim 5 , wherein estimating a background image in the subregion, further comprises applying a low-pass two-dimensional Gaussian filter.
7. The method of claim 4 , wherein image enhancing each subregion further comprises:
enhancing contrast in the subregion; and
smoothing to reduce step effects.
8. The method of claim 1 , wherein segmenting the candidate microaneurysms from a background image and from other structures in each subregion using shape features of microaneurysms and directional morphological operations, comprises:
using a “Top Hat” filter to identify microaneurysms.
9. The method of claim 1 , wherein segmenting the candidate microaneurysms from a background image and from other structures in each subregion using shape features of microaneurysms and directional morphological operations, comprises:
dilating structures in the image using a linear structuring element.
10. The method of claim 1 , further comprising:
identifying a structure other than the candidate microaneurysms in the digital ocular fundus image; and
discounting a false microaneurysm upon detecting that a location of the false microaneurysm corresponds to a location of the identified structure other than the candidate microaneurysms.
11. The method of claim 10 , wherein the identified structure other than the candidate microaneurysms is one or more of an optic disk, and a hard exudate, and identifying the hard exudate comprises combining a color feature and a texture feature of the hard exudate.
12. A computer program product comprising a non-transitory computer readable recording medium having recorded thereon a computer program comprising code means for, when executed on a computer, instructing said computer to control a method for detection of candidate microaneurysms in a digital ocular fundus image, the method comprising:
subdividing the image into a plurality of subregions;
image enhancing each subregion to correct for background variations within the subregion, wherein each subregion is assigned a threshold value equal to a lesser of a locally adapted threshold value based on properties of the subregion and a pre-defined maximum threshold value, and the candidate microaneurysms are identified in the subregions; and
segmenting the candidate microaneurysms from a background image and from other structures in each subregion using shape features of microaneurysms and directional morphological operations.
13. The computer program product of claim 12 , further comprising:
separating the subregions into overlapping regions; and
discounting a false microaneurysm by determining that a structure detected in a first overlapping region is a part of a structure other than the candidate microaneurysms detected in a second overlapping region.
14. The computer program product of claim 13 , wherein the digital ocular fundus image is 1024×1280 pixels, the subregions are 120 pixels per side and the regions overlap by 10 pixels per side.
15. The computer program product of claim 12 , wherein image enhancing each subregion to correct for background variations within the subregion, further comprises:
correcting for shading effect in each subregion.
16. The computer program product of claim 15 , wherein correcting for shading effect in each subregion, further comprises:
estimating a background image in the subregion; and
subtracting the estimated background image from an original image.
17. The computer program product of claim 16 , wherein estimating a background image in the subregion, further comprises applying a low-pass two-dimensional Gaussian filter.
18. The computer program product of claim 15 , wherein image enhancing each subregion further comprises:
enhancing contrast in the subregion; and
smoothing to reduce step effects.
19. The computer program product of claim 12 , wherein segmenting the candidate microaneurysms from a background image and from other structures in each subregion using directional morphological operations, comprises:
using a “Top Hat” filter to locally identify microaneurysms.
20. The computer program product of claim 12 , wherein segmenting the candidate microaneurysms from a background image and from other structures in each subregion using shape features of microaneurysms and directional morphological operations, comprises:
dilating structures in the image using a linear structuring element.
21. The computer program product of claim 12 , wherein the method further comprises:
identifying a structure other than the candidate microaneurysms in the digital ocular fundus image; and
discounting a false microaneurysm upon detecting that a location of the false microaneurysm corresponds to a location of the identified structure other than the candidate microaneurysms.
22. The computer program product of claim 21 , wherein the identified structure other than the candidate microaneurysms is one or more of an optic disk, and a hard exudate, and identifying the hard exudate comprises combining a color feature and a texture feature of the hard exudate.