Laser Scar Enhancement with before/after comparison in Retinal Images

Laser surgery is a widely used treatment of microaneurysm and hemorrhage in ophthalmologic clinic. Laser scar is blurry light yellowish spots in retinal image (see Figure 1). The purpose of laser scar detection is to .....

Figure 1

We are using before after comparison technique to enhance the laser scars. Images of patients are taken periodically, and two adjacent images are compared to see if there are laser scars available, and whether the number of laser changes.

The procedure contains 4 steps:

• Image quality checking

Images quality affect the detection result significantly. There are three kinds of unqualified images: under exposed, over exposed and unbalanced exposed (see Figure 2). The


Under exposure	Over exposure	Unbalanced exposure

                                 

• Blood vessel map detection



Blood vessel map is extracted using our blood vessel detection algorithm in this journal paper, details can be found here.



• Image registration
  

  Two images need to be registered to perform the comparison. The registration is based on matching blood vessel bifurcation points. A bifurcation point on blood vessel map is where two or more blood vessels join. We choose bifurcation points as the land mark because they are stable in all images of the same patient and the number of them is small, therefore efficient in computation. Image registration quality is evaluated with "registration score". The skeletons of two blood vessel map are taken and registered. For each pixel of one skeleton in the common area, if there's any pixel on the other skeleton available in a small neighborhood, we define this pixel has a "match". The ratio between number of "matched" pixels and number of all pixels in the common area is called "registration score". If this score is lower than 50%, the registration is considered to be failed.
  

  
  

• Background extraction, image subtraction and post-processing

  Background of each image are extracted by a morphological opening operation with disk structure element. Let I₁ and I₂ be green channels of the two original images. and B₁ and B₂ be the background extracted. Let R_k=(I_k-B_k) be the residue image. The difference between two residue image (R₂-R₁) should be the brighter parts that appear in the later image and not in the older one, which is what we want - the brighter laser scar spots.

  A post-processing procedure is employed to pick up only laser scars. All non-round shape objects are removed, and the number of scars are counted.
  

Experiment Results

Pre	Post	Result

Number of laser scars found: 11

More results can be found here