Registration of retinal angiograms using self organizing maps

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate registration of multimodal retinal images is crucial for diagnosing and monitoring various eye conditions.
• Manual registration methods are time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop and evaluate an automated method for registering multimodal retinal images.
• To improve the accuracy and efficiency of retinal image registration.

Main Methods:

• A three-step registration algorithm was developed, involving vessel centerline detection and bifurcation point extraction in the reference image.
• Automatic correspondence of bifurcation points between images was achieved using a novel Self Organized Maps (SOMs) implementation.
• Affine transform parameters were extracted based on the established correspondences.

Main Results:

• The automated registration method was tested on 24 multimodal retinal image pairs.
• The algorithm demonstrated advantageous performance in terms of accuracy compared to manual registration.
• The use of SOMs facilitated robust and accurate correspondence of bifurcation points.

Conclusions:

• The proposed automatic method offers a more accurate and efficient alternative to manual registration for multimodal retinal images.
• This technique has the potential to enhance clinical diagnosis and treatment monitoring in ophthalmology.
• Further validation on larger datasets is warranted to confirm its clinical utility.