Research on registration method for enface image using multi-feature fusion

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Choroidal Neovascularization (CNV) is a significant cause of vision loss.
• Accurate registration of Optical coherence tomography (OCT) enface images is crucial for monitoring disease progression.
• CNV lesions can interfere with feature-based image registration.

Purpose of the Study:

• To develop a fast and accurate method for registering sequential OCT enface images.
• To overcome challenges posed by CNV lesions in OCT image registration.
• To improve the monitoring of retinal diseases and assist in clinical diagnosis.

Main Methods:

• A multi-feature registration strategy using a combined feature (com-feature) incorporating 3D, intersection, and SURF features.
• Simplification of vascular coordinate sets using the Douglas-Peucker algorithm.
• Registration of enface coordinate point sets using the coherent point drift algorithm.

Main Results:

• The proposed method significantly enhances the success rate of global registration for vascular networks in OCT enface images.
• The simplification step improves registration speed while preserving essential vascular features.
• Achieved Mean Squared Error (MSE) of 0.07993, Dice Similarity Coefficient (DSC) of 0.9693, and a time complexity of 42.7016 seconds.

Conclusions:

• The novel registration method improves accuracy and significantly reduces time complexity for OCT enface images.
• This technique offers efficient monitoring of CNV progression and aids in clinical diagnosis.
• The method demonstrates effective registration performance in clinical settings, providing a valuable tool for diagnosis and treatment.