PROJECT TITLE :

OCTRexpert A Feature-Based 3D Registration Method for Retinal OCT Images

ABSTRACT:

Data from longitudinal and cross-sectional studies can be analysed using medical image registration, which can also be used to guide computer-assisted diagnosis and treatment. However, the development of deformable registration for retinal optical coherence tomography (OCT) pictures has been slow. OCTRexpert, a new 3D registration method for retinal OCT data, is proposed in this work. There are currently no other 3D registration methods for retinal OCT pictures that can be applied to both normal and serious diseased individuals, according to our knowledge. After removing eye motion artefacts, a novel design-detection-deformation mechanism is used for registration. Each voxel in the image is given a few features to work with in the design phase. The point-to-point correspondences between the subject and template images are established in the detection step. The image is deformed hierarchically based on the multi-resolution correspondences found in the deformation step. A dataset of OCT images from 20 healthy people and four patients with significant Choroidal Neovascularization is used to test the suggested approach (CNV). Both the Dice similarity coefficient and average unsigned surface error are statistically significantly better when compared to other registration methods.