Alignment algorithm for retinal optical coherence tomography images based on improved deep feature matching

We propose an improved Deep Feature Matching (DFM) method for alignment of retinal optical coherence tomography (OCT) images. A squeeze-and-excitation module is integrated within the DFM framework, which adjusts the weight of each feature channel adaptively to enhance the capacity of DFM to capture key anatomical features relevant to the alignment task. It compensates for the shortcomings of DFM and improves alignment accuracy and robustness on noisy and artefact-heavy OCT images significantly. Experimental results indicate that the improved DFM outperforms the original DFM and traditional methods significantly in terms of matching accuracy and feature distribution.