Strategies to improve the generalizability of deep learning-based OCT despeckling methods

OCT speckle hinders image interpretation. Hardware-based suppression is impractical and slow for in vivo imaging. Software-based methods, including deep learning, offer alternatives. However, deep learning models often lack generalization across different OCT systems. This study investigates adaptive normalization as a data augmentation technique to enhance the generalizability of deep learning-based speckle suppression for data from OCT systems not included in the training dataset.