Deep-learning–based optical coherence tomography reconstruction for high-speed and contrast morphology and vasculature imaging

Abstract. Significance Spectral-domain optical coherence tomography (SD-OCT) has been widely used in clinical ophthalmic imaging for high spatial resolution and phase stability. The implementation of multiple spectrometers could help resolve the challenges of SD-OCT, including limited imaging speed and sensitivity. However, these two improvements cannot be achieved concurrently. Aim We propose a deep-learning–based approach to enhance both imaging speed and sensitivity of SD-OCT systems using a modified U-Net architecture. Approach This network adopts a visual state space model to synthesize the high signal-to-noise ratio (SNR) OCT and OCTA from high-speed acquisitions, which bypasses the hardware restriction. The model performance is evaluated both qualitatively and quantitatively using the multiscale structural similarity index measure (MS-SSIM) and contrast-to-noise ratio (CNR). Results The results demonstrate effective performance in high-SNR OCT/OCTA reconstruction, providing better contrast between the retinal layers and improved delineation of layer boundaries. Fine structures in both the inner and outer retina, such as microcapillaries and choroid, are successfully restored. Conclusions We proposed an effective approach to improve the OCT image quality while maintaining the high acquisition speed using the DNN-based architecture, enabling simultaneous benefits of high imaging speed and enhanced sensitivity.