Enhanced Framework for Concurrent correction and Segmentation in Retinal Optical Coherence Tomography

A revolutionary imaging technique called optical coherence tomography of visible light (VIS-OCT) of the individual uses shorter visible light wavelengths than traditional near-infrared (NIR) light. To more accurately discern stratified retinal layers, it offers microvascular oximetry together with one-micron level axial resolution. Since the allowed illumination power is significantly lower than NIR OCT due to practical limits regarding laser safety and comfort, it might be difficult to generate VIS-OCT images of a high enough quality for further image processing. As a result, denoising VIS-OCT images is a crucial step in the whole workflow for clinical applications involving VIS-OCT. The first retinal image collection is presented in this study from normal eyes obtained using VIS-OCT. We offer a simultaneous self-denoising and segmentation system based on deep learning. Both tasks complement one another inside the same network to increase each other’s productivity. Annotation-efficient training is demonstrated by a discernible increase in When the available annotation falls below 25%, the Dice coefficient was 2% higher than with the segmentation-only method.) is accomplished. Additionally, we demonstrated how well the denoising model learned on our dataset could be applied to an alternative scanning method.