Extended Data Fig. 9: RLN outperforms RCAN when attempting confocal-to-STED cross modality prediction.
From: Incorporating the image formation process into deep learning improves network performance
a) Lateral views (top) and axial views (bottom) of U2OS cells immunolabeled with a primary antibody against Tomm20 and an anti-rabbit secondary antibody conjugated with Alexa Fluor 594, comparing the raw input collected by confocal microscopy, STED microscopy images (ground truth), RCAN output, and RLN output. b) Line profiles across the yellow and red lines in the lateral view and axial view in a, demonstrating that RCAN and RLN improve resolution compared to the input, yet not to the extent of the ground truth. c) SSIM and PSNR analysis for data shown in a), means and standard deviations are obtained from N = 10 slices. Scale bars: 4 μm.
