UC Riverside

Adaptive optics (AO) has enabled in vivo imaging of the living human retina with diffraction-limited spatial resolution and thereby can image the retinal structure at the cellular level. However, AO may not readily be available for all imaging modalities, and in some cases its performance may be compromised by incomplete compensation of ocular wave aberrations. We investigate a deep learning based method to enhance spatial resolution of retinal images without or with AO. Twelve high-resolution retinal images were obtained using AO scanning laser ophthalmoscopy (AOSLO) as the ground truth. To model the image blur induced by ocular optical defects, we introduced various wave aberrations (with varying Zernike coefficients up to the 6th order). To expand the dataset, the AOSLO images were split into small patches (2820 average patches/image). With 8 different blurring kernels applied, there were 270,736 image patches for training and testing. In application of the trained networks, the corrected patches were combined to form images of their original sizes. The artificial neural network was based on a U-net architecture. We performed a 4-fold cross validation study with a quarter of images reserved for testing per cross validation. Normalized mean squared error (NMSE) and structural similarity index measure (SSIM) were calculated for the images before and after correcting for the blurring effects, using the ground truth images as reference. After correction, the NMSE was reduced by 51% on average, from 0.059 + 0.019 to 0.029 + 0.012. The SSIM was improved by 155% on average, from 0.22 + 0.09 to 0.56 + 0.12. The improvements were significant (by paired t-tests, p< 0.001). In another experiment, we trained the network on non-retinal images and fine-tuned on retinal images using 6-fold cross validation to explore the feasibility of applying transfer learning. Our results showed that deep learning may be useful in correcting the retinal image blur caused by aberrations. We outline the next possible phase for our work, where we intend to apply deep learning to retinal images taken without AO in order to demonstrate the removal of aberrations through processing in real-world examples.