Quality improvement of adaptive optics retinal images using conditional adversarial networks

Abstract

The adaptive optics (AO) technique is widely used to compensate for ocular aberrations and improve imaging resolution. However, when affected by intraocular scatter, speckle noise, and other factors, the quality of the retinal image will be degraded. To effectively improve the image quality without increasing the imaging system's complexity, the post-processing method of image deblurring is adopted. In this study, we proposed a conditional adversarial network-based method for directly learning an end-to-end mapping between blurry and restored AO retinal images. The proposed model was validated on synthetically generated AO retinal images and real retinal images. The restoration results of synthetic images were evaluated with the metrics of peak signal-to-noise ratio (PSNR), structural similarity (SSIM), perceptual distance, and error rate of cone counting. Moreover, the blind image quality index (BIQI) was used as the no-reference image quality assessment (NR-IQA) algorithm to evaluate the restoration results on real AO retinal images. The experimental results indicate that the images restored by the proposed method have sharper quality and higher signal-to-noise ratio (SNR) when compared with other state-of-the-art methods, which has great practical significance for clinical research and analysis.

ERICA: Emulated Retinal Image CApture - A tool for testing, training and validating retinal image processing methods

“…Previous approaches for generating synthetic images have performed image generation in a similar way to the method we have developed. These approaches model cone cells as Gaussian intensity profiles and convolve the retinal image with the PSF 53,54 . ERICA fully replicates the image generation process, as shown in Figure 4, and we include a step to separately specify the input and output PSFs to include, for example, differences in the input and output pupil diameters.…”

“…Such an approach has been demonstrated for testing the accuracy of motion extraction from an AOSLO using synthetic data created by resampling real images 50,51 . Synthetic images have also been used to test the accuracy of cone detection algorithms 52,53 and, recently, to train deep learning models for image enhancement 54 . These approaches have generated synthetic retinal mosaics by randomly placing cells 52 or by modifying a hexagonal packing matrix 53,55,56 but stop short of simulating the full data capture pipeline.…”

The optics of the human eye at 8.6 µm lateral resolution

“…The applications of ocular aberrometry are ubiquitous: they include refractive surgery (i.e., LASIK) [28], the diagnosis and monitoring of diseases [29], myopia control [30], intraocular lenses [31], light- adjustable intraocular lenses [32], objective refraction [33], intraoperative aberrometry [34], and vision science [35]. Additionally, combining a wavefront sensor with a compensating element, such as a liquid crystal spatial light modulator, has dramatically improved in vivo observations of the human retina [36]. It has also enabled the possibility of improved visual perception, so-called super-vision, which is a visual acuity far beyond the normal values [37].…”

Emulated retinal image capture (ERICA) to test, train and validate processing of retinal images

“…Previous approaches for generating synthetic images have performed image generation in a similar way to the method we have developed. These approaches model cone cells as Gaussian intensity profiles and convolve the retinal image with the PSF 41 , 42 . ERICA fully replicates the image generation process, as shown in Fig.…”

“…Such an approach has been demonstrated for testing the accuracy of motion extraction from an AOSLO using synthetic data created by resampling real images 22 , 39 . Synthetic images have also been used to test the accuracy of cone detection algorithms 40 , 41 and, recently, to train deep learning models for image enhancement 42 . These approaches have generated synthetic retinal mosaics by randomly placing cells 40 or by modifying a hexagonal packing matrix 41 , 43 , 44 but stop short of simulating the full data capture pipeline.…”