We present a robust deep learning framework for the automatic localisation of cone photoreceptor cells in Adaptive Optics Scanning Light Ophthalmoscope (AOSLO) split-detection images. Monitoring cone photoreceptors with AOSLO imaging grants an excellent view into retinal structure and health, provides new perspectives into well known pathologies, and allows clinicians to monitor the effectiveness of experimental treatments. The MultiDimensional Recurrent Neural Network (MDRNN) approach developed in this paper is the first method capable of reliably and automatically identifying cones in both healthy retinas and retinas afflicted with Stargardt disease. Therefore, it represents a leap forward in the computational image processing of AOSLO images, and can provide clinical support in on-going longitudinal studies of disease progression and therapy. We validate our method using images from healthy subjects and subjects with the inherited retinal pathology Stargardt disease, which significantly alters image quality and cone density. We conduct a thorough comparison of our method with current state-of-the-art methods, and demonstrate that the proposed approach is both more accurate and appreciably faster in localizing cones. As further validation to the method’s robustness, we demonstrate it can be successfully applied to images of retinas with pathologies not present in the training data: achromatopsia, and retinitis pigmentosa.

Precise measurements of photoreceptor numerosity and spatial arrangement are promising biomarkers for the early detection of retinal pathologies and may be valuable in the evaluation of retinal therapies. Adaptive optics scanning light ophthalmoscopy (AOSLO) is a method of imaging that corrects for aberrations of the eye to acquire high-resolution images that reveal the photoreceptor mosaic. These images are typically graded manually by experienced observers, obviating the robust, large-scale use of the technology. This paper addresses unsupervised automated detection of cones in non-confocal, split-detection AOSLO images. Our algorithm leverages the appearance of split-detection images to create a cone model that is used for classification. Results show that it compares favorably to the state-of-the-art, both for images of healthy retinas and for images from patients affected by Stargardt disease. The algorithm presented also compares well to manual annotation while excelling in speed.

The field of view of high-resolution ophthalmoscopes that require the use of adaptive optics (AO) wavefront correction is limited by the isoplanatic patch of the eye, which varies across individual eyes and with the portion of the pupil used for illumination and/or imaging. Therefore all current AO ophthalmoscopes have small fields of view comparable to, or smaller than, the isoplanatic patch, and the resulting images have to be stitched off-line to create larger montages. These montages are currently assembled either manually, by expert human graders, or automatically, often requiring several hours per montage. This arguably limits the applicability of AO ophthalmoscopy to studies with small cohorts and moreover, prevents the ability to review a real-time captured montage of all locations during image acquisition to further direct targeted imaging. In this work, we propose stitching the images with our novel algorithm, which uses oriented fast rotated brief (ORB) descriptors, local sensitivity hashing, and by searching for a ‘good enough’ transformation, rather than the best possible, to achieve processing times of 1–2 minutes per montage of 250 images. Moreover, the proposed method produces montages which are as accurate as previous methods, when considering the image similarity metrics: normalised mutual information (NMI), and normalised cross correlation (NCC).

No abstract

No abstract