The National Eye Institute (NEI) seeks research co-development partners and/or licensees for an automatic deep learning-based algorithm to detect and quantitate ellipsoid zone (EZ) loss in Spectral Domain Optical Coherence Tomography (SD-OCT) images.
The present disclosure generally relates a method of automatically detecting ellipsoid zone (EZ) loss in spectral domain optical coherence tomography (SD-OCT) imaging. EZ band represents outer segments of photoreceptors in the retina, and its loss reflects a deterioration of the photoreceptors. EZ loss has been proposed to be evidence of progression of several retinal degenerative diseases including, but not limited to, retinitis pigmentosa and hydroxychloroquine (HCQ)-induced retinal toxicity. HCQ is a first-line drug used to treat autoimmune diseases such as systemic lupus erythematosus, Sjögren’s syndrome, and rheumatoid arthritis. One of the major side-effects that can occur in long-term users of HCQ, is EZ loss that can result in retinal toxicity and permanent damage to photoreceptors and retinal pigment epithelium (RPE), eventually leading to irreversible loss of central vision. The American Academy of Ophthalmology (AAO) recommends two main screening modalities including SD-OCT imaging and functional tests such as visual fields with the goal of recognizing early definitive signs of HCQ-induced retinal toxicity to prevent vision loss. Although this side-effect is estimated to occur in 7.5% of patients taking the drug for more than 10 years, we currently have no treatment for this serious side effect that tends to continue even after the cessation of the drug.
Researchers at the NEI have developed a method that can automatically detect and quantitate EZ loss in SD-OCT images immediately after image acquisition. The method includes a deep learning framework with a two-step approach. In the first stage, the method detects and annotates EZ loss regions in individual OCT B-scans. A 2D map is constructed twice in a dual architecture to enhance robustness, where horizontal and vertical slices extracted from the 3D image are trained separately. The second stage of the model operates on these two 2D maps and estimates the final EZ loss map representing the 3D OCT volume. Compared to other screening methods, the algorithm demonstrated excellent performance in diagnosing toxicity even as a stand-alone test, with an F1 score, a measure of test accuracy, of 0.91. This indicates the utility of the tool in assisting with screening for toxicity in an automatic, accurate, time-effective, cost-effective, and objective manner. Addition of this methodology onto current SD-OCT screening could assist the clinician in making diagnostic and treatment decisions immediately after SD-OCT acquisitions.
Protected claims for this invention include the method of detecting and outlining the region of EZ loss directly from acquired OCT images, associated algorithms, and the device containing these algorithms and capabilities. This technology is available for licensing.