Deep Learning-based retinal optical coherence tomography markers for optic neuropathies

PROJECT SUMMARY/ABSTRACT Optic neuropathies, a wide-ranging group of eye diseases, primarily target the optic nerve and frequently result in irreversible vision loss. Globally affecting millions, these conditions stem from a variety of causes such as glaucoma, ischemic, inflammatory, compressive, toxic, and hereditary factors, each with distinct pathophysiological traits. Characterized by the degeneration of retinal ganglion cells, these diseases lead to changes in the optic nerve head and visual field defects. The diverse onset and myriad underlying causes of these neuropathies complicate early diagnosis, underscoring the necessity for advanced diagnostic tools. This project proposes leveraging deep learning-based morphometric analysis in optical coherence tomography (OCT) to develop innovative biomarkers for optic neuropathies. OCT, a non-invasive imaging technique, has revolutionized the diagnosis and management of these conditions, providing detailed retinal and optic nerve head imagery. However, its efficacy is limited by device-dependent variability, signal quality dependency, and insufficient sensitivity of current thickness markers in chronic disease monitoring. This project proposes to overcome these limitations by employing advanced computational techniques such as deep learning, neural fields, and geometric modeling. These methods excel in extracting complex patterns from medical images and enhancing the accuracy of morphometric analyses. Geometric deep learning adapts to the neuroretina's geometric structures, offering novel insights into the optic nerve head and macula. Neural field image registration enhances OCT image co-registration accuracy, crucial for longitudinal disease monitoring. The project aims to revolutionize the diagnosis and management of optic neuropathies through three interconnected objectives. The first aim is to employ deep generative models for mapping structure-function relationships in optic neuropathies, focusing on predicting visual field outcomes and tracking disease progression using advanced deep learning techniques. The second aim is to pioneer the next generation of retinal morphometric OCT biomarkers using deep learning, enhancing the precision in identifying retinal changes and improving the longitudinal analysis of optic nerve head and macular structures. Finally, the third aim is dedicated to leveraging deep learning models for the classification of various types of optic neuropathies directly from retinal OCT scans, aiming to significantly increase the accuracy and efficiency in distinguishing these conditions. This research proposes a multidimensional strategy to significantly improve the diagnostic and management capabilities in the field of neuro-ophthalmology through innovative applications of deep learning to OCT imaging. By integrating advanced imaging techniques with deep learning models, it aims to unveil novel biomarkers and predictive models, offering insights into the progression and treatment of these complex eye diseases. The success of this project could lead to earlier detection, personalized treatment strategies, and ultimately, better outcomes for patients with optic neuropathies.