THE EYE AS A "WINDOW TO THE BRAIN": A COMPREHENSIVE REVIEW OF MACHINE LEARNING FOR OCULAR-BASED NEUROLOGICAL DISEASE DETECTION

Abstract

Neurological disorders are increasingly recognized as a significant global health challenge, thereby necessitating the development of diagnostic tools that are non-invasive, easily deployable, and affordable. This review proposes a novel approach: utilizing machine learning (ML) to identify neurological diseases via ocular imaging. Based on the neuroanatomical connection between the retina and the central nervous system (CNS), the retina is conceptualized as a dynamic "window to the brain," where systemic neuropathology’s manifest visibly. The subsequent analysis identifies retinal biomarkers associated with particular neurological conditions: amyloid-beta (Aβ) and phosphorylated-tau (pTau) aggregates in Alzheimer ’s disease (AD); dopaminergic neuronal loss and retinal thinning in Parkinson’s Disease (PD); and axonal degeneration following optic neuritis in Multiple Sclerosis (MS).  The computational domain is subsequently delineated across the machine learning spectrum, encompassing fundamental image segmentation techniques utilizing U-Net architectures and progressing to sophisticated network paradigms like Vision Transformers (ViT) and Graph Neural Networks (GNNs), which are adept at capturing intricate morphological and relational characteristics. Contemporary models have exhibited remarkable effectiveness, as evidenced by certain Alzheimer's disease detection systems attaining accuracies surpassing 98%. Nevertheless, the practical application of these models in clinical settings is presently impeded by considerable challenges, including the opacity of algorithmic processes, the presence of biases within data environments, and the ethical implications associated with patient privacy. Therefore, the future necessitates a multifaceted strategy Explainable AI to improve interpretability, federated learning to enable secure data sharing, and multimodal fusion to ensure comprehensive diagnostic integration—thereby progressing the field towards a truly transparent and clinically relevant area within neurodiagnostics.