Abstract
Neurocognitive Disorders (NCDs) pose a substantial public health challenge, demanding timely detection and intervention to enhance patient outcomes. Recent strides in medical imaging have unveiled choroidal morphological alterations as a novel avenue for early NCD diagnosis. This study investigates the effectiveness of deep learning algorithms, particularly Residual neural networks (ResNet) and Visual Geometry Group 19 (VGG19) architectures, in analyzing choroidal images to identify subtle morphological changes indicative of underlying neurodegenerative processes. A comprehensive dataset encompassing choroidal images from individuals with NCDs and healthy controls was employed for model training and evaluation. Among the architectures assessed, ResNet152 exhibited superior performance, achieving an impressive area under the curve (AUC) of 0.97 in distinguishing between NCD patients and healthy individuals. The findings suggest that choroidal morphological changes hold promise as a potential biomarker for early NCD detection and monitoring. Integration of deep learning-based choroidal morphology analysis into clinical practice has the potential to enhance early intervention strategies, thereby contributing to improved patient outcomes in neurocognitive disorders. This research underscores the importance of exploring innovative approaches for early diagnosis and management of NCDs, aiming to enhance the quality of life for affected individuals and alleviate the societal burden associated with these disorders. Further studies are warranted to validate the utility of choroidal morphology analysis in clinical settings and to elucidate its role in the broader context of NCD diagnosis and management.
Keywords: AUC, Choroidal Morphology, Neurocognitive Disorders, ResNet, VGG19.