To evaluate the diagnostic performance and clinical applicability of artificial intelligence (AI) models in analyzing retinal images (RI) for the assessment of neurodegenerative diseases (NDDs).

NDDs are characterized by progressive damage to neural structure and function, which mandates the need for early diagnosis and monitoring. The retina, as an anatomical extension of the brain, offers noninvasive evaluation of central nervous system pathology.  With recent technological developments, AI-driven analysis of RI has been emerging as a promising approach, however, a systematic evaluation of its capabilities and applications across NDDs is needed.

A search was conducted following PRISMA guidelines. Databases (PubMed, Web of Science, Scopus, Embase) were searched from January 2015 to May 2025 for studies applying AI to RI for NDD diagnosis, classification, or progression monitoring. Of 326 identified records, 29 studies met the inclusion criteria.

AI models demonstrated exceptional diagnostic accuracy for specific NDDs. For Alzheimer’s disease, a deep learning model (DL) achieved an area under the curve (AUC) of 0.9978, with models leveraging biomarkers such as retinal vascular attenuation (reduced fractal dimension, vessel density) and inner retinal layer thinning, which correlated with cerebrospinal fluid Aβ42/Aβ40 ratios and cognitive scores. In Parkinson’s disease, AI identified thinning of the inner nuclear and outer plexiform layers as important features, with a top model achieving an AUC of 0.918. DL (convolutional neural networks and transformers) was the predominant technique, outperforming classical machine learning.

AI-based RI analysis represents a pioneering and high-accuracy tool for NDD assessment. It quantifies disease-specific retinal biomarkers that mirror central neuropathology, showing immense potential for scalable and non-invasive screening and monitoring. Future work must focus on prospective validation and standardization for clinical translation.