Retinal spectral imaging shows early Parkinson signals but no diagnostic signature yet

Retinal hyperspectral imaging, multicolor imaging, and diffuse reflectance spectroscopy are attracting growing attention as possible windows into Parkinson disease, but a new scoping review finds that the evidence base remains remarkably thin. Published in Discover Neuroscience, the review identified only four in vivo reports worldwide that met its strict eligibility criteria: one small human study, two peer-reviewed animal articles, and one animal conference abstract. Despite intriguing optical differences between Parkinson-related conditions and controls in every report, the review concludes that no reproducible, disease-specific retinal spectral signature has been established, and that none of these techniques is ready for clinical use.

The review, led by Rocío Mesa Martínez, was deliberately narrow in scope. It searched PubMed/MEDLINE, Embase, Scopus, and Web of Science from inception to July 2026, looking for original in vivo studies that used wavelength-resolved retinal reflectance imaging or spectroscopy in people with clinically diagnosed Parkinson disease or in live mammalian models of Parkinson disease or experimental parkinsonism. From 131 database records, 43 duplicates were removed and 88 records were screened; seven reports were assessed in full, and four qualified for the primary synthesis. Ex vivo studies, fluorescence-based techniques such as fluorescence lifetime imaging ophthalmoscopy, and conventional optical coherence tomography were mapped separately as context but excluded from the core analysis.

The single human study came from Ueda and colleagues, who examined 20 participants with Parkinson disease and 20 age- and sex-matched controls using a snapshot hyperspectral camera integrated into a fundus camera. The system acquired sixteen spectral bands between 463 and 598 nanometers in a 45-degree field centered between the temporal optic-disc edge and the fovea. After removing vessels with a difference-of-Gaussians filter and omitting two noisy bands, the researchers found that mean relative reflectance was lower in the Parkinson group at shorter wavelengths, most notably in the superonasal region below 490 nanometers and the inferonasal region below 510 nanometers.

Those group differences, however, translated into weak diagnostic performance. A linear discriminant analysis with nested leave-one-out cross-validation yielded an area under the receiver-operating-characteristic curve of 0.60, with a 95 percent confidence interval of 0.43 to 0.78, an interval that includes chance-level performance. Sensitivity was 60 percent, specificity ranged from 50 to 55 percent depending on the region, and accuracy was between 55 and 57.5 percent. The review stresses that the regions and wavelength intervals were identified within the same dataset later used for classification, that no external validation was performed, and that the study did not provide complete analytical validation for pupil effects, cataract, refraction, test-retest variability, or cross-device transfer.

The animal evidence was heterogeneous and, in several respects, pointed in different analytical directions. Trlin and colleagues used a custom sequential ophthalmoscopic hyperspectral system spanning 320 to 680 nanometers in A53T alpha-synuclein and TauKO mouse models, with wild-type aging and 5xFAD Alzheimer’s disease comparators. Advanced A53T and TauKO groups showed higher normalized short-wavelength reflectance than wild-type controls, extending broadly across approximately 450 to 600 nanometers, although earlier disease stages did not show the same pattern. Notably, healthy aging produced lower short-wavelength reflectance, and the 5xFAD Alzheimer’s comparator showed a similar direction to the advanced Parkinson models, a similarity that weighs against interpreting the pattern as specific to synucleinopathy.

A third report, available only as a conference abstract, came from Richard and colleagues, who evaluated TauKO and wild-type mice using the Spectralis multicolor imaging module and optical coherence tomography. The confocal system acquired blue, green, and infrared reflectance, and at 18 months the short-wavelength-to-infrared ratio showed genotype and interaction effects, with a greater ratio in TauKO mice. The review cautions that radiometric calibration and repeatability were incompletely described, and that independence from the TauKO program reported by Trlin’s group could not be established from the abstract, making this adjacent rather than independent validation.

The fourth report, from Munro and colleagues, used a targeted diffuse reflectance spectroscopy system in four ovariectomized female cynomolgus macaques before and five months after MPTP exposure, when moderate-to-severe parkinsonian features were stable. The device sampled spectra from a 1.5-degree region while simultaneously displaying a fundus image, with measurements taken from the optic nerve head and temporal retina. The largest pre-post difference in spectral shape appeared between 480 and 525 nanometers, and the slope of the normalized log spectrum in that interval was lower after MPTP in both regions. The review emphasizes that this should not be described simply as lower reflectance: a change in spectral slope is a different mathematical quantity from the relative reflectance level reported in the human study or the band ratios used in the mouse work. The within-animal design reduces inter-animal variability, but the sample of four animals, the post hoc selection of the wavelength interval, the absence of an untreated longitudinal group, and the toxin-based model all constrain interpretation.

A central contribution of the review is its dissection of why these findings cannot be simply pooled or compared. Fundus reflectance is a composite signal generated by wavelength-dependent absorption, scattering, and reflection across ocular media, blood, pigment, and multiple retinal and choroidal interfaces, and it is not molecule-specific. Illumination spectrum, detector response, acquisition geometry, pupil and media, refraction, chromatic focus, sampled retinal location, vessel masking, baseline correction, normalization, and feature selection can all alter the final result. A 2026 confocal multicolor imaging study in healthy mice demonstrated that depth and longitudinal chromatic aberration materially affect selected-band ratios, meaning a difference in intensity versus wavelength can arise from depth and focus even without any disease-specific molecular absorber or scatterer. The three peer-reviewed studies were not replications of one another: they examined different species and disease constructs, sampled different retinal locations, used different optical systems, and analyzed different spectral quantities.

For clinicians, the review’s message is deliberately conservative. Retinal hyperspectral imaging, multicolor imaging, and diffuse reflectance spectroscopy remain research tools and should not be used to diagnose or exclude Parkinson disease, identify prodromal disease, distinguish Parkinson disease from mimics, stage disease, monitor progression, guide treatment, or provide prognosis. Current diagnosis and management should continue to rely on established clinical assessment and validated investigations. The authors propose a four-level validation framework: analytical validation of calibration, repeatability, and cross-device transfer; anatomical and biological validation combining spectral acquisition with co-registered structural imaging and molecular measurement; clinical validation in larger independent cohorts including early and prodromal disease and biomarkers such as alpha-synuclein seed amplification; and finally demonstration of clinical utility against patient-relevant outcomes.

There is nonetheless a genuine biological foundation for continued work. Post-mortem evidence of phosphorylated alpha-synuclein in the Parkinson retina, structural and vascular differences measured with optical coherence tomography and angiography, retinal alpha-synuclein-associated pathology in animal models, and population-scale OCT data associating retinal features with prevalent and incident Parkinson disease all support the plausibility of a retinal optical signal. Four ongoing study registrations identified through supplementary searching indicate that the field is expanding, although none had posted results as of August 2026. The review’s central finding stands as both a warning and a roadmap: the small number of eligible studies is itself the headline result, and translating these optical measurements into anything clinically meaningful will require harmonized acquisition, prespecified analyses, and validation in independent, longitudinal, biomarker-characterized cohorts before any reproducible Parkinson-specific retinal signature can be claimed.

Subject of Research: Retinal spectral imaging and spectroscopy as emerging research approaches in Parkinson disease

Article Title: Retinal spectral imaging and spectroscopy are emerging research approaches in Parkinson disease

Article References: Mesa Martínez, R. (2026). Retinal spectral imaging and spectroscopy are emerging research approaches in Parkinson disease. Discover Neuroscience, 21(1), Article 114. https://doi.org/10.1186/s13064-026-00347-5

Image Credits: AI Generated

DOI: 10.1186/s13064-026-00347-5

Keywords: Parkinson disease, retinal imaging, hyperspectral imaging, multicolor imaging, diffuse reflectance spectroscopy, biomarkers, scoping review, alpha-synuclein, MPTP macaques, optical coherence tomography, diagnostic accuracy, neurodegeneration