For ophthalmologists, the idea that the retina could provide a window into neurodegeneration is no longer science fiction. Alzheimer’s disease (AD) is increasingly recognized as having measurable ocular correlates. However, translating that biology into a practical screening tool has proved difficult. In a new Scientific Reports study, researchers at the Istituto Italiano di Tecnologia, Rome, describe a streamlined approach: label-free tri-spectral retinal imaging, designed to plug into existing fundus camera workflows and generate an interpretable optical biomarker for AD.
The team developed a custom tri-spectral imaging module that can be integrated with a commercial fundus camera and captures retinal reflectance simultaneously across three optimized bands: blue, green, and red, all acquired with a single flash. Compared with standard RGB imaging, the authors argue this design reduces spectral overlap, improves signal-to-noise ratio, and boosts sensitivity at shorter wavelengths – the region where AD-associated light scattering effects (known as the "spectral signature”) are thought to be most pronounced.
To validate the concept, the team conducted a single-center, biomarker-confirmed case–control study including 38 patients with mild AD and 28 age-matched healthy subjects. AD diagnosis was supported by amyloid PET in 21% and CSF biomarkers in 79%, with a mean MMSE (Mental State Examination) score of 19.5.
At first glance, averaged tri-spectral images did not reveal obvious morphological differences between groups. But when the researchers calculated a blue-to-green (B/G) ratiometric map, a signal emerged: AD eyes showed a relative increase in blue reflectance compared with controls.
To push performance further, the team trained an XGBoost-based machine learning model combining raw spectral intensities (rather than ratios), total reflectance in each band, and basic demographic/clinical features. In leave-one-out cross-validation, the model achieved an AUC of 0.83; in an independent test set, it reached 0.91 with balanced sensitivity and specificity.
The take-home message for clinicians is pragmatic: tri-spectral imaging offers a potentially scalable, non-invasive approach to AD screening that fits within routine ophthalmic imaging. However, this remains an early case–control study, and the authors rightly emphasize the need for larger datasets and broader validation before any clinical deployment.
The team developed a custom tri-spectral imaging module that can be integrated with a commercial fundus camera and captures retinal reflectance simultaneously across three optimized bands: blue, green, and red, all acquired with a single flash. Compared with standard RGB imaging, the authors argue this design reduces spectral overlap, improves signal-to-noise ratio, and boosts sensitivity at shorter wavelengths – the region where AD-associated light scattering effects (known as the "spectral signature”) are thought to be most pronounced.
To validate the concept, the team conducted a single-center, biomarker-confirmed case–control study including 38 patients with mild AD and 28 age-matched healthy subjects. AD diagnosis was supported by amyloid PET in 21% and CSF biomarkers in 79%, with a mean MMSE (Mental State Examination) score of 19.5.
At first glance, averaged tri-spectral images did not reveal obvious morphological differences between groups. But when the researchers calculated a blue-to-green (B/G) ratiometric map, a signal emerged: AD eyes showed a relative increase in blue reflectance compared with controls.
To push performance further, the team trained an XGBoost-based machine learning model combining raw spectral intensities (rather than ratios), total reflectance in each band, and basic demographic/clinical features. In leave-one-out cross-validation, the model achieved an AUC of 0.83; in an independent test set, it reached 0.91 with balanced sensitivity and specificity.
The take-home message for clinicians is pragmatic: tri-spectral imaging offers a potentially scalable, non-invasive approach to AD screening that fits within routine ophthalmic imaging. However, this remains an early case–control study, and the authors rightly emphasize the need for larger datasets and broader validation before any clinical deployment.
