Author Identifier
Date of Award
2026
Keywords
Alzheimer’s disease, amyloid-beta, biomarkers, brain, early detection, ocular, retina, machine learning, diagnosis, hyperspectral imaging
Document Type
Thesis
Publisher
Edith Cowan University
Degree Name
Doctor of Philosophy
School
School of Medical and Health Sciences
First Supervisor
Eugene Hone 
0000-0001-6708-3718
Second Supervisor
Ralph Martins 0000-0002-4828-9363
Abstract
Alzheimer's disease (AD) is the most prevalent form of dementia, characterised by the presence of extracellular senile plaques of amyloid-beta (Aβ) surrounding neuronal cells and the intracellular accumulation of hyperphosphorylated tau (p-tau) proteins within neurons. Common symptoms of AD dementia include memory impairment, behavioural changes, and a diminished capacity to perform daily tasks.
Early detection is critical because biochemical changes in the central nervous system (CNS) precede the onset of clinical symptoms by decades, and available disease-modifying treatments perform best when applied early. An affordable, non-invasive, and scalable screening method is crucial for the early diagnosis and intervention of AD. However, current gold-standard techniques, such as positron-emission tomography brain imaging and cerebrospinal fluid analysis, are unsuitable due to expense and invasiveness.
The retina, being an extension of the CNS, offers an alternative, non-invasive window into neurological changes. Studies have suggested that AD may manifest in the retina as pathological and/or morphological changes. Retinal imaging techniques, such as fundus photography, scanning laser ophthalmoscopy, and optical coherence tomography, have been employed to assess structural and functional retinal changes. A significant limitation of these approaches is a lack of specificity for AD, as ocular changes can also occur due to other neurological and ocular conditions. The emerging retinal hyperspectral imaging (rHSI) technique provides a unique capability to examine both spatial and spectral features, with the potential to identify AD-specific changes in the retina.
This thesis explores the potential of rHSI as a screening tool for AD, reporting on our studies in well-characterised cohorts, including participants with high and low brain Aβ burden, and individuals with genetic mutations for AD. A hyperspectral retinal camera was employed to capture a series of retinal images spanning the visible and near-infrared spectrum (450–905 nm) at the Alzheimer’s Research Australia, Perth. This approach enabled the extraction of not only conventional spatial features but also the spectral characteristics of the retinal tissue. The studies analysed changes in retinal reflectance, texture, and vascular structure associated with brain Aβ levels.
The results show clear differences in retinal spectral, textural, and vascular structural features between participants with high and low brain Aβ burden. Among these features, retinal reflectance was more informative than textural or vascular structural measures for distinguishing between groups. Similarly, within a cohort with a familial history of AD, reflectance patterns around blood vessels differed significantly between mutation carriers of specific genes (APP, PSEN1, and PSEN2) and non-carriers. Together, these findings suggest that retinal changes may serve as a promising biomarker for the early and more specific detection of AD. These results have been published, submitted, or are currently under preparation for publication in peer-reviewed journals.
This thesis begins by introducing AD and exploring how retinal imaging may help with its early detection. Chapter 2 reviews existing evidence showing that several eye structures, particularly the retina, reflect AD-related changes, while highlighting the need for larger, long-term studies. Then Chapter 3 focuses on retinal reflectance as a predictor of high brain Aβ burden, developing a machine learning classification model. After that, Chapter 4 compares the performance of different machine learning models to distinguish brain Aβ positive and -negative individuals, with the multilayer perceptron model as the top performer. Additional retinal texture and vascular features are also examined in Chapter 5. Finally, Chapter 6 provides a comprehensive evaluation of these findings and discusses their potential clinical implications. Overall, the findings suggest retinal imaging could support earlier identification of AD, timely intervention, and cost-effective, non-invasive screening for future AD clinical trials.
Access Note
Access to this thesis is embargoed until 10th October 2029
Recommended Citation
Poudel, P. C. (2026). Retinal hyperspectral imaging biomarkers of Alzheimer’s disease. Edith Cowan University. https://doi.org/10.25958/6src-f506