Eying up Parkinson’s: early disease detection with AI
Early markers of Parkinson’s have been discovered 7 years before symptoms through retinal imaging and AI. The study utilized data from the AlzEye project to uncover neuro-links and high-resolution retinal scans to search for retinal thinning.
Parkinson’s disease is a progressive neurological disorder marked by decreasing dopamine levels, and primarily affects the movement of the body. Autopsies of Parkinson’s patients have revealed differences in the inner nuclear layer (INL) of the retina, and previous optical coherence tomography (OCT) scans have discovered associated potential structural irregularities, though with variations.
A team of researchers from University College London (UK) and Moorfields Eye Hospital (London, UK) has identified indicators that predict the presence of Parkinson’s disease about 7 years before clinical manifestation. These findings mark the first instance of discoveries years ahead of standard diagnosis and were facilitated by the largest study conducted so far using retinal imaging in connection to Parkinson’s disease.
The study, published in Neurology, the medical journal of the American Academy of Neurology (MN, USA), has unveiled identifying markers for Parkinson’s disease through the utilization of AI in eye scans. The examination of the AlzEye dataset was replicated using the broader UK Biobank database, consisting of data from healthy volunteers, which resulted in replication of the findings.
The strategic use of these two datasets has helped the research team to pinpoint these subtle markers, despite Parkinson’s disease having a relatively low occurrence rate of 0.1-0.2% of the population. The creation of the AlzEye dataset was enabled by INSIGHT, the world’s most extensive database of retinal images and their corresponding clinical information.
“This work demonstrates the potential for eye data, harnessed by the technology, to pick up signs and changes too subtle for humans to see. We can now detect very early signs of Parkinson’s, opening up new possibilities for treatment,” said Professor Alastair Denniston, consultant ophthalmologist at University Hospitals Birmingham, professor at the University of Birmingham (UK) and part of NIHR Moorfields Biomedical Research Centre.
The use of eye scan data has previously brought to light indications of other neurodegenerative disorders, such as Alzheimer’s, multiple sclerosis and schizophrenia. This growing and captivating field of study, known as “oculomics,” has also demonstrated the capability of eye scans, and related data to find predispositions to conditions such as high blood pressure and cardiovascular diseases, including strokes and diabetes.
Medical professionals have long been aware that the eye serves as a ‘window’ to various aspects of our overall health, offering valuable insights into our well-being. High-resolution imagery of the retina has become a standard component of eye care, specifically employing the 3D scanning technique, OCT. The OCT method is extensively utilized in both eye clinics and optical stores, and in less than a minute, an OCT scan generates an intricately detailed cross-sectional view of the retina, of up to one-thousandth of a millimeter.
While these images serve as a valuable tool for monitoring ocular health, their significance extends beyond this. They hold the unique ability to non-invasively reveal layers of cells beneath the skin’s surface, a capability that was otherwise inaccessible. Recently, researchers have begun utilizing powerful computers to meticulously analyze extensive collections of OCT scans and other ocular images, completing the task in a fraction of the time a human would require. Utilizing machine learning, computers can now unearth concealed insights about the entire body solely from these images.
Louisa Wickham, Moorfields’ Medical Director, stated, “Increasing imaging across a wider population will have a huge impact on public health in the future, and will eventually lead to predictive analysis. OCT scans are more scalable, non-invasive, lower cost and quicker than brain scans for this purpose.”
This research upheld earlier findings by confirming a notably reduced thickness of the ganglion cell–inner plexiform layer (GCIPL), while also revealing, for the first time, a diminished thickness of the INL. Additionally, it established a connection between the decreased thickness of these layers and an increased likelihood of developing Parkinson’s disease.
Further studies will be necessary to ascertain whether the decline in GCIPL thickness is prompted by changes within brain characteristics, or if the thinning of the INL occurs before GCIPL atrophy. Exploring this aspect could provide insights into the underlying mechanism and determine the potential role of retinal imaging in aiding the diagnosis, prognosis and intricate management of individuals grappling with Parkinson’s disease.