Degenerative brain diseases such as Parkinson's and Alzheimer's are costly and difficult both for people living with these conditions and for society. Increasingly, treatments are available, but these are most effective when begun early in the disease process, when relatively little damage has occurred. Early diagnosis of these disorders is therefore essential. Yet the first changes occur deep inside the brain, and often do not cause noticeable symptoms until damage is already widespread.

Scientists have recently learnt that these diseases also cause changes in the retina, the light-sensitive tissue at the back of the eye. This is not as surprising as it sounds, because the retina develops from the same type of cells as the brain. In fact, the retina is effectively an outgrowth of the brain, and the only part which can be viewed from the outside. Over the last decade or so, the treatment of eye disease has been revolutionised by a new imaging technique, called optical coherence tomography or OCT. In just a few seconds, OCT scans a patient s eye and shows doctors the 3D structure of the patient's retina. OCT imaging is now standard in NHS eye clinics, and increasingly widespread in high-street opticians.

OCT scans reveal almost microscopic levels of detail, but their very richness makes them hard for humans to interpret. We hope to overcome this challenge by using another revolutionary innovation – new techniques for programming computers which enable them to learn different tasks by studying vast amounts of data. This “machine learning” is what makes it possible for Alexa to understand your instructions, or Facebook to tag your friends in photos. The OCTAHEDRON project aims to use machine learning to detect early signs of neurodegenerative disease - such as Parkinson's - in OCT scans of the retina. To achieve this aim, NHS doctors will teach our computer system how to recognise the different layers of the retina in OCT images. As it learns, the computer will in turn help teach junior doctors, providing valuable training. Then, we will exploit the power of the NHS s vast data-sets. By studying thousands or even millions of NHS OCT scans, the computer becomes familiar with their structure. By combining this with a smaller number of scans from patients who went on to be diagnosed with neurological conditions, the computer can then learn how these patients' retinas are different.

Ultimately, the software will help the NHS to diagnose more efficiently and without overstretching the limited number of human experts trained to interpret OCT scans. Neurological disorders will be detected sooner, so treatment can begin earlier. The reduced disability and improved outcomes will benefit patients and reduce costs to wider society.