The project aims to develop an automated, machine learning (ML) enabled spontaneous venous pulsation (SVP) detection system, first for widely available optical coherence tomography (OCT) fundal videography, and second for videos acquired from hand-held retinal fundus cameras.
Elevated intracranial pressure (ICP) can lead to blindness, brain injury and death, and occurs in many serious neurological conditions including idiopathic intracranial pressure (IIH), subarachnoid haemorrhage, major ischaemic stroke, meningitis, intracranial tumours and severe traumatic brain injury (TBI). Presence of SVPs is sensitive for the exclusion of raised intracranial pressure (ICP). Automated SVP detection could therefore reduce the need for invasive lumbar punctures, transforming ICP testing in clinical and community settings. Positive results would support prospective clinical trials, further system development and expansion to portable retinal imaging devices, in turn broadening ambulatory and community applications. The project is interdisciplinary, translational and with a high potential for impact.
Prior studies have demonstrated that SVPs are easily identifiable in high-quality OCT fundal videos, but expert interpretation is required. In its first part, the project will automate the detection of SVPs on OCT fundal videos and quantify the extent of pulsation.
In its second part, the project will focus on videos from hand-held retinal fundus cameras, namely Zeiss Visuscout 100. Such cameras are a cost-effective solution to visualising the retina, but are limited in terms of image quality. Domain adaptation approaches will be applied to allow ML models trained on OCT videos to utilise Zeiss Visuscout 100 videos to maximise the use of the available data.