Non-invasive cerebrospinal fluid pressure estimation using multi-layer perceptron neural networks.

Cerebrospinal fluid pressure (CSFp) provides vital information in various neurological abnormalities including hydrocephalus, intracranial hypertension and brain tumors. Currently, CSFp is measured invasively through implanted catheters within the brain (ventricles and parenchyma) which is associated with a risk of infection and morbidity. In humans, the cerebrospinal fluid communicates indirectly with the ocular circulation across the lamina cribrosa via the optic nerve subarachnoid space. It has been shown that a relationship between retinal venous pulsation, intraocular pressure (IOP) and CSFp exists with the amplitude of retinal venous pulsation being associated with the trans-laminar pressure gradient (i.e. IOP-CSFp). In this study we use this characteristic to develop a non-invasive approach to estimate CSFp. 15 subjects were included in this study. Dynamic retinal venous diameter changes and IOP were measured and fitted into our model. Artificial neural networks (ANN) were applied to construct a relationship between retinal venous pulsation amplitude, IOP (input) and CSFp (output) and develop an algorithm to estimate CSFp based on these parameters. Results show a mean square error of 2.4 mmHg and 1.27 mmHg for train and test data respectively. There was no significant difference between experimental and ANN estimated CSFp values (p>0.01).This study suggests measurement of retinal venous pulsatility in conjunction with IOP may provide a novel approach to estimate CSFp non-invasively.