Optic nerve sheath diameter (ONSD) measurement has shown promise as a non-invasive method for estimating intracranial pressure (ICP). However, its practicality is limited due to the need for expert operators, as well as significant heterogeneity in imaging and measurement techniques reported within the current literature. The use of machine-learning to automate ONSD measurement might help standardize the measurement process and make ICP estimates more accessible.

A database containing 52 de-identified transorbital sonographic scans from 46 patients admitted to the Baylor St. Luke’s Neurocritical Care Unit (NCCU) were imported into an AI based automatic annotation and segmentation platform. Two annotators independently selected images suitable for measurement of ONSD and automatically segment the optic nerve, optic nerve sheath including the subarachnoid space and dura, retina, and vitreous using AI.

To evaluate the performance of the AI method, the annotations and segmented images will be reviewed by a third expert (auditor) to check the concordance between the auditor and AI based annotations, segmentations, and measurements.

The development of an automated machine-learning algorithm for ONSD measurement has the potential to improve the accuracy of ONSD measurements which can be used as a proxy for diagnosis and management of elevated ICP in patients with acute brain injury and several other neurological pathologies, such as hydrocephalus, subarachnoid hemorrhage and intracranial hematoma.