Measurement of Optic Nerve Sheath on Ocular Ultrasound Image Based on Segmentation by CNN

Recent studies have revealed the correlation between the optic nerve sheath diameter (ONSD) measured at 3.0 mm behind the eyeball and the occurrence of increased intracranial pressure (ICP) caused by head trauma. To conduct this measurement, in clinical diagnosis, emergency departments take B-mode ultrasound images as reference, which contains lots of pathology information not limited to ophthalmology, and is non-ionizing radiation and in low-cost. This repetitive and onerous measurement is usually conducted by experienced physicians, however, can be facilitated by sophisticated computer-aided diagnosis (CAD) systems. The measurement involves extraction of eyeball regions and optic nerve sheath from the B-mode ocular ultrasound images. In this paper, we propose a novel method which extracts eyeball regions using convolutional neural networks first, and measures the diameter based on the segmentation results. We construct a dataset of B-mode ocular ultrasound images with manual annotations of the eyeball and optic nerve regions in each image. Our segmentation network achieves the intersection over union (IoU) accuracy of 97.20%. Then, the measurement of ONSD can be easily conducted on the segmentation results. This study may be embedded in CAD systems to assist clinical diagnosis in the future.