Bruch's membrane opening segmentation in cases of optic disc swelling using deep learning

The optic disc is the region at the back of the eye where the retina connects to the optic nerve. Swelling in this area can have multiple different causes. One cause is raised intracranial pressure, resulting in a condition called papilledema. This raised intracranial pressure can be a consequence of issues like brain tumors, meningitis, or idiopathic intracranial hypertension. Left untreated, papilledema can have effects as detrimental as visual loss or neurological damage. Another condition is anterior ischemic optic neuropathy which is caused by restricted blood flow to the front portion of the optic nerve. Though both diseases present as optic disc edema, the contrasting etiologies and their respective treatments make identification of the proper cause vital.

One non-invasive approach that has been helpful in differentiating the source of swelling is analysis of the shape of Bruch’s membrane (BM), a retinal structure that can be seen in scans of the back of the eye. To perform proper shape analysis, an accurate segmentation of Bruch’s membrane opening (BMO) is vital. However, shadowing in the scans of extremely swollen eyes makes identification of the BMO difficult, even when attempted manually. Luckily, localization of the BMO becomes much easier once the swelling has reduced.

In this thesis, we propose an approach that utilizes longitudinal data to aid in the segmentation of Bruch’s membrane opening. First, a registration approach was used to align less swollen scans to their swollen counterparts. Then, an accurate segmentation from the less swollen scan was transferred back to the swollen scan. This segmentation was used to train a computer to automatically predict the BMO in severely swollen cases through a deep learning model. The effective segmentation of BMO in cases of optic disc edema can aid in automatic shape analysis of Bruch’s membrane for differentiation of cause and evaluation of swelling.