Retinal Segmentation of Intraoperative B-Scan Optical Coherence Tomography Using Deep Learning

The epiretinal membrane is a condition in which a small bro cellular scar tissue forms on the surface of the retina. This membrane can distort, reduce or block completely the patient eyesight. The ophthalmologist can intervene by peeling the scar tissue from the retina surface. Removing an object close to the retina can have complications. Small movements of surgical tools can touch and damage the retina. Moreover, peeling the scar tissue can produce unwanted stress on the retina surface, with possible retina detachments, that can impair the normal eyesight of the patient. Due to these complications, support tools are needed during the surgical operation in order to reduce possible accidents. Nowadays one of the most used tools is the intraoperative Optical Coherence Tomography (iOCT). The iOCT serves the same purpose as a standard Optical Coherence Tomography (OCT) but is enhanced for real-time usage. Unfortunately, this speed enhancement reduces the quality of the image artifacts produced by the iOCT, thus making it challenging to identify the retina location. The iOCT produces a cross-section pair of images of the retina, and displays them to the surgeon in real-time. We propose a support tool that utilizes the cross images produced by iOCT and automatically detects the retina in real-time, helping the surgeon identify correctly the retina boundaries. We will show the execution times and accuracy of the proposed algorithm, together with possible future expansions to the project.