Surgical Instrument Tracking for Intraoperative Vitrectomy Guidance Using Deep Learning and Stereo Vision

Surgeries are always challenging procedures. In ophthalmology in particular, the main difficulties are represented by limited space, complicated viewpoints, and bad light conditions, which make it even harder for surgeons to operate safely. This means avoid touching the retina, i.e. the back surface of the eye, perform secure movements and do not apply too much force on surfaces. This thesis deals with the problem of retinal collision avoidance through a proposed real-time pipeline utilizing both Deep Learning and Computer Vision to provide ophthalmologists with additional information about depth perception.