Tracking of Surgical Instrument Based on Neural Network and Optical Flow for Videos of Simulated Intrascleral Intraocular Lens Fixation

Intrascleral intraocular lens (IOL for short) fixation is the latest cataract surgery, and a number of ophthalmologists desire developing the surgery training system for it. In this paper, as part of developing such system, a method of tracking a main instrument appearing in videos of simulated intrascleral IOL fixation is presented, using a convolutional neural network and optical flow. The proposed method first detects the main instrument appearing in some frame by applying the network referred to as scSE-FC-DenseNet40. For several continuous frames following to the frame to which network-based detection is applied, optical flow works to find the main instrument. The proposed method keeps fast and accurately tracking the main instrument appearing in each frame by repeating the network-based stage and optical-flow-based stage. It is finally established that the proposed method can achieve not only high accuracy but also fast processing speed in tracking the main instrument, even if it is implemented on CPU machines available in operation rooms.