Ensemble learning for improvement of imaging speed with coherent Raman scattering rigid endoscopy

Abstract

Injuries of peripheral nerves during surgery should be avoided to prevent a poor prognosis. Nevertheless, it is difficult to recognize peripheral nerves. Therefore, a system to identify nerves during surgery is required. We had developed a coherent Raman scattering rigid endoscope to visualize nerves without staining. By using coherent Raman scattering, substances are visualized selectively without staining based on information of molecular vibration. We had successfully visualized nerves in the rabbit prostate fascia, but it had been necessary to reduce the present exposure time of several minutes to several seconds. Last year we reported that improving the signal-to-noise ratio using deep learning shortens the exposure time. In the conference, we show the results of ensemble learning to achieve significantly higher accuracy comparing with the previous learning results.