A Study on Ultrasound Nerve Image Segmentation with multi-densely-layer supervision mechanism

Abstract

Ultrasound Guided Regional Anesthesia (UGRA) is widely used in pain management of surgical procedure, which can reduce the risk of general anesthesia side-effect and nerve trauma complications. Whereas a successful UGRA requires accurate semantic segmentation of nerve ultrasound images. In this paper, several classical semantic segmentation methods will be introduced and tested at first. Then a novel encoderdecoder based segmentation framework is proposed, that is integrated with densely connection, attention mechanism and multi-layer supervision mechanism. Finally, the proposed method is tested with a public ultrasound nerve benchmark dataset. The experiment with varieties of evaluation methods demonstrates promising performance against several other deep learning models.