Deep Learning Based Real-time Segmentation in Ultrasonic Imaging Following the Doctor's Voice Guide

Deep learning models such as convolutional neural network have been widely used in clinical diagnosis biomedical segmentation and achieve state-of-the-art performance. However, most of them often adapt a single modality or stack multiple modalities as different input channels. They all analyze medical images that have already been obtained. To better make use of the real-time and portable advantage in ultrasonic instrument, we propose a model called Voice U-Net was proposed for real-time segmentation guided by the doctor's voice instruction during ultrasonic scanning. Through the instructions, the network can identify and segment organs in different directions. Experimental results on phantom and public dataset show that our method outperforms state-of-the-art biomedical segmentation approaches.