Neural Network-based Fast Liver Ultrasound Image Segmentation
Mohammed Yusuf Ansari, Iffa Afsa Changaai Mangalote, Dima Masri, Sarada Prasad Dakua
Abstract
Ultrasound is quite popular among the clinicians’ fraternity, because the machine is cheap, easy to use, and mobile. However, the image is not that easy to study properly, if the examiner does not have adequate expertise. Image segmentation, being the first step to ease the study, is considered crucial. In this paper, we have presented a neural network-based image segmentation that is based on Pyramid Scene Parsing. Additionally, we have studied the importance of removing noise before the actual segmentation. We have tested the method on the data obtained from Hamad Medical Corporation and found Dice coefficient of 0.913 ± 0.024 while delivering a real-time performance of 37 frames per second.